A Short-Term Load Forecasting Technique Using Extreme Gradient Boosting Algorithm

In the framework of this study, an effective extreme gradient boosting algorithm is implemented for the prediction of short term electric load. It contributes significantly in scheduling of power plant, spinning reserve and unit commitment. It also helps the energy providers by giving the information of the future electrical energy demand. In recent days' numerous methodology such as artificial neural network, deep learning methodology, support vector regression machine algorithm is mostly used for the prediction of short term load. But the existing approaches may not give accurate prediction of load. That's why, to deal with the precise short term load forecasting (STLF), it is motivated to develop an extreme gradient boosting algorithm (XGBoost) in this research. The developed methodology was validated using historical load data set of Bangladesh Power System (BPS). The overall performance of the developed algorithm is compared with the existing conventional approaches in terms of forecasting errors. It is seen that outcomes obtained using XGBoost algorithm in STLF is more stable and precise with respect to conventional approaches such as linear regression method, long short term memory network (LSTM), radial flow neural network (RBFN).