Edinburgh Research Explorer

Two-stage Optimisation of Hybrid Solar Power Plants

Citation for published version:

Bravo, R & Friedrich, D 2018, 'Two-stage Optimisation of Hybrid Solar Power Plants', Solar Energy, vol.
164, pp. 187-199. https://doi.org/10.1016/j.solener.2018.01.078

Digital Object Identifier (DOI):

10.1016/j.solener.2018.01.078

Link:
Link to publication record in Edinburgh Research Explorer

Document Version:
Peer reviewed version

Published In:
Solar Energy

General rights
Copyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s)
and / or other copyright owners and it is a condition of accessing these publications that users recognise and
abide by the legal requirements associated with these rights.

Take down policy

The University of Edinburgh has made every reasonable effort to ensure that Edinburgh Research Explorer
content complies with UK legislation. If you believe that the public display of this file breaches copyright please
contact [email protected] providing details, and we will remove access to the work immediately and
investigate your claim.

Download date: 19. May. 2021

 Two-stage Optimisation of Hybrid Solar Power Plants

R.Bravo1 , D.Friedrich∗
School of Engineering, Institute for Energy Systems, The University of Edinburgh, UK

Abstract
Hybrid solar power plants which combine concentrated solar power (CSP) and photovoltaic (PV) systems with thermal
energy storage (TES) have the potential to provide cost competitive and dispatchable renewable energy. The integration
of energy storage gives dispatchability to the variable renewable generation while the combination of different generation
technologies can reduce the costs. However, the design of reliable and cost competitive hybrid solar power plants
requires the careful balancing of trade-offs between financial and technical performance. This is made more complicated
by the dependence on a larger number of parameters compared to conventional plants and due to the integration of
TES which requires that the operational profile is optimised for every design. This contribution presents a two-stage,
multi-objective optimisation framework which combines multi-objective linear programming methods for the operational
optimisation with multi-objective genetic algorithms for the design optimisation. The operational optimisation which
is performed for every design point needs to be performed with linear programming methods. Here an automated
scalarisation method is developed for the linear programming method which enables the multi-objective optimisation of
the operational profile. This enables the evaluation of the trade-offs between financial and technical performance in both
the design and operational optimisation, which is required to design reliable and cost competitive sustainable energy
systems. The two-stage multi-objective optimisation is applied to analyse and improve the design of the hybrid solar
power plant Atacama-1. It is demonstrated that balancing the trade-off between financial and technical performance
is key to increase the competitiveness of solar energy and that it is possible to simultaneously increase dispatchability
and decrease the levelised cost of energy. This shows that the operational and design optimisations have to be directly
linked in order to exploit the synergies of hybrid systems. Thus the optimisation framework presented in this study can
improve the decision making in the design of hybrid solar power plants.
Keywords: Hybrid energy systems, Thermal energy storage, Linear programming, Two-stage, multi-objective
optimisation

1. Introduction 50% that the maximum increase in temperature will be not

more than 2◦ C by 2100) [2]. Moreover, the use of renew-
During the last year, the new installed capacity of re- able power generation reduces air pollution and increases
newable energy projects in the power sector was greater the energy independence, among others [1]. On the other
than the development of conventional energy systems [1], hand, because of the variability of the renewable energy
and nowadays renewable energy systems are one of the resource, a high proportion of renewable generation added
most used technologies to cover the increase of the demand to the electrical system will result in large supply fluctua-
[2]. Moreover, the implementation of new renewable power tions to the power system and a mismatch between supply
plants has increased more rapidly compared with other en- and demand [3]. To avoid fluctuations, the intermittent
ergy technologies, and it is estimated that this will further generation from renewable energy can be integrated with
increase by 36% to 2021 [1]. energy storage in order to accumulate energy during hours
The growth in the use of renewable energy in the elec- with excess of generation and use it when energy is needed,
tricity market has many advantages both in the present providing a dispatchable or baseload generation from re-
and in the future. For instance, renewables reduce the newable energy technologies [4].
carbon emissions of the power sector, and its quick imple- In power systems, energy can be stored in different
mentation is key to accomplish the decarbonisation neces- forms: Mechanical, Electrochemical, Electrical, Chemical
sary for the 2SD scenario (get a probability not less than or Thermal [5]. Nowadays the most used technologies in
the electrical grid, due to its technical and financial per-
∗ Corresponding author formance in large scale integration, are different kinds of
Email addresses: [email protected] (R.Bravo),
[email protected] (D.Friedrich) mechanical energy storage (pumped hydro, compressed air
1 The author is supported by a PhD student scholarship from BE- energy storage, flywheel) and chemical energy storage (hy-
CAS CHILE, CONICYT.

