Up to year 2020 load forecasting using neural nets

Abstract

Prediction of peak electric loads in Japan up to year 2020 is discussed using the artificial neural networks (ANNs). In this study, total system load forecast reflecting current and future trends is carried out for nine power companies in Japan. Two ANNs, a three-layered back-propagation and a recurrent neural network, were designed and tested for the purpose. Predictions were done for target years 1999, 2000, 2005, 2010, 2015, and 2020, respectively. Two case studies, preservation of the status and structure reform, were also tested for predicting the loads of years 2010 and 2020. Unlike short-term load forecasting, long-term load forecasting is mainly affected by economical factors rather than weather conditions. This study focuses on economical data that seem to influence long-term electric load demands. Here, 10 factors are selected as inputs for the proposed ANNs: (1) gross national product, (2) gross domestic product, (3) population, (4) number of households, (5) number of air-conditioners, (6) amount of CO 2 pollution, (7) index of industrial production, (8) oil price, (9) energy consumption, and (10) electricity price. The data used are: actual yearly, incremental growth rate from the previous year, and both together (actual and incremental growth rate from the previous year). As a result, the demands for 2010 and 2020 are predicted to be 225.779 and 249.617 GW, respectively (preservation of the status). With structure reform, the demands for 2010 and 2020 are predicted to be 219.259 and 244.508 GW.