<p>Under the long-term use of industrial pipe network, due to the external environment, the nature of internal media, and other factors, it is easy to appear various abnormal situations, which undoubtedly bring great economic losses to major enterprises. Therefore, it is urgent to adopt effective anomaly detection technology to effectively detect the application safety of industrial pipe networks (e.g., water pipe networks, oil pipe networks, etc.) to eliminate the danger in the bud. Specifically, this work proposed an anomaly detection framework for industrial pipe network time series data based on an LSTM Autoencoder. Firstly, this work obtains the training data based on the real-time pipeline signal collected by the sensor. In order to solve the problem of a lack of anomaly samples in field data, this work designed multiple enhancement attacks to generate real anomaly samples, then feed the enhanced data set into an autoencoder designed to learn the identity function in an unsupervised manner to reconstruct the original input while compressing the data. In addition, in order to capture the temporal dependence between industrial pipe network data, this work also introduced the LSTM layer (LSTM Autoencoder), which can effectively improve the detection effect. Finally, according to the comparison between the output results of the trained anomaly detection model and the threshold value, the signal of the oil or water pipe network is determined to be abnormal.</p>