Short-Term Probability Density Function Forecasting of Industrial Loads Based on ConvLSTM-MDN

Abstract

Load forecasting for industrial customers is essential for reliable operation decisions in the electric power industry. However, most of the load forecasting literature has been focused on deterministic load forecasting (DLF) without considering information on the uncertainty of industrial load. This article proposes a probabilistic density load forecasting model comprising convolutional long short-term memory (ConvLSTM) and a mixture density network (MDN). First, a sliding window strategy is adopted to convert one-dimensional (1D) data into two-dimensional (2D) matrices to reconstruct input features. Then the ConvLSTM is utilized to capture the deep information of the input features. At last, the mixture density network capable of directly predicting probability density functions of loads is adopted. Experimental results on the load datasets of three different industries show the accuracy and reliability of the proposed method.

Files

Additional details