Multi-agent cooperative control system and method for process industry

The invention provides a multi-agent cooperative control system and method for the process industry. A neutral network model from state values to observation values is arranged for feature extraction,important features of data are better extracted, and processing of mass data or even high-dimensional data can be adapted to. Then the minimax deep deterministic policy gradient algorithm is adoptedfor learning, wherein actor and critic networks in the minimax deep deterministic policy gradient algorithm conduct learning from a new neural network instead of conducting learning from initial observation. By means of the method for conducting state representation learning by utilizing the neutral network, the network can well capture features, and the adaptability to data is higher. 本公开提出了一种用于流程工业的多智能体协同控制系统及方法，设置进行特征提取的状态值到观测值的神经网络模型，更好的提取数据的重要特征，从而可以适应海量数据，甚至是高维数据的处理。然后采用极大极小深度确定性策略梯度算法进行学习，极大极小深度确定性策略梯度算法中的actor和critic网络从新的神经网络中学习，而不是从最初的观察中学习。这种利用神经网络进行状态表示学习的方法，使网络本身能够很好地捕捉特征，对数据的适应性更强。