Nonlinear moving target tracking based on LightGBM model

To enhance the capability of assessing and forecasting the spatial-temporal situation of nonlinear moving targets, a nonlinear moving target tracking method based on LightGBM gradient boosting decision tree is proposed. The LightGBM model is used to learn the features of the target trajectory, establish the mapping relationship between the incremental target trajectory motion and the input features, and realize the high accuracy forecast of the nonlinear motion target trajectory. The experimental results show that the method is significantly better than the Kalman filter model in terms of prediction accuracy and prediction robustness by predicting the future points of the trajectory of a typical precision-guided weapon guided bomb and comparing the prediction results with the Kalman filter model. On each track of the test set, the mean absolute errors of the predictions of the LightGBM model in azimuth, altitude and oblique distance are 4.27%, 4.33% and 5.43% of the predictions of the Kalman filter model, respectively, and the mean square deviation are 4.11%, 4.24% and 5.21% of the predictions of the Kalman filter model. The importance of the features is analyzed by the SHAP method, and it is found that the derived features of target trajectory motion increments predicted by Kalman filter model contributed most to the model forecasts.