Gatalytic Combustion Gas Sensing System Based On Pulse Power

Gas explosion is an important hidden danger of energy production safety issues such as gas drilling and coal mining.At present,detection methods of methane gas include optical interference,catalytic combustion and infrared absorption.Compared with other detection methods,catalytic combustion has the advantages of high precision and low price.However,catalytic element of the catalytic combustion gas sensor in the traditional detection method was greatly affected by temperature.Especially in the high concentration of the gas under test,catalyst was easily deactivated under high temperature combustion.Deactivation of the catalyst has caused damage to the sensor and the problems of low detection range and large temperature drift.In this paper,a methane detection system with pulse power supply,multisensor information fusion and feedback regulation technology was designed.STM32F051 was used as main control chip in the pulse powered methane detection system.Methane,temperature and humidity,carbon dioxide and oxygen collection terminals were collected different parameters in the environment.A special infrared methane detector was used to record the standard concentration of methane in the pulse-powered methane detection platform in real time.In this paper,data collected by multiple sensors often has abnormal values in the collected data due to performance of sensor itself,stability of detection circuit,and loss of data frame during data transmission.Moreover,the feature dimension of the data is high.It is necessary to remove outliers,reduce computational overhead and improve algorithm performance.First,local outlier factor(LOF)algorithm was used to preprocess realtime data collected by multiple sensors.Second,random forest(RF)algorithm was used to classify features and extract important features.Third,collinear features were removed.Fourth,extreme random forest(ERF)algorithm was used to train data to obtain a methane catalytic combustion sensor prediction model.Prediction model trained by background management system of pulse powered methane detection was called by the upper computer to realize the real-time display of methane,carbon dioxide,oxygen,temperature and humidity in the environment.At the same time,collected data of multiple sensors was stored in the My SQL database.Feasibility of above methods for high concentration methane detection was verified by setting up a pulse powered methane detection platform.The results show that non-pulse power supply system has a low accuracy after the methane concentration is higher than 50000 ppm.When methane concentration is less than 73000 ppm,average accuracy of the pulse power supply system is 69.3%,while average prediction accuracy of the system under pulse power supply combined with the ERF algorithm is 85.9%.After reducing voltage of the pulse power supply through negative feedback adjustment technology,system detected the highest methane concentration of 90,000 ppm,and average prediction accuracy when concentration is less than90,000 ppm is 89.1%,which achieved a higher concentration of methane detection.