Time-varying Identification And Saturation Characteristic Analysis Of Data-driven Control Systems

In recent years, the concept of data-driven control, which stemmed from the area of computer science, was posed in control area. Today, in the situation of large accumulation in the production data and facing the challenge of the difficulty to achieve the mechanism model, how to effectively use the great information in these data to research data-driven control method is paid much more attention.Predictive control, which is considered to be the only advanced technology to deal with online optimal control of the constrained multivariable system in a systematic and intuitive approach, has been widely and successfully applied in the complex industrial production process with the control strategies of predictive model, multi-step rolling optimization and feedback correction. Because of the direct use of the observed input and output data to identify the system model and the robustness of numerical calculation, subspace method is widely used in multi-variable system without knowing the prior structural information. For all of these advantages, there have been lots of data-driven control methods based on subspace online identification and predictive control.In this paper, considering online identification and control of the linear time-varying (LTV) system with saturation constraints, the improvement of the identification algorithm, impact due to saturation data and design of predictive controller and other aspects are deeply discussed and studied. The main contents include:(1) In the existence of large changes in dynamic characteristics, a complete data-driven method is proposed based on subspace identification and predictive control with feedback correction varying forgetting factor. In order to adjust the weight of history data and form a strategy of feedback correction, the variable forgetting factor is structured by error of the real and predictive outputs value. Thus the current characteristics are better reflected and the identification sensitivity and control effect are improved. Finally, a simulation example is given to demonstrate the efficiency of the proposed algorithm.(2) Saturation character exists widely in the actual system, and a large number of articles take the constraints condition into account during control process. However, relative to the research on the actuator saturation, study on the output (sensor) saturation has not been taken seriously and research results are very little, especially in the aspect of effect on the identification due to the saturation outputs. While unknowing the change of the system information because the output signals are locked, the relation among the output saturation step, prediction horizon and subspace matrix is obtained. Finally, a simulation example is given to demonstrate the correction of the conclusion.