Research On Clustering Prediction And Intention Recognition Technology Based On Trajectory Data

With the increasing complexity of air warfare scenarios,trajectory prediction and intent recognition have become key technologies to ensure air superiority.In this context,trajectory prediction is used to predict the future movement of hostile aircraft,and intent recognition infers the targets of enemy aircraft based on their trajectories and clustering results.These techniques can significantly improve the situational awareness of air defense forces and provide effective countermeasures against enemy aircraft.However,due to the dynamic nature of air combat scenarios,there are still many challenges to overcome in order to achieve accurate and reliable prediction,clustering,and intent recognition.In this thesis,the main work carried out for trajectory prediction and intent recognition in the context of air warfare is summarized as follows:1.The traditional trajectory prediction algorithms mainly use single model and hybrid model,which have insufficient ability to extract trajectory features and poor prediction effect,while the hybrid model has many parameters and the accuracy performance is unstable under insufficient training.To address this problem,this thesis constructs a hybrid model for trajectory prediction based on the attention mechanism.By introducing convolutional neural network as a pre-processing layer,the local features of the data are explored; for the problem of unstable accuracy due to the expansion of parameters of the hybrid model,an attention mechanism is introduced to enable the model to pay attention to more important information and improve the stability of the model.The experiments show that using convolutional neural network as the preprocessing layer can improve the accuracy of the original single model,and adding the attention mechanism can improve the stability of the hybrid model.2.Due to the dynamic changes of air target trajectories and frequent trajectory shortcuts in the context of air warfare,the clustering results obtained by the traditional Euclidean distance-based clustering method are easily affected by hyperparameters and it is difficult to find the optimal hyperparameters.To address this problem,a trajectory clustering algorithm based on the edit distance method is proposed in this thesis.The similarity of two trajectories is calculated by the idea of adding,deleting and changing in the edit distance method,which solves the problem that the traditional method is difficult to identify the shortcut trajectory effectively and requires a large number of parameters in the algorithm.Experiments show that the proposed algorithm has better performance than the original TRACLUS(TRAjectory CLUStering)algorithm in trajectory clustering.3.Most of the existing airborne target intent recognition methods directly feed target data into the model for training and prediction without considering the influence of geographic location factors on airborne target intent.To address this problem,this thesis proposes an intention recognition model based on trajectory clustering,which extracts the geographic location information implied in the data by trajectory clustering method.In the training phase,the trajectory cluster dimension of the training data is expanded by designing the trajectory backtracking method,i.e.,the geographic location information dimension; in the prediction phase,the geographic location information of the predicted data is determined and then the intention recognition is performed.Experiments show that the proposed intention recognition model based on trajectory clustering improves the overall index and has better performance in the intention dimension with small data volume compared with the original model.