Hybridization of GREY-PPO and ARIMA-GRNN Algorithms with Rolling Mechanism for the Prediction of Electrical Energy Consumption

This article deals with the prediction of energy consumption of certain loads whose behavior is difficult to predict using artificial intelligence tools.It examines the prediction of electricity consumption of a Southern Interconnected Grid (SIG) industrial load using the hybrid GM (1,1)-PPO model on the one hand, and that of the electricity consumption of a SIG household load using the hybrid ARIMA-GRNN model on the other hand.The input data taken by our GM (1,1) model is submitted to the Accumulated Generating Operator (AGO) and then to the Inverse Accumulated Generating Operator (IAGO) to determine the forecast values.Finally, a rolling mechanism is applied to enhance the performance of the GM (1,1) configuration.The hybridization of the GM (1,1)-PPO algorithm helps determine and optimize parameters a and b of the PPO.The results of the hybrid GM (1,1)-PPO model show high accuracy according to the LEWIS criteria: MAPE=3.8%against MAPE=11.41%for the GM (1,1) model alone.As for the second tool, the ARIMA model receives the input data, performs a regression and provides the predicted values from the generalized differential equation.Finally, given the random nature of the parameters p, d and q, the ARIMA model is combined with the Generalized Regression Neural Network (GRNN).The prediction results give good accuracy: MAPE = 9.33% versus MAPE = 92.5% for the ARIMA model alone.