基于PSO-BP神经网络的固冲发动机推力估计器设计  

Design of thrust estimator in the solid rocket ramjet based on PSO-BP neural network

作　　者：王昭 田小涛 黄萌 张博 WANG Zhao;TIAN Xiaotao;HUANG Meng;ZHANG Bo(Xi'an Modern Control Technology Research Institute,China North Industries Group Corporation Limited,Xi'an 710065,China;Key Laboratory for Modern Control Technology,China North Industries Group Corporation Limited,Xi'an 710065,China)

机构地区：[1]中国兵器工业集团有限公司西安现代控制技术研究所,西安710065 [2]中国兵器工业集团有限公司现代控制技术重点实验室,西安710065

出　　处：《航空动力学报》2022年第7期1487-1494,共8页Journal of Aerospace Power

摘　　要：为了对固体火箭冲压发动机的推力进行直接控制,提出了基于PSO(粒子群优化算法)优化BP(back propagation)神经网络的固体火箭冲压发动机推力估计方法。采用SPSO(标准PSO)和三种不同的BBPSO(骨干PSO)寻优神经网络权值,而后以最优权值进行BP网络训练,对其进行精调,如此便可获取推力和燃气流量、飞行马赫数以及飞行高度之间的非线性关系,从而完成推力估计器的设计。利用240组训练集数据对网络进行训练,并用180组测试集数据对网络进行验证。仿真结果表明:在SPSO、BBExp(exploiting BBPSO)、ABPSO*(modified adaptive BBPSO)和SNPSO(simplified pruning strategy based BBPSO)等四种不同的PSO中,基于SNPSO优化BP神经网络实现推力估计器设计是最为便捷有效的方法,不仅形式简单,而且对于测试集数据而言,其能够将推力相对误差控制在5%以内。A solid rocket ramjet thrust estimation method based on BP(back propagation)neural network optimized by PSO(particle swarm optimization)was proposed for direct control of the thrust of solid rocket ramjet.In detail,SPSO(standard PSO)and three different BBPSO(bare bones PSO)methods were adopted to optimize the network weights.Then,the optimal weights as the initial value were tuned finely by BP neural network training.Therefore,the nonlinear relationship between thrust and gas flow,flight Mach number as well as flight height was obtained such that the design of thrust estimator was completed.240 sets of training data were used to train the network,and 180 sets of testing data were used to verify the network.The simulation results showed that among four different PSO methods such as SPSO,BBExp(exploiting BBPSO),ABPSO*(modified adaptive BBPSO) and SNPSO(simplified pruning strategy based BBPSO),the design of thrust estimator based on BP neural network optimized by SNPSO is the most convenient and effective method not only due to its simple form but also due to its capacity of controlling the relative thrust error within 5% for test set data.

关 键 词：固体火箭冲压发动机 推力估计器 粒子群优化算法 神经网络 非线性关系 

分 类 号：V435.5[航空宇航科学与技术—航空宇航推进理论与工程]