Preprint submitted to Elsevier January 24, 2018

drogen, synthetic natural gas) [5], [6]. Moreover, depend- ing the energy storage system, the two-tanks molten salt
ing on the required application (time-shifting, electric sup- system has been used in most of the CSP plants [13]. Fur-
ply capacity, load following, regulation, etc.), energy stor- thermore, depending on the design, molten salts can work
age systems can be integrated in different areas of the elec- as both the HTF and also as the storage medium.
trical grid: generation, transmission, distribution, or in the In order to reach the desired performance, CSP tech-
customer side [6]. In renewable energy power plants, en- nologies need high values of direct normal irradiation, for
ergy storage systems can be applied to the system under instance, to produce around 1 kW he per m2 per day, the
two objectives: injection profiling (time-shifting) or injec- solar field of the CSP plant needs a DNI greater than 7
tion smoothing (capacity firming) [7]. kWh m−2 day −1 [10]. Areas with clear skies close to the
Energy storage technologies that are suitable or un- Tropic of Capricorn and Cancer, between north or south
der developed for renewable energy projects focusing on latitudes of 15 and 40, present the best conditions for its
both time-shifting and injection smoothing, are batteries, operation [10]. Nowadays, large power plants that are in
flywheels [6] and thermal energy storage [4]. One of the study, development and under construction are located in
prominent technologies that are plausible in the near fu- these zones, for example, the south-western United States
ture to be included in large scale renewable energy power (California, Arizona), Tunisia, Chile, among others [11],
plants are batteries such as Lithium-ion technology [1]. [14], [15]. Some of these projects integrate thermal energy
Nowadays, a large scale battery infrastructure is at least storage, while other designs consider hybridisation. For in-
one or two orders of magnitude more expensive than ther- stance, Atacama-1 or Cerro Dominador Solar Power Plant,
mal energy storage [8]. In fact, thermal energy storage located in Northern Chile, will supply firm electricity by
(TES) is a key technology that has been implemented in combining CSP with thermal energy storage, capable to
concentrating solar power plants (CSP) to store heat and deliver energy at full working capacity for 17.5 hours. In
deliver energy in form of heat or electricity, increasing the addition, hybridisation was designed by integrating a pho-
dispatchability of solar power plants and promoting the tovoltaic (PV) power plant [16]. While energy storage sys-
integration of renewable energy power plants [9]. tems allows full dispatchability, hybridisation offers perfor-
Large scale commercial concentrating solar power plants mance benefits and synergies. It improves both technical
have been operating in California since the 1980s and some and financial performance by integrating a cheaper tech-
of these power plants are still in operation [10]. Four dif- nology, e.g. PV, with a more expensive, but dispatchable
ferent concentrating solar power technologies are commer- technology, e.g. CSP with TES [17], [18]. In the long
cially available and have been developed and implemented term, due to cost reduction of batteries, the integration
from small scale to utility scale projects around the world, of battery energy storage systems with PV power plants
i.e. solar tower, parabolic trough, linear Fresnel reflectors could be key to develop dispatchable power plants with
and dish/engine systems [11]. During recent years, solar improved financial performance. However, due to the cur-
tower technology has shown an interesting development, rent high cost of batteries for PV compared with TES for
and the largest solar power plants in operation or under CSP, batteries will not be evaluated in the current research
development are based on this technology. For instance, [8].
the Crescent Dunes power plant, located in Nevada, which As a pathway to a cost-competitive decarbonisation for
started its operation in 2015, is one of the first large scale electricity generation, a co-firing option can be included
CSP power plants to supply almost continuous electricity into a CSP plant in order to get a firm power supply,
by using a single tower, a 110 MW power block, and ther- working even with no solar irradiation, hence, increasing
mal energy storage equivalent to 10 hours of full power its dispatchability. However, the current model focuses
[12]. on the performance of power generation only from solar
The process in these CSP plants begins in the solar technologies. Some research demonstrate that hybrid sys-
field, where a large number of strategically located he- tems integrating high cost CSP with TES and low cost PV
liostats (two axis tracking mirrors) concentrate the sun- power plants can be key to provide competitive dispatch-
light in a chamber located on the top of a tower. In this able large scale energy generation [17], [19]. Moreover, the
chamber, known as receiver, the energy from the solar ra- operational optimisation of solar tower systems integrated
diation is transferred to a heat transfer fluid (HTF). Then, with thermal energy storage and hybridised with photo-
the two-tank energy storage system allows the possibility voltaic power plants allows to reach high capacity factors
to store the energy from the hot HTF to be used later. [20], [21]. Many authors agree that solar energy systems
Hence, after leaving the chamber, the HTF is pumped to integrated with energy storage is one of the most suit-
the hot tank to be transferred to the storage medium for able sustainable technologies to provide economical, reli-
later use or used directly as a heat injection in a Rankine able, and dispatchable power [4], and that the hybridisa-
cycle, through a heat exchanger. Next, the ”cold” HTF tion of such systems allows even better performance [17],
is pumped directly to the tower and heated through the [18]. Hence, the hybridisation of firm generation from CSP
receiver, or it is used to reduce the temperature in the cold plants with thermal energy storage and lower cost genera-
tank. In the Rankine cycle, superheated steam is produced tion from PV power plants enables excellent features like
in order to run a turbine and generate electricity. Regard- dispatchability, decreases the intermittent generation from
2
renewables, match supply and demand, as well as decrease problem intractable with optimisation methods for nonlin-
the levelised cost of electricity from solar power plants. ear problems. However, the standard linear programming
In pursuance of reaching high dispatchability as well methods are only capable of single objective optimisation
as low cost of energy generation, several studies focusing and thus previous studies have only considered single ob-
on implementation of CSP with TES in different areas jective, two-stage optimisation of PV and CSP with TES
have been published in recent years. Some of them are systems. The aim of this research is to fill this gap, by
focused on the optimisation of the design and its oper- optimising at the same time the design and operation of
ation through different modelling approaches (linear pro- a hybrid solar power plant composed of a CSP plant with
gramming, neural networks, evolutionary algorithms, non- TES and a PV power plant with respect to multiple ob-
linear modelling). For instance, [22] developed different jectives.
thermodynamics configurations in order to model the be- In this paper a two-stage optimisation in which a multi-
haviour of the thermal energy storage in a small-scale objective operational optimisation is used to supply the
power plant that supplies constant heat for industry ap- operational information to a multi-objective design opti-
plications, minimising the thermal losses of the system. misation is presented. The results of the optimisations
In order to optimise the design of a solar power plant are Pareto fronts which show the trade-offs of different de-
by sizing its elements, [23] employs artificial neural net- signs with respect to different technical and financial per-
works and a genetic algorithm to maximise the financial formance metrics of hybrid solar power plants. The Pareto
performance of the project. [24] evaluated the possibil- fronts show the effects of different designs and enable the
ity to supply electricity and heat in an off-grid scheme for design of hybrid solar power plant that balance economic
a large scale copper mine, Collahuasi, one of the largest and reliability requirements.
copper mines located in Chile [25]). This research was fo-
cused in the optimal selection of the best technology of an
Abbreviations
integrated solar power plant (CSP and PV) with energy
storage (pumped hydro energy storage, advanced adiabatic DNI: Direct normal irradiation
compressed air storage, and thermal energy storage), min- GII: Global incidence irradiation
imising the investment of the complete system to ensure TMY: Typical meteorological year
the supply of energy. Other studies demonstrate the suit- CSP: Concentrating solar power
ability of CSP with TES, for example in Northern Chile, to PB: Power block
work as baseload power as well as the benefits for the grid PV: Photovoltaic
[26]. Moreover, the Atacama Desert is one of the most ap- TES: Thermal energy storage
propriate places to develop CSP plants due to high levels SM: Solar multiple
of direct normal irradiation [27], [21]. StH: Storage hours
Previous studies do not exploit the synergies of large CF: Capacity Factor
scale hybrid renewable power plant systems by simulta- LPS: Loss of power supply
neously optimising financial and technical performance in LPSC: Loss of power supply capacity
both, the design and the operational optimisation stages. LPSP: Loss of power supply probability
In order to optimise the design of a power plant regard- LCOE: Levelised cost of electricity
ing defined objectives, the operation of the power plant TLCC: Total life cycle costs
has to be optimised focusing on the same objectives. For CRF: Capital recovery factor
instance, if the objectives are related with financial and SoC: State of Charge
reliability performance, i.e. the trade-off between the cost
of the energy and the reliability of the power plant, the
optimal design of a project requires a multi-objective op-
timisation of both the operation as well as the design in Nomenclature
a two stage process. The first optimisation stage should
i: period (hours)
find the optimal operational profile for a given design, e.g.
t: period (years)
maximising the energy delivered of a defined power plant
DNIi : Direct normal irradiation period i
(that means minimising the cost of the energy) and max-
GIIi : Global incidence irradiation period i
imising its reliability. In the second optimisation stage,
ACSP : Solar field area
the design is modified with the aim of designing a power STO
Emax : Storage capacity
plant with low cost of energy and high reliability. PB
Pmax : Power block capacity
While the small number of design variables enables the PV
Pmax : Photovoltaic power plant capacity
use of nonlinear, multi-objective methods for the design
∆ti : delta time period i
optimisation, the operational optimisation requires the use
Pigen : Power generation period i
of linear programming methods. This is due to the large gen
Etot : Total energy generation
number of optimisation variables, e.g. 8760 for a yearly demand
Pi : Power demand period i
operation profile with hourly resolution, which make the
3
 demand
Etot : Total energy demand model by the optimisation of the capacities of four subsys-
CFCSP : Capacity factor CSP plant tems: solar field area of the CSP, storage system, power
CFP V : Capacity factor PV plant block, and PV power plant, and these are optimised in a
η DN I→CSP : Efficiency CSP solar field (from irradiation in second stage. Finally, the model focuses in supply energy
heliostats to receiver) to a given Load, which is considered as a commitment,
η CSP →P B : Efficiency CSP plant (from receiver to power moreover, the excess of generation has no penalty and can
block) be delivered to the network
η ST O→P B : Efficiency TES (from TES to power block) The parametric model developed in this study con-
η T ES : Efficiency TES (self discharge) siders a typical meteorological year (TMY) with hourly
η P B : Efficiency PB (th to el) resolution to represent the long time solar resource per-
η GII→P V : Efficiency PV array formance of the location under consideration. According
η IN V : Efficiency INV (DC to AC) to [28], designers and developers, in order to evaluate the
LPSi : Loss of power supply period i feasibility of a solar power plant project in a particular
location, frequently use the TMY, which represents the
condition of the site in study through an annual data set.
Moreover, the TMY is included in open data source in the
2. Framework description
countries mentioned above (Chile, United States) [29], [8].
In the present section, a description of the power flow However, while TMY is a valuable indicator that repre-
model of a hybrid solar power plant is explained. Then, de- sents the conditions of the location during a long period
tails on the implementation of the optimisation of the op- of time, it does not show the performance of a particular
eration of such power plants are given. After that, a two- year. Thus simulations based on the TMY are not able to
stage design and operational optimisation is presented. evaluate worst case scenarios of the performance of solar
power plants [28].
2.1. Power flow model The parameters used to model the system are:
In order to optimise a hybrid solar power plant, it is • DNI data of the location
necessary to study the thermodynamic performance of the
process for a given location. The performance of the hy- • GII data of the location for a defined slope
brid solar power plant for different configurations and con- • The efficiencies of each component
ditions is calculated with a power flow model.
The structure used in modelling a hybrid solar power • Demand that the power plants should dispatch: Pidemand
plant which integrates a concentrating solar tower plant (in order to calculate LPSi )
with a two-tank thermal energy storage system and a pho-
tovoltaic power plant is shown in Figure 1. This power flow • Operational and local limits, e.g., capacity of the
model exposes the processes involved in the operation of network.
this type of solar power plant in terms of power flows, The power flow model is used in the optimisation of the
energy balances, energy losses, and capacities of compo- design of the hybrid solar power plant. The optimal plant
nents. In the power flow model presented, every block is is reached by selecting the best sizes of each subsystem, i.e.
one of the main subsystems of the power plant: solar field solar field area, thermal energy storage, power block, and
of the CSP, thermal energy storage system, power block photovoltaic array. At the same time, the optimised opera-
(PB), PV power plant, inverter, and network. Every line tion of a particular design has to be performed. Therefore,
that connect two subsystems represent heat or electricity a two-stage optimisation model is necessary in order to si-
transfer. Parameters are given by blue letters, and these multaneously optimise the operation and the design of the
are associated with solar resource (DNI, GII) and efficien- hybrid solar power plant. A schematic of this is shown in
cies of the pipes, power lines, or subsystems. Constraints Figure 2.
are related with the capacities of different subsystems or
components. The three heat flows (from the CSP to the 2.2. Operational optimisation
PB, the CSP to the TES, and from TES to the PB) are
The aim of the operational optimisation is to find the
optimisation variables which are optimised with respect to
specific operation at each time step that enhances the per-
the objectives of the operational optimisation. Other vari-
formance of a hybrid solar power plant for a given design.
ables like the amount of curtailed energy in the CSP plant
For example, the objectives can be related with financial
when the storage system is full and when the power block
(cost of the electricity, investment), and/or technical (re-
is working at full capacity, as well as the thermal losses in
liability) performance.
the TES system or in the lines and the energy dispatched
As detailed previously, and shown in Figure 1, the re-
by the PV power plant are calculated by energy balances
sults of the operational optimisation are related to power
according to the results of the optimisation. The optimisa-
flows from each element considering energy balances and
tion of the design of the power plant is represented in the
the capacity of its components. The constraints are related
4
 Figure 1: Power flow model hybrid system

