动态结构优化神经网络及其在溶解氧控制中的应用

Dynamic Structure Optimization Neural Network and Its Applications to Dissolved Oxygenic(DO) Control

摘要: 提出一种神经网络结构动态优化设计方法，其实质是通过判断隐含层神经元在学习过程中对输出的影响，利用竞争机制在信息处理过程中增加或删减隐含层神经元，实现神经网络结构的动态优化调整．将结构动态优化设计神经网络应用于污水处理过程中溶解氧的控制，实验结果表明该控制器较之固定结构神经网络控制器在超调、调节时间、自适应能力方面都更优．

Abstract: A dynamic structure optimization design method is proposed for the neural network.This dynamic structure optimization neural network(DSONN) can add or prune the hidden nodes based on the competition mechanism which estimate the influence of hidden nodes on the network output in the learning process.Then it can adjust the architecture of neural network automatically.In the end,the proposed DSONN is used to control the DO concentration in the wastewater treatment processes.The simulation results compared with the fixed structure neural network controller suggest that the proposed DSONN is more effective in the overshoot,the adjusting time and the self-adaptability.