AdaBoost-PSO-LSTM网络实时预测机动轨迹

针对自主空战中轨迹预测难以同时保持高预测精度和短预测时间的问题, 提出一种自适应增强的粒子群优化长短期记忆网络预测方法。首先，建立三自由度无人机动力学模型, 解决机动轨迹的数据来源问题。其次，分析长短期记忆网络, 并引入在线预测的滑动模块输入矩阵, 利用粒子群优化算法代替传统基于时间的反向传播算法进行网络内部权值更新; 同时为解决优化算法非定向性问题, 提出数据共享方法。然后，为进一步提高预测精度, 采用自适应增强算法搭建外框架, 通过控制弱预测器的数量平衡预测精度与预测时间。最后, 在一段变化较为频繁的轨迹进行预测, 与5种神经网络预测方法进行比较, 结果表明所提方法能够较好地满足精度和时间要求。

Real time prediction of maneuver trajectory for AdaBoost-PSO-LSTM network

To solve the problem that trajectory prediction is difficult to maintain high prediction accuracy and short prediction time in autonomous air combat, an AdaBoost particle swarm optimization for long and short term memory network prediction method is proposed. Firstly, the dynamic model of unmanned aerial vehicle with three degrees of freedom is established to solve the problem of data source of maneuver trajectory. Secondly, the long and short term memory network is analyzed, and the sliding module input matrix of online prediction is introduced. Particle swarm optimization algorithm is used to replace the traditional back propagation algorithm through time to update the internal weights of the network. Meanwhile, in order to solve the problem of non-orientation of the optimization algorithm, a data sharing method is proposed. Then, in order to further improve the prediction accuracy, AdaBoost algorithm is used to build the outer framework, and the prediction accuracy and prediction time are balanced by controlling the number of weak predictors. Finally, compared with five neural network prediction methods in a period of more frequent changes in the trajectory prediction, the results show that the proposed method can meet the requirements of accuracy and time.