Research On Neural Network Inverse Control In The Main Steam Temperature System

As the development of the electricity production is in the directionof the large capacity and high parameters, the inert and delay of the mainsteam temperature become larger which leads to the increase ofdifficulty in control. Coupled with its time-varying and its uncertainty,PID control is not able to meet the high control requirements of the mainsteam temperature. Therefore, the research of the main steam temperaturecontrol strategy has highly important significance.Inverse dynamics of thermodynamic system is a new research fieldwhich has become one of the hot points in many research and applicationfields. It is a open-loop control system when putting the trained inversemodel in front of the plant which will avoid the instability caused by thefeedback. The neural network is combined with inverse dynamics to obtain the non-analytic inverse model. While the characteristics of theinverse system is subject to the physical characteristics of the originalsystem. Therefore, for the use of inverse dynamics, the original systemshould be processed. So in this paper, the delay time was obtained byusing adaptive linear element.The positive model without delay and itsneural network inverse model were obtained by the three-layer BPnetwork.A delay separated neural network inverse control scheme wasproposed in this paper. When the effectiveness of this control system isverified, the improvement is made to eliminate the static error. Off-lineidentification and on-line identification are combined to get the inversemodel in order to reduce the steady-state error and make the system havefine adaptive capacity and ensure the accuracy of the system’s output.This control system is applied to a300MW main steam temperature system. Simulation results show that this control system has fasterresponse， better robustness and anti-interference performance. It issignificantly improved compared with the case of original cascade PIDcontrol scheme.