A gas identification method, system and device based on edge computing

Abstract

The present invention relates to the field of gas identification technology, and specifically to a gas identification method, system and device based on edge computing. The gas identification method comprises the following steps: step 1: sensor array construction; step 2: data acquisition and response value calculation; step 3: time calibration; step 4: temperature and humidity calibration; step 5: deep learning model training; step 6: model output and association; step 7: model deployment and edge computing. The present invention uses a multi-sensor array to capture more and richer gas features, and at the same time performs time calibration and environmental calibration on the sensor to avoid the inconsistency caused by changes in the sensor over time and the environment. The classification-guided concentration-associated DNN model can accurately predict the classification and concentration of the gas at the same time, and its classification accuracy reaches 97.1%, the concentration prediction error is within 5%, and it is easy to deploy to the edge computing device end.

Description

Technical Field

[0001] The present invention relates to the field of gas identification technology, and specifically to a gas identification method, system and device based on edge computing. Background Art

[0002] Traditional gas identification uses a single gas sensor, and as long as the sensor responds, it will give an alarm or concentration warning for the gas. However, the selectivity of a single sensor is poor. In addition to responding to the target gas, it will also respond to other gases, resulting in more false alarms.

[0003] In addition, there is also a method of using multiple gas sensors to form an array for gas identification, but using traditional machine learning methods, the calculation is simple and the accuracy is low.

[0004] The sensor itself will also change with time, temperature, humidity, etc. Traditional gas identification methods do not consider the impact of these factors on the sensor, resulting in low gas identification accuracy.

[0005] Although deep learning models are effective, they are difficult to deploy on edge computing devices because they require more computation than traditional machine learning. Summary of the invention

[0006] The purpose of the present invention is to provide a gas identification method, system and device based on edge computing to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a gas identification method based on edge computing, the gas identification method comprising the following steps:

[0008] Step 1: constructing a sensor array, using a plurality of sensors of different types to form a sensor array, through which matrix data of gas is collected;

[0009] Step 2: Data collection and response value calculation, collecting the response data of the target gas sample through the sensor and calculating the response value of each sensor;

[0010] Step 3: Time calibration, study the decay law of sensor response value over time and fit the time decay model;

[0011] Step 4: Temperature and humidity calibration, consider the impact of ambient temperature and humidity on the sensor response value, and calibrate the temperature and humidity separately;

[0012] Step 5: Deep learning model training, building a classification-guided concentration-associated DNN model, and using the calibrated sensor response values ​​to train the model to ensure that the model has the ability to classify gases and predict concentrations;

[0013] Step 6: Model output and association. In the output layer of the DNN model, the classification confidence of the target gas is output through Head1, the concentration of the target gas is output through Head2, and the classification features are associated with the concentration features through the association matrix;

[0014] Step 7: Model deployment and edge computing: Deploy the trained DNN model on the edge computing device to achieve real-time gas identification and concentration prediction.

[0015] Preferably, the response value of the sensor is calculated by the following formula:

[0016]

[0017] Wherein, Rd represents the response value of the sensor, Ra represents the air baseline resistance, and Rg represents the target gas response resistance.

[0018] Preferably, the sensor includes a methane sensor, a carbon monoxide sensor, a hydrogen sulfide sensor, and a Bosch constant temperature sensor, wherein the attenuation multiple formula of the methane sensor response value over time is as follows:

[0019]

[0020] Among them, α represents the attenuation multiple, T represents the number of days the sensor is used, and the methane sensor can be calibrated using this formula to obtain the calibrated methane sensor response value:

[0021]

[0022] Where Rd represents the original response value of the methane sensor, represents the time-corrected response value.

[0023] Preferably, the response value of the sensor is calibrated by the temperature and humidity change value as follows:

[0024]

[0025] in, represents the time-calibrated response value, represents the response value after environmental calibration, Indicates the percentage change in humidity. Indicates the percentage change in temperature.

[0026] Preferably, the concentration feature is obtained by multiplying Head1 and the response feature by the correlation matrix:

[0027]

[0028] Among them, feature1~feature4 indicate the concentration characteristics of carbon monoxide, methane, ammonia, and hydrogen sulfide respectively, conf1~conf4 represent the confidence of carbon monoxide, methane, ammonia, and hydrogen sulfide gases predicted by the network respectively, and the matrix in the middle is the correlation matrix.

[0029] A gas identification system is applied to the above-mentioned gas identification method based on edge computing, and the gas identification system includes a sensor array module, a data acquisition module, a data preprocessing module, a deep learning model module, a dynamic weight adjustment module, an attenuation law analysis module, an edge computing processing module, a data storage and log module, and a self-check module;

[0030] Sensor array module, used to collect concentration data of multiple gases;

[0031] Data acquisition module, used to receive sensor output signals in real time and form a gas concentration matrix;

[0032] Data preprocessing module, used to calibrate the collected data in terms of time and environment;

[0033] A deep learning model module for training and deploying classification models to improve the accuracy of gas identification;

[0034] Dynamic weight adjustment module, used to dynamically adjust the weight of each sensor in the model according to the response characteristics of the sensor and the classification prediction results;

[0035] Attenuation law analysis module, used to study and model the attenuation law of sensor response values;

[0036] Edge computing processing module, used to process data and run deep learning models in real time on edge devices;

[0037] Data storage and log module, used to save historical data and detection records;

[0038] The self-test module is used to adjust the environment of the sensor array module to standard temperature and humidity conditions, and compare the actual detected data under these conditions with the attenuation predicted data to perform sensor calibration and update the sensor attenuation model.

