Flow Prediction and Analysis of Aviation Hydraulic System Based on XGBoost and Feature Attribution

In order to improve the fault diagnosis accuracy of airborne hydraulic system and enhance the safety and reliability level of aircraft, aiming at the problem that the flow rate of hydraulic system is difficult to be measured, a flow prediction model using eXtreme Gradient Boosting (XGBoost) method is proposed. In addition, three feature attribution methods are used to analyze the interpretability of the flow prediction model, and the key features affecting the flow prediction are obtained. SHAP (SHapley Additive exPlanations) algorithm is used to calculate the contribution of different features at a single sample in the flow prediction model, which improved the interpretability of the prediction model. The experimental result shows that the model possesses high prediction accuracy, and the important features obtained by three methods are accordant, which verifies the effectiveness of this method.