Analysis and design of a linear input /output data-based predictive control

Abstract

A subspace identification algorithm is reformulated from a control point of view. Model Predictive Control (MPC) is defined as a category of control system that determines the control trajectory that will result in an optimized future behavior of a plant. Typical design of predictive controllers starts with identification of parametric model using plant input and output data. Then, predictor matrices can be obtained from the model. Next, the predictor matrices are used to obtained predictions for the process output which are used in the controller design. In this work, however, the predictive controller can be obtained directly from input and output data by using the subspace matrices. Therefore, this method eliminates the intermediate step of parametric model identification. The comprehension of this concept is discussed and the implementation of predictive controller is done virtually by using MATLAB simulation. The proposed linear input/output data-based predictive controller is applied to the property control of an activated sludge plant. A typical control problem such as variation of set-point tracking has shown that the proposed controller demonstrates satisfactory control performance.