A Back Propagation Based Optimized Regression Learning Algorithm for Load Estimation in Power Systems

The strengths of soft computing and machine learning based algorithms are being harnessed globally to improve and optimize power system performance. One such application happens to be load forecasting in power systems, with the objective of minimizing the difference between demand and generation of electrical power. However, it is a nontrivial challenge as electrical load exhibits high volatility owing to several variables such as date, time, time of the year, socio economic factors, atmospheric conditions etc. Hence it is extremely difficult to estimate future trends in electrical load with high accuracy. Previous, conventional statistical techniques have been used in conjugation with existing load curves to estimate future load trends. However with emerging machine learning and deep learning algorithms rendering high accuracy of pattern recognition, forecasting electrical load employing machine learning and deep learning algorithms has become an active area of research. This paper presents a combination of data preprocessing and back propagation based optimized regression learning for electrical load forecasting. It has been shown that the proposed approach outperforms existing baseline approaches in the domain and attains a mean absolute percentage error (MAPE) of only 1.3% and an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}{}^{\wedge}\boldsymbol{2}$</tex> value of 0.97.