POWER SYSTEM TRANSIENT
STABILITY ANALYSIS BY
NEURAL NETWORKS
Under the guidance of : Prof. Sushil Chauhan
Submitted by:
Chirag Goyal (14281), Shivam Gupta (14278),
Piyush Verma (14241), Reshma Azad (14239)
Overview
The transient stability analysis is one of the main studies carried out in
EPS (Electric Power Systems). This analysis can be carried out, e.g. by
simulation (numerical solution of nonlinear differential equations that
describe the system dynamic).
An alternative procedure consist of obtaining the analysis without
solving such differential equations, as it is the case of LDM
(Lyapunov Direct Method). Aiming to analyze a great number of
contingencies the simulation as well the LDM are procedures that
present yet computational cost little compatible with the real time
applications.
This work aims to investigate the use of artificial neural network in the
diagnosis of transient stability of EPS, having as its main objective the
computational cost reduction.
Introduction
Now-a-days, it has become a necessity to maintain synchronism because
the system is expanding day-by-day and these results in installation of larger
machines. Due to this, transient disturbances are increasing continuously in
power system. The transient disturbances are caused by the changes in the
load, switching operations, faults and loss excitations. Thus, it is very
important to regain synchronism or equilibrium after disturbances in the
electrical utilities. Hence, thorough analysis of transient stability is required
to reduce problems such as blackouts, loss of synchronism, etc.
The term stability refers to maintenance of synchronism and stability limit
refers to maximum power flow possible in the system or a part of system at
which system is operating with stability. Power system stability is the property
of power system that enables it to remains in a state of equilibrium under
normal operating conditions and to regain equilibrium after being
subjected to disturbances.
What is Transient Stability
Transient stability is the ability of system to remain in
synchronism during the period of disturbance and prior
to the time that the governors can act. The transient
stability analysis is carried out for a short time period that
will be equal to the time of one swing. This analysis is
carried out to determine whether the system losses
stability during the first swing or not.
Whenever there is a change in the configuration of the
system then the need for transient stability analysis arises.
Transient stability analysis
For Transient stability analysis we need to consider three states
1. Pre-fault - Before the fault occurs the system is assumed to be at an
equilibrium point.
2. Faulted - The fault changes the system equations, moving the system
away from its equilibrium point.
3. Post-fault - After fault is cleared the system hopefully returns to a new
operating point.
Swing equation
The behaviour of a synchronous machine during transients is described by
the swing equation.
Where ,
M = angular momentum of the rotor
Ps = mechanical power input
PE = Pmax sin = electrical power output
Pa = Ps - PE = accelerating power
Transient stability solution methods
There are two methods for solving the transient stability problem
1. Numerical integration
This is by far the most common technique, particularly for the large systems,
during the fault and after the fault the power system differential equations are
solved using numerical methods.
2. Direct or energy methods
These are the direct methods to determine transient stability using energy
functions. These include Lyapunov’s Direct Method, Potential Energy Boundary
Surface Method (PEBS) etc.
Mostly used to provide an intuitive insight into the transient stability problem.
PowerWorld Simulator
PowerWorld Simulator (Simulator) is a power system simulation
package designed from the ground up to be user-friendly and
highly interactive. Simulator has the power for serious engineering
analysis, but it is also so interactive and graphical that it can be
used to explain power system operations to non-technical
audiences.
Simulator is actually a number of integrated products. At its core is a
comprehensive, robust. Power Flow Solution engine capable of
efficiently solving systems of up to 60,000 buses. This makes Simulator
quite useful as a stand alone power flow analysis package.
Unlike other commercially available power flow packages,
however, Simulator allows the user to visualize the system through
the use of full-color animated one line diagrams with full zooming
and panning capability. Moreover, system models may be modified
on the fly or even built from scratch using Simulator’s full featured
graphical case editor. Transmission lines may be switched in or out
of service, new transmission or generation may be added, and new
transactions may be established, all with a few mouse clicks.
Simulator’s extensive use of graphics and animation greatly
increases the user’s understanding of system characteristics,
problems, and constraints, as well as of how to remedy them.
Here we have simulated IEEE defined Standard 9-BUS system in
PowerWorld Simulator to evaluate the different sets of inputs and
outputs.
It consists of 3 generator buses, 3 transformer buses and 3 load
buses.
The available feature of transient stability analysis in the PowerWorld
Simulator is used for evaluating the stability of system under different
faults and different fault clearing times.
Transient stability of the system is determined by creating the most
severe fault at the most critical point and observing the variation of
power angle with time.
If power angle settles to a value (new operating point) then the
system will be stable for applied transient disturbance, else the
system will be unstable.
Multilayered Perceptron Neural Network
The neural network we will use is of feedforward type made up of neurons of
McCulloch-Pitts with sigmoid non-linearity
The network training is carried out by the Back-Propagation (B-P) with multi layer
perceptron algorithm .
A multilayer perceptron (MLP) is a class of feedforward artificial neural network. An MLP
consists of at least three layers of nodes. Except for the input nodes, each node is a
neuron that uses a nonlinear activation function. MLP utilizes a supervised learning
technique called backpropagation for training . Its multiple layers and non-linear
activation distinguish MLP from a linear perceptron. It can distinguish data that is not
linearly separable.
The input is compounded by the nodal active power vector and the fault data is,
represented in a similar way to the binary code. The output (stable or unstable) is
determined through a computational program based in.
