基于孤立森林与KDE-LOF的冷水机组故障检测

FAULT DETECTION FOR CHILLER BASED ON ISOLATED FOREST AND KDE-LOF

为了提升冷水机组微小故障的检测率,提出一种核密度估计的局部异常因子算法(KDE-LOF)结合孤立森林(iForest)的冷水机组故障检测策略。该策略通过使用孤立森林对实验数据异常值进行剔除,计算正常数据的LOF值作为统计量,并使用KDE确定控制限来完成模型的训练。通过监测数据LOF值是否超过设定的控制限进而判断是否出现故障。采用ASHRAE RP-1043数据集进行验证,并分析了该方法与主元分析和单类支持向量机的方法的优劣,结果表明该方法检测效果要优于其他两种模型,该方法在微小故障下检测率超过80%,性能最佳。

To improve the minor fault detection rate of the chiller, this paper proposes a local outlier factor algorithm based on kernel density estimation(KDE-LOF) combined with isolated forest(iForest) fault detection strategy of water chillers. This strategy removed outliers from experimental data by using isolated forests, and calculated LOF values of normal data as statistics. And it used KDE to determine the control limit to complete the training of the model. A fault was judged by monitoring whether the LOF value of the data exceeded the set control limit. ASHRAE RP-1043 data sets were adopted for validation. And this paper analyzed the advantages and disadvantages among this method, PCA and SVM. The results show that the proposed method is better than the other two models in detection performance. Its detection rate is over 80% for minor faults, and its performance reaches the best.