Hyper-Parameter Tuned Short Term Load Forecasting Using Stochastic Classifier-Regression Mapping For Power System Operator

Abstract

One of the important aspect in operation of power system is management of Intermittent parameters like load. Short Term Load Forecasting (STLF) is done in this manuscript using Classifier-Regression mapping. The historical data of the assumed load is classified into 5 bands. Bands of the data are predicted using classifier and exact value of the predicted band is forecasted using regression analysis. Three classifiers techniques viz Decision Trees, Ensemble and SVM(Support Vector Machine) and three Regression Techniques viz Tree, Neural Network and GPR (Gaussian Process Regression) are used. Nine different combinations are formed by mapping each classifiers with each regression. Data is trained using these nine techniques and the best method is determined using statistical tools. The best method is further optimized by tuning parameters with an objective of reducing MSE using three methods viz Bayesian, Grid Search and Random Search Optimization. The whole analyses is done on Python and MATLAB platforms. It is observed that best combination for STLF is obtained using Tree as classifier and Neural Networks for regression analysis with RMSE (Root Mean Square Error), R2, MSE(Mean Square Error), MAE(Mean Absolute Error) and training time of 0.083, 0.85, 0.0069, 0.058, 4.325sec respectively. Grid Search is observed as the best method for hyperparameter tuning with reduced MSE of 0.0064.