Research And Application Of Gas Usage Abnormal Detection Based On Transfer Learning

With the advancement of science and technology and the improvement of living conditions,my country’s demand for energy is growing.Traditional fossil energy has been unable to meet the needs of production and life.In order to protect the environment and insist on sustainable development,new energy came into being.The development trend of replacing high-carbon and high-polluting coal with natural gas is inevitable,so the consumption of natural gas has also received attention.However,due to various factors,the development of the natural gas market is not perfect.The behavior of gas use is not only related to the storage of energy in my country,but also an important part of the production and operation of natural gas enterprises.Therefore,it is of great practical significance and application value to study the abnormal behavior of natural gas consumption.In this paper,we first find a suitable anomaly detection method to label the natural gas data set,and select two different unsupervised anomaly detection methods to preprocess the data.Dataset for detection,integrating the detection results of the two models and using a small number of manually labeled label corrections to generate final result labels.Secondly,a method of abnormal gas use behavior detection based on transfer learning is researched and proposed,which can use the information,knowledge and features of relevant datasets in the source domain to complete the task of the target domain.This paper selects similar time series as the source domain data.The feature vector extracted by the one-dimensional convolutional neural network is input into the long-term and short-term memory network model with the attention mechanism added,and the dynamic time warping algorithm is used to assign weights to the trained base learner,fine-tune the model,and finally realize the application of natural gas.Anomaly detection of gas behavior.Finally,through the comparison test with other models,effective evaluation is carried out according to five evaluation indicators: root mean square error,mean absolute error,mean relative error,F1 score and confusion matrix.The results show that the model proposed in this paper has the best effect.The model is applied to anomaly detection system of natural gas consumption behavior.The system is based on the hybrid model based on transfer learning proposed in this paper,and provides anomaly detection services for natural gas stations.It mainly includes abnormal detection,meteorological information,gas behavior analysis and information modules.Accurately detect the gas consumption behavior of the station,and improve the operation efficiency of the station’s intelligent terminal.