Short-term probabilistic building load forecasting based on feature integrated artificial intelligent approach

Abstract

Due to various influential factors that lead to instability and volatility of the building load, short-term building load forecasting is a gruelling task. This paper proposes a hybrid short-term building load probability density forecasting method based on Orthogonal Maximum Correlation Coefficient (OMCC) feature selection and Convolutional Gated Recurrent Unit (CGRU) quantile regression. Firstly, the optimal feature set is selected by OMCC. Then Value-At-Risk (VAR) is determined from fitting Copula model to construct indicator variables. Next, the data from the feature selection stage is used as input to the quantile regression model of CGRU for building load forecasting. Finally, the building load probability density distribution is fitted by kernel density estimation. The forecasting performance is evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Arctan Absolute Percentage Error (MAAPE). Simulation results across all three buildings validate the reliability of the proposed model for the short-term building-level probabilistic load forecasting tasks.