Enhanced grid integration through machine-learning optimized bidirectional EV chargers

Pulkit Kumar, Harpreet Kaur Channi, Sita Rani, Aman Kataria, Punam Rattan,
Enhanced grid integration through machine-learning optimized bidirectional EV chargers,
Sustainable Energy, Grids and Networks,
Volume 44,
2025,
102007,
ISSN 2352-4677,
https://doi.org/10.1016/j.segan.2025.102007.
(https://www.sciencedirect.com/science/article/pii/S2352467725003893)
Abstract: Large-scale electric vehicles (EVs) implementation depends on reliable and stable bidirectional charging that can be effectively implemented, efficient, and practical. This paper proposes a Bidirectional Learning-based Electric Vehicle Charger (BLEVC) with Battery Energy Storage (BES) that boosts grid stability during the Grid-to-Vehicle (G2V) and the Vehicle-to-Grid (V2G) operation modes. The innovation is the methodical comparison of 3 machine learning (ML) controllers: Dynamic Time Reversal (DTR), Recurrent Neural Network (RNN), and Support Vector Machine (SVM) with the traditional Proportional-Integral (PI) controller under the same testing parameters. Findings indicate the definite merits of ML strategies. RNN lowered G2V charging time (PI 35 mins to 8 mins) and voltage ripple at 48 G2V, and DTR showed a stable steady state response, although its computational requirement was high. On the other hand, SVM had infinite settling time and large ripple time; poor evidence of using it on dynamic duty-cycle regulation. In V2G mode, RNN and DTR have quicker and more constant energy dispatch than PI. Integration of the BES enhanced peak shaving 22 % and could smooth the state of charge to within 5 %, confirming its usefulness in grid support and demand reshaping. This contributed work offers a validated architecture of BLEVC and a comparative framework, which is a gap in the literature. Future work directions will be typhoon hardware in-loop (HIL)-hybrid ML-PI controllers, hyperparameter optimization, and pilot-scale tests with utilities to enable secure, scalable EV-grid integration.
Keywords: Bidirectional EV chargers; Grid integration; Machine learning optimization; Electric vehicles; Control strategies; Vehicle-to-Grid (V2G); Grid-to-Vehicle (G2V)