2019 年 40 巻 5 号 p. 144-153
Short-term load forecasting (STLF) using artificial intelligence (AI) methods has become an active and important research topic with the recent rapid advancement in data processing capabilities of computers．In this paper, we attempted to predict next-day hourly electric loads in Chubu area of Japan using artificial neural networks (ANNs). The model used in this study exploits principal component analysis (PCA) and the selective ensembling (SE) method to achieve a good predictive performance. The effects of changing several hyperparameters, such as the number of hidden layers and neurons, as well as the selection of the input data and the activation function, were tested on the load data obtained from the website of Chubu Electric Power Company. We obtained the following results:
1) With the optimal model configuration, mean absolute percentage error (MAPE) was reduced by 0.19 percentage points,from2.32% to 2.13%.
2) Using relative humidity forecasts as part of the input data to ANN reduced MAPE by 0.34 and 0.19 percentage points in Augustand September, respectively.
