Random Forest Ensemble of Support Vector Regression for Solar Power Forecasting
Download as PPTX, PDF•1 like•450 views
This document describes a methodology for generating combined solar power forecasts using an ensemble of support vector regression models. The methodology includes: 1. Developing 24 individual SVR forecasting models from two different solar power datasets. 2. Using random forest regression to combine the forecasts from the 24 SVR models. 3. Evaluating the combined forecasts against individual model forecasts and a simple average combination, finding the ensemble approach improved accuracy in most months.
Random Forest Ensemble of Support Vector Regression for Solar Power Forecasting
- 1. Random Forest Ensemble of Support Vector Regression Models for Solar Power Forecasting 1 Mohamed Abuella Prof. Badrul Chowdhury Energy Production and Infrastructure Center Department of Electrical and Computer Engineering University of North Carolina at Charlotte Charlotte, USA April 24, 2017
- 2. 2 Presentation Outline Solar Power Forecasting Combining Solar Power Forecasts Results and Evaluation
- 3. Renewables Generations (Wind and Solar) are Too Variable High Efficiency and Large Energy Storage Still not Exist Reducing Cost and Pollution Why Forecast? Variable Generations (V.G.) Forecasting
- 4. Variable Generations (V.G.) Forecasting The usage of VG forecasting in US. electric utilities and ISOs such as CAISO, MISO, NYISO,…etc. Botterud, J. Wang, V. Miranda, and R. J. Bessa, “Wind power forecasting in US electricity markets,” The Electricity Journal, vol. 23, no. 3, pp. 71–82, 2010. Elke Lorenz, “Solar Resource Forecasting” International Solar Energy Society (ISES) Webinar, 2016. 4 Forecast Horizons Where Forecast? Post operating reserve requirements Clear DA market using SCUC/SCED Rebidding for RAC Post-DA RAC using SCUC Prepare and submit DA bids Clear RT market using SCED (every 5min) Intraday RAC using SCUC Prepare and submit RT bids Post results (RT energy and reserves) Post results (DA energy and reserves) Day Ahead: Operating Day: 11:00 16:00 17:00 -30min Operating hour DA: Day Ahead. RT: Real Time. SCUC: Security Constrained Unit Commitment. SCED: Security Constrained Economic Dispatch. RAC: Reliability Assessment Commitment. Wind / Solar Power Forecasting
- 5. 5 Flowchart of the Combined Approach (Physical and Statistical) of the Solar Power Forecasting How Forecast? ; Solar Plant and Terrain Characteristics Numerical Weather Prediction (NWP) Atmospheric variables SCADA Data Spatial Refinement Local Roughness, topology. Atmospheric Stability Solar Power Plant Modeling Solar Power Conversion Equation/s Model Output Statistics (MOS) Systematic Error Correction Solar Generation Forecast Conversion to Power Downscaling Regression Extrapolation Solar Power Forecasting
- 6. 6 Normalizing the data is of paramount importance since the scale used for the values for each variable might be different. The best practice is to normalize the data and transform all the values to a common scale: Xstandardized = x − mean X std(𝑋 XScaled = a + x − min X max X − min X ∗ b − a where x is a sample from data variable X, {a, b} is the desired range of the normalized data, such as {0, 1}, and X (min, max)=the minimum and maximum values of the observed data. DATA NORMALIZINFG These two techniques of data normalizing are used also to build different forecasting models There is also a standardization technique, especially when the variance of the data is high, which is making the data to have a zero mean and a unit standard deviation, as follows:
- 7. 7 The Model: The Support Vector Regression (SVR) is a Machine Learning tool that is deployed here for short-term solar power forecasting. Flowchart of Machine Learning (Black Box) New Input X Predicted Output Y SVR is robust and reliable, to that it was chosen to carry out some different investigations that need reliable results. FORECASTING MODEL
- 8. 8 From Sklearn library, shows the effects of the optimal Hyperparameter of SVR model The best parameters are {C=1, gamma=0.10} Leads to 97% of classification accuracy γ (Gamma) C 1 100 0. 1 10 0.01 1 The Complexity of the SVR model changes by changing its parameters (C, gamma) Support Vector Regression (SVR) and its Hyperparameters (C and γ Gamma) FORECASTING MODEL
- 9. 9 Grid Search of SVR’s Hyperparameters Contour Plot of Grid Search of SVR’s Hyperparameters, C and Gamma RMSE (as a Score of SVR) C γ (Gamma) FORECASTING MODEL
- 10. 10 Construction of 12 SVR models from a dataset. Another 12 SVRs from the other dataset. The total number of SVR models is 24 VARIOUS FORECASTING MODELS The available data is divided into two sets: One dataset is consisting of all 26 months and another dataset is consisting of the most recent 12 months only.
