基于噪声参数最优ELMD方法多核多特征融合的风电机组齿轮箱状态监测关键技术研究

The safe operation of wind turbine has been attracted much attention. The project targets vibration signals in wind turbine gearbox. In order to improve the accuracy of wind turbine gearbox condition monitoring, exploratory study based on optimal noise parameter ensemble local mean decomposition (ELMD) method and based on feature fusion in multi-kernel relevance vector machine (RVM) pattern recognition method is presented. For modal aliasing phenomenon in ELMD method, the search of the noise mechanism for more accurate parameters, to find the optimal parameters of adding noise and to minimize the modal aliasing are done; Based on these researches, as feature vectors energy entropy, singular value entropy and approximate entropy of the PF component will be get, and then a suitable kernel function respectively will be found. After building combination kernel function, the classification of conditions, in which informations of conditions are more comprehensive and classification is more accurately, will be fulfilled; In the process of multiple features fusion, a new kind of particle swarm optimization (PSO), which is based on human cognitive psychology named self regulating particle swarm optimization (SRPSO), will be presented to find the combination kernel function and to build the relevance vector machine model. In the project, based on the demonstration of optimal noise parameters targeting technology role in promoting the restraint of modal aliasing phenomenon and the discussion of a new strategy of a new method of combined kernel function, the process of wind turbine gearbox condition monitoring with high accuracy will be explored.

风力发电机组的安全运行问题已受到广泛关注。本项目以风电机组齿轮箱振动信号为研究对象，探索性地研究基于噪声参数最优的总体平均局部均值分解（ELMD）方法和基于多核多特征融合的相关向量机（RVM）模式识别方法，以期提高风电机组齿轮箱状态监测的精度。针对ELMD方法中的模态混叠现象，探寻更为准确的噪声参数机理，找到加入噪声的最优参数，使模态混叠程度降到最低；以此为基础，将得到各个PF分量的能量熵、奇异值熵和近似熵作为特征向量，分别找到适合的核函数，构建组合核函数，完成状态信息更加全面、分类更加准确的状态分类；在多特征融合过程中，研究一种基于人类认知心理学学习原则的新粒子群算法—自我调节粒子群优化算法（SRPSO），找到组合核函数，构建多核多特征融合的相关向量机模型。以论证最优噪声参数对抑制模态混叠的提升作用，探讨基于SRPSO算法的组合核函数构建的新方法，以期完成高精度风电机组齿轮箱的状态监测。

风力发电机组的安全运行问题已受到广泛关注。本项目以风电机组齿轮箱振动信号为研究对象，研究了基于噪声参数最优的总体平均局部均值分解（ELMD）方法和基于多核多特征融合的相关向量机（RVM）模式识别方法，提高了风电机组齿轮箱状态监测的精度。针对ELMD方法中的模态混叠现象，探寻更为准确的噪声参数机理，找到加入噪声的最优参数，使模态混叠程度降到最低；以此为基础，将得到各个PF分量的能量熵、奇异值熵和近似熵作为特征向量，分别找到适合的核函数，构建组合核函数，完成状态信息更加全面、分类更加准确的状态分类；在多特征融合过程中，研究了一种基于人类认知心理学学习原则的新粒子群算法—自我调节粒子群优化算法（SRPSO），找到组合核函数，构建多核多特征融合的支持向量机模型。论证了最优噪声参数对抑制模态混叠的提升作用，探讨基于SRPSO算法的组合核函数构建的新方法，完成了高精度的风电机组齿轮箱的状态监测过程。