Abstract

The diversity of machine learning algorithms influences power load forecasting in energy sectors. The forecasting of power load decides the policymaking of power generation and distributions. The forecasting of power load depends on various factors of non-linear characteristics of data. Recently several single and multiple predictive models have been developed. Machine learning algorithms and the artificial neural network have played a significant in forecasting in the current decade. This paper study of machine learning algorithms such as support vector machines, LSTM, ensemble classifiers, recurrent neural networks and deep learning. The various authors suggest that the combination of RNN and LSTM outperforms forecasting of power load instead of existing machine learning algorithms. The deep learning results also surprise the accuracy of forecasting. Deep learning processing is very complex instead of RNN and LSTM networks. However, the combined task of extracting attributes and classification has several advantages over specific limitations. This paper extensively analyses short-term power load forecasting using Chandīgarh UT electricity consummation for the last five years. The accuracy of forecasting estimates as RMSE, NMSE, MAE and MI. these parameters results suggest the pros and cons of machine learning algorithms. The hyper-parameters of the algorithm decide as per data transformation of energy consumption. The experimental validation uses MATLAB 2017 software.

Keywords: Forecasting, Power Load, Machine Learning, SVM, LSTM, RNN, Deep Learning, EM