Empowering Industrial Energy Management: Advancing Short-Term Load Forecasting with LSTM and CNN Deep Learning Models - Insights from a Moroccan Case Study

Abstract

Self-consumption of electricity plays an important role in the energy transition and using green, sustainable energy sources for industrial self-sufficiency and electricity bills, meeting part of their own energy needs and even generating 20% of the annual surplus that could be sold to the grid. This study aims to forecast short-term load spanning between 2022 and 2023, employing deep learning models, specifically Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. Various performance metrics have been used to evaluate and compare the accuracy of these two models, including mean squared error, root mean squared error, mean absolute error, and mean absolute deviation. Results reveal LSTM's superior performance over CNN, with LSTM demonstrating adeptness in capturing underlying patterns while CNN tends to learn noise, leading to a divergence in performance metrics between training and validation data. The findings underscore the significance of LSTM for accurate load forecasting and suggest the inclusion of additional hyperparameter optimization to enhance the reliability of short-term load predictions, distinct from previous studies.