Wind Tunnel Valve Opening Prediction Based on Multi-Target Mutual Information and Progressive Dilated TCN

Accurate valve opening prediction is important for improving the stability and control performance of wind tunnel systems. To address the challenges of strong variable coupling, nonlinear dynamics, and complex temporal dependencies, this paper proposes a mutual information-based progressive dilated temporal convolutional network, named MI-PD-TCN. The proposed method first preprocesses multi-source sensor and valve operation data, constructs candidate features such as historical lag and differential features, and selects informative variables using mutual information. Then, phase space reconstruction and a sliding-window strategy are used to generate temporal input tensors. Finally, a progressive dilated TCN is developed to extract multi-scale temporal dependencies and predict valve openings. Experimental results show that the proposed method improves prediction accuracy and generalization performance, providing an effective data-driven solution for valve opening prediction in wind tunnel control systems.