基于核密度估计的船舶柴油发电机组冷却水温度异常检测方法

Abnormal Temperature Detection Method of Marine Diesel Generator Cooling Water Based on Kernel Density Estimation

对船舶柴油发电机组基于冷却水高温阈值开展异常检测的方法进行改进,基于核密度估计法采用多阈值构建全工况下的异常检测模型。根据冷却系统工作原理,借鉴故障树分析法(FTA)对引发冷却水温度高报警的原因进行分析,确定本文使用的异常检测特征参数。在此基础上,基于核密度估计法(KDE)分别求取各特征参数在不同运行工况下的报警阈值,进而通过拟合构建全工况的异常检测模型;利用柴油发电机组的冷却水温度异常升高的实际运行数据进行验证,选择混淆矩阵和接受者操作特性(ROC)曲线作为检测准确度的性能验证指标。结果表明,文中提出的冷却水温度异常检测方法有较高准确度,可以进一步指导实际工程应用。

The abnormal detection method of Marine diesel generator set based on the high temperature threshold of cooling water temperature is improved,and the abnormal detection model under all working conditions is constructed by using multi-threshold based on kernel density estimation method.According to the working principle of cooling system,fault tree analysis(FTA)is by reference to analyze the causes of high alarm of cooling water temperature,and to determine the abnormal detection characteristic parameters.On this basis,the alarm thresholds of each characteristic parameters in different operating conditions are obtained based on kernel density estimation(KDE),and then the anomaly detection model of the whole operating conditions is constructed.The model is verified by comparing with the actual operation data of abnormal increasing of cooling water temperature of diesel generator set,confusion matrix and the ROC curve are chosen as the detecting accuracy index of performance verification.The results show that the implementation of this study of cooling water temperature anomaly detection method has relatively high accuracy,which can be the guideline to the practical engineering application.