摘要
针对多级同步感应线圈发射器建模涉及多物理场耦合、现有优化方法迭代时间长等问题,基于粒子群优化-循环神经网络(Particle Swarm Optimization-Recurrent Neural Network,PSO-RNN)算法建立多级同步感应线圈发射器非参数模型,并进行电枢出口速度预测。通过正交结合随机实验的方法,获得以线圈匝数、触发时间、触发位置为输入,出口速度为输出的样本集;采用循环神经网络算法对样本集进行训练并建立非参数模型;通过粒子群优化算法进一步优化RNN神经网络参数,提高非参数模型的预测性能;采用建立的模型预测出口速度并与实验结果对比。结果表明:所建立非参数模型的均方预测误差、平均绝对百分比误差、均方根误差分别为0.0028、0.036、2.18,且经过PSO优化后模型的3项评价指标分别降低39%、38%、46%,提高了预测性能;PSO-RNN非参数模型的一致性较好且预测的平均值与实验测得的出口速度相差1.2 m/s,误差百分比为1.8%,小于标准值5%。将PSO-RNN算法用于同步感应线圈发射器的非参数建模可行且对出口速度的预测较为准确,可为多级同步感应线圈发射器的工程设计提供新思路。
An non-parametric model of multi-stage synchronous induction coilgun(MSSICG)based on the particle swarm optimization and recurrent neural network(PSO-RNN)algorithm is proposed to solve the problems such as multi-physics field coupling and long iteration time of existing optimization methods.And the ejection velocity of the armature is also predicted by the model.A sample set with the turns per coil,triggering time and trigger position as inputs and the ejection velocity as output is obtained through the orthogonal and random experiments.The RNN algorithm is used to train the sample set and the non-parametric model is established.The parameters of the RNN model are further optimized by the PSO algorithm,to improve the prediction performance of the non-parametric model.The ejection velocity of the armature is predicted using the proposed PSO-RNN model and compared with the experimental result.The MSPE,MAPE,and RMSE of the non-parametric model are 0.0028,0.036,and 2.18,respectively,which are reduced by 39%,38%,and 46%after the optimization of PSO.The difference between the predicted and experimental velocities is 1.2 m/s with the error percentage of 1.8%,which is less than 5%.The study provides a novel idea for the modelling and engineering design of MSSICG.
作者
秦涛涛
季思源
雷琳
郑占锋
QIN Taotao;JI Siyuan;LEI Lin;ZHENG Zhanfeng(National Key Laboratory of Transient Physics,Nanjing University of Science and Technology,Nanjing 210094,Jiangsu,China;Zaozhuang Power Supply Company,State Grid Shandong Electric Power Company,Zaozhuang 277000,Shandong,China;State Grid Electric Power Research Institute Co.,Ltd.,Nanjing 210014,Jiangsu,China)
出处
《兵工学报》
北大核心
2025年第7期87-97,共11页
Acta Armamentarii
基金
国家自然科学基金项目(51707094)。
关键词
多级同步感应线圈炮
非参数模型
循环神经网络
粒子群优化
出口速度预测
multi-stage synchronous induction coilgun
non-parametric model
recurrent neural network
particle swarm optimization
muzzle velocity
