The Improvement of Data Selection Method in Data-Driven Model Predictive Control of Uncertain Systems

Data quality is pivotal for the performance of data-driven model predictive control systems. While traditional data processing methods, such as cleaning and enrichment, are necessary, they do not always improve the controller performance. To tackle this problem, this paper introduces a new data selection method specifically designed for data-driven model predictive control. It includes a controller performance evaluation approach that assesses the quality of the current data, enabling the controller to select a set of informative data elements that enhance its performance. Additionally, a sliding window mechanism is implemented to compare current data against historical data, preventing the loss of significant patterns or trends. A simulation example illustrates the effectiveness of the proposed data selection method.