基于HI-HMM的风电机组齿轮箱运行状态评估

Operation State Evaluation ofWind Turbine Gearbox Based on HI-HMM

齿轮箱是风电机组的重要组成部件,准确获取其运行状态及劣化趋势对提高风电机组的运行可靠度至关重要。本文提出基于健康指数(health index,HI)与隐马尔可夫模型(hidden Markov model,HMM)的风电机组齿轮箱状态评估与分析的新方法。利用HI获得齿轮箱历史温度数据的特征观测序列,分别对不同状态的HI-HMM模型进行训练,建立模型库。利用HI-HMM模型库对待评估样本进行识别,根据最大对数似然概率值判断齿轮箱的状态。最后,利用该方法对齿轮箱故障前的状态进行评估与分析,其状态变化与实际完全吻合。与随机森林、支持向量机、概率神经网络和BP神经网络方法进行对比,本文方法状态识别的准确度达到97%。该方法能准确识别齿轮箱状态,获取劣化趋势,为风场确定故障预警时间提供参考。

The gearbox is an important component of the wind turbine.Operating state and deterioration trend are obtained accurately and it is essential to improve the operating reliability of the wind turbine.A new evaluation and analysis method based on health index and hidden Markov model was proposed.HI was used to obtain the observation sequence of historical data.The HI-HMM containing different states was trained.The HI-HMM library was established.The samples to be evaluated were identified based on HI-HMM library and the state of the gearbox was decided according to the maximum log-likelihood value.Finally,the method was used to analyze the state of the gearbox before failure.The state change of the gearbox was consistent with the actual.Compared with the random forest,support vector machines,probabilistic neural network and BP neural network,the accuracy of state identification is 97%.The method can accurately identify the state of the gearbox and obtain its deterioration trend.It also provides a reference for the wind farm to determine the warning time of failure.