Research on UAV swarm effectiveness evaluation method based on deep learning

UAV swarms have shown significant advantages in complex scenarios such as fire rescue due to their flexibility and collaboration capabilities, but their effectiveness evaluation still faces challenges such as dynamics, nonlinearity and real-time performance. In this study, a method of UAV cluster effectiveness evaluation based on deep learning is proposed. The evaluation system covering six indicators such as flight speed, task completion time and cluster coverage is constructed by analytic hierarchy process. The multi-dimensional task data is generated by NetLogo simulation platform, and the Z-score standardization and Min-Max normalization strategy are used for preprocessing. A deep learning model is constructed based on fully connected neural network (DNN), and L2 regularization and Dropout technology are introduced to suppress overfitting. Experiments show that the mean absolute error (MAE) of the model in the test set is 5.32 %, and the prediction error of the task completion time is less than 0.5 %, which verifies its robustness in dynamic scenarios. This study provides a quantifiable theoretical framework for the effectiveness evaluation of UAV swarm, and provides a reference for the design of intelligent decision support system in complex environment.