Target Trajectory Prediction Based On CMAC Neural Network

It is widely used to predict the trajectory of a target by using the historical position information of the target,such as digital battlefield,intelligent traffic,logistics monitoring,electronic commerce and so on.Predicting target trajectories in many application scenarios plays a key role in further decision making.In this thesis,the trajectory prediction problem of fast moving target is studied.The improved cerebellar model neural network is used to improve the speed and accuracy of the prediction system.The method used in the thesis solves the problem that the existing prediction system has low prediction accuracy when the acquired data interval is not uniform,and it is difficult to quickly give accurate prediction results when the received data is limited.On the basis of the study of the existing linear target trajectory prediction method and the nonlinear target trajectory prediction method,the cerebellar model neural network is used to predict the target trajectory.The main contents are as follows:Firstly,on the basis of summarizing various nonlinear target trajectory prediction methods and linear target trajectory prediction methods,the advantages of cerebellar model neural network in the prediction of target trajectory are explained in terms of the structure and principle of the network.Then,according to the actual application scenario,the trajectory of the missile is simulated according to the actual missile flight condition.The missile trajectory model is used to extract training data and verification data of cerebellar model neural network.Application examples are used to analyze the defects of the cerebellar model: One is that,local convergence may reduce the generalization ability of the network and reduce the nonlinear approximation capability of the network;The other is the increase of input dimension will increase the accuracy of the network but at the same time may reduce the learning speed of the network.In view of the above defects a recursive unit with variable weights is added to the traditional cerebellar model neural network structure to solve the problem of the decline of the nonlinear approximation ability of the network;Combined with the idea of transfer learning,the offline learning method is used to train the network,and the weights of the offline training network are directly used as the initial weights of the online prediction system to improve the convergence speed of the network when predicting different targets.The network uses the simulated target trajectory to train,and then gives the prediction results.The trajectory of the target is simulated on the MATLAB platform.At the same time,the prediction results are compared with other forecasting methods through various data indicators.The experimental results show that the improved CMAC combined with the transfer learning method has better performance in target trajectory prediction.