| In modern warfare,accurate and rapid prediction of the impact point is the key to improving the accuracy of shooting,and it is also the premise of trajectory correction and guidance.The traditional impact point prediction method is solving the ballistic equation with numerical integration method.However,in order to obtain a high precision solution value,a smaller integration step and a more complex trajectory model are required,which will increase the solution time and the difficulty of obtaining the flight state parameters.How to predict the impact point accurately and quickly when it is easy to obtain the flight state parameters has become a problem worthy of study.Therefore,this paper takes the flight state parameters as the input characteristics of the machine learning model and takes the impact point information as the output characteristics of the machine learning model by combining the machine learning theory to establish the impact point prediction model without calculating the ballistic equation explicitly.In this paper,two kinds of impact point prediction models based on machine learning,BP neural network prediction model and support vector machine prediction model,are established.The former is a machine learning model based on connectionism,and the latter is a machine learning model based on statistical learning theory.Through the simulation,the impact point prediction of the two machine learning models is studied when the current acquisition points of flight state parameters are combined with 1 to 4 acquisition points in front of it.The results show that the prediction errors of the two models don’t always decrease with the increase of the number of acquisition points: when the number of acquisition points is 2,the prediction error of the BP neural network model is the smallest;when the number of acquisition points increase,the prediction error of the support vector machine model doesn’t change too much.In practical application,the number of state parameters acquisition points can be reasonably selected,so that the impact point can be predicted faster and more accurately.The prediction errors of lateral deviation of both models are smaller than the prediction errors of range,and both models can predict range and lateral deviation accurately.The root-mean-square errors of range and lateral deviation of the BP neural network model are 3.83 m and 0.95 m respectively,and the maximum errors of range and lateral deviation of the BP neural network model are 10.04 m and 2.71 m respectively;the root-mean-square errors of range and lateral deviation of the support vector machine model are 7.05 m and 1.11 m respectively,and the maximum errors of range and lateral deviation of the support vector machine model are 14.52 m and 2.32 m respectively.At the same time,the prediction time of the two models is statistically analyzed,and the results show that the prediction time of the two models is both shorter than the time used for the numerical integration of the 6-DOF ballistic equation.The parameters of the BP neural network model and the support vector machine model are optimized based on particle swarm optimization algorithm and the simulation results show that the prediction performance of the two models has been improved to some extent after optimization.Therefore,the two machine learning prediction models established in this paper can predict impact point accurately and quickly,which can provide a certain reference for practical application. |
