摘 要：为提高汽轮机转子故障诊断的准确率和识别效率，提出基于云粒子群优化算法（CPSO）优化支持向量机（SVM）的故障诊断方法。首先利用补充总体平均经验模态分解（CEEMD）对转子振动信号进行分解，利用能量法筛选出更为有效的固有模态分量（IMF）并计算对应的排列熵（PE）作为故障特征值；其次将云理论引入到粒子群优化算法（PSO）中得到CPSO算法，通过CPSO算法优化SVM得到诊断模型。在ZT-3试验台对汽轮机转子常见4种故障（正常状态、转子不平衡、转子不对中和动静碰磨状态）状态进行模拟实验，获取故障数据后进行故障识别研究。研究表明：在相同测试样本的条件下， CPSO-SVM诊断模型的识别准确率为95%，比PSO-SVM诊断模型提高了5%，运行时间为22.055 s，比PSO缩短了14.5 s。研究结果验证了CPSO-SVM算法在汽轮机转子故障诊断方面的优越性。

Abstract:To enhance the accuracy rate and identification efficiency to diagnose any fault occurred to rotors of steam turbines,a fault diagnostic method based on the cloud particle swarm optimization algorithmoptimized supporting vector machine was put forward.Firstly,the complementary ensemble empirical mode decomposition (CEEMD) was utilized to decompose the vibration signals from a rotor and the energy method was used to sift out more effective intrinsic mode function (IMF) components and calculate the corresponding permutation entropy (PE) to serve as the fault characteristic value.Secondly,the cloud theory was introduced into the particle swarm optimization algorithm to obtain a cloud particle swarm optimization (CPSO) algorithm and also acquire a diagnostic model through optimizing the supporting vector machine by using the CPSO algorithm.On the ZT-3 test stand,four faulty states commonly seen in rotors of steam turbines (normal state,rotor unbalance,rotor misalignment and rotorstator rubbing state) were simulated and tested to obtain faulty data and perform a study of the fault identification.It has been found that under the condition of identical test specimens,the identification accuracy rate by using the CPSO-SVM diagnostic model can hit 95%,increasing by 5% when compared with that by using the PSO-SVM diagnostic model and the calculation operation time duration being 22.055 s,14.5 s shorter than that by using the PSO diagnostic model.It has been verified that the CPSO-SVM algorithm is superior in diagnosing any fault occurred to rotors of steam turbines.

TK267

国家自然科学基金（51576036）；吉林省科技发展计划（20100506）