Abnormal Flow Detection Of Gas Meter Based On Improved Generative Adversarial Network

In recent years,the global attention to the gas industry has been increasing,and the proportion of gas in primary energy continues to rise.Accurate monitoring of the flow rate of users’ gas meters can help gas companies avoid economic disputes and reduce operation and maintenance costs.This paper combines the sensor data of gas meters collected by a gas company in a real scenario to conduct the following research on the detection of gas meter flow rate anomalies:Firstly,this paper applies the Savitzky-Golay filter to denoise the three-dimensional gas data and extract better sample features.To address the issue of imbalanced positive and negative samples,WGAN is used for data augmentation.Moreover,in order to better capture the dependency relationships between the three-dimensional data,a convolutional neural network is chosen as the classification model to enable cross-channel information interaction.In addition,in the process of anomaly detection,due to the large number of model parameters,manual parameter tuning requires a lot of time and increases model uncertainty.Therefore,this paper uses the particle swarm optimization algorithm to optimize the key parameters of the network model,thereby improving the accuracy of the model in detecting gas meter flow anomalies.Secondly,to address the issue of being unable to obtain abnormal samples or unknown abnormal samples in actual industrial scenarios,a method combining improved generative adversarial networks and a feature extraction module was adopted,using unsupervised learning.During the training process,the method learned the deep connections of the normal operation data of the three sensors in the gas meter,and further learned the data distribution characteristics and pattern features of the normal samples.During the detection process,the normal and abnormal samples were separated by setting a reconstruction error threshold,thus achieving abnormal detection.In order to make the network focus more on channels with high importance,a feature extraction module was added to improve the detection rate of the network for abnormal gas meter flow.Finally,this paper conducted simulation experiments on the dataset and compared it with other common algorithms.The experimental results show that the proposed method can construct an accurate and reliable framework for detecting anomalies in gas meter flow,more accurately detect anomalies in industrial users’ gas meter flow,and provide effective strategies for the operation and maintenance of gas companies.