风电测控系统的集值观测故障检测与虚拟容错方法研究

Sensors and actuators for measurement and control of large wind turbine usually have high rate of fault, which causes the decrease of both quantity and quality of power generation, also increases mechanical load. Concerning the fact that strong nonlinear features, the test error, difficulty of parameterized description and the uncertainty caused by model error and etc., high-precision detection of sensor actuators in wind power system is a tough problem to be solved in this field. Project the main research contents are as follows. Firstly, we study the functional redundancy of the wind power equipment, and build an information database about the failure and the functional redundancy, and explore the unified linear time-varying parameter description model connecting with model error, stochastic disturbance, test error, fault and etc. Secondly, according to the difficulty in setting appropriate thresholds, dealing with the model error, and noise and disturbance parameterization in fault detection method focused on threshold of residual, our research propose a fault detection and isolation method based on set observer, which applies to unknown bound hypothesis. Meanwhile, in order to meet the needs of online detection, we make the research of active fault excitation and detection, together with the improved fast algorithm. Thirdly, in order to make the early failure of wind turbine can be self-healing, we use virtual sensor actuators to deal with fault tolerance, instead of changing the original controller. This research will build the theoretic, methodological and technological system involving the unified linear time-varying description model, set-value fault detection, virtual tolerance and etc. to support the improvement for reliability and availability of wind turbine.

风电测控传感器执行器故障率很高，导致发电量、电能品质下降，机械载荷增大。强非线性特征、测试误差和扰动参数化描述困难、以及模型误差不确定性等因素，使其故障的高效准确检测一直是风电行业亟待解决的难题。本项目将主要研究：（1）揭示风电测控系统功能冗余和故障机理，建立其故障及功能冗余信息库；探索综合模型误差、随机扰动、无规律测试噪声、故障因素等的线性时变参数统一描述模型。（2）针对残差阈值故障诊断方法中设定合理阈值、处理模型误差、参数化噪声及扰动的难题，以未知有界形式研究基于集值观测的故障检测、故障隔离方法，并利用主动故障检测和快速算法实现工程在线检测。（3）探索不改变原有控制器，以虚拟传感器和执行器实现容错控制的方法，实现早期故障时风力机自愈可用。项目将建立风电测控环节线性时变参数统一描述模型、集值故障诊断、虚拟容错等理论方法技术体系，为提高风电设备可靠性、可用性提供理论和技术支撑。

针对风电系统存在的非线性、噪声及扰动难以参数化、模型存在不确定性等问题，开展大型风电机组典型故障检测和容错处理的理论及方法研究。主要工作有：（1）在系统故障、功能冗余分析及模型方面。利用模型分解、故障传播分析、失效模式效应分析、故障评估、结构分析、故障分类、维护行为选择等方法，揭示测控系统传感器和执行器的故障机理，建立故障模型和故障信息库。利用机电分析动力学理论，探索风力机非线性机电耦合机理，建立控制系统动力学模型，利用线性时变参数建模方法，确定调度参数和边界，建立非线性、模型误差、随机扰动、无规律测试噪声、故障状况等因素的LPV统一描述模型。为诊断风电传动系统故障提出了风电机齿轮、轴承弹流润滑仿真建模方法。（2）在风电故障诊断方法方面。研究扰动测试误差随机性、模型误差及故障因素的线性时变参数描述方法，设计故障正常状态集值观测器，实现基于LPV集值观测的故障检测、故障隔离。研究了分类时频图像特征张量、方向自适应核分布、定向循环解调的用于风机主轴系统故障诊断方法。研究了自适应特征选择和新模型的变预测模型、盲变分模态分解结合倒谱包络、递归定量分析和投票法多变量预测模型等用于风机轴承故障诊断方法。（3）在风电故障虚拟容错方法及评价方面。研究了风电齿轮箱故障非线性特征测度及状态TWSVM辨识、小波核极值学习自适应粒子群优化故障严重度识别的风电齿轮箱故障性能衰退评估方法。研究在时变参数统一框架下，采用虚拟传感器和执行器实现容错处理的理论和方法，实现无需重新设计控制器的故障容错处理。项目研究成果将从源头上减少风电机故障，避免风电系统重大事故的发生和早期故障的扩大化；将提高风力机的可靠性和可用性，具有重要的科学理论意义和工程应用价值。