PI Parameter Setting Of Intelligent Positioner Based On BP Neural Network

In the process of production process,the control valve is the terminal execution unit of the whole process control system.Its performance has a direct impact on the stability of the process flow,product quality,and production safety.As a control valve accessory,valve positioner has greatly improved the applicability of the control valve and control performance.Valve positioner has become a standard configuration and the ‘brain' of the subsystem of control valve.In the market,intelligent valve positioner has been widely used,whose mainstream products are still imported.Domestic research and development,especially the core positioning control algorithm shown by the dynamic and static performance of the product still has a large room for continuous improvement.This paper belongs to the applied research,aiming at studying the intelligent electric valve positioner which consisted of torque motor I/P conversion unit,pneumatic amplifier and PI controller.In this paper,a parameter setting method of PI location control algorithm based on BP neural network is proposed.As the valves of different specifications and sizes have different structure parameters and dynamic characteristics,the parameters of the control algorithm need to be automatically adjusted by the intelligent positioner.However,it is not easy to measure the dynamic online.The method proposed in this paper is convenient for setting parameter automatically.The effectiveness of the control algorithm and the applicability of different control valves are verified by the simulation study and a large number of experiments.First of all,we have established a perfect control valve mechanism model CVM(including pneumatic film actuator,straight through single seat valve and intelligent locator).On the basis of the model,different simulation control valves were established by changing the control valve parameters including the actuator diaphragm area,actuator spring stiffness,valve travel,etc.Open-loop step experiment was implemented in the simulation control valve and the open loop characteristic parameters were extracted.The optimal PI parameters of the simulation control valve based on certain rules are found out.Then,a large number of data samples were obtained.Secondly,the collected data samples were preprocessed.Through the correlation analysis,the data with strong correlation in the open-loop characteristic parameters were removed to ensure that there was no correlation or weak correlation between the remaining open-loop characteristic parameters.The preprocessed partial data samples were used as the training set of BP neural network to determine the structure and specifications of BP neural network.Levenberg-Marquardet algorithm was used to train BP neural network,and the reserved samples were tested on the trained network to verify the accuracy and stability of the network.Finally,different specifications of the actual control valve were tested using the LabVIEW test environment in the laboratory control valve test platform and enterprise production site.Through the comparison closed-loop test with the mainstream intelligent positioner AVP301,the validity and applicability of the PI parameter tuning method proposed in this paper was verified.