Deep Learning Based Real-Time Approach for Optimal Power Flow in Unbalanced Distribution Systems with High-Level EV Penetration

With the emergent and rapidly increasing de-mands for transport decarbonization onto the grid for mobility and freight. This becomes an increasing stochastic optimization challenge for grid network operators. Traditional optimization methods struggle to address the real-time requirements of these highly dynamic systems, particularly under ultra-rapid electric vehicle (EV) charging scenarios. To tackle these issues, this paper proposes a high-resolution optimization framework leveraging a Deep Neural Network (DNN)-based approach for solving the AC Optimal Power Flow (AC-OPF) problem in three-phase unbalanced power distribution networks. The proposed framework incorporates practical EV charging load models and Soft Open Points (SOPs) to enhance flexibility and stability in network operations. By decomposing the optimization problem into quasi-real-time intervals, the method enables rapid computation of control actions while ensuring operational constraints are met. Case studies conducted on the IEEE 13-node system, show a reduction in computation time to less than 1 second for ultra-rapid charging scenarios, compared to over 67 seconds using traditional methods. Additionally, network losses are reduced by up to 52% with the integration of SOPs, with voltage profiles maintained within safe operational limits, even under 90% EV penetration level. These findings highlight the proposed framework's potential to improve the reliability and efficiency of future power distribution networks.