基于粗集—神经网络的风电机组状态评估及齿轮箱异常预警

发布时间：2018-04-02 06:39

本文选题：风电机组　切入点：状态评估　出处：《华侨大学》2014年硕士论文

【摘要】：随着全球风电产业的迅速发展，风电机组装机容量大幅增加，结构日趋复杂，运行环境复杂多变，使得机组的运行与维护面临着前所未有的挑战。实时评估机组状态，对异常运行状态及时识别并进行预警，，可以保障机组安全高效地运行，对优化检修策略具有重大意义。在充分利用SCADA系统提供的大量监测数据，且不增加额外传感器等其他监测设备的情况下，综合考虑监测项目之间的关系，建立了风电机组实时状态评估模型及齿轮箱异常预警模型。利用粗糙集约简方法选择模型的特征参数，即在平凡约简的基础上，利用属性重要度为启发信息再进行二次约简，并结合自然算法、半自然算法及等频率法的离散化方法和遗传算法、约翰逊算法的约简算法的约简结果提取出模型的特征参数。以有功功率为决策属性，SCADA系统其他监测量为条件属性，约简得到机组实时状态评估模型的特征参数；分别以齿轮箱输入轴温度、齿轮箱输出轴温度以及齿轮箱油温等三个齿轮箱监测项目为决策属性，SCADA系统中其他监测量为条件属性，约简后得到齿轮箱异常预警模型的特征参数。建立机组实时状态评估模型，结合SCADA越限报警系统，与有功功率预测模型得到的预测值与实际值的差值来实时评估机组状态。经实际监测数据验证，基于粗糙集与TS模糊神经网络建立的预测模型优于BP神经网络和SVM模型的预测精度，并可以准确识别机组处于异常运行状态。齿轮箱异常预警模型基于齿轮箱输入轴温度、输出轴温度以及齿轮箱油温等三个温度变量的预测模型构建。与以往研究齿轮箱温度量变化趋势预测仅从该温度量前一段时间的若干个时间序列值入手分析不同，本文利用粗糙集属性约简算法引入了对齿轮箱温度量有影响的其他监测量，同时也加入其上一时刻的温度值作为特征参数，建立TS模糊神经网络预测模型，以输出温度量预测值与实际值的差值是否越限来实现齿轮箱异常预警目的。本文提出的机组实时状态评估方法和齿轮箱异常预警方法均是基于风电机组SCADA系统的真实的监测数据，可操作性强，且具有良好的推广性。
[Abstract]:With the rapid development of the global wind power industry, the installed capacity of wind turbine units is increasing dramatically, the structure is becoming more and more complex, and the operating environment is complex and changeable, which makes the operation and maintenance of wind power units face unprecedented challenges.Evaluating the status of the unit in real time, identifying and warning the abnormal operating state in time can ensure the safe and efficient operation of the unit, and it is of great significance to optimize the maintenance strategy.Taking full advantage of the large amount of monitoring data provided by the SCADA system and without adding additional sensors and other monitoring equipment, taking into account the relationship between the monitoring items,The real-time evaluation model of wind turbine and the gearbox anomaly warning model are established.The rough set reduction method is used to select the characteristic parameters of the model, that is, on the basis of trivial reduction, using attribute importance as the heuristic information, the quadratic reduction is carried out, and the natural algorithm is combined.The discretization method and genetic algorithm of semi-natural algorithm and equal frequency method, and the reduction result of Johnson algorithm are used to extract the characteristic parameters of the model.Taking active power as decision attribute and other monitoring parameters of SCADA system as conditional attributes, the characteristic parameters of the real-time state evaluation model of the unit are obtained, and the temperature of the shaft is input to the gearbox, respectively.Three gearbox monitoring items, such as gearbox output shaft temperature and gearbox oil temperature, are decision attributes and other monitoring quantities in SCADA system are conditional attributes. After reduction, the characteristic parameters of gearbox anomaly warning model are obtained.The real-time state evaluation model of the unit is established, and the difference between the predicted value and the actual value obtained from the prediction model of active power and the difference between the predicted value and the actual value are combined with the SCADA over-limit alarm system to evaluate the unit state in real time.The prediction model based on rough set and TS fuzzy neural network is better than that of BP neural network and SVM model, and it can accurately identify the unit in abnormal operation state.The gearbox anomaly warning model is constructed based on three temperature variables, namely, the temperature of the gearbox input shaft, the temperature of the output shaft and the oil temperature of the gearbox.Different from the previous research on the prediction of gearbox temperature change trend only from several time series values of the previous time series of the gearbox temperature quantity, this paper introduces other monitoring quantities which have influence on the gearbox temperature quantity by using the rough set attribute reduction algorithm.At the same time, the temperature value at the last moment is added as the characteristic parameter, and the TS fuzzy neural network prediction model is established to realize the goal of gearbox anomaly warning by whether the difference between the predicted value of output temperature quantity and the actual value exceeds the limit.Both the real-time status evaluation method and the gearbox anomaly warning method proposed in this paper are based on the real monitoring data of wind turbine SCADA system. They are easy to operate and have good generalization.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM315

【参考文献】