A Novel Adaptive Subspace Predictive Control Approach with Application to Continuous Stirred Tank Heater

Subspace predictive control (SPC) is a widely utilized data-driven control technique in various industrial applications. However, its static nature restricts its ability to effectively track nonlinear dynamic systems, resulting in diminished performance. To address this problem, an adaptive subspace predictive control approach is proposed, incorporating an adaptive mechanism to continuously update the subspace predictor. The designed adaptive mechanism mitigates the negative impact of historical data by sliding the data window. It simultaneously employs the addition and deletion of data vectors in the data matrix through recursive matrix transformation, simplifying computational complexity while maintaining accuracy. In addition, the developed subspace predictor enables online learning and effectively handles the dynamic nature of industrial processes, requiring little prior knowledge. The theoretical analysis of the proposed control approach includes recursive feasibility and stability, along with a discussion on determining relevant parameters. The effectiveness of the proposed control approach is demonstrated through its application to a continuous stirred tank heater benchmark. The results exhibit significant improvements in tracking control performance, leading to enhanced efficiency and cost reduction. Overall, this research presents a promising solution for addressing the challenges of predictive control in industrial processes.