Comparative Analysis of Deep and Convolutional Neural Networks for Short-Term Load Forecasting

Abstract

This study in order to provide effective power system operations, this study compares Deep Neural Networks (DNN) with Convolutional Neural Networks (CNN) for short-term load forecasting (STLF). Traditional load forecasting methods, Regression and time series models, for example, frequently fall short in addressing the intricate, nonlinear patterns present in power demand data. AI-driven models, particularly DNN and CNN, offer advanced capabilities in capturing intricate patterns within large-scale, high-dimensional data. In this research, both DNN and CNN models were trained using historical load data with an emphasis on accuracy metrics, including Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). DNN's flexibility allows it to model complex nonlinear relationships, while CNN's design enables it to capture temporal dependencies in time-series data effectively. Validation results reveal that CNN achieves superior forecasting accuracy with the lowest error values, registering an RMSE of 1.2, MSE of 1.44, and MAE of 0.77, thus demonstrating a competitive advantage over DNN in short-term load forecasting. This comparative analysis underscores the effectiveness of AI techniques in load forecasting, with CNN showing notable strengths in sequential data modelling. The findings contribute to the ongoing advancement of AI in energy management, providing utility providers with reliable tools for optimizing grid operations and decision-making.