to energy balances and maximum power flows given by the Linear programming methods are used to find the power
capacities of the components. For instance, the maximum flows which optimise the given objective. This objective
power dispatched by the CSP plant has to be equal to the and further calculated properties, e.g. LPSP, are returned
capacity of the power block. The variables that the model and can be used in the design optimisation. The opera-
optimises are associated with the operational profile of the tional optimisation needs to be performed by linear pro-
system, i.e. power flows between each subsystem. gramming due to the large number of optimisation vari-
The objectives of the optimisation of the operation of a ables: 8760 optimisation variables for each connection in
given system are related to the optimisation of its technical Figure 1 for an annual operational profile with hourly res-
performance, some of the properties or indicators that can olution. However, linear programming solvers can only
be used as objectives are: handle a single objective so that multiple objectives can
gen only be added as constraints or through weighting of ob-
• Total energy generation: Etot
jectives.
• Capacity factors of the CSP and PV plants: CFCSP ,
CFP V 2.3. Design plus operational optimisation
PI
• Loss of power supply capacity: LPSC = i=1 LPSi ∆ti The aim of the design optimisation, defined as a two-
stage optimisation problem, is to have the best or a range
LPSC of designs of hybrid solar power plants that optimise the
• Loss of power supply probability: LPSP = demand
Etot pursued objectives. The design optimisation needs to si-
In order to calculate the LPSC and LPSP, the loss multaneously optimise the operation of each candidate and
of power supply of period i (LPSi ) with i ∈ [1, 8760] is focus on the selection of the best designs. This is done by
necessary. This variable is defined as 0 when generation the operational optimisation described above, and its re-
exceed demand, and by the difference between the demand sults are the input data to select the best range of designs.
that should be dispatched in period i (Pidemand ) and the In other words, the operation of each configuration of the
generation of the same period Pigen when demand exceed design optimisation is optimised by the operational opti-
generation: misation.
( Depending on the number of objectives, the design
Pidemand − Pigen , Pidemand > Pigen , plus operational optimisation can be modelled as single
LPSi =
0 , otherwise.
5
or multi-objective optimisation. Whilst the single objec- • Creation of local jobs during the construction and
tive optimisation gives the best design that optimises the operational phases
objective, the multi-objective optimisation reach a range
of non-dominated or Pareto optimal solutions,which rep- • Visual impact
resent the best design for a defined trade-off between ob- • Security level of the sustainable energy system
jectives. In other words, every point in the Pareto frontier
is valuable and a potential candidate, hence, the next step Because the present model is a numerical approxima-
to finish the multi-objective optimisation is a posteriori tion of the best design, each objective has to be quantita-
selection of the best design regarding the desired target, tive, hence, qualitative social or environmental goals have
which has to be done by the user of the model. to be represented by numerical values.

2.3.1. Objectives 2.3.2. Decision Variables

The objectives of the design optimisation can be re- The decision variables of the design optimisation are:
lated to technical (described for the operational optimi-
sation), financial, environmental or societal performance • Solar field area: ACSP m2
metrics. This research is focused on technical and finan- STO
• Storage capacity: Emax MWh
cial performances; however, other indicators related to en-
vironmental or societal performance can be calculated for PB
• Power block capacity: Pmax MW
each design outside the design optimisation and used in
PV
the a-posteriori selection in order to pick the most ap- • PV power plant capacity: Pmax MW
propriate design from all the potential candidates of the
Some of these variables are related to each other under
multi-objective optimisation.
the following indicators, which can be key to understand
Financial performance is fundamental to evaluate the
and define the features of optimised plants:
system and to compare it with the market as well as other
available technologies. Some examples of financial perfor- • SM, Solar multiple. Defined by [4] as the relation
mance metrics that can be used to evaluate the project between the design capacities of the solar field and
during its lifetime are: the power block.
• Investment • StH, hours of storage. Is the ratio between the total
• Levelised cost of energy: LCOE (for a given annual capacity of the storage system (MWh) and the power
discount rate and lifetime) block capacity (MW).

The levelised cost of energy is the present value of the cost 2.3.3. Optimisation method
of every unit of energy produced during all the lifetime In order to model the design optimisation, a genetic al-
of the power plant. Considering a constant annual energy gorithm was used. This mathematical process is shown in
production during the lifetime of the project, the LCOE Figure 2, explaining a two-stage mathematical optimisa-
can be simplified by [30]: tion model of the design of the power plant by genetic
TLCC algorithms and of the operation of the power plant by
LCOE = gen · CRF(i, n) linear programming . The optimisation starts generating
Etot
a random population of a defined number of individuals,
Where, TLCC is the total life cycle costs, and CRF is this means that each individual is a particular solar power
the capital recovery factor which depends on r (annual plant, i.e. a hybrid solar power plant with a defined size
interest rate) and N the lifetime of the project CRF = of the solar field area (m2 ), the storage size (M W hth ), the
r/[1 − (1 + r)−N ]. power block (M We ), and/or the photovoltaic power plant
Environmental performance should be an important in- capacity (M We ). After that, the algorithm evaluates each
dicator of any sustainable project. Key indicators of a individual under the operational optimisation model pre-
project can be given by: viously described and measures the performance of each
individual regarding the objectives of the design optimi-
• Reduction of carbon emissions (compared with the
sation problem. Then, the genetic algorithm defines the
emissions of the local network)
best offspring by crossing and mutating the population in
• Use of water (important in locations with restricted which power plants with better performance have higher
availability) chance to evolve. Finally, the end condition is associated
with a defined number of generations which is given to the
• Reduction of other contaminants model as an instruction.
Societal indicators are key in projects that are focusing
on social development as well as economics and environ-
mental performance. Some societal targets are:
6
 for low temperature heating operations mainly in copper
refining and hydrometallurgical processes [33]. Because
the mining industry leads the electricity consumption in
Northern Chile, the demand is quite flat, with no signifi-
cant variations between day and night, hence, the North-
ern Chile electricity market needs to supply a steady en-
ergy demand 24 hours and 7 days per week. Regarding the
Chilean Center for Economic Load Dispatch (CDEC) of
the Northern Interconnected System (SING), 75.4% of the
electricity generated during 2015 was generated in coal-
powered power plants, 21% from other fossil-fuelled power
plants (Natural gas, Diesel, Fuel Oil), and just 3.6% from
renewable resources (solar, wind, hydro) [34]. According
to these numbers, the Chilean Ministry of Energy reported
that the carbon intensity of the SING in 2015 was 0.764
tCO2eq MWh−1 [35]. On the other hand, one of the biggest
challenges of the Chilean mining industry is to get econom-
ical, reliable, and sustainable energy resources, as well as
the efficient use of them [25]. To apply and prove the
model, the Atacama-1 or Cerro Dominador Solar Power
Plant, a hybrid solar power plant under construction in
the Atacama Desert in Chile has been studied. Regard-
ing the published information by the constructor company
[16] and by the Chilean Ministry of Environment [36], some
features of the project are:
Figure 2: Multi-objective optimisation framework
• Location: Antofagasta Region, Chile ≈ S 22◦ W 69◦
2.4. Computer information • CSP Plant
All simulations and optimisations were performed with
– Heliostats: 10, 600 ≈ 148.4 ha
the following hard- and software:
– StH: 17.5 h
• PC: Intel Core i7-6700 3.4 GHz and 16 GB RAM
– Power Block Capacity: 110 MW
• Operating system: 64-bit Windows 7
• PV Plant: Capacity: 100 MW
• Programming language: Python 3.5.3
• GHG emissions avoided: 870, 000 tCO2eq year−1
• Optimisation packages: Pyomo with CPLEX and
• Total Investment: between 1, 300 and 1, 500 MUSD
Gurobi, DEAP
3.1.1. Solar Irradiation Data
3. Results The DNI and the GII (for a panel slope ≈ latitude)
data in the location of the project was obtained from the
3.1. Case Study Chilean Ministry of Energy and University of Chile solar
In order to set a case study, the Atacama Desert resource data centre [29]. This open source information
will be considered. This arid region which covers around includes weather and irradiation data of the Chilean ter-
300, 000 km2 is located in Northern Chile, and is one of the ritory.
sunniest places on Earth [31], where in a typical year, the Figure 3 shows the moving average of 1 and 2 days in
annual Direct Normal Irradiation (DNI) is near or more the location. In both cases, the 5th percentile is around
than 3, 500 kWh m−2 [29], [15]. In this zone are located 23 330 Wh m−2 , this means that 95% of the time the DNI is at
of the 30 bigger copper mines of the Chilean copper in- least 330 Wh m−2 day−1 , in other words, the daily DNI has
dustry [25], which accounts for around 73% of the copper no great variation during the year. The present study con-
production in Chile, whereas this Chilean industry con- siders the typical meteorological year, hence, the results
tributes more than 30% of the total copper production represent the long time performance of the project. Never-
of the world [32]. This copper industry is a continuous theless, the irradiation variability in the Atacama Desert is
and energy intensive process, in fact, during 2015, Chilean influenced by El Niño-Southern Oscillation (ENSO). While
copper mines used around 23, 600 GWh of electricity and La Niña has a high correlation with high precipitations
additionally around 2, 700 GWh was burned as a fossil fuel

