Gas anomaly data detection based on hybrid neural network

Abstract:
The accurate measurement of gas flow is directly related to the economic benefits and social interests of gas enterprises. Moreover, as a kind of time-series data, there is a serious imbalance in the distribution of positive and negative samples in the data, which leads to low accuracy and difficulty in detecting gas anomaly data. To address these problems, a cost-sensitive hybrid network model is proposed in this paper. The model combines the stronger local feature learning ability of CNN network and the better sequence feature learning ability of GRU network, and then introduces a cost-sensitive loss function to solve the problem of inaccurate classification accuracy due to skewed data distribution. The experimental results show that using this model can find the outliers in the measurement data more accurately, which has some reference significance for the accurate measurement of traffic.