Short-Term Load and Price Forecasting based on Improved Convolutional Neural Network

Abstract

This research work focuses on two major issues in smart grid, i.e., accurate load and price forecasting. While literature has focused on forecasting at national and industrial level, this paper focuses on short-term load and price forecasting in residential area. To achieve aforementioned objective, a model is proposed which consists of two stages; feature engineering and forecasting. Feature engineering includes feature extraction and feature selection to reduce dimensionality. Mutual Information (MI) is used for feature extraction. Random forest and Recursive Feature Elimination (RFE) is used for feature selection. For prediction of load and price, Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN) and Improved Convolutional Neural Network (ICNN) are used. However, ICNN beats MLP and CNN with 81% accuracy. Prediction performance is evaluated by Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Mean Squared Error (MSE). Further, UMASS data set is used for price and load prediction.