Research On Intelligent Adaptive Robust Control For Sensorless PMSM Servo Systems

The sensorless control is one of the important research direction for permanent magnet synchronous motor(PMSM) control, because the control method without using the speed and position sensor, simplifies the system structure, improves the reliability and reduces the cost. In fact, these advantages are obtained at the expense of increasing the computational complexity and decreasing the control performance. Therefore, the sensorless control method is still in the research stage and has not been used in the application fields required a high precision. The accurate estimation of position and speed of the rotor, and the ideal controller design are the important problems in the study of sensorless PMSM control. However, system parameter un-certainties and external disturbances will cause the non-convergence and vibration of the estimation and control results. Intelligent control methods such as neural net-work (NN) and fuzzy logic system provide the powerful technique supports to solve the complexity, time-variance and uncertainty of the sensorless PMSM system, and to obtain a high-performance control effectiveness. At present, many scholars have proposed the adaptive robust estimation and control methods based on a variety type of NN and fuzzy logic system, but they have not yet resolved many problems, as the following problems:when NN is used in the design of the controller and estimator, they have not eliminated an effect of NN reconstruction error, when the robust position and speed estimation are realized by PMSM mathematical model, they have not proposed the method that not robust to one or two parameter uncer-tainty, but achieve robustness to the whole parameter uncertainty, when the particle swarm optimization(PSO) algorithm is used to optimize the parameters of fuzzy PI controller, they have not solved the problem that the PSO algorithm easily falls into local optimum, when NN is used in the controller design, they have not proposed the methods that obtain a most suitable adaptive robust control law for the dynamics of sensorless PMSM, and calculate an optimal learning rate. For solving the above problems, the main work of this paper are as follows:1. A robust speed estimation method is proposed based on NN adaptive observer for sensorless PMSM. The influence of system parameters uncertainties and load disturbance on the speed estimation performance are eliminated through the adaptive observation of system dynamics. Because NN used in the adaptive observer has a NN reconstruction error to a certain degree, which can cause the vibration and non-convergence in the estimation result, an adaptive law and a robust compensation scheme are proposed to solve that problem. NN based robust speed estimator estimates the rotor speed by using the system state variables identified by NN adaptive observer. The rotor position is obtained by integrating speed estimation. The simulation results show that the proposed speed estimation scheme can achieve high dynamic response, robustness and asymptotic convergence.2. A robust position estimation method is proposed based on PMSM complex number model. The influence of whole system parameters uncertainties to the position estimation are eliminated by the complex analysis that the estima-tion result can be separated into the real and imaginary parts. The estimated position value is divided by the real and imaginary parts, the real part repre-sents the actual position estimation, the imaginary part represents the position estimation error caused by the system uncertainties. In order to reduce the estimation error to zero, the position type PI controller is applied in the imag-inary part of position estimation, so the estimation error converges to 0. The robustness of this method can only be satisfied in the field of rated speed, be-cause the demagnetization in the flux weakening region affects the estimation strategy. In order to ensure the feasibility and robustness of this scheme in the flux weakening region, EKF is used to identify the permanent magnet flux linkage. The rotor speed is obtained by differential position estimation. The simulation results show that the scheme can achieve the fast dynamic perfor-mance and robustness for the rotor position and speed estimation in the entire control range.3. In order to design an adaptive robust speed controller of sensorless PMSM based on fuzzy theory, a parameter optimization method of fuzzy PI con-troller is proposed based on adaptive PSO. In order to avoid the reduction of particle swarm search space by constraint conditions of fuzzy membership function parameters, and PSO algorithm easily fall into local optimal, a new particle composition method is proposed. The particle is random in the range [0,1], and provides the maximum search space to obtain the optimal member-ship function parameters. In order to improve the adaptive rate of parameter optimization, presents an improved adaptive PSO algorithm. The simulation results show that the proposed optimization scheme can achieve high dynamic response and robustness for the sensorless PMSM control system.4. In order to realize an adaptive robust speed control of sensorless PMSM based on NN, a speed controller design method is proposed based on recurrent Elman NN(RENN). In this method, the influence of system parameters uncertainties and load disturbance to the speed tracking performance are eliminated by the improved RENN structure and the adaptive law. The adaptive robust speed controller is composed by the RENN controller and the compensation controller. RENN controller estimates an ideal PMSM speed controller. The compensation controller eliminates the error between the ideal speed controller and actual RENN controller. The stability of RENN based adaptive robust speed controller and the asymptotic convergence of speed tracking results are proved by Lyapunov theory.5. A calculation method of optimum learning rate is proposed to improve the adaptive performance of RENN speed control system. First, proposed a most suitable online training algorithm for the improved RENN. Next, in order to improve the accuracy and rate of the weight learning, the calculation method of optimal learning rate is proposed. The optimal learning rate is determined by the rotor speed tracking error and the maximum value of RENN weight norm. The simulation results show that the optimal learning rate can achieve the faster adaptibility and the tracking convergence.