As electric vehicle adoption accelerates globally, charging station operators face a complex optimization problem: setting prices that maximize revenue and utilization while minimizing stress on power distribution networks. Researchers have developed an advanced machine learning solution that automates this pricing challenge through a sequential game framework.
The proposed system treats charging operators as strategic leaders who adjust pricing, while EV users act as cost-conscious followers selecting stations based on fees, location, and queue length. Rather than using fixed tariffs, operators employ a soft actor-critic (SAC) algorithm—a type of reinforcement learning—to continuously refine prices based on real-time demand patterns and network conditions.
To improve training stability and convergence speed, the research team integrated three custom modules: a demand-aware experience pool that learns from diverse traffic scenarios, a critic-disagreement regularizer that reduces overfitting, and an adaptive update mechanism triggered by network constraints. These enhancements address common challenges in deploying machine learning to power systems, where instability during training could translate to poor real-world decisions.
Testing in Wuxi, China across five independent runs showed compelling results. Average revenue increased 45.7% for participating charging operators, while utilization improved and power distribution network imbalance metrics fell 18.4%. Charging times also decreased under baseline demand scenarios.
The framework's ability to coordinate multiple objectives—operator profit, equipment utilization, and grid health—makes it valuable for utilities managing rapid EV infrastructure expansion. Unlike static pricing models, the dynamic approach adapts to temporal variations in demand and grid conditions, potentially reducing infrastructure overinvestment while improving service quality.
As EV charging networks mature into critical grid infrastructure, AI-driven pricing could enable operators to self-optimize pricing without central coordination, improving overall system efficiency while maintaining network stability.