7
 Table 1: Efficiencies used in the model

CSP Plant PV Plant

solarf ield
η =0.48735 η panel =0.1806
η pipelines =0.99 η P V →IN V =0.85048
η T ES =0.99 η IN V →N ET W =0.9726
η P B =0.3708

3.2. Operational optimisation - CSP and PV plant

The design of the CSP plant of Atacama-1 has a so-
lar multiple of 2.59, and a storage capacity equivalent to
17.5 hours of full capacity. The operational optimisation
model was first applied to the CSP plant of Atacama-
1. The parameters used were the typical meteorological
year of the location reported by [29], a discounted rate
equal to 7% and 25 years of lifetime (in order to com-
pare the results with the IEA report [38]). The model
Figure 3: Moving average of the solar irradiation for the typical gen
meteorological year
was run twice with two different objectives: M ax{Etot }
and M in{LPSC} (for a fixed constant supply of 110 MW,
which correspond to the maximum capacity of the power
during summer (Dec-Jan-Feb-March) of the Southern Hemi- block), these results are shown in column CSP in Table 2.
sphere in the Atacama Desert between a six years interval, Because Eidemand is equal to the capacity of the power
gen
high rainfalls during winter are associated with El Niño plant, both M ax{Etot } and M in{LPSC} obtain the same
[37]. These phenomena will result in years with solar ir- results. Nevertheless, if Eidemand were variable and not
radiation significantly different from TMY. The effect this equal to the capacity of the power plant, or if the local
has on the hybrid solar power plant will be evaluated in a electrical market had variable electricity cost during the
future publication. day, the optimisations with different objectives would pro-
duce different results. These different cases, that depend
3.1.2. Solar Power Plant Features on the market in which the system will operate, can be
Technical and financial information of solar power plants modelled by adjusting the input parameters of the pro-
in operation and under construction is necessary to built gramme.
the model. The System Advisor Model (SAM) [8] was used In the case of the PV plant of Atacama-1, the one
to get financial (unitary costs) and technical (efficiencies) year operational optimisation results, similarly to the CSP
gen
data to built the model. In the model, the investment cost plant, are the same for both objectives: M ax{Etot } and
is calculated by scaling the unitary cost of the compo- M in{LPSC} (for a constant supply of 100 MW which is
nents of the CSP and the PV power plants, e.g. USD m−2 the maximum capacity of the plant), these results are
for the solar field, USD MWh− 1 for the thermal energy shown in the column PV in Table 2.
storage system, USD MW−1 for the power block, among
others. Around 70% of the total annual daytime hours 3.3. Multi-objective Operational Optimisation - Hy-
the DNI is greater than 800 Wh m−2 , moreover, the op- brid plant
erational optimisation is focused on maximising the use The objectives defined in the model depend on the tar-
of the power plant, hence the power block will work near get pursued by the owner of the power plant. For instance,
full capacity most of the time. For these reasons, and in the user of the model can be focused just on a financial
order to simplify the model, the efficiencies used for each perspective by selling the maximum quantity of energy,
subsystem are constant for every hour and every design, thus, reaching the lowest LCOE. On the other hand, the
these are shown in Table 1. These efficiencies are used to case that the owner of the power plant is a large consumer,
estimate the sizes of the components and the operational its objectives might be focused on both financial and re-
performance of the system. In order to validate the model, liability performance. In addition, these will be different
results of the model were compared with both, the System for grid connected or off-grid power plants. Finally, the
Advisor Model and information published by the IEA in electricity market operator will be focused on both, a low
the report: Projected Cost of Generating Electricity 2015 price of the electricity, as well as a firm electricity supply,
edition [38]. which can be evaluated through the LPSC. The opera-
tional optimisation developed in this model will focus on a
combination of these two objectives, the maximisation of

8
 Table 2: Operational optimisation Atacama-1

CSP PV Hybrid
gen
KPI unit M ax{Etot } M in{LPSC} Autom.Scalarisation
gen
Etot GWh year−1 864.3 261.4 1,125.7 953.8 1,109
LPSC GWh year−1 99.3 615.9 97.5 9.8 9.9
LPSP % 10.30 72.30 10.12 1.016 1.02
CFCSP % 89.69 - 89.69 71.85 87.98
CFPV % - 27.12 27.12 27.12 27.12
Investment MUSD 1,192 262 1,455 1,455 1,455
LCOE USD MWh− 1 132.06 92.40 122.85 144.04 124,60

