To effectively control the output voltage of solid oxide fuel cells (SOFCs) and improve the operating efficiency of SOFC systems, an SOFC output voltage data-driven controller based on multi-agent large-scale deep reinforcement learning is proposed, whereby a discrete–continuous hybrid action space large-scale multi-agent twin delayed deep deterministic policy gradient (DHASL-MATD3) is used as the control algorithm for this controller. To solve the low robustness problem of deep reinforcement learning-based conventional controllers, this algorithm adopts a hybrid action space multi-agent policy that achieves parallel exploration by using double deep Q-learning (DDQN) agents with discrete space and deep deterministic policy gradient (DDPG) agents with continuous action space, thus improving exploration efficiency and realizing excellent robustness. In addition, many techniques are adopted by this algorithm to solve the problem of Q-value overestimation. Ultimately, an SOFC output voltage controller with stronger robustness is obtained. Simulation results show that this controller can effectively control the output voltage of a SOFC by regulating the fuel flux and maintaining its fuel utilization within a reasonable range.

为了有效控制固体氧化物燃料电池（SOFCs）的输出电压并提高SOFC系统的运行效率，提出了一种基于多智能体大规模深度强化学习的SOFC输出电压数据驱动控制器，其中离散-连续混合动作空间大规模多智能体孪生延迟深度确定性策略梯度（DHASL-MATD3）用作该控制器的控制算法。为了解决基于深度强化学习的传统控制器鲁棒性低的问题，该算法采用混合动作空间多智能体策略，通过使用具有离散空间和深度确定性策略梯度的双深度 Q 学习（DDQN）智能体实现并行探索（ DDPG）具有连续动作空间的代理，从而提高探索效率并实现出色的鲁棒性。此外，该算法采用了许多技术来解决Q值高估的问题。最终，获得了具有更强鲁棒性的 SOFC 输出电压控制器。仿真结果表明，该控制器通过调节燃料流量，将燃料利用率保持在合理范围内，可以有效地控制SOFC的输出电压。