神经网络自适应动态规划在污水处理过程控制中的应用

摘要

Considering the control of dissolved oxygen and nitrate nitrogen concentration in wastewater treatment plant (WWTP) , a neural network-based adaptive dynamical programming (NNADP) scheme was proposed , which employs neural network to approximate the existing scheme' s evaluation function and the optimal control policy, and makes use of gradient descent-based online algorithm to train the weights of neural networks. Testing NNADP' s control performance with BSM1 (Benchmark Simulation Model No. 1) indicates that NNADP has strong decoupling ability and higher control precision in comparison with PID.%针对污水处理过程的溶解氧及硝态氮浓度控制问题,提出一种基于神经网络的自适应动态规划(Neural network-based adaptive dynamical programming,NNADP)方法.该方法采用神经网络逼近当前策略的评价函数以及最优的控制策略.采用梯度下降算法对各神经网络权值进行在线训练.基于污水处理过程国际标准模型BSM1(Benchmark Simulation Model no.1)对NNADP控制性能进行了测试,结果表明:与PID控制相比,NNADP具有较强的解耦能力,控制精度也有较大提高.