基于改进PSO-SVM参数优化的发动机起动过程辨识  

Engine Start Identification Based on Parameter Optimization of Improved PSO-SVM

作　　者：王冠超 杨春[2] 徐波[3] 王文博[4] 孙波[2] 

机构地区：[1]海军驻长春地区航空军事代表室 [2]海军航空工程学院研究生管理大队 [3]91467部队 [4]92514部队

出　　处：《燃气涡轮试验与研究》2011年第1期35-41,共7页Gas Turbine Experiment and Research

摘　　要：针对影响支持向量机辨识性能的核函数及相关参数,找出使辨识结果最佳的核函数;结合两种措施改进粒子群算法,优化相关参数,选择最佳的参数组合。对比BP神经网络和支持向量机对发动机起动过程的辨识结果,得到支持向量机的辨识精度和收敛时间优于BP神经网络,与起动数据基本一致。在训练样本存在噪声的情况下,验证了所建辨识模型具有很强的泛化能力。基于所建模型,计算了发动机的起动性能,其结果与起动数据吻合较好。该方法对发动机起动性能计算具有一定的理论指导和应用价值。For kernel function and relative parameters of affecting SVM identification performance, the best result has been found. Relative parameters are optimized by combining improved PSO with two methods. Parameters affecting identification performance and the best parameter selection are defined. Contrasting engine start identification results of BP neural network and SVM, identification precision and converage time of SVM which are consistent with starting data are superior to BP neural network. And generalized performance of identification model is good. Based on the model, engine start performance is calculated, and the results are good agreement with test-drive data. This method is instructional for start performance calculation.

关 键 词：发动机起动 BP神经网络 支持向量机 辨识 改进粒子群 

分 类 号：V235.13[航空宇航科学与技术—航空宇航推进理论与工程]