Synthesis Of Data-driven Controller Based On Multirate Sampled-data System

In the modern industrial production process,the digitization of the control system has promoted the widespread application of the sampled-data system in the industrial world.Therefore,the research on the control method of the sampled-data system has always been a hot research topic in the field of control.The sampler and its sampling period are the basic elements of the sampled-data control system.For general sampled-data systems,engineers generally believe that all samplers and zero-order or first-order retainers use the same cycle.But in the case of multi-sensor fusion,different sensors do have inconsistent sampling periods.That is,there may be a period inconsistency between the sampler and the holder of the sampled-data system,between the sampler and the sampler,and between the holder and the holder.A sampled-data system that has this phenomenon is called a multirate sampled-data system.This thesis studies the data-driven control method for this kind of multirate sampled-data system.From the 1950 s to the present,multirate sampled-data systems have always attracted the attention of scientific researchers,and related controller design methods range from traditional control methods to advanced control theories.However,there are many dynamic characteristics that are difficult to solve in the industrial production process,such as strong coupling,strong time variability,and nonlinearity.These characteristics make it difficult for researchers to linearize descriptions with accurate mathematical models.The method of obtaining controller parameters through formula derivation is also difficult to solve the above engineering problems.For this problem,one solution is to directly use the input and output data with multirate sampled-data characteristics to design the control algorithm,that is,the data-driven controller.Aiming at the practical engineering problems of multirate sampled-data systems,this thesis progressively proposes a variety of data-driven controller design methods to meet the increasing complexity and performance requirements of the system.The introduction describes the background and development status of multirate sampled-data systems,data-driven control methods,and analyzes the limitations of current multirate sampled-data system controller design methods.The thesis first introduces the modeling of general multirate sampled-data systems and the analysis method of multirate sampled-data characteristics,which provides mathematical foundation and knowledge for the design of data-driven controllers in subsequentchapters.Taking the active suspension system with multirate sampled-data characteristics as an example,the paper presents the whole process of multirate sampled-data system modeling,analysis and frequency domain controller design.Through the limited frequency domain H? output feedback controller,the active suspension system has better interference suppression performance in the operating frequency domain.The modeling and analysis of the active suspension system also provides a reference for the analysis of the multirate sampled-data characteristics of the input and output data in the subsequent chapters.In terms of data-driven controller design,this paper firstly designs the data-driven controller from the two aspects of multirate sampled-data self-stabilizing system,from system model regression and system identification.The former combines the multirate sampled-data characteristics with the partial least-squares method to return to the system prediction model.The latter combines a multirate sampled-data system with subspace identification to identify relevant parameters of the system and directly derive a prediction model.Combining these two models with model predictive control methods,the paper proposes a data-driven model predictive controller for multirate sampled-data systems based on partial least squares and a data-driven model predictive controller for multirate sampled-data systems based on subspace identification.Chapter 3 applies the algorithm to the continuous stirred tank heating reactor to solve its output tracking problem when the system parameters are unknown.The thesis further addresses the problem of single-performance data-driven control of multirate sampled-data non-stable systems.A new LQR controller design method is proposed through a new dimension expansion technique.Based on the Bellman equation,the paper successively derives multi-rate system online and offline controller parameter optimization algorithms.Using the least square method,the paper derives the data-driven parameter optimization method of the multi-rate system controller based on the offline optimization algorithm.The algorithm is applied to a three-degree-of-freedom helicopter attitude system with multirate sampled-data characteristics.The thesis solves the problem of controller parameter optimization when the system parameters are unknown.Through the controller optimized by the algorithm,the three-degree-of-freedom helicopter has better performance in angle tracking.Aiming at the problem of mixed performance data-driven control of multirate sampleddata non-stable systems,this paper proposes an iteratively optimized data-driven controller design algorithm based on the policy gradient descent algorithm.This kind ofalgorithm describes the mixed performance index expected by users by designing a reward function module.The reward function module will help the policy gradient descent algorithm to optimize the neural network controller according to the mixed performance index,so that the trained controller meets the desired control performance index.The thesis applies this algorithm to a three-degree-of-freedom helicopter attitude system with multirate sampled-data characteristics to solve its data-driven controller design problems when the system parameters are unknown,the performance requirements are mixed and complex,and the input limiting is applied.Through the controller obtained by the algorithm,the three-degree-of-freedom helicopter can quickly track the angle while meeting various performance requirements and limiting conditions.Based on the data-driven controller based on policy gradient descent,the paper proposes a data-driven controller design algorithm with higher data utilization.That is,the neural network model based on set probability is used to estimate the next state,and the neural network is iteratively optimized under the loss function.The trained neural network model can effectively predict the output of the next moment.Combining neural network model and predictive control can provide a new method for data-driven predictive control of multirate sampled-data systems.Combining the model with the neural network controller can obtain a data-driven controller with high data utilization.Through the controller,the improved version of a continuous nonlinear stirred tank heating reactor can control the liquid temperature,liquid flow rate,and liquid level height to reach the desired steady state quickly when the parameters are unknown.