基于Bagging集成策略和多元状态估计的风电机组齿轮箱状态监测

风电机组齿轮箱的故障率和维护成本相对较高,有必要对其运行状态进行实时监测。多元状态估计(multivariate state estimate technique, MSET)是一种常用的状态监测方法,但在记忆矩阵规模较大时,MSET在线计算的实时性较差。为此,提出一种基于Bagging集成策略和MSET的新方法:首先基于Bagging集成策略,对训练数据进行多次随机抽样,构造多个记忆矩阵规模较小的MSET子模型,最终将子模型的结果平均后作为集成模型的输出。以某2 MW风电机组的运行数据为算例,对集成MSET的性能进行了对比实验。结果表明:在精度相当的前提下,集成方法的计算时间仅为常规方法的60%;结合统计过程控制技术设计了预警阈值和滑动窗口异常率,并对集成MSET的故障预警能力进行验证,结果表明,集成方法能够提前约10 d预警齿轮箱的实际故障。

Wind Turbine Gearbox Condition Monitoring based on Bagging Ensemble Strategy and MSET

The failure rate and maintenance cost of wind turbine gearbox are relatively high, it is necessary to monitor its operation condition in real-time. Multivariate state estimation technique (MSET) is a commonly used condition monitoring method, he real-time performance of MSET online computing is poor when the memory matrix is large. To this end, a novel method based on Bagging ensemble strategy and MSET was proposed. Firstly, the training data are sampled randomly for several times based on Bagging ensemble strategy, and a subset of MSET with smaller memory matrix is constructed. Finally, the results of the subset are averaged as the output of the ensemble model. Taking the operational data of a 2 MW wind turbine as an example, the performance of ensemble MSET was compared. The results show that the calculation time of ensemble method is only 60% of conventional method under the premise of the same accuracy. The warning threshold and sliding window exception rate were designed based on statistical process control (SPC), and the fault warning ability of ensemble MSET was verified. The results show that the ensemble method can warn the actual faults of gearbox about 10 days in advance.