Sulfurized rust is the production of corrosion in crude oil tanks. It will be oxidized and self-heating when contacting with air, and the rise of temperature can cause severe accidents. This paper focuses on the temperature measurement of distributed optical fiber sensor (DOFS) and the research on anomaly detection methods aided by deep learning. An experimental apparatus was set up to simulate the temperature change during sulfurized rust self-heating, then some artificial ambient temperature was added to interference anomaly detection. The DOFS returned normal temperature, artificial ambient temperature and self-heating temperature data for analysis. Furthermore, four Auto-Encoder (AE) based algorithms and several traditional machine learning methods were tested on the collected temperature data for anomaly detection. Test revealed that Convolutional Neural Networks Auto-Encoder (CNN-AE) was successful in detecting the anomaly situations at an accuracy level of 0.98. The study demonstrates that DOFS and deep learning would be a potential solution for detecting anomaly temperature change to prevent self-heating accident caused by sulfurized rust.

硫化锈是原油罐腐蚀的产物。与空气接触会被氧化并自热，温度升高会造成严重事故。本文重点研究分布式光纤传感器（DOFS）的温度测量和深度学习辅助的异常检测方法研究。搭建实验装置模拟硫化铁锈自热过程中的温度变化，然后加入一定的人工环境温度进行干扰异常检测。DOFS 返回常温、人工环境温度和自热温度数据进行分析。此外，在收集的温度数据上测试了四种基于自动编码器 (AE) 的算法和几种传统的机器学习方法，以进行异常检测。测试表明，卷积神经网络自动编码器 (CNN-AE) 成功地以 0.98 的准确度检测异常情况。该研究表明，DOFS 和深度学习将是检测异常温度变化以防止硫化锈引起的自热事故的潜在解决方案。