- 11. General diagram of combining different models Model B Model A Model C Model N Method of Combining The Models Combined Forecasts Individual forecasting models 11 Combining Various Models Methods of Combining The Models The random forest is chosen to be the learning ensemble method for combining the various models’ outcomes. ENSEMBLE LEARNING The simple average of the various models’ outcomes. Fcomb=WA*MA+ WB*MB + WC*MC ….+ WN*MN WN is a weight is assigned to the outcome of a model MN
- 12. 12 Flowchart of an illustrative example of using the decision trees for the solar power forecasting The trees model is trained with historical data to find the rules that will be set and then used for combining other models’ outcomes and obtain Combined Forecasts. Decision Trees for the Solar Forecasting Flowchart of Machine Learning (Black Box) Hypothesis could be regression or classification ENSEMBLE LEARNING New Input X Predicted Output Y Temp>75 Pred_value=AX1+BX2+.. Cloud Cover>0.5 Pred_value=AX1+BX2+.. Temp<75 Pred_value=AX1+BX2+.. Cloud Cover<0.5 Pred_value=AX1+BX2+.. Solar Irradiance>400 Pred_value=AX1+BX2+.. Solar Power<1500 Solar Irradiance<400 Pred_value=AX1+BX2+.. Root Node Terminal Node Terminal Node Terminal Node Terminal Node
- 13. METHODOLOGY SCHEMES Day Month 1 : June : July : : : : : : : April 30 May 31 00:00 : 23:00 Weather Data : : : : : : : : : : : : : : : : : : : : : : PV Power : : (PastObservations) : : : : : : Forecasts (Model’s Outcomes) at 00:00 Day Month 1 : June : July : : : : : : : April 30 May 31 00:00 : 23:00 Weather Data : : : : : : : : : : : : : : : : : : : : : : Models’ Outcomes : : : : : : : : : : : : : : : : : : : : : : : : Forecasts PV Power : : (PastObservations) : : : : : : Combined Forecasts (a) (b) (a) Day-ahead forecasting by Support Vector Regression models 13 (b) Combining by Random Forest Day-ahead forecasts and combining them, for May 31st
- 14. Data Description: 14 The solar power system is in Australia The panel type is Solarfun SF160-24-1M195, consisting of 8 panels, its nominal power of (1560W), and panel orientation 38° clockwise from the north, with panel tilt (of 36°). The historical observed solar power data are normalized to the rated capacity (i.e., 1560W). https://crowdanalytix.com/contests/global-energy-forecasting-competition-2014-probabilistic-solar-power-forecasting Benchmark Data Scatter & Box plots the Data Cleansing the Data Correlation and Sensitivity Analysis Select most Effective Variables Flowchart of Data Preparation CASE STUDY Training Testing
- 15. 15 Flowchart of Solar Power Combined Forecasts
- 16. 16 RESULTS and EVALUATION Root Mean Squared Error The lower RMSE, the better forecasts accuracy The Performance evaluation are carried out by: Some statistical metrics, plots, and comparison. Best Model (4) =All-months dataset + Normalize (A) + Parm10_08 + Orig12ins The comparison is carried out over the entire year, between: The ensemble learning model vs.: The Simple Average methodand The higher positive improvement rate, the better ensemble forecasts Improvement % = 1 − RMSEEnsemble Model RMSEOther Model ∗ 100 (also called Skill Score)
- 17. 17 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22 Days RMSE The RMSE of the Models M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14 M15 M16 M17 M18 M19 M20 M21 M22 M23 M24 Combined Forecasts Average Forecasts Daily RMSEs of different models and Ensemble forecasts, October RESULTS and EVALUATION
- 18. 18 Monthly RMSE of All Forecasts RESULTS and EVALUATION 0.000 0.020 0.040 0.060 0.080 0.100 0.120 RMSE Monthly RMSE Simple Average Best Model Ensemble Root Mean Squared Error The lower RMSE, the better forecasts accuracy
- 19. 19 Improvement % = 1 − RMSEEnsemble Model RMSEOther Model ∗ 100 (also called Skill Score) The higher positive improvement rate, the better ensemble forecasts June July August September October November December January February March April May Best. Model -4% 3% 3% 3% 18% 11% 14% 7% 5% -6% 1% 0% Simple Average 3% 3% 5% 12% 28% 20% 19% 13% 10% -3% 2% 0% -7% -2% 3% 8% 13% 18% 23% 28% 33% Improvement(%) Improvement (Skill Score ) of Ensemble Forecasts over: Best. Model Simple Average RESULTS and EVALUATION
- 20. 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 Importance of Features, October Weather Features ImportanceScore 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 0.2 0.4 0.6 0.8 1 1.2 1.4 Importance of Features, March Weather Features ImportanceScore 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Importance of outcomes, October Models’ Outocmes ImportanceScore 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Importance of outcomes, March Models’ Outocmes ImportanceScore The estimation of weather features importance by Random Forest (a) October (b) March The estimation of models’ outcomes importance by Random Forest (a) October (b) March RESULTS and EVALUATION
- 21. 21 The standard deviation and the correlation of different models 0.950 0.955 0.960 0.965 0.970 0.975 0.980 0.985 0.990 0.995 1.000 0.000 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 0.018 Correlation Std.Dev. Std. Dev. and Correlation of Models Std.Dev. Correlation The performance of the ensemble forecasts is better with models of higher variance (higher std. dev.) or less correlated models. RESULTS and EVALUATION -7% -2% 3% 8% 13% 18% 23% 28% 33% Improvement(%) Improvement (Skill Score) of Ensemble Forecasts over: Best. Model Simple Average
- 22. 22 CONCLUSION • Combining the forecasts yields accurate forecasts and a stable performance; • The ensemble learning is efficient for assigning the weights of combined forecasts; • The combined forecasts by the simple average are not as accurate as by ensemble learning; • The combined forecasts out of diverse models are more accurate; • Adding the past generated forecasts increases the accuracy of the combined forecasts.
- 23. 23 Thanks for Listening Any Question? http://epic.uncc.edu/ Energy Production and Infrastructure Center Department of Electrical and Computer Engineering University of North Carolina at Charlotte Mohamed Abuella [email protected]