Systematic evaluation of abnormal detection methods on gas well sensor data

Natural gas, as a kind of clean energy, has attracted significant attention in the global market. However, how to ensure the safety and high efficiency in natural gas production becomes a hot research problem in the gas industry. The real-time abnormal status detection of the natural gas well empowers the decision-maker to prevent potentially catastrophic damage and correct unexpected situations. In this paper, we systematically evaluate the 9 state-of-the-art machine learning methods to detect such anomalous status on large sensor data collected from 4 natural gas wells. In addition, we have identified the most important features that can improve anomaly detection performance. The challenges and potential research directions have been discussed. This is the first work to investigate different types of anomaly detection methods on natural gas well sensor data. Our research results provide valuable insights for developing specific anomaly detection systems in the natural gas industry.