基于PSO-LSSVR的CNG气瓶损伤监测方法

A CNG Cylinder Damage Monitoring Method Based on PSO-LSSVR

针对车载气瓶损伤诊断问题,对气瓶诊断方法进行研究,提出基于最小二乘支持向量回归(LSSVR)以及粒子群优化算法(PSO)的气瓶损伤诊断方法,通过PSO进行LSSVR参数寻优,得到优化的LSSVR模型。采用随机振动的有限元计算方法,模拟多组损伤气瓶在运输中的情况,并收集气瓶振动时损伤位置的垂直加速度以及等效应力作为模型输入,得到气瓶损伤的诊断结果。以诊断值与实际值的均方根误差作为评判,采用PSO进行模型参数优化,将诊断误差稳定在1%以内,得出较为合适的优化模型。将该模型与未优化的LSSVR算法进行对比,可得出,该模型在低损伤以及高损伤位置识别精度较高。为进一步体现该模型优势,将其与BP神经网络以及支持向量回归(SVR)算法进行比较,结果显示,该模型在识别精度以及稳定性上都较高。

In this paper,focusing on the on-board cylinder damage diagnosis,the cylinder damage diagnosis method is studied,and a cylinder damage diagnosis method based on least square support vector regression(LSSVR)and particle swarm optimization algorithm(PSO)is proposed.The LSSVR parameters are optimized by PSO,and the optimized LSSVR model is obtained.The finite element method of random vibration is used to simulate the situation of multiple groups of damaged cylinders during transportation,and the vertical acceleration and equivalent stress of the damaged location are collected as the inputs to the model to obtain the diagnosis results of cylinder damage.Taking the root mean square error between the diagnostic value and the actual value as the criterion,PSO is adopted to optimize the model parameters,and the diagnostic error is stabilized within 1%,and a more appropriate optimization model is obtained.Compared with BP neural network,support vector regression(SVR)algorithm and unoptimized LSSVR algorithm,the results show that this model exhibits higher accuracy and stability in identification.