基于局部离群因子和神经网络模型的设备状态在线监测方法研究

Study on Online Monitoring of Equipment Condition Based on Local Outlier Factor and Artificial Neural Networks Model

核电厂重大设备状态在线监测是保障核电厂安全和经济运行的重要技术,针对传统阈值监测的固有缺陷,提出一种基于局部离群因子(LOF)和神经网络模型的设备状态在线监测方法。此方法属于多参数动态阈值监测方法,首先分析监测对象的故障模式和故障现象,选择一组可覆盖故障现象的传感器测点;根据设备运行特点采集足够长时间的历史运行数据,筛除异常数据;计算历史运行数据的LOF,以历史运行数据为输入、LOF为输出,建立并训练得到神经网络模型;最后基于神经网络模型和传感器测点实时数据计算设备健康指数,监控当前设备健康状态。将本文的监测方法用于循环水泵泵体健康状态的监测,并采集了一段时间的正常数据和异常数据以验证其监测效果,验证结果表明,本文提出的监测方法可以提前10 d进行预警,降低误报率,大幅提升监控效能。

The centralized online monitoring technology plays the most important role in nuclear power plants for the safety of major equipments and economic operation.In order to solve the false alarm and alarm failure problems in the traditional online monitoring,a new artificial intelligence monitoring method based on the local outlier factor and artificial neural networks model is put forward in this paper.This method is one of the multiple parameter dynamic threshold detection method.Firstly,a group of monitoring parameters of equipment is selected by analyzing the failure modes and failure phenomena of equipment.Secondly,enough data of this group of parameters needs to be collected and the abnormal data needs to be screened out.Thirdly,all the selected data is used to calculate the local outlier factor,and then the neural network model will be established by inputting the selected data and the local outlier factor.Finally,the neural network model can be used to assess the equipment health index with the real-time data of equipment parameters as input,and the health index represents the real-time health of equipment.In this paper,this method is used to develop a monitoring model of circulating water pump.In order to verify the validity of the model,enough monitoring data of healthy equipment and malfunction equipment are used to verify the monitoring results.The results show that the method can provide a pre-alarm for the early failure of the equipment with low false alarm rate,greatly improving the monitoring efficiency.