Ensemble Technique-Based Short-Term Supply and Demand Forecasting with Features Selection Approach in Decentralized Energy Systems

Article Ensemble Technique-Based Short-Term Supply and Demand Forecasting with Features Selection Approach in Decentralized Energy Systems Adugna Gebrie Jember 1, Ruiyu Bao 1, Zijia Yao 1, Zhao Wang 1,*, Zhenyu Zhou 1 and Xiaoyan Wang 2 1 State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources, North China Electric Power University, Beijing 102206, China 2 Graduate School of Science and Engineering, Ibaraki University, Mito 310-8512, Japan * Correspondence: [email protected] Received: 21 August 2024; Revised: 2 December 2024; Accepted: 2 December 2024; Published: 31 December 2024 Abstract: Decentralized energy systems (DESs) present a paradigm shift toward more sustainable and resilient electricity networks with the increasing integration of renewable energy sources. Accurately forecasting energy supply and electricity demand is crucial for the efficient operation of DES. This paper focuses on accurate short-term energy supply and demand forecasting using machine learning (ML) algorithms across different seasons within small-scale DES, including photovoltaic (PV) generation, wind generation, and load demand. Initially, we develop multiple base model ML algorithms including long short-term memory (LSTM), convolutional neural networks (CNN), eXtreme gradient boosting (XGBoost), and recurrent neural networks (RNN) with feature selection approaches to improve forecasting accuracy and reduce model complexity. These base models leverage key temporal and spatial features and seasonal variations process to improve the model forecasting accuracy and reduce overfitting across different seasons. To further reduce forecast errors, we propose robust ensemble forecasting techniques including simple averaging (SA), weighted averaging (WA), and Stacking. In ML algorithms, the ensemble forecasting technique combines the multiple base models’ forecasts to produce a more accurate and robust final forecast by leveraging the strengths and compensating for the weaknesses of individual base models. Finally, numerical simulations are conducted using Python, Keras, and TensorFlow libraries to develop, train, evaluate, and validate the effectiveness and accuracy of the developed forecasting models and the proposed ensemble forecasting techniques. The results demonstrate that the proposed approach offers a robust solution for short-term supply and demand forecasting problems in DES. The work is both novel and effective from the perspectives of application, ML algorithms combination, and performance improvement.