Enhancing Short-Term Load Forecasting using ES-dRNN with Context Vector

Abstract

Electrical load forecasting is an integral part of power system planning, operation, and control. Accurate load forecasting is beneficial for making various operational decisions such as energy generation, reliability analysis, and dispatch scheduling of generated energy. However, short-term load fore-casting is difficult due to the complexity posed by the nature of load time series as it expresses multiple seasonality and nonlinear trend. In this paper, we propose an extension of a novel hybrid hierarchical deep learning-based forecast model which incorporates multiple seasonality. The original groundbreaking hybrid forecasting model is developed by Smyl. The model presented in this paper is based on a dilated recurrent neural network with a context vector by integrating exponential smoothing (EScdRNN). Exponential smoothing performs the adaptive time series processing whereas dilated recurrent neural network using context vector helps in cross-learning. This helps in the selection of useful input information which leads to improved accuracy. The results of the proposed methodology are compared with different statistical machine learning methods which show the potential of our proposed approach in terms of increased accuracy.