Which control management strategies are best suited for EV charging stations? A comparison of price optimization and machine-learning approaches

The increasing adoption of electric vehicles (EVs) in urban areas poses overloading challenges to the power distribution grid. Therefore, previous research has focused on developing different smart charging architectures, strategies and models to mitigate potential threats in the electrification of the transportation sector. However, many existing studies focused on either optimization-based or machine learning (ML)-based approaches separately, often within centralized architectures. This paper presents a comparative assessment of smart charging control strategies in a distributed architecture, considering both optimization and ML-based models, as well as their combination. The proposed system operates within a receding horizon framework, ensuring adaptability for continuous real-life applications. A comprehensive set of scenarios is analyzed, varying upper-level control strategies, forecasting time horizons, energy allocation strategies, and feedback mechanism between upper- and lower-level distributed architecture control layers. The results show that for those systems where feedback from EVs cannot be implemented, the most suitable operation control strategy in terms of satisfying different involved parties is an ML prediction- driven approach. For the scenarios where feedback can be implemented, the optimal strategy is an optimization model with fixed energy allocation per day. All scenarios with feedback mechanism show more than 95% delivery of energy requests, while for the scenarios without feedback, the best delivery is only 79%. The obtained results can help charging point operators (CPOs), EV users and distribution system operators (DSOs) to analyse EV charging strategies and choose the operational approach based on charging limitations and possibilities.