Hybrid LSTM-Graph Convolutional Neural Network with Wavelet Transform and Correlation Analysis for Electrical Demand Forecasting

Accurate electrical demand forecasting is essential for power system efficiency, renewable energy investment, and cost-effective electricity production. For electrical demand consumption time series forecasting, this article proposes a novel deep learning architecture, wavelet transform and correlation-based hybrid LSTM-GCNN, that integrates long short-term memory (LSTM) and graph convolutional neural network (GCNN) layers. A GCNN captures dynamically distributed features and temporal correlations from graph data generated by wavelet decomposition and correlation analysis. The temporal patterns of the electrical demand consumption time series are captured by an LSTM. The proposed hybrid LSTM-GCNN architecture is evaluated using Indian Northern Regional Load Despatch Centre (NRLDC) electrical demand consumption data from 2018–2021 with a 15-min resolution of states Uttar Pradesh (U.P.) and Jammu and Kashmir (J &K). Hybrid LSTM-GCNN outperforms ARIMA, LSTM-univariate, LSTM-convolutional neural network and LSTM-multivariate prediction algorithms in universality, reliability, and accuracy. The proposed hybrid LSTM-GCNN architecture offers an efficient and promising method for forecasting time series of electrical demand consumption.