Parameter tuning method for model predictive control of uncertain systems based on machine learning

Abstract

The invention discloses a method for setting parameters for predictive control of uncertain system model based on machine learning, comprising the following steps: 1) obtaining an m-dimensional output weight matrix Q and an n-dimensional input weight matrix R in the cost function of the system model predictive control; 2) Obtain the output sequence and mn×L 3 Group performance indicators, in which each group of performance indicators is a set of column vectors constructed from the output overshoot and adjustment time. For each group Q and R, the worst overshoot and worst adjustment time are obtained respectively, and then the The worst overshoot and worst adjustment time taken are stored in the matrix F; 3) Build an RBF neural network, and then use the established RBF neural network to calculate the optimal performance index; 4) Build a BP neural network, and then use the BP neural network to find Take the performance label; 5) Using the performance label as the basis for optimization, the PSO optimization algorithm is used to adjust the model predictive control parameters of the uncertain system. This method can more accurately realize the setting of the model predictive control parameters of the uncertain system.

Description

technical field [0001] The invention relates to a parameter tuning method for predictive control parameters of an uncertain system model, in particular to a parameter tuning method for predictive control parameters of an uncertain system model based on machine learning. Background technique [0002] Model predictive control (Model Predictive Control, MPC) is an advanced control strategy, and its mechanism can be described as: at each sampling moment, according to the current measurement information obtained, a finite time-domain open-loop optimization problem is solved online, and the obtained The first element of the control sequence acts on the plant. At the next sampling instant, the above process is repeated: the optimization problem is refreshed with new measurements and re-solved. [0003] Predictive control is quite different from algorithms such as traditional PID control and optimal control. In particular, it has many controller parameters, which need to be selecte...

AI Technical Summary

Problems solved by technology

[0004] In addition, the controlled object model in industrial applications is generally obtained by identifying the input and output data obtained during the actual operation of the system. Due to the time-varying characteristics of the system working conditions and the existence of interference such as measurement noise, the obtained model is different from the actual controlled object. There is generally an error between them, that is, there is model uncertainty, which requires the control system to have good robust performance

The parameter tuning of an uncertain system introduces additional complexity, and often can only be tuned based on expert experience about the system, which greatly increases the cost of tuning

Images