Multivariate machine learning algorithms for energy demand forecasting and load behavior analysis

doi:10.1016/j.ecmx.2025.100903

Return

Multivariate machine learning algorithms for energy demand forecasting and load behavior analysis

2025-04-01

Farhan Hussain

M. Hasanuzzaman *

Nasrudin Abd Rahim

DOI：10.1016/j.ecmx.2025.100903

Original

Save

View PDF

Abstract

En 中文

This article presents deep learning frameworks for predicting electricity demand in the Western region of Bangladesh, utilizing Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS) algorithms. Critical hyperparameters, such as the number of hidden neurons in ANN and the number of input membership functions in ANFIS, are optimized to effectively capture uncertainties in load patterns. Short-term load forecasts from September 2023 to February 2024 incorporate meteorological and economic factors, achieving a low average MAPE of 1.72 % with the ANN network having sixteen hidden neurons, supporting dynamic grid operations. For strategic planning, medium- and long-term electricity projections integrate ANN and ANFIS with an econometric approach aligned with national policy, providing area-wide developmentconsistent forecasts. The annual peak load demand in Khulna and Rajshahi is projected to account for 70 % of the total load in the Western region, reaching 11.5 GW and 10 GW, respectively, by 2050. In contrast, Barisal is expected to have the lowest demand, requiring only 3 GW by 2050. Seasonal trend analyses indicate peak electricity demand during summer, while autumn exhibits greatest vulnerability in prediction accuracy. Forecast performance improves during winter, with average RMSE and MAPE reductions of 14.61 MW and 1.53 %, respectively, attributed to consistently lower load demand. On non-working days, energy demand decreases as economic activity slows, with total reductions averaging 71 MW more across the Khulna, Rajshahi, and Rangpur areas combined, owing to their higher concentration of industrial and commercial activities.

Keywords:

Energy demand forecasting

Adaptive neuro-fuzzy inference system

Artificial neural network

Levenberg-marquardt backpropagation

Sugeno fuzzy inference system

Artificial intelligence

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.