Least Squares Support Vector Machines Inverse Control Of Two-motor Variable Frequency Speed-regulation Systems

Two-motor variable frequency speed-regulating system is widely used in the modern industrial manufacturing plants. In order to improve quality and quantity of the products, the high-performance control system is needed. In recent years, the inverse system method has been developed to a complete control theory for the general form of nonlinear systems. It is easy in practical use and strict in theory. Since least squares support vector machines (LSSVM) has a strong potential to approximate a nonlinear function, a new LSSVM inverse model is obtained by introducing the LSSVM into the inverse system. The LSSVM inverse model can overcome the difficulty in implementing the inverse system by analytic means and break through the bottleneck in engineering applications. According to the characteristics of multi-variables, high nonlinearity and strong-coupling, the model identification and the controller design for the two-motor system are investigated based on the LSSVM inverse system. The main contents and fruits of this thesis are as follows:1) The fundamental concept of the inverse system theory and the system reversible principle are introduced. The inverse system includes the left and the right inverse system. By cascading the inverse system with the original system, the whole system is decoupled to a pseudo-linear combined system in order to implement linearization and decoupling control. The LSSVM regression algorithm, which is adopted to approximate nonlinear mapping of the inverse system, is depicted. Experimental platform of two-motor variable frequency speed-regulating system is presented, which provides necessary support.2) Inversibility of the two-motor system in vector control mode is verified. By using the extended structure of the inverse system, the LSSVM inverse control strategy based on active disturbances rejection control (ADRC) is proposed. A pseudo-linear system is completed by combining the LSSVM inverse model with the two-motor system. ADRC is used as a closed-loop controller for the pseudo-linear system, and the uncertainty estimated by the extended state observer is used to construct the LSSVM inverse model. The proposed method can reduce the interference caused by load disturbance and modeling error. The good decoupling applicability and strong robustness can be achieved meanwhile.3) To solve the problems such as high cost and poor reliability caused by tension sensors, tension identification based on the LSSVM left inverse is proposed. The left-invertibility of the tension model is proved. Speed signal which is relatively easy to be measured has been used to build the "assumed inherent sensor" left inverse soft-sensing model for tension identification. The proposed method can identify the actual tension quickly and accurately, independent of dynamic model and specific parameters of the two-motor system.4) Taking full advantage the left inverse soft-sensing model in estimating the system state, a new generalized combined inverse controller is proposed. Based on the generalized inverse theory, the combination of the left inverse and the right inverse forms a whole combined inverse controller. LSSVM is used to implement it, resulting in the LSSVM combined inverse controller. Sensorless operation of the two-motor system can be achieved through the LSSVM combined inverse controller.5) The characteristics of the two-motor system by the rotor magnetic field-oriented vector control are analyzed. A new modeling method is proposed based on the data-driven principle. The affinity propagation (AP) clustering algorithm is successively applied to group the input and the output data into clusters. LSSVM is used to construct local models. The estimation of every LSSVM local model is fused to build the speed model and the tension model. The proposed method can fit the nonlinear characteristics of the system with high precision and good generalization.6) To mend the defects associated with conventional inverse model identification methods for two-motor system, a multi-model LSSVM modeling method based on the modified recursive least-squares (RLS) algorithm is proposed. The initial, off-line, inverse model of the system is deduced by computing the weighted sum of every LSSVM local model. The modified RLS algorithm is adopted to adjust the weights according to system deviation adaptively. The proposed method is feasible, effective, and suitable for two-motor system.Finally, some conclusions and further research projects are raised.