Research Of Event-triggered Data-driven Predictive Control

Predictive control,as one of the main control methods of modern control theory,has been widely used in many practical industrial control systems because of its superior performance in handling multivariate complex optimal control problems with constraints.However,with the rapid development of modern production technology,on the one hand,the structure of actual industrial systems has become increasingly complex,with multiple variables,strong coupling and other complex characteristics,making it difficult to establish accurate mechanistic models of production systems,thus limiting the application of conventional model predictive control methods;on the other hand,the actual production process saves a large amount of offline and online production data,which implies the key information about the system production operation.Therefore,the data-driven predictive control methods have emerged and received widespread attention.This paper investigates the theory of data-driven predictive control,and combines the core idea of event-triggered control theory,proposes a solution to replace the receding horizontal optimization with an event-triggered law for the problem that the conventional subspace predictive control method imposes a huge computational load and data transmission load on the controller in the operation process.By making full use of the measurable data of the system,different event-triggered laws are designed for discrete nominal linear constant system and for practical control systems with multiple constraints,which significantly reduce the data transmission and computation of the subspace predictive control method while ensuring the control performance of the system,and provide a theoretical basis for the practical application of data-driven predictive control.The main research contents are as follows:1)For discrete linear time-invariant systems,an event-triggered subspace predictive control algorithm is proposed by combining the event-triggered control idea.Based on the robust stability theory,the required event-triggered law is designed using only measurable data of the system when some of the system model parameters are unknown,thus reducing the receding horizon optimization process of the original algorithm and demonstrating the stability of the proposed algorithm.Finally,a numerical simulation example of a Continuous Stirred Tank Heater verifies that the proposed algorithm can effectively reduce the computational load and transmission load of the original data-driven predictive controller while ensuring the system control performance.2)For the constraints with the actual system,the connection between the constraints and the event-triggered law is studied in depth,and the event-triggered data-driven predictive control algorithm with constraints is designed comprehensively.Taking the actual data center energy management system as the research object,we first establish the thermodynamic model and equipment power consumption model of the data center,analyze the control problem of the data center energy management system,and obtain the specific objective function.After that,the event triggering law under the constraints is designed for this control problem and the actual constraints contained in the data center.Finally,the control performance of the event-triggered data-driven predictive control algorithm and the effectiveness of the event-triggered law in improving the computational efficiency of the algorithm are verified for a small air-cooled data center.