the energy supplied (low cost of electricity) and the min- the correct value of the scaling or constraint parameters
imisation of the loss of power supply. in order to get suitable solutions. Moreover, the result of
In this context, the previous results show that CSP a multi-objective optimisation problem is a Pareto fron-
with TES plants have higher LCOE but lower LPSC than tier (or surface) which represents a set of points that are
PV plants (because energy can be stored in the CSP plant solutions to the problem, ergo, every point is an optimal
and used during night hours), hence, the combination of candidate with exclusive features regarding the objectives.
these two solar power plants should mean both better fi- These methods, which are modelled and applied to the hy-
nancial and technical performances. Consequently, a de- brid power plant, are described below.
crease in both LCOE and LPSC is expected by the hy-
bridisation. In this case the maximum capacities of both 3.3.1. Scalarisation method
PB PV
power plants are different, Pmax = 110 MW and Pmax = The following function describes the new single objec-
demand
100 MW. Besides, the Ei was defined fixed and con- tive optimisation problem:
stant at every hour and equal to the maximum capacity I
of the CSP power plant (110 MW), because unlike the PV X
maximize {Pigen ∆ti − wLP Si ∆ti }
plant, the CSP plant with thermal energy storage system
i=1
can deliver energy during the night. Therefore, both meth-
gen Where the positive parameter w is the scaling factor ap-
ods: M ax{Etot } and M in{LPSC} get different results,
which are summarised in Table 2. plied to the second objective. Regarding the results shown
gen in Table 2, the scaling factor that balances the second ob-
First, the M ax{Etot } method achieves the highest to-
tal energy generated (1, 125.7 GWh year−1 ), consequently, jective (LPSi ) with respect to the first one (Pigen ) can be
the minimum LCOE, but the LPSP is high (10.12%). Sec- approximated by
ond, the M in{LPSC} results in a very low value in both Emax − Emin 1, 125.5 − 953.8
gen
LPSP (1.016%) and Etot (953.8 GWh year−1 ), consequently, = = 1.96
LPSCmax − LPSCmin 97.5 − 9.8
a higher LCOE. The hybridisation of the power plant should
be able to give a better performance. However, the right Hence, in order to built the Pareto frontier, and compare
operational strategy of the hybrid solar power plant is es- the solutions of this method with the previous results of
sential in order to simultaneously maximise the energy de- the single objective optimisations, w was evaluated in the
livered to the network (which influences the LCOE) and range w ∈ [0, 100], which is large enough to cover the
minimise the LPSC (which influences the reliability of its range between the two single objective optimisations. The
operation). For that reason, a multi-objective optimisa- Pareto frontier generated from this method is presented in
tion algorithm, showing the trade-off between financial and Figure 4.
technical performance is needed, in order to reach better
results. 3.3.2. Epsilon constraint method
Two techniques have been applied to solve multi-objective In this method the optimisation problem is formulated
optimisation problems in the linear programming oper- as:
ational optimisation [39]: the weighted-sum or scalari- gen
sation method, and the ε-constraint method. Whilst in maximize Etot
the scalarisation method, the multi-objective optimisation subject to LPSC ≤ ε
problem is transformed to a single objective optimisation
problem by combining and weighting both objectives [40], Where ε varies between the values of the LPSC given by
in the ε-constraint method, one objective is considered as a both previous single objective optimisations shown in Ta-
constraint in the formulation of the optimisation problem ble 2: ε ∈ [9.8, 97.5] GWh year−1 . The Pareto optimal val-
[41]. One important challenge of these methods is to define ues generated from this method are also shown in Figure
4.
9
 gen
jective optimisation of M ax{Etot } and it gives the point
I1 = (Emax , LPSCmax ) in Figure 4. Then, the second it-
eration is a single objective optimisation of M in{LPSC}
which produces the point I2 = (Emin , LPSCmin ) (Figure
4). The purpose of these first two iterations is to get both
an estimation of w and the line I1 I2 . Then, the initial
value of w is calculated by
Emax − Emin
w0 =
LPSCmax − LPSCmin
in order to give the same relative weight to both objectives.
After that, 5 iterations are carried out to get an improved
w (w = αw0 ) which is used as input to the a-posteriori
automated selection. This last step, the automated a-
posteriori selection can be modelled by different methods
depending on the purpose of the user. In this case, it is
gen
Figure 4: Etot vs LPSC, summary all methods done by selecting the α value for which the result (Ii =
(Ei , LPSCi )) of the ith iteration is furthest from the line
I1 I2 . This is calculated by finding the maximum distance
From Figure 4 it is possible to appreciate the behaviour di which is the perpendicular line that connects Ii with
of each of the 4 different methods studied: First, the I1 I2 ). The following algorithm describes this procedure:
gen
M ax{Etot } method maximises the energy delivered, get- start
ting the lowest LCOE (122.85 USD MWh−1 ), nevertheless,
it presents the highest LPSP 10.12% which is not an at-
tractive value for the reliability of the system. Second, the 1st it.obj :
M in{LPSC} method is outside of the Pareto frontier, and gen I1 = (E1 , LPSC1 )
M ax{Etot }
it presents a very low value of the total energy delivered
to the network, so, the LCOE associated to this method
is high and not attractive for the financial optimisation 2nd it.obj :
I2 = (E2 , LPSC2 )
of the system. Finally, both the scalarisation and the ε- M in{LPSC}
constraint methods show similar Pareto optimal solutions.
Because the purpose of the model is focused on a strate-
E1 − E2
gic point of view in which the generation company is look- calculate w0 w0 =
ing for a maximisation of its profit (a minimum LCOE), LPSC1 − LPSC2
and the market operator is focusing on a reliable energy
system, the objective of the operational optimisation will 3rd − 7th it.obj :
M ax{E − αwo LPSC} : Ii = (Ei , LPSCi )
focus on a combination of both objectives, thus, the tar-
get of the optimisation should reach the zone highlighted α = 10{−4,−2,0,2,4}
gen
with an ellipse in Figure 4. Both the M ax{Etot } and the
M in{LPSC} methods do not reach this zone, on the other
hand, the scalarisation and the ε-constraint method, with select
a good definition of values of w and ε, respectively, can α|di maximised
best w
reach this area. The main difference between the scalarisa-
tion and the ε-constraint methods is that each iteration of
the scalarisation method takes a few seconds to calculate,
compared with each iteration of the ε-constraint method gen
Results Etot , LPSC
which takes almost 10 minutes to be processed. Hence, to
continue with the optimisation, the scalarisation method This automated scalarisation method was applied to
is analysed in detail and automated, to ensure that the find the best operational profile for Atacama-1. Figure 4
optimal value of w is chosen. summarises this method, in which the best result, or the
maximum value of segment d is found for α = 1. As a
3.3.3. Automated scalarisation method clarification, segments d and I1 I2 are perpendicular, nev-
In order to have an automated decision system ensur- ertheless, because the scale of both axes are different in
ing that the solution of the operational optimisation is the Figure 4, these do not seem to be perpendicular in the
in the desired zone, an automated scalarisation method diagram.
was developed and applied. This autonomous algorithm The results of the automated scalarisation are shown
requires 7 iterations. The first iteration is a single ob- in Table 2. These results, compared with the method in
10
 Figure 5: Annual energy flow diagram

gen
which the objective is the M ax{Etot } means an increase
in 1.4% in the LCOE, nevertheless, the LPSP is just 10%
of the original (1.02% instead of 10.12%). As expected,
comparing the CSP plant with the hybrid plant, the LCOE
Figure 6: State of charge of the TES system
decreases from 139.06 to 124.6 USD MWh−1 and the LPSP
decreased from 10.30% to 1.02%.
Another useful output of the operational optimisation which are applied to the operational optimisation. Third
are the hourly power flows between each component, and the objectives can be related with the goals that the user
the losses in each subsystem. With these it is possible to is focusing on. Finally, the variables are related with the
get the state of charge (SoC) of the thermal energy storage capacities of each component, these are the solar field area
system, which is shown in Figure 6. The storage system of and the capacities of the thermal energy storage system,
Atacama-1 has a StH of 17.5 h, this means that when the the power block and the photovoltaic power plant.
storage is fully charged (100%) the power plant can work at
full capacity for 17.5 h with no solar irradiation. However, Table 3: Design Optimisation: list of parameters and constraints
regarding Figure 6, the maximum state of charge of the and all potential objectives and variables
TES system is 83%, with a mean of 36.3%. This suggests
Param. Constr. Obj. Var.
that it is possible that the TES system of Atacama-1 is
oversized and its capacity could be reduced in order to DNI Energy LCOE ACSP
STO
reduce investment costs in the design step, which opens the GII balances Investment Emax
PB
possibility to improve the design of Atacama-1. Regarding Efficiencies Power LPSC Pmax
PV
the features of the thermal energy storage system, there is Unitary costs capacities Pmax
a lower limit of temperature that the molten salts must not Financial param.
reach in order to avoid its solidification. Hence, as shown
in Figure 6, during cold months, where the minimum state The following results describe different methods ap-
of charge is 0 every day, special attention has to be put plied to the same design optimisation problem. These
on the perfect operational control needed to avoid this depend on the number of objectives, and the number of
problem. variables. First, the number of variables considered in the
optimisation can be from one variable up to four variables,
3.4. Multi-objective design optimisation - Upgrad- related with the capacities of each subsystem. The case in
ing Atacama-1 which one variable is considered can be thought as an up-
The optimisation model was applied and run for the grade to the existing power plant in order to improve its
same location of Atacama-1. The purpose of this step is performance. On the other hand, four variables can be
to get the best design of a hybrid solar power plant in the considered in order to develop a brand new power plant
same location of Atacama-1 and with similar features. Ta- defined by the given parameters. Second, the number and
ble 3 summarised the components considered in the design kind of objectives considered by the user. For instance,
optimisation. First, the parameters are related with the a generator company might want to increase the revenues
solar resource of the location, the efficiencies and unitary of the power plant by reaching the lowest LCOE. Other
cost of the components given by data from power plants users could be the market operator or a large consumer,
under construction and in operation, as well as other finan- which could be interested not just in the financial perfor-
cial parameters like the lifetime, the discount rate, among mance but in the reliability as well. The most complex
others. Second, the constraints are associated with the ap- situation is a multi-objective optimisation, in which finan-
plication of energy balances and energy or power capacities cial, technical, environmental and/or social objectives are
11
pursued. of the project with the objective of decreasing the LCOE.

