朱敏1 季艳2 刘紫薇3
（1.云和县天然气有限公司，浙江 丽水 323600
2.浙江浙能数字科技有限公司，浙江 杭州 310012
3.山东理工职业学院，山东 济宁 272000）
摘要：城市天然气日用量数据的异常检测对能源系统安全运行至关重要。针对其非线性强、异常样本稀缺和变量间依赖复杂等问题，本文提出一种预测增强型卷积自编码异常检测（PE-CAEAD）模型。该模型由表征网络（卷积自编码器、反卷积解码器）和预测网络组成，可在无监督条件下实现城市天然气日用量数据的异常检测。在某城市天然气日用量数据集上的实验结果表明，该模型的精确率、召回率、ROC曲线下的面积、F1分数分别为0.94、0.96、0.96、0.95，验证了其在城市天然气日用量数据异常检测中的有效性与适用性。
关键词：城市天然气；异常检测；时间序列；自编码器；交叉注意力机制；双向LSTM
中图分类号：TE832 文献标志码：A 文章编号：1674-2605(2026)03-0002-08
DOI：10.12475/aie.20260302 开放获取
Anomaly Detection of Daily Natural Gas Consumption Data in Cities Based on Autoencoder Networks
ZHU Min1 JI Yan2 LIU Ziwei3
(1.Yunhe County Natural Gas Co., Ltd., Lishui 323600, China 2.Zhejiang Zheneng Digital Technology Co., Ltd., Hangzhou 310012, China 3.Shandong Polytechnic College, Jining 272000, China)
Abstract: Anomaly detection in urban daily natural gas consumption data is crucial for the safe operation of energy systems. To address the challenges of strong nonlinearity, scarce anomaly samples, and complex inter-variable dependencies, this paper proposes a prediction-enhanced convolutional autoencoder anomaly detection model, named PE-CAEAD. The model consists of a representation network, including a convolutional autoencoder and a deconvolutional decoder, as well as a prediction network, enabling unsupervised anomaly detection for urban daily natural gas consumption data. Experimental results on an urban daily natural gas consumption dataset show that the proposed model achieves a precision of 0.94, a recall of 0.96, an area under the ROC curve of 0.96, and an F1-score of 0.95, demonstrating its effectiveness and applicability in anomaly detection for urban daily natural gas consumption data.
Keywords: urban natural gas; anomaly detection; time series; autoencoder; cross attention mechanism; Bi-LSTM
