Unconstrained MPC tuning for prediction accuracy in Networked Control Systems

Abstract

Model Predictive Control (MPC) has received wide attention from both the academic and industrial societies, mainly in the chemical industries, and is fast gaining popularity in Electro-Mechanical industrial systems. Great success stories of applications of MPC to industrial systems have been reported. However, the model based nature of this control methodology raises numerous questions of robustness, mainly towards prediction accuracy. A large amount of research has been conducted in MPC tuning, both mathematically inclined and heuristic. However, most of the works were intended at improving performance and robustness for control systems that are not interconnected over shared networks. In this paper, analysis of the standard tuning parameters of MPC and their effects on prediction accuracy are investigated, implications of such accuracy on Networked Control Systems (NCS) with random data packet dropouts are demonstrated by simulation and experimental studies.

Tuning Guidelines for Model-Predictive Control

“…Figure shows the fractionator input and output responses to a + 0.6 step change in the y 2 set point. It depicts that the tuning approaches of refs and provided similarly better y 1 , y 2 , and y 3 responses. Figure shows the fractionator input and output responses to a + 0.35 step change in the y 3 set point.…”

“…Figure shows the fractionator input and output responses to a + 0.7 step change in the y 1 set point; the four tuning approaches provide similarly good (in terms of the integral of squared error) y 1 responses. However, they yielded different y 2 and y 3 responses; the approach of ref provided the best y 2 response, while the approaches of refs and yielded equally superior y 3 responses. Figure shows the fractionator input and output responses to a + 0.6 step change in the y 2 set point.…”

“…This approach was applied to a quadrotor reference tracking problem and yielded satisfactory set point tracking. Conversely, Tsoeu and Koetje 52 recommended setting ρ to a value close to 1.…”

“…The manipulated and controlled variable constraints are – 0.4 ≤ u i ≤ 0.4, i = 1, 2, 3. Simulation studies were carried out to study the performance of MPC tuned using the guidelines given in refs , , , and in the presence of a step change in each set point at t = 1 min, while the other two set points remained unchanged. Two cases were considered: nominal case (no process-model mismatch) and uncertainty case (when all nine steady-state gains of the model used in the controller were higher by 10%).…”

“…In MATLAB, Simulink and the MPC toolbox were used to simulate the process under model-predictive control. For this example, the MPC tuning guidelines given in refs , , , and yielded the controllable tunable parameter values given in Table .…”