A novel optimized machine learning algorithm approach for solar-powered xEV charging stations

• Published in: Multiscale and Multidisciplinary Modeling, Experiments and Design Vol. 8; no. 10; p. 426
• Main Authors: Rajender, Jatoth, Dubey, Manisha, Kumar, Yogendra, Srinivasulu, Gumpu
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.11.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Alternative energy sources; Artificial intelligence; Blockchain; Characterization and Evaluation of Materials; Control algorithms; Electric power demand; Electric vehicle charging; Electric vehicles; Electricity; Energy demand; Energy management; Engineering; Forecasting; Infrastructure; Machine learning; Mathematical Applications in the Physical Sciences; Mechanical Engineering; Neural networks; Numerical and Computational Physics; Optimization; Original Paper; Polynomials; Renewable energy; Renewable resources; Simulation; Solar energy; Solid Mechanics; Space vector modulation; Systems stability; Solar photovoltaic (S-PV); Rosenbrock function based sea-horse optimization (RF-SHO); Direct current (DC); Charging system (CS); Generalized space vector modulation based switching regulators (GSVM-SR); Alternative current (AC); Electric vehicles (EV)
• ISSN: 2520-8160; 2520-8179
• DOI: 10.1007/s41939-025-01014-8

The rapid adoption of Electric Vehicles (EVs) has significantly increased energy demand, necessitating efficient and stable charging infrastructure powered by renewable energy. This paper proposes a novel, season-aware forecasting and optimization framework for solar-powered EV charging stations using Joint Fusion Layer–Bidirectional Long Short-Term Memory (JFL-BiLSTM) and Rosenbrock Function-based Sea-Horse Optimization (RF-SHO). The proposed system smooths DC power using Laguerre Polynomial-based Ramp Rate Method (LP-RRM) and manages grid interruptions through Generalized Space Vector Modulation-based Switching Regulators (GSVM-SR). Seasonal classification and feature extraction are performed on a real-world dataset, and the JFL-BiLSTM model forecasts EV charging demand with 98.61% accuracy, 98.01% precision, 98.95% recall, 98.02% specificity, and 98.10% F-measure. The RF-SHO algorithm selects optimal charging stations within 7815 ms, outperforming baseline algorithms (e.g., SHO, SSO, ACO, EHO) in convergence speed and selection accuracy. Simulation results demonstrate significant improvements in load prediction, power smoothing, ripple reduction, and system stability. The proposed framework proves to be an effective and intelligent solution for dynamic EV charging management under seasonal and operational uncertainties, paving the way for robust, user-centric charging infrastructure.