An abnormal diagnosis method for gas-metering systems based on wavelet transform and deep learning

In urban gas-distribution systems, gas-metering systems are critical for trade settlement and operational safety, and their performance directly affects billing accuracy and system reliability. However, due to limitations in measurement range and sensor precision, some abnormal conditions, such as the “gas flows but not recorded” phenomenon, are difficult to detect in time, which may cause economic losses and safety risks. This paper proposes an intelligent diagnosis method that combines continuous wavelet transform (CWT) with a deep neural network. Multi-pipeline temperature and pressure signals are transformed into multi-scale time-frequency representations by CWT, and a diagnostic model is constructed using Inception modules and a hybrid SimAM+SE attention mechanism to improve abnormal-pattern recognition and robustness to missing data. Experiments on gas-usage data from more than 700 industrial and commercial users show that the proposed model achieves an accuracy of 97.00%, a precision of 96.30%, a recall of 98.19%, and an F1-score of 97.24%, outperforming long short-term memory (LSTM), particle swarm optimization-support vector machine (PSO-SVM), TimesNet, and PatchTST. Expert verification further confirms its effectiveness and engineering applicability in practical gas-metering-anomaly diagnosis.