Abstract: To address the huge challenges of high time-sensitivity, behavioral complexity, and massive data volumes in the combat effectiveness evaluation of unmanned aerial vehicle(UAV)cluster, a data-driven effectiveness evaluation method based on the particle swarm optimization(PSO)-extreme gradient boosting(XGBoost)-Shapley additive explanations(SHAP)model was proposed. An effectiveness evaluation model was established using the XGBoost algorithm based on an evaluation index system. The mapping relationship between index data and combat effectiveness was analyzed and verified using simulation data to mine the combat effectiveness of the UAV cluster system. The PSO algorithm was employed to optimize the hyperparameters of the XGBoost model to enhance evaluation accuracy and efficiency. To balance the predictability and interpretability of the evaluation work, the SHAP mechanism was utilized to interpret the effectiveness evaluation process and identify directions for index optimization. Verified by UAV cluster combat simulation data collected via the Vensim platform, the proposed model demonstrated superior accuracy and interpretability compared to support vector regression(SVR), random forest(RF)algorithm, light gradient boosting machine(LightGBM), and back propagation(BP)neural network methods.

Key words: UAV cluster, effectiveness evaluation, machine learning, XGBoost, SHAP