基于遗传算法优化LSSVM的初速预测

Muzzle Velocity Prediction Based on Genetic Algorithm Optimized LSSVM

利用两台初速雷达测试弹丸炮口初速的试验中,当一台雷达的数据出现缺失时,通过建模预测出缺失的数据成为一项必要的工作。预测初速主要采用GM(1,1)模型,但是该模型并不完全适合预测初速,所以预测精度不理想。通过深入分析两台雷达之间的关联性,选择把两台雷达的数据进行融合,同时根据弹丸初速自身的特征,选择建立遗传算法优化LSSVM对缺失的数据进行预测。实验验证时,选择ARIMA模型、GM(1,1)模型、支持向量回归机、BP神经网络作为对比模型,两次验证的结果表明,遗传算法优化LSSVM的预测精度最高,误差远小于1‰,达到了初速雷达测试弹丸初速的误差标准,是预测弹丸初速的最佳模型。

In the test of using two muzzle velocity radars to measure muzzle velocities of projectiles, when the data of one radar is missing, it is necessary to predict the missing data through modeling. GM(1,1) model is mainly used to predict muzzle velocity, but the model is not completely suitable for the job, and the prediction accuracy is not ideal. Through an in-depth analysis of the correlation between two radars, we chose to fuse the data of the two radars. At the same time, according to the characteristics of projectile muzzle velocity, we chose to use the genetic algorithm optimized LSSVM(GA-LSSVM) to predict the missing data. During the experimental verification, the ARIMA model, GM(1,1) model, support vector regression machine and BP neural network were selected as comparison models. The results of two verification tests show that the GA-LSSVM has the highest prediction accuracy, and the error is far less than 1‰, which meets the error standard of muzzle velocities measured by muzzle velocity radars and is the best model for predicting muzzle velocity.