Simultaneous model prediction and data-driven control with relaxed assumption on the model.

Area of Science:

• Control Systems Engineering
• Time Series Analysis
• Optimization Theory

Background:

• Model Predictive Control (MPC) is crucial for advanced system regulation.
• Accurate system identification and control signal design are often challenging due to unknown parameters and dynamics.
• Existing methods may struggle with complex, non-convex optimization problems.

Purpose of the Study:

• To develop a novel Model Predictive Control (MPC) law using system input-output time series data.
• To simultaneously identify unknown Auto-Regressive Integrated Moving Average (ARIMA) model parameters and the controller signal sequence.
• To reformulate a non-convex optimization problem into a more tractable form for efficient solving.

Main Methods:

• Utilizing time series data from system inputs and outputs.
• Formulating an optimization problem to determine unknown model parameters and controller signals within a data window.
• Transforming a non-convex optimization problem with non-convex constraints into an equivalent problem with convex constraints and a non-convex objective function.

Main Results:

• The proposed transformation simplifies the optimization problem, making it solvable with current solvers.
• The effectiveness of the developed MPC approach was demonstrated through various examples.
• The method achieved satisfactory results in simultaneously estimating model parameters and designing control signals.

Conclusions:

• The presented approach offers an effective way to design MPC laws from time series data.
• The transformation technique significantly eases the solution of complex optimization problems in system identification and control.
• The study validates the practical applicability and convincingness of the proposed method in real-world scenarios.