Short-Term and Medium-Term Load Forecasting Using Principal Component Analysis (PCA)-Based Recurrent Neural Network
Abstract
Electricity is one of the most significant discoveries of man. At present, everything relies on electricity and as development continues, demand also increases requiring a need for a reliable supply of power. With this, load forecast had always played an essential role in ensuring the adequacy of electricity to consumers, thus having a good load forecasting model is of utmost importance. This paper presents a short-term and medium-term load forecasting model by combining Principal Component Analysis (PCA) and Recurrent Neural Network to forecast the hourly and daily peak demand of the Luzon Grid. The ability of the PCA to transform data into principal components made the visualization and analysis, as well as deciding the number of variables to retain easier. The use of Mahalanobis Distance made it possible to filter and exclude outliers in the dataset enabling the model make more accurate forecasts. The use of Gated Recurrent Neural Network produced excellent predictions as compared to the traditional Neural Network. For STLF, the proposed PCA-RNN model was able to forecast the hourly demand with a MAPE of 0.72159%, FAR of 97.2845% and a MSE of 0.00749; while for the MTLF, the proposed model was able to predict the daily peak demand with MAPE of 0.8250% and MSE of 0.039642.