JoWUA

Abstract

Electricity consumption predictions for a long period are critical in the institutions that distribute the electricity and governmental or private entities that supply the electricity. It guarantees optimum energy utilization and aids in making strategic decisions for improving the energy production quality. This need is especially important in nations like Iraq, which has suffered from energy crises for many years. This study uses daily household electricity consumption data acquired from the Ministry of Electricity in Iraq, namely the Rusafa area of Baghdad, from 2022 to 2024. Weather data for the same years was also included, which contains external weather factors such as temperature, humidity, and solar radiation that directly influence consumption patterns. This paper proposes a hybrid forecasting model that utilizes advanced deep learning architectures LSTM and CNN-based deep learning architectures for forecasting along with an upgraded stacked hybrid model that employs CNN, GRU, Stacked Bi-LSTM, and machine learning regressors, such as XGBoost Regressor, and LightGBM Regressor. These models are being trained to improve accuracy in the forecast and to improve energy acoustic production strategies. The 30 epochs were trained and evaluated on the proposed model using the mean relative absolute error (MAPE) and mean root mean square error (RMSE) to examine the prediction quality. Among all models tested, the best performance was achieved using LightGBM regressor in our hybrid model with MAPE and RMSE of periodic forecasts for the next spilled of time being 0.185155 and 0.094603, respectively. The results show the potential of hybrid modeling techniques for energy forecasts and electricity distribution systems optimization.