A Hybrid Forecasting and Optimization Framework for Electric Vehicle Scheduling in Virtual Power Plants

Abstract

The present study proposes a hybrid energy management (EM) framework for vehicle‐to‐grid (V2G) enabled virtual power plants (VPPs), integrating Green Anaconda Optimization (GAO) with a Pyramidal Dilation Attention Convolutional Neural Network (PDACNN) for cost‐efficient and adaptive scheduling. The PDACNN module forecasts renewable energy (RE) generation and load demand using multiscale feature extraction and attention mechanisms, while GAO optimizes EV charging (EVC)/discharging schedules under grid constraints. Two scenarios are evaluated: with and without IoT‐enabled coordination. Simulation results show that the proposed GAO‐PDACNN method reduces the total operating cost (OC) to $499 626/year, which is 1.7% lower than MOPSO, 14.2% lower than the GA, and 34.6% lower than the HPOA. Further, CO 2 emissions are lowered by 9.0 × 10 6 tons/year, which is 53.8% less than conventional strategies using HPOA. The model also has a load cover accuracy of 92.4%, and a convergence in only 62 iterations, which is more efficient and stable than other benchmark models. This is to validate the efficiency of the framework to facilitate intelligent coordination in V2G for smart grid (SG) with higher RE penetration.