Application Of Neural Network Controller In Thermal Power Plant

In thermal power units,most of the controlled objects have large delay,strong coupling and time-varying characteristics.In order to ensure the safe and stable operation of thermal power units,the more conservative PID control is often used in engineering,which is difficult to achieve better control effect.With the application of distributed control system(DCS),it creates conditions for the application of intelligent control algorithm in thermal power units.In recent years,artificial neural network has played an important role in the field of intelligent control with its strong function fitting ability,self-learning ability,generalization ability,good robustness and many other advantages.Therefore,the intelligent control method based on neural network is considered to be used in the control of thermal power units.Firstly,the inverse controller based on neural network is designed.The field data are collected,the inverse model of the object is trained as the controller,and the real-time update is carried out to ensure the robustness.In order to solve the widespread delay problem of the actual object,the predictor based on Elman neural network is designed.This control method overcomes the time-varying and nonlinear characteristics of the actual industrial field,and has good anti-interference ability.At the same time,it also solves the problem that the inverse controller is not easy to obtain in the traditional internal model control,and it does not need to carry out complex controller parameter tuning.It has been applied to the actual water level system of No.2 high pressure heater and achieved good results.Another control scheme is PID controller based on reinforcement learning,which combines reinforcement learning algorithm with PID control,adopts A3C algorithm framework in the field of reinforcement learning,trains the action network offline,and uses the action network to dynamically adjust PID parameters to obtain PID controller based on reinforcement learning.This control scheme improves the traditional PID control algorithm,and uses the action network to adjust the PID parameters in real time and dynamically.After off-line training,the action network learns the PID parameter regulation rules for a class of objects,so it has better robustness and simplifies the design of the controller to a certain extent.The controller is applied to 330MW coordinated control system,which effectively reduces the fluctuation of main steam pressure and has good adaptability to coal quality disturbance.