机械

为解决泵系统节能控制优化过程中多设备协同控制的问题，提出一种基于深度Q网络（Deep Q-Network,DQN）和深度确定性策略梯度（Deep Deterministic Policy Gradient,DDPG）算法的多智能体强化学习泵系统节能控制优化策略。将泵送系统构建为马尔可夫决策过程，采用DQN算法构建泵启停离散动作空间，DDPG算法构建电机转速连续动作空间，并在DQN和DDPG算法中嵌入长短期记忆网络（Long Short-Term Memory,LSTM）用于增强记忆历史运行数据能力，提高智能体训练和控制性能。实验结果表明，基于多智能体强化学习控制的泵系统较人工调控节能15.81%，具有较好的节能控制效果。

To address the issue of multi-equipment cooperative control in the energy-saving optimization process of pump systems, this paper proposes a multiple-agent reinforcement learning energy-saving control optimization strategy for pump systems based on the Deep Q-Network(DQN) and Deep Deterministic Policy Gradient(DDPG) algorithms. The pump system is modeled as a Markov Decision Process(MDP), where the DQN algorithm is employed to construct the discrete action space for pump start/stop operations, and the DDPG algorithm is used to build the continuous action space for motor speed control. Additionally, Long Short-Term Memory(LSTM) networks are embedded into both the DQN and DDPG algorithms to memorize historical operational data, thereby enhancing agent training and control performance. Experimental results demonstrate that the pump system controlled by the multi-agent reinforcement learning approach achieves a 15.81% energy saving compared to manual regulation, exhibiting superior energy-saving control effectiveness.