3.4.1. Single variable, single objective

As a first approximation, a single variable single objec-
tive optimisation was developed and applied to Atacama-
1. In this model just one design variable is considered as
a variable and the other three are considered parameters
(fixed). This problem was developed as a deterministic
global optimisation problem, and an improved design was
reached. Table 4 shows the results reached by the de-
sign optimisation of the power plant focusing on the min-
imization of the LCOE, while the operation of each itera-
tion (which correspond to a different power plant design)
was optimised by the automated scalarisation method de-
scribed previously, that simultaneously maximises the en-
ergy delivered and minimises the LPSC.
Variable: Solar field area. Keeping TES, PB, and
PV capacities fixed, the deterministic global optimisation
was run getting an improved design of the plant when the
solar field area is 161.4 ha, this means 930 more heliostats
than the original design (11, 530 instead of 10, 600), hence,
a larger SM. As a results, despite the investment increase, Figure 7: Results showing the sensitivity analysis of the design op-
timisation with 1 variable and 1 objective.
more energy can be delivered, and more energy is available
during night, thus, both LCOE and LPSC decrease
Variable: Thermal energy storage capacity. Keep-
ing the solar field area, the PB and the PV capacities 3.4.2. Multi-variable, multi-objective optimisation
STO PV
fixed, the best design of the plant is given by a decrease The multi-variable (ACSP , Emax , and Pmax ) and multi-
of the TES capacity from 5243 MWh (Sth= 17.5 h) to objective (LCOE, Investment, LPSC) design optimisation
3503 MWh (StH= 11.7 h), achieving an LCOE= 118.78 is performed with the previously described genetic algo-
USD MWh−1 . This result agrees with the previous analy- rithm. This heuristic optimisation method starts with a
sis of the state of charge of the TES system. The reduction random population (here 200 individuals), and applies the
in TES capacity directly reduces the investment costs and NSGA-II algorithm through a number of generations (here
thus has a positive impact on the LCOE. Moreover, the 80). Every individual is composed of 3 variables, corre-
lower TES capacity produces two effects: (i) a decrease in sponding to a defined design. In the algorithm, the in-
the energy losses in the storage system (because less en- vestment related to each individual is calculated, then its
ergy is stored), as a result more energy can be delivered optimal operational performance is achieved by applying
(which has a positive effect on the LCOE), (ii) because the automated scalarisation method described previously.
less energy is available during night hours, less energy is This simultaneously optimises the LCOE and LPSC. The
delivered during this time (which has a negative effect on selection of the best individuals is executed by the algo-
the LPSC), therefore, LPSC increases. The combination rithm under two or three objectives: Min LCOE, Min In-
of all these interactions results in a small increase on the vestment and/or Min LPSC.
energy delivered during the day, a decrease in the energy The optimisation produces a range of different points
available during night, and a considerable decrease on the that represents different options of the design of the hy-
investment, as a result, LCOE decreases and LPSC in- brid solar power plant, and the respective performance
creases. during its lifetime (based on the TMY). Each design on
Variable: Power block capacity. By keeping the this Pareto frontier represents a potential solution and the
solar field area, TES and PV capacities fixed, the best final choice will depend on the aims of the developer. In
design of the plant is reached by a decrease in the power every case, in order to calculate the LPSC in the oper-
block from 110 to 108 MW. This small change in the power ational optimisation step, and to get results comparable
block should mean a small decrease in the investment and with Atacama-1, the power that the generation should sup-
thus a decrease in the LCOE, but an increase in the LPSC ply was considered fixed, i.e. Eidemand = 110 MW. For this
because the demand is fixed, i.e. Eidemand = 110 MW. reason, because the CSP plant can deliver energy during
In summary, all the iterations of the optimisation for the night, the power block capacity was fixed and equal to
PB
the three cases are shown in the sensitivity analysis dia- the demand, Pmax = 110 MW.
gram in Figure 7, which shows the convexity of the model Three variables, two objectives:
for single variables and single objectives. Moreover, this The design optimisation is extended to three variables
STO PV
diagram illustrates the best options to improve the design (ACSP , Emax , Pmax ) and two objectives (LCOE and In-
12
 Table 4: Design Optimisation - single variable, single objective

Indicator unit Atac-1 CSP Storage PB

New value - 161.4 ha 3503 MWh 108.3 MW
SM - 2.59 2.82 2.59 2.63
StH h 17.5 17.5 11.68 17.8
gen
Etot GWh year−1 1,109 1,157 1,110 1,105
LPSC GWh year−1 9.9 6.3 69.1 16.1
LPSP % 1.02 0.65 7.17 1.67
CFCSP % 87.98 93.0 88.1 88.9
CFPV % 27.1 27.1 27.1 27.1
Investment MUSD 1,455 1,513 1,381 1,450
LCOE (objective) USD MWh−1 124.6 123.95 118.78 124.57

vestment cost). Figure 8a shows the results of the opti- Table 5: Design Optimisation - Three variables, Two objectives
misation, the Pareto Frontier as well as the performance
item unit Atac-1 A B
of Atacama-1 in order to make a quick comparison be-
tween the results. For instance, Figure 8a highlights two ACSP ha 148.4 115.3 92.2
STO
points (A and B) belonging to the non-dominated solu- Emax MWh 5243 2314 1230
PV
tions. These solutions are detailed in Table 5, including Pmax MW 100 164 69
the performance of Atacama-1 to make a quick compari- SM − 2.59 2.01 1.61
son. These points are related with the best performance StH h 17.5 7.72 4.1
gen
that can be reach with similar Investment or similar LCOE Etot GWh year−1 1,109 1,119 742
presented by Atacama-1. The first point, A, shows a de- LPSC GWh year−1 9.9 215 363
sign with a decrease of the LCOE and Investment, never- LPSP % 1.02 22 37
theless, its LPSP is 22%. The second point, B, displays CFCSP % 87.98 71.4 58.28
that a similar LCOE than Atacama-1 can be reached with CFPV % 27.1 27.1 27.1
just 65% of the original investment, but the LPSP is 37%, Investment MUSD 1,455 1,353 951
a very high value compared with Atacama-1. LCOE USD MWh−1 124.6 115.9 125.3
Figure 8b shows the parameters and objectives of ev-
ery point on the Pareto frontier with LPSP< 30%. On the
horizontal axis are shown the 4 components of the design • LPSP ≤ 3%
PB STO PV
(Pmax , ACSP , Emax , Pmax ), the two objectives of the de-
sign optimisation (LCOE, Investment cost), and the LPSP • LCOE ≤ 130 USD MWh−1
which is calculated from the operational profiles. The ver- • INV ≤ 1, 700 MUSD
tical axis is the normalised value of the variables, where
minimum and maximum values are indicated in the fig- In Figure 9a, the three dimensional performance (LCOE,
ure. This figure explains that the model can reach simul- Investment, LPSC) of the multi-objective optimisation is
taneously better LCOE and Investment cost for a design represented in a two dimensional diagram (LCOE, Invest-
similar to Atacama-1, nevertheless, because the technical ment), in which the third objective (LPSP) is illustrated
performance (represented by the LPSC) is not included in through different ranges and symbols. Near to the cen-
the design optimisation stage, the values of the LPSP are tre is Atacama-1, which divides the plane into four quad-
very high. As a consequence of these results, in order to rants. The crosses and the stars have a LPSP lower than
reach better financial (LCOE, Investment) and technical Atacama-1, thus, any of them located in Quadrant I have
(LPSC, LPSP) performance from the design optimisation, both better financial and technical performance than Atacama-
the LPSC is incorporated as a third objective. 1. Second, any point in Quadrant I has better financial
Three variables, three objectives: performance than Atacama-1, but its LPSP varies between
In this step, the complexity of the model is increased 0.65% to 3%. Third, in order to reach lower values of LPSP
through a third objective, that is related with the technical (shown by crosses and stars in the diagram), similar or
performance of the hybrid power plant. Here the ability higher investments are needed, nevertheless, lower values
to dispatch energy when it is needed (LPSC) is added. of LCOE can be reached simultaneously. Fourth, whilst
Figure 9a shows some of the points that belong to the lower values of LCOE (Quadrants I and II) are possible
Pareto surface as well as the performance of Atacama-1. with similar investments than Atacama-1, their LPSP can
Only points with a performance similar to or better than be even near 0.285%. For instance, five of these interesting
Atacama-1 as defined by the following ranges are shown: points are summarised in Table 6 and shown in Figure 9a
bounded in a circle and defined by the letters A, B, C, D
13
 corr(X, Y ) =
LPSP SM StH
 
