风电机组变桨距系统中基于并行非线性滤波的故障诊断方法研究

The fault diagnosis of variable pitch system is the fundamental issue for guaranteeing the stable operation of the wind power system. How to obtain the accurate online fault diagnosis in complex condition has become a key scientific issue to be solved urgently. The nonlinear filtering method is adopted in this project, and the fault dynamic diagnosis model of parallel nonlinear filtering is proposed to solve the accuracy of fault diagnosis under the condition of uncertain parameters and time-varying noise. Through the establishment of distributed parallel diagnostic model, the real-time performance of dynamic diagnosis system is to be solved. The fuzzy synthetic evaluation model, which is to be established with the residual correlation analysis by introducing the fuzzy logic inference theory, to construct the joint decision-making mechanism. The goal of this project is to establish a real-time diagnostic model for the variable pitch system of wind turbine and to provide an innovative idea and a novel method for the on-line diagnosis of the variable pitch system of megawatt rated wind power generation system. The research are expected to have an important value and significance to improve the application level of intelligent diagnosis technology in wind turbine..This study belongs to the interdisciplinary domain of mechanical system dynamics, control theory and mathematics, which will play an important foundation role for the establishment of the theoretical bases on dynamic fault diagnosis for our country.

变桨距系统的故障诊断对于保障风电系统的稳定运行至关重要，如何在复杂工况下获得准确的在线故障诊断方法，已经成为迫切需要解决的关键科学问题。本项目采用非线性滤波方法，拟构建并行非线性滤波的动态故障诊断模型，解决参数不确定和噪声时变等条件下故障诊断的准确性问题。通过建立分布式并行故障诊断模型，解决动态诊断系统的实时性。基于模糊逻辑推理理论，建立基于残差关联度分析的故障模糊综合评判模型，形成故障的联合决策机制。预期目标是建立风电机组变桨距系统的实时故障诊断模型，欲为我国兆瓦级风电机组变桨距系统的在线诊断提供一种理论方法。预期研究成果对提升我国智能诊断技术在风电机组中的应用水平具有重要价值及意义。.本研究属于机械系统动力学、控制理论和数理学科等学科交叉问题。它对建立我国动态故障诊断建模的理论体系，有重要的基础作用。

复杂工况下研究准确实时的故障诊断方法对于风力发电机组以及各类工业设备的状态监测和诊断具有重要的研究意义和应用价值，多年来均受到了国内外学者的广泛重视。本项目主要针对故障诊断中复杂噪声干扰和故障状态的不确定性以及诊断的实时性问题进行了深入研究。首先，分析了实际工况中噪声的未知和时变特性，研究了噪声统计特性自适应更新的非线性滤波方法，并从重采样和粒子退化问题出发，提出了多种改进算法。然后，在此基础上，在 CUDA 技术架构下进行并行程序设计，优化内存访问原则、解决任务间的数据依赖性，完成并行滤波算法的优化设计。以IMM算法为基础，优化设计了自适应故障决策机制，实现了多故障的检测和识别，并以贝叶斯推理为基础，考虑故障特征的耦合性和不确定性，探索了基于贝叶斯网络的故障融合决策方法。最后，基于SCADA数据和建立的风机仿真平台进行了实验分析和说明，为复杂设备在线故障诊断技术的发展提供了一定的理论基础和借鉴。