Hybrid Machine Learning–Time Series Forecasting Framework for Energy Demand Prediction under Weather Variations

Abstract

Precise energy consumption forecasting is essential in the rapidly evolving electric power market that is adjusting to a future without regulation and with intense competition. Establishing the Scene: Utilities, energy dealers, and system operators have come to rely more and more on short-term forecasts due to the deregulation of energy markets, while accurate demand projections have always been crucial for efficient power system planning and operation. To tackle this issue, this study investigates Energy Demand Prediction in the presence of weather changes by the application of data smoothing techniques, specifically the Savitzky-Golay filter and Gaussian kernel density estimation, both of which are optimised with PSO. A new hybrid BiLSTM-FCN design that takes use of feature extraction and temporal dependencies is also suggested after extensive investigation into sophisticated deep learning models like BiLSTM and FCN. Findings: The suggested models surpass the current state-of-the-art methods in experimental evaluation, which includes testing for classification error and comparison with other top-tier approaches. Finally, some thoughts: The BiLSTM-FCN model showcases its potential as a strong instrument for efficient power system planning and management under weather-induced changes with performance metrics of RMSE 2.50, MAE 2.10, and MSE 6.24. It is extremely effective for accurate energy demand forecasts.

https://doi.org/10.65470/james.3