LCOE −0.635 0.468 0.791
INV 
 −0.629 −0.29 0.42 

LPSP 
 1 −0.236 −0.76 

SM  −0.236 1 −0.073 
StH −0.76 −0.073 1

These results simplify the understanding of the design

of the power plant, moreover, some of them are key to
develop future approximations. For instance, because SM
and StH are related with the installed capacities of the so-
lar field and the storage system, both have a positive cor-
relation with the LCOE. In relation with the LPSP, as ex-
pected, this has a negative correlation with the LCOE and
Investment, in other words, in order to increase the tech-
(a) Pareto optimal solutions
nical performance of the power plant (decrease the LPSP
or increase the reliability of the system), a decrease in the
financial performance is expected (a higher value of the
LCOE and/or the initial Investment). This also can be
explained by the negative correlation between LPSP and
both SM and StH, suggesting that lower LPSP are reached
in oversized power plants. For that reason, the trade-off
between technical and financial performance is important.
Another interesting point is the correlation between SM
and StH, which suggest that there is a positive correlation
between the solar field capacity and the storage capacity
observed in optimised designs.
Finally, in order to have a complete picture for the a-
posteriori decision, the particular design of every individ-
ual, i.e. all the capacities of the components, can be com-
bined with key indicators of their optimised operational
(b) Parameters, objectives and final performance performance during its lifetime. In summary, Figure 9a
and detailed results of the model can be used together
Figure 8: Design Optimisation: 3 variables, 2 objectives to make a very accurate a-posteriori decision in order to
select the best design with optimised performances. Be-
cause Figure 9a is a representation of the non-dominated
and E. Designs A and B reach lower LCOE, nevertheless, solutions, the best decision should be made by the user.
high investments are necessary to get higher reliability,
this is explained by the performance of designs A and B.
Design C represents a power plants with similar invest- 4. Conclusion
ment but a lower LCOE than Atacama-1. Designs D and
In order to make renewable energy systems economical
E are examples of very reliable power plants, which can be
and reliable, the design and operation of hybrid renewable
developed with low LCOE, nevertheless, high investment
energy systems have to consider the financial and techni-
is necessary.
cal performance of the system and the synergies of differ-
The 12 stars located in Quadrant I in Figure 9a, which
ent technologies. To achieve this, the design optimisation
have better financial (LCOE, Investment) and technical
needs an internal routine which optimises the operational
(LPSC) performance than Atacama-1, are detailed in Fig-
profile with respect to multiple and often conflicting design
ure 9b. This last diagram shows the design parameters,
objectives. However,the operational optimisation is usu-
the results of the three objectives and some key design in-
ally performed with single objective linear programming
dicators (SM, StH, and CFCSP ) as well as its comparison
methods. This contribution presents a two-stage optimisa-
with Atacama-1. Moreover, for these 12 individuals a cor-
tion for the design of hybrid solar power plants which uses
relation matrix between each objective and key indicators
an automated scalarisation method for multi-objective lin-
of the design of the power plant (SM and StH and CFCSP )
ear optimisation of the operational profile.
was calculated and shown below:
14
 (a) Pareto optimal solutions

(b) Parameters, objectives and final performance

Figure 9: Design Optimisation: 3 variables and 3 objectives

15
 Table 6: Design Optimisation - multi-variable, multi-objective

item unit Atac-1 A B C D E

ACSP ha 148.4 151.5 154.6 146.6 157.5 164.7
STO
Emax MWh 5243 4276 4658 3956 5040 4717
PV
Pmax MW 100 75 95 125 121 126
SM − 2.59 2.64 2.7 2.56 2.75 2.88
StH h 17.5 14.27 15.55 13.2 16.81 15.74
gen
Etot GWh year−1 1,109 1,057 1,120 1,173 1,201 1,239
LPSC GWh year−1 9.9 19 8.9 27.5 6 5.8
LPSP % 1.02 2.0 0.9 2.86 0.6 0.6
CFCSP % 87.98 89.44 90.51 87.58 91.64 94.06
CFPV % 27.1 27.1 27.1 27.1 27.1 27.1
Investment MUSD 1,455 1,361 1443 1,461 1,544 1,577
LCOE USD MWh−1 124.6 122.87 122.66 118.57 121.89 120.6

