Abstract: Aiming at problem of few samples and non-linear characteristics in diesel engine fault diagnosis, Least Squares Support Vector Machine（LSSVM） can be better to diagnostic studies, but the results of diagnosis are greatly influenced by the penalty factor and the selection of kernel parameters, it is necessary to optimize its parameters, LSSVM algorithm based on Binary Differential Evolution（BDE） is proposed. For the data of diesel engine valve vibration signal used as the characteristics values of model, and wavelet transformed, the fault diagnosis model based on BDE-LSSVM is established. Compared with LSSVM model based on the particle swarm algorithm and genetic algorithm, the results show that LSSVM with BDE has better fitness value and stability, as well as more perfect accuracy and speed in the diagnosis classification.

Key words: Least Squares Support Vector Machines（LSSVM）, Binary Differential Evolution（BDE）, fault diagnosis, wavelet transform

摘要： 针对柴油机气门故障的诊断样本少和非线性数据特征等问题，最小二乘法的支持向量机（LSSVM）能够较好地进行诊断研究，但由于惩罚因子[C]和内核参数[σ]的选取对诊断结果影响较大，有必要对其进行参数优化，因此提出了基于二进制微分进化算法（BDE）的最小二乘法支持向量机算法。利用柴油机气门振动信号作为数据，经小波变换作为模型特征，建立了基于BDE-LSSVM故障诊断模型，并与基于遗传和基于粒子群算法的LSSVM模型进行柴油机气门故障诊断的性能对比。比较结果证明，基于BDE优化的LSSVM模型在故障特征选取前后具有更好的适应度值和稳定度，故障分类准确性高且运算速度更快。

关键词: 最小二乘支持向量机, 二进制微分进化, 故障诊断, 小波变换