Methods On Fast And Accurate Prediction Of Trajectory Correction Projectile Impact Points

Trajectory Correction Projectiles are the higher cost performance of ammunitions, and they will play an important role today and in the future wars. Trajectory correction technology is the key technology that trajectory correction projectiles strike the targets precisely, and forecasting the impact point of the projectiles is the core of trajectory correction technology. In order to achieve real-time trajectory correction, it needs to forecast projectile impact point fast and accurately. Therefore, several impact point prediction methods were proposed.I. With reasonable assumptions for linear trajectory model, the circular motion of the projectile was regarded as linear time-invariant systems, and the Gradient approximation method was used to derive differential equations in order to get analytic solutions of the linear trajectory equation. Dimensionless arc length was estimated to predict the projectile impact point. After the simulation, the results showed that lateral deviation prediction error was smaller, but the accuracy of forecasting the range was not ideal. Generating the causes was analyzed, and several improvements were proposed. Then on the predictable time, the option 2 proposed to increase the trajectory solver speed.II. The way of projectile impact point prediction based on BP neural network was proposed. Combined with the projectile impact point prediction equation, the range and the lateral deviation prediction model of BP neural network were established. After forecasting model was trained by the gradient descent method and the model was simulated, the results showed that the prediction accuracy of lateral deviation was well, but the range!s was poor. To solve this problem, the improved methods was to improving BP network hidden layer nodes, and to adjust the weights in gradient descent method with momentum term, and the range and lateral deviation of information were regarded as the output of BP network at the same time. Then the simulation test results showed that the prediction accuracy of range and lateral deviation had been greatly improved, and the average prediction time was 6.629 ms.III. The method of projectile impact point prediction method based on interpolation Radial Basis Neural Network was proposed. The prediction model RBF neural network of the range and lateral deviation was established, respectively. And using the model interpolation strict training, this prediction model was simulated, and the results showed that the largest prediction error of range was 5.5706 m,and the largest lateral deviation forecast error was 0.0057 m, and the average time of forecasting projectile impact point was 39.048 ms. Although this prediction model had simple theory, strong generalization ability, good fault-tolerant, but the number of hidden layer neurons was still a lot. Using optimization principle was to optimize the network structure of the prediction model. After the model simulation test, the results showed that the maximum error of the range did not exceed 0.8m, and lateral deviation error did not exceed the maximum 6 10-3m, and the average forecast time was to reach a placement8.216 ms.IV. The method of projectile impact point prediction method based on generalized regression neural network was proposed. This prediction model was not good. To solve it, the particle swarm algorithm was used to optimize the smooth factor of the model. After simulation, the results showed that the prediction error of range from the original maximum of 6000 m down to the maximum of 40 m, and lateral deviation prediction error from the original maximum error 40 m down to the current maximum error 0.2m.V. The impact point prediction method based on multidimensional interpolation was presented. Then adopting a Gaussian function as the base of its multidimensional interpolation equation was to obtain showing projectile prediction model. After simulation test, the results showed that the prediction error of range was less 0.03 m, and lateral deviation forecast error is about 4 10-5m, and the average prediction time was 3.506 ms.In summary, the prediction accuracy of the high dimension interpolation method and the improved RBF neural network is most accurate, and the average prediction time of method based on linear trajectory and the high dimension interpolation method is the shortest. Therefore, it is feasible and effective for the proposed methods to predict projectile impact point in this paper, and they can provide a theoretical reference value.