Research On Speed Observer Of Permanent Magnet Synchronous Linear Motor Based On RBF Neural Network

Permanent Magnet Linear Synchronous Motor(PMLSM)has the advantages of large acceleration,high accuracy,fast response,low noise,and simple structure.It is widely used in civil,industrial,military and other fields.In order to achieve accurate PMLSM control,it is important to obtain accurate motor speed and position information.The traditional speed measurement method using mechanical sensors such as grating ruler or magnetic scale ruler will bring a series of problems such as increased cost,reduced reliability and difficult maintenance to the servo system,hindering the development of PMLSM's high precision.Speed observation technology can realize the estimation of speed and position without the help of mechanical sensors.On the one hand,it can be used to assist mechanical sensors,to compensate the detection accuracy of mechanical sensors,or as redundant control after mechanical sensors fail;on the other hand,it can also It completely replaces the sensor and realizes the sensorless control of the motor,so it has become a research hotspot in recent years.The purpose of this paper is to design a speed observation method based on BRF neural network to realize the speed and position estimation of PMLSM control system.The main work done is:Firstly,the structure and working principle of the PMSM are introduced.Based on the principle of coordinate transformation,the mathematical model of PMLSM in three coordinate systems is established,and the vector control strategy selected as the main control method in this paper.Secondly,the basic structure and working principle of the RBF neural network are introduced.Based on the study and analysis of the neural network learning algorithm,it is clear that the neural network learning method used in this paper is: initializing the parameters of neural network in the off-line training link,and dynamically adjusting the parameters in the on-line training link by gradient descent method.According to Lyapunov theory,the convergence of neural network is analyzed,and the range of learning rate is obtained.In MATLAB / Simulink simulation software,a simulation example is set up to demonstrate the learning mode of neural network,and the influence of different learning rates on the convergence of neural network is compared and analyzed.It is proved that the learning rate obtained by convergence analysis is an ideal choice.Then,the nonlinear relationship of the PMLSM dynamic model is studied and analyzed,and a speed observer based on the RBF neural network is designed based on this.Using the current observation value obtained from the current equation as an indirect feedback quantity,real-time correction of the observation error of the speed observer is realized.Aiming at the problem of how to obtain the Jacobian matrix of the velocity observer,two methods are proposed that use a switching function and construct a parameter identifier to approximate the Jacobian matrix.The speed observer is applied to the speed sensorless control of PMLSM,and the corresponding simulation model is established in MATLAB / Simulink simulation software.According to the simulation results,the two methods are compared and analyzed,and the speed observation method with the parameter identifier is determined as the final method in this paper.Finally,the design of the control system software was completed on the experimental platform,and the dynamic and static characteristics and anti-disturbance experiments of the speed observer were established.The experimental results and simulation results confirm each other and prove the effectiveness of the proposed method.