The two-stage optimisation framework developed in 1 can be simultaneously improved by an optimised de-
this research simultaneously optimises the design and op- sign. For example, with an investment of 1443 MUSD
eration of a hybrid solar power plant with respect to mul- (lower than Atacama-1) a decrease in the LCOE from
tiple objectives. The results of the design optimisation 124.6 to 122.66 USD MWh−1 and a decrease in the LPSP
stage produce the best configuration (sizes, capacities) of from 1.02% to 0.9% can be reached. Moreover, the op-
a hybrid solar power plant while the operational opti- timisation produces a Pareto frontier of non-dominated
misation simultaneously produces the optimal operation solutions which show the trade-off between different ob-
of the power plant. The latter can be used to analyse jectives. Thus the specific design needs to be selected by
the hourly power flows between each component as well the developer based on further criteria, such as a limited
as the estimated losses in each subsystem. Two meth- budget available for the project (investment cost), as well
ods for the operational optimisation were evaluated and as other environmental or societal performance indicators
it was found that the linear scalarisation method achieves that could be key for the user of the model.
the same results but is much faster than the ε-constraint The large number of potential solutions enabled the
method. Then, the linear scalarisation method was auto- development of correlations between the different design
mated by valuing the trade-off between the two objectives. parameters (e.g. SM, StH) and objectives (LCOE, invest-
This enabled the integration of the multi-objective linear ment cost, LPSP) of hybrid solar power plants. These
optimisation in the two-stage multi-objective optimisation correlations can be used to get a first approximation for
framework. the design of hybrid solar power plants. For instance,
The optimisation framework was applied to analyse the reverse correlation between LPSP with LCOE and In-
and improve the design of the hybrid solar power plant vestment, indicates the trade-off between technical and fi-
Atacama-1. The results show that both the financial and nancial performance, in addition, the reverse correlation
technical performance can be optimised. First, it was between LPSP with SM and StH suggests that oversized
shown that both, the LCOE and the LPSP of the CSP power plants reach better reliability.
plant can be improved by its hybridisation with a PV This investigation can be applied to any location and
power plant. Whilst the LCOE decreased by 5.6%, the for different scenarios: for instance, power plants that sup-
LPSP was reduced by 90%. By varying a single design ply electricity to the network, as well as off-grid power
variable from Atacama-1 it was shown that the energy plants that supply energy to a large end-user company.
storage system could be reduced by 33% (from 17.5 to Besides, the optimisation of the technical and financial
11.7 h), whereby the LCOE decreases by almost 5% (from performance can be tailored to particular perspectives of
124.6 to 118.78 USD MWh−1 ), but the LPSP increased the owner of the power plant. In addition, the optimisa-
from 1.02% to 7.17%. A two-objectives optimisation shows tion framework can be used to evaluate the technical and
that both the LCOE and investment cost can be reduced financial performance of solar power plants as well as the
simultaneously. However, its reliability was reduced sig- operation of such systems, in order to get key information
nificantly because it was not considered as an objective in that supports decision and policy making.
the design optimisation.
A three-objective optimisation (LCOE, investment cost Acknowledgements
and LPSP) was prompted by the increase in LPSP of
the previous optimisations. This optimisation shows that Ruben Bravo is supported by a PhD Scholarship from
both, the technical and financial performance of Atacama- Becas Chile, National Commission for Scientific and Tech-
nological Research (CONICYT-Chile).
16
References [21] A. Starke, J. M. Cardemil, R. Escobar, S. Colle, Assessing the
performance of hybrid CSP+PV plants in northern Chile, in:
[1] IEA, Tracking Clean Energy Progress 2017, International En- AIP Conference Proceedings, Vol. 1734, 2016. doi:10.1063/1.
ergy Agency, Organisation for Economic Co-operation and 4949230.
Developmentdoi:10.1787/energy_tech-2014-en. [22] A. S. Wallerand, A. Selviaridis, A. Ashouri, F. Maréchal, Tar-
[2] IEA, Energy Technology Perspectives 2017, International En- geting Optimal Design and Operation of Solar Heated Indus-
ergy Agency, Organisation for Economic Co-operation and trial Processes: A MILP Formulation, Energy Procedia 91 (91)
Developmentdoi:10.1787/energy_tech-2014-en. (2016) 668–680. doi:10.1016/j.egypro.2016.06.229.
[3] P. Denholm, M. Hand, Grid flexibility and storage required to [23] S. A. Kalogirou, Optimization of solar systems using artificial
achieve very high penetration of variable renewable electricity, neural-networks and genetic algorithms, Applied Energy 77 (4)
Energy Policy 39 (3) (2011) 1817–1830. doi:10.1016/j.enpol. (2004) 383–405. doi:10.1016/S0306-2619(03)00153-3.
2011.01.019. [24] O. O. Amusat, P. R. Shearing, E. S. Fraga, Optimal integrated
[4] P. Denholm, J. Jorgenson, M. Miller, E. Zhou, Methods for An- energy systems design incorporating variable renewable energy
alyzing the Economic Value of Concentrating Solar Power with sources, Computers & Chemical Engineering 95 (2016) 21–37.
Thermal Energy Storage, Tech. Rep. July, National Renewable doi:10.1016/j.compchemeng.2016.08.007.
Energy Laboratory (2015). [25] Mining Council Chile, Consejo Minero, Reporte Anual 2015,
[5] IEC, Electrical Energy Storage - White Paper, Tech. rep. (2011). http://www.consejominero.cl (2015) 54.
arXiv:arXiv:1011.1669v3, doi:10.1002/bse.3280020501. [26] M. Grageda, M. Escudero, W. Alavia, S. Ushak, V. Fthenakis,
[6] A. Abbas, G. Huff, A. B. Currier, B. C. Kaun, D. M. Rastler, Review and multi-criteria assessment of solar energy projects
S. B. Chen, D. T. Bradshaw, W. D. Gauntlett, DOE/EPRI 2013 in Chile, Renewable and Sustainable Energy Reviews 59 (2016)
electricity storage handbook in collaboration with NRECA, 583–596. doi:10.1016/j.rser.2015.12.149.
Tech. Rep. July (2013). doi:SAND2013-5131. [27] G. Cáceres, N. Anrique, A. Girard, J. Degrève, J. Baeyens,
[7] G. Zini, Energy Storage as a Value Proposition, in: Green Elec- H. L. Zhang, Performance of molten salt solar power towers
trical Energy Storage: Science and Finance for Total Fossil Fuel in Chile, Journal of Renewable and Sustainable Energy 5 (5).
Substitution, McGraw Hill Professional, Access Engineering, doi:10.1063/1.4826883.
2016. [28] M. Sengupta, A. Habte, S. Kurtz, A. Dobos, S. Wilbert,
[8] NREL, System Advisor Model ( SAM ) (2011). E. Lorenz, T. Stoffel, D. Renné, C. Gueymard, D. Myers,
URL https://sam.nrel.gov/ S. Wilcox, P. Blanc, R. Perez, Best Practices Handbook for
[9] K. M. Powell, K. Rashid, K. Ellingwood, J. Tuttle, B. D. Iver- the Collection and Use of Solar Resource Data for Solar Energy
son, Hybrid concentrated solar thermal power systems: A re- Applications, Tech. rep. (2015).
view, Renewable and Sustainable Energy Reviews 80 (2017) [29] Ministry of Energy - University of Chile, Explorador Solar Chile
215–237. doi:10.1016/j.rser.2017.05.067. - Department of Geophysics (2016).
[10] IEA, Technology Roadmap: Solar Thermal Electricity, Interna- URL http://www.minenergia.cl/exploradorsolar/
tional Energy Agency, Organisation for Economic Co-operation [30] D. Y. Goswami, Principles of solar engineering, 3rd Edition,
and Developmentdoi:10.1007/SpringerReference_7300. Boca Raton, FL : CRC Press, 2015.
[11] NREL, Concentrating Solar Power Projects. [31] R. R. Cordero, A. Damiani, G. Seckmeyer, J. Jorquera, M. Ca-
URL https://www.nrel.gov/csp/solarpaces/ ballero, P. Rowe, J. Ferrer, R. Mubarak, J. Carrasco, R. Ron-
[12] Solar Reserve LLC, Crescent Dunes (2012). danelli, M. Matus, D. Laroze, The Solar Spectrum in the Ata-
URL http://www.solarreserve.com/en/global-projects/ cama Desert, Scientific Reports 6 (2016) 22457. doi:10.1038/
csp/crescent-dunes srep22457.
[13] I. Rodrı́guez, C. D. Pérez-Segarra, O. Lehmkuhl, A. Oliva, Mod- [32] S. Northey, N. Haque, G. Mudd, Using sustainability reporting
ular object-oriented methodology for the resolution of molten to assess the environmental footprint of copper mining, Jour-
salt storage tanks for CSP plants, Applied Energy 109 (2013) nal of Cleaner Production 40 (2013) 118–128. doi:10.1016/j.
402–414. doi:10.1016/j.apenergy.2012.11.008. jclepro.2012.09.027.
[14] M. Balghouthi, S. E. Trabelsi, M. B. Amara, A. B. H. Ali, [33] Comisión Chilena del Cobre, Yearbook: Copper and other Min-
A. Guizani, Potential of concentrating solar power (CSP) tech- eral Statistics 1996-2015, Tech. rep. (2016).
nology in Tunisia and the possibility of interconnection with [34] CDEC-SING, Annual Report and Operational Statistics 2015,
Europe, Renewable and Sustainable Energy Reviews 56 (2016) Tech. rep. (2015).
1227–1248. doi:10.1016/j.rser.2015.12.052. [35] Ministry of Energy - Chilean Gobernment, Indicadores Ambi-
[15] C. Parrado, A. Marzo, E. Fuentealba, A. G. Fernandez, 2050 entales del Sector Energı́a.
LCOE improvement using new molten salts for thermal energy URL http://www.minenergia.cl/indicadoresambientales/
storage in CSP plants, Renewable and Sustainable Energy Re- [36] Servicio de Evaluación Ambiental, Sistema de Evaluación de
views 57 (2016) 505–514. doi:10.1016/j.rser.2015.12.148. Impacto Ambiental e-seia (2014).
[16] Abengoa Solar, Abengoa: About Atacama-1 (2016). URL http://seia.sea.gob.cl
URL http://www.abengoa.com/web/en/novedades/atacama-1/ [37] J. Houston, Variability of precipitation in the Atacama Desert:
acerca/factsheet/ Its causes and hydrological impact, International Journal of
[17] M. Petrollese, D. Cocco, Optimal design of a hybrid CSP- Climatology 26 (15) (2006) 2181–2198. arXiv:joc.1492, doi:
PV plant for achieving the full dispatchability of solar en- 10.1002/joc.1359.
ergy power plants, Solar Energy 137 (2016) 477–489. doi: [38] IEA, NEA, OECD, Projected Costs of Generating Electricity
10.1016/j.solener.2016.08.027. 2015, International Energy Agency, Nuclear Energy Agency, Or-
[18] C. A. Pan, F. Dinter, Combination of PV and central receiver ganisation for Economic Co-operation and DevelopmentarXiv:
CSP plants for base load power generation in South Africa, Solar arXiv:1011.1669v3, doi:10.1787/cost_electricity-2015-en.
Energy 146 (2017) 379–388. doi:10.1016/j.solener.2017.02. [39] B. H. Gebreslassie, G. Guillén Gosálbez, L. Jiménez, D. Boer,
052. Design of environmentally conscious absorption cooling sys-
[19] G. Srilakshmi, N. Suresh, N. Thirumalai, M. Ramaswamy, Pre- tems via multi-objective optimization and life cycle assessment,
liminary design of heliostat field and performance analysis of Applied Energy 86 (9) (2009) 1712–1722. doi:10.1016/j.
solar tower plants with thermal storage and hybridisation, Sus- apenergy.2008.11.019.
tainable Energy Technologies and Assessments 19 (2017) 102– [40] A.-T. Nguyen, S. Reiter, P. Rigo, A review on simulation-based
113. doi:10.1016/j.seta.2016.12.005. optimization methods applied to building performance analy-
[20] A. Green, C. Diep, R. Dunn, J. Dent, High Capacity Factor sis, Applied Energy 113 (2014) 1043–1058. doi:10.1016/j.
CSP-PV Hybrid Systems, in: Energy Procedia, Vol. 69, 2015, apenergy.2013.08.061.
pp. 2049–2059. doi:10.1016/j.egypro.2015.03.218.

17
[41] S. Fazlollahi, P. Mandel, G. Becker, F. Maréchal, Methods
for multi-objective investment and operating optimization of
complex energy systems, Energy 45 (1) (2012) 12–22. doi:
10.1016/j.energy.2012.02.046.

18