A Bidirectional Vehicle-to-Grid Controller for DC Microgrid EV Charging Network Stations

This paper suggests an effective hybrid approach that combines Multi-fidelity Deep Neural Networks (MFDNN) with Clouded Leopard Optimization (CLO) for bidirectional energy transfer between Electric Vehicles (EVs) and the grid in DC Microgrids (MGs) with integrated EV Charging Stations (EVCS). The solution aims to reduce the issues associated with high operational expenses, charging duration, and emissions when grid power is used during periods of insufficient photovoltaic (PV) power or high load. The CLO-MFDNN method optimizes energy exchange schedules and resource utilization, ensuring smooth operations and efficient energy distribution. MFDNN helps in forecasting energy demand patterns and renewable energy (RE) generation, thereby supporting proactive decision making for efficient EVCS management. The suggested technique is compared with existing methods such as Assailant Inspired Chimp Optimization Algorithm (AIChOa), Enhanced Multi-Agent Neural Network (EMANN), Dung Beetle Optimizer-Binarized Spiking Neural Networks (DBO-BS4NN), Artificial Neural Network-Particle Swarm Optimization (ANN-PSO), and Multi-Agent Deep Neural Network (MADNN) in MATLAB. The results indicate that the CLO-MFDNN approach achieves emissions of 60.5 kg CO2, a charging time of 34 min, and an operational cost of $1506 under identical operating conditions. These values are better than the benchmark methods regarding environmental impact, operational efficiency, and cost management for EV charging in DC MGs.