Machine Learning for Gas Leak Detection and Forecasting

Abstract

The deployment of numerous sensors across several equipment and processes plays a pivotal role in identifying and mitigating potential hazards, particularly those associated with gas leaks. The industry has increasingly embraced a data-centric approach to validate alarm signals and initiate appropriate responses. This paper delves into the applications of gas sensors, emphasizing their capability to detect extremely low concentrations of target gases. The primary objective is to enhance false alarm detection by focusing on the precise identification of gas presence. The study explores various scenarios where machine learning based analysis of gas sensor data proves instrumental in maintaining a safe working environment across industries.Gas sensors, designed to detect target gases with extremely low concentrations, face challenges posed by environmental factors such as temperature and humidity variations. This study addresses the critical issue of leveraging machine learning and forecasting techniques to enhance the detection of hazardous gases and reduce false alarms in the presence of environmental fluctuations. The research focuses on the application of machine learning methods to predict the presence of dangerous gases in the petroleum industry based on sensor outputs. The study is structured into two distinct tasks, each contributing to the robust detection performance of gas sensors. Task 1 involves predicting the concentration of carbon monoxide (CO) and the relative humidity percentage, while Task 2 focuses on predicting the presence of ethylene and methane gases. By strategically employing machine learning methodologies, the researchers aim to determine the accuracy of gas detection and false alarm mitigation in the petroleum industry, thereby advancing safety measures and operational reliability.

Keywords: gas, Leak, Regression, Detection, machine learning, Methane