Explicit Model Predictive Control of a Hybrid System Using Support Vector Machines

Abstract

The objective of this work is to study the capabilities of support vector machines for approximating the complex control law arising from model predictive control of a hybrid MIMO-system. By approximating the control law, an explicit formulation can be obtained, which is computationally less intensive for on-line use. The explicit model predictive control approach is applied to a simulated hybrid system consisting of two tanks in series. The system has real-valued, integer-valued, and binary-valued control inputs. A model predictive controller is first designed to control the entire system. This controller is then approximated using support vector machines, with separate approximators for each control input. Reasonable control results were achieved with the approximators.

Networked explicit hybrid predictive control of an industrial benchmark process

“…The two tanks can be set cascaded with a constant flow rate and all inputs and outputs of the tanks are controlled by valves to adjust their levels. 8,9 Another configuration of the double-tank hybrid system is one tank is set at a different height from the other tank either the upper tank is cylindrical as same as the bottom one, 10 or has a spherical shape. 11,12 Also, the two tanks can be set at the same height, connected by two valves, and the outputs of the tanks are controlled by valves to control the second tank.…”