Research On The Methods Of Fast Impact Point Prediction For Guided Ammunition

Guided munitions have become one of the main munitions for target strikes in modern battlefields.In order to solve the problems of traditional ground monitoring command guidance,limited monitoring time and large errors in prediction models,this paper introduces wavelet neural network theory into the application of fall point prediction and establishes the applicable The non-linear mathematical model for forecasting the impact point of projectiles and rockets based on wavelet neural network realizes the rapid and accurate forecast of guided munitions and effectively solves the problem of real-time impact point prediction.The main research contents are as follows:A real-time projectile impact prediction method based on wavelet neural network is proposed.The MATLAB programming language is used to realize the WNN algorithm function,the forecast model of the guided munition drop point is established,and the gradient descent method is used to train a single reference trajectory.The simulation results show that the maximum range prediction error is 12.1795 m and the average absolute error is 4.1146 m for the reference trajectory at the predicted firing angle of 51°;the maximum lateral deviation prediction error is 0.0375 m,and the average absolute error is 0.0222m;The lateral deviation is good,and the range prediction error needs to be improved.It takes an average of 0.7076 ms to complete an impact forecast,which meets the requirements of real-time projectile forecasting,which meets the requirements of real-time projectile forecasting.Compared with BP algorithm,WNN algorithm is more superior.By forecasting the benchmark trajectory at other firing angles,the universal adaptability of the forecasting method is verified.In order to effectively solve the problem of large errors in the range prediction of the gradient algorithm,the method of adding momentum factor is used to improve the gradient algorithm.The simulation results show that the improved algorithm has a maximum range prediction error of 0.72 m and an average absolute error of 0.22 m,which effectively avoids the local optimization problem of network parameters and greatly improves the prediction accuracy.Aiming at the problem of long training time caused by the fixed learning rate,the gradient descent method of the segmented learning rate with momentum term is used to adjust the parameters of the network.Through simulation analysis,under the same target error of 10-6m2,The average iteration of the momentum method can reach 406.7 steps,while the segmented learning momentum method only needs 128.7 steps,which shows that the improved algorithm can effectively shorten the training time and converge quickly.Aiming at the uncertainty of the initial parameters of the neural network on the training results,this paper combines the particle swarm optimization algorithm to improve the wavelet neural network.Through simulation analysis,the results are more stable than the gradient method,the convergence speed is faster,the training result error is small,and the real-time solution is satisfied.demand.Developed wavelet neural network visualization application software on MATLAB/App Designer platform.The article mainly studies the impact point prediction of a single trajectory.This visualization application software greatly simplifies the operation steps.Through parameter adjustment,the factors that affect network performance can be found intuitively,and the range of parameters can be narrowed,so as to obtain network parameters that are more conducive to point-of-fall prediction,which is useful for engineering research and teaching.Provide certain convenience and reference.