Akademik Veri Yönetim Sistemi
Electrical Engineering, cilt.108, sa.3, 2026 (SCI-Expanded, Scopus)
This paper presents a machine learning (ML)-based solution proposal model for electric vehicle (EV) energy flow management for a vehicle-to-grid (V2G) topology in residential microgrids. In contrast to conventional methods that depend exclusively on the status of charger connections, this innovative approach entails the prediction of which electric vehicles (EVs) are most likely to maintain a connection during grid faults, thereby ensuring power quality in islanded mode. A novel, behavior-based dataset was developed by recording the parking in/out timestamps of 20 fuel-based vehicle users over a one-year period between 01/01/2022 and 12/31/2022. This dataset simulates the behavior of future vehicles when they are EVs. When power problems arise in the microgrid, the aim is to maintain the power quality of the microgrid until the power problem is solved by supporting the microgrid with the batteries of some of the vehicles in the parking lot. In V2G topology, determining which of the 20 EVs are in the parking lot until the main grid problem is resolved has become a classification problem with this novel dataset. Twenty-two ML methods and three deep learning (DL) methods were employed to address the classification problem, with the objective being to predict the vehicles with the highest probability of maintaining connectivity with the microgrid until the grid’s issue is resolved. Bagged tree is applied to the V2G topology for the first time. A mathematically formulated vehicle selection algorithm that considers dynamic grid conditions and a real-time deployable ML-based V2G energy dispatch model is presented.