To achieve the online precise and advanced identification of abnormal gas volume fraction,an anomaly identifica-tion method of gas volume fraction based on multivariate heterogeneous data fusion(KPCA-BiLSTM-iForest)was pro-posed.The kernel principal component analysis(KPCA)was applied for dimensionality reduction and feature extraction of nonlinear data,then the main information was extracted,and the computation amount was reduced.The prediction of gas vol-ume fraction was conducted on the reduced-dimension data by using the bidirectional long short-term memory neural network(BiLSTM),and the anomaly detection was carried out according to the prediction results and actual values by using the isola-tion forest(iForest).The results show that this method can detect the gas volume fraction anomaly 20 minutes in advance,and the accuracy of anomaly identification can be improved by more than 3 percentage points compared with the KPCA-LSTM-iForest method,the KPCA-iForest method and the KPCA-BiLSTM-LOF method.The results of the study can pro-vide a basis for identifying gas volume fraction anomalies and providing early warnings.
维普
万方数据