Machine learning-based prediction of power demand and fuel consumption of a power plant: A case study from Bangladesh

Abstract

This study aims to address the issue of power and fuel shortages in developing economies like Bangladesh, where, despite having sufficient capacity to generate more electricity than needed, the country often faces challenges due to limited fuel availability. To mitigate this problem, the study proposes a predictive model that enables power plants to accurately estimate the required power output at specific times, along with the corresponding fuel needs. Unlike models that rely on extensive sensor networks, this study develops a solution that remains effective under sparse instrumentation, making it suitable for low-resource environments. Machine learning (ML) models were applied to predict power demand and fuel consumption (FC) for a 150 MW heavy fuel oil (HFO) power plant. The research utilized 7 years of operational data (2017–2023) and evaluated the performance of ML algorithms, including K-nearest neighbors (KNNs), artificial neural networks (ANNs), and gradient-boosted regression trees (GBRTs). Power demand was predicted based on 6 input parameters: working hours, FC, auxiliary consumption, atmospheric temperature, relative humidity, and atmospheric pressure. The GBRT algorithm outperformed the others, achieving the highest accuracy with a coefficient of determination (R²) of 0.9994 and a root mean square error (RMSE) of 1.102. The findings highlight the potential of ML in enhancing energy management, with the GBRT model offering precise predictions that can support proactive fuel procurement strategies and help mitigate energy shortages.