| 引用本文: | 赵洪山,刘辉海.基于性能改善深度置信网络的风电机组主轴承状态分析[J].电力自动化设备,2018,(2): |
| ZHAO Hongshan,LIU Huihai.Condition analysis of wind turbine main bearing based on deep belief network with improved performance[J].Electric Power Automation Equipment,2018,(2): |
|
| 摘要: |
| 针对风电机组数据采集与监视控制系统采集的状态数据具有大容量、多样性的特点，充分利用该数据研究风电机组主轴承的状态分析方法成为了重要问题。采用深度学习方法分析风电机组主轴承变量间的特征规则，提取反映主轴承状态的特征变量；通过指数加权移动平均法设定阈值检测特征变量的变化趋势，判定异常状态的发生；根据深度置信网络的特点，从数据集变量的异常数据剔除、训练数据批次的选择、参数调优的迭代周期以及在线学习训练等方面对模型性能进行优化和改善，从而使得深度置信网络能够充分挖掘数据集的信息特征，达到有效地反映主轴承状态的目的。通过对主轴承发生故障前、后记录的数据进行仿真分析，结果验证了深度置信网络方法对主轴承状态监测的有效性。 |
| 关键词: 风电机组 主轴承 状态分析 深度学习 深度置信网络 |
| DOI：10.16081/j.issn.1006-6047.2018.02.006 |
| 分类号:TM614 |
| 基金项目:国家科技支撑计划项目(2015BAA06B03) |
|
| Condition analysis of wind turbine main bearing based on deep belief network with improved performance |
|
ZHAO Hongshan, LIU Huihai
|
|
School of Electrical & Electronic Engineering, North China Electric Power University, Baoding 071003, China
|
| Abstract: |
| The condition data collected by supervisory control and data acquisition system of wind turbine has the characteristics of large capacity and diversity, based on which to analyze the condition of wind turbine main bearing has becoming an important research point. The characteristic rules of the main bearing variables of wind turbine are analyzed by deep learning method to extract the characteristic variables reflecting the condition of main bearing. Threshold value is set by exponentially weighted moving average approach to detect the variation trend of characteristic variables and determine the occurrence of abnormal conditions. According to the characteristics of deep belief network, the performance of the model is optimized and improved from aspects of: abnormal data exclusion from data sets, selection of training data batches, iteration cycles of parameter optimization, on-line learning and training and so on, so that the information characteristics of data sets are fully excavated by deep belief network to achieve the purpose of reflecting main bearing's condition effectively. The recorded data of the main bearing with and without faults are simulated, and the results verify the effectiveness of the deep belief network method on bearing's condition monitoring. |
| Key words: wind turbines main bearing condition analysis deep learning deep belief network |
