A neural network for evaluating the concentration of leaked gas clouds detected by TDLAS at oil and gas stations

Due to the significant advantages of methane sensitivity and area-type leakage detection, tunable diode laser absorption spectroscopy (TDLAS) gas detection has been promoted as an effective method for microleakage monitoring in oil and gas stations. However, the output of the TDLAS detector is the integral concentration (IC). Based on the relevant research, the alarm threshold and risk are assessed via the gas concentration rather than the IC. How to evaluate the concentration of leaked gas clouds based on IC from TDLAS detectors is still a challenge. To address this problem, the characteristics of IC and the influence of the laser path, wind speed and leakage rate were studied via computational fluid dynamics (CFD). A neural network classification model (NNCM) was proposed to obtain the probability distribution of the maximal concentration along the laser path (Cmax). The results indicated that the IC is strongly correlated with the Cmax. Considering the accuracy and operability, the NNCM with input features of IC, wind speed and angle of laser path was selected. Field tests showed that the developed model achieved the concentration evaluation of leaked gas clouds. Additionally, the NNCM can also quantify the uncertainty of the results, which avoids misjudgments caused by deviations.