Short- and long-term electricity load forecasting using classical and neural network based approach: A case study for the Philippines

摘要

Higher quality of power service is needed in order to sustain the increasing economic activities of a growing country such as the Philippines. An important step in achieving such goal is to have a good forecasting model that could accurately model the load behavior that must be met at all cost in order to have a secure and optimal electrical power system. This paper compared the performance of Holt-Winters' method and neural network both in short and long term Philippine electricity demand forecasting. The results show that although all methods can model the data under study well, the Holt-Winters' method yielded the most promising results both on short and long term forecasting with a mean absolute percentage error(MAPE) of about 9% and 3% respectively. A simple combination of the two models was also made and tested for hour-ahead load projection. The corresponding mean absolute percentage error (MAPE), mean absolute error (MAE) and root mean square error (RMSE) were measured and served as the performance metrics for all methods considered in this paper.