Mid Term Daily Load Forecasting using ARIMA, Wavelet-ARIMA and Machine Learning

The load forecasting is one of the important topics of discussion while studying the power system. The importance of load forecasting becomes even more important with the expansion of the horizon of the power system operation, may it be with the changing climatic patterns or increasing uncertain renewable penetration in the grid. This paper takes up the midterm daily load forecasting using the three models, namely, ARIMA, Wavelet-ARIMA and Machine Learning. The first two models are time series based, while the third one is the application of the Artificial Intelligence and combines the past data along with the climatic patterns. The Wavelet decomposition is performed in order to show the effect of decomposition on the forecasting of time series and to check the performance of the different discrete wavelets. The machine learning uses averaging and boosting ensemble methods in order to combine the single regression techniques. The results show that the performance of machine leaning algorithm was found to be better than the time-series algorithms, thus providing the idea that inclusion of climatic factors is very important in the load forecasting models. Moreover, the averaging type of ensemble models were found to perform better for most of the months than the boosting type. Thus, the best forecasting accuracy is obtained by combining the climatic patterns in the study as well as using the ensemble models.