[0039] A gas identification device is provided, equipped with the above-mentioned gas identification system, the device comprises an identifier body, a lifting seat is arranged on the identifier body through a through slot, a gas sensor is arranged on the lifting seat, a slider is arranged on the lifting seat, and the slider is slidably mounted on a spring guide rod, a magnetic sheet is arranged on the slider, and an electromagnetic seat is installed in the identifier body, the lifting seat is received into the identifier body by energizing the electromagnetic seat, a sealing plate is rotatably mounted on the identifier body through a side seat, and a transmission structure driven by the slider is arranged in the side seat, when the lifting seat and the gas sensor are received into the identifier through the transmission structure, the sealing plate closes the through slot, and a temperature and humidity control device is installed in the identifier body.

[0040] Preferably, the through groove is arranged on the top surface of the identifier body, and the gas sensor includes a methane sensor, a carbon monoxide sensor, a hydrogen sulfide sensor, and a Bosch constant temperature sensor, and the gas sensor array is arranged on a lifting seat.

[0041] Preferably, the spring on the spring guide rod provides elastic force to make the lifting seat extend from the through slot, the magnetic sheet is embedded in the side of the slider, and the electromagnetic seat is installed at the bottom of the identifier body.

[0042] Preferably, a torsion spring shaft is provided on the side seat, and the sealing plate is connected to the torsion spring shaft, a rotating wheel is installed on the torsion spring shaft, and a winding rope is wound around the rotating wheel, a sleeve is vertically installed on the bottom of the side seat, and the winding rope is arranged through the sleeve, a base is installed at the bottom end of the winding rope, and a side pressure block is installed on the slider, and after the slider descends, the base is pushed down by the side pressure block to drive the sealing plate to rotate, and a sealing gasket is provided on the sealing plate.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The present invention uses a multi-sensor array to capture more and richer gas features. At the same time, the sensors are time-calibrated and environmentally calibrated to avoid the inconsistency of the sensors over time and over the environment. The classification-guided concentration-associated DNN model can accurately predict the classification and concentration of the gas at the same time. The classification accuracy is 97.1%, the concentration prediction error is within 5%, and it is easy to deploy on the edge computing device.

[0045] 2. The present invention improves the traditional random forest model into a deep learning model. The model structure takes into account the different response values ​​of sensors to different gases, and performs weighted calculation of gas concentration adjustment and optimization for different sensors according to the classification prediction results. The deep learning model is locally deployed, and a positive and negative sample mechanism is added to the model. After user feedback, the samples are automatically added to the training set for model adjustment.

[0046] 3. The gas identification system of the present invention can complete the construction of the classification-guided concentration-associated DNN model, and realize data acquisition and corresponding time calibration and environmental calibration. At the same time, the gas identification system has a self-test module that can adjust the environment of the sensor array module to standard temperature and humidity conditions, and compare the actual detected data under these conditions with the attenuation predicted data to achieve the effect of sensor calibration and update the sensor attenuation model of the system.

[0047] 4. The gas identification device of the present invention can cooperate with the self-test module to achieve the effect of model calibration of the identification system. Through the automatic lifting and closing structure, the gas sensor can enter the cavity of the identifier body, thereby utilizing the temperature and humidity control device in the identifier body to create a standard temperature and humidity environment to complete the self-test work. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the response values ​​of methane gas on four sensors.

[0049] Figure 2 Schematic diagram of the response values ​​of carbon monoxide gas on four sensors.

[0050] Figure 3 Schematic diagram of ammonia gas on four sensors.

[0051] Figure 4 Schematic diagram of hydrogen sulfide gas on four sensors.

[0052] Figure 5 This is the attenuation law of the methane sensor response value over time.

[0053] Figure 6 Schematic diagram of the classification-guided concentration-associated DNN model.

[0054] Figure 7 Schematic diagram of carbon monoxide label concentration and predicted concentration.

[0055] Figure 8 Schematic diagram of methane label concentration and predicted concentration.

[0056] Fig. 9 Schematic diagram of nitrogen label concentration and predicted concentration.

[0057] Fig.10 Schematic diagram of hydrogen sulfide label concentration and predicted concentration.

[0058] Fig.11 FIG. 4 is a block diagram of a gas identification system according to the present invention.

[0059] Fig.12 Schematic diagram of the gas identification device of the present invention.

[0060] Fig.13 It is a schematic diagram of the internal structure of the identifier of the present invention.

[0061] Fig.14 Schematic diagram of the sensor array structure of the present invention.

[0062] Fig.15 Schematic diagram of the sealing plate structure of the present invention.

[0063] Fig.16 It is a schematic diagram of the sealing plate driving structure of the present invention.

[0064] In the figure: 1. identifier body; 2. lifting seat; 3. gas sensor; 4. spring guide rod; 5. slider; 6. magnetic sheet; 7. electromagnetic seat; 8. side pressure block; 9. side seat; 10. sealing plate; 11. sealing gasket; 12. torsion spring shaft; 13. rotating wheel; 14. winding rope; 15. sleeve; 16. base. DETAILED DESCRIPTION

[0065] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment. It should be noted that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0066] See also Figures 1 to 16 The present invention provides a technical solution: a gas identification method, system and device based on edge computing, which uses multiple sensors to form a sensor array and collects matrix data of gas, which can effectively deal with the problem of poor selectivity of a single sensor, and uses time calibration, temperature and humidity calibration and other methods to effectively solve the influence of sensor response values ​​on environmental changes such as time and temperature and humidity, and uses a deep learning model structure to train the model, which can greatly improve the accuracy of recognition. Finally, the model is deployed on the edge of the gas detection and identification device, realizing edge computing and effectively solving the problem of the difficulty of deploying deep learning models on edge devices. At the same time, the gas identification method combines the characteristics of the difference in the response values ​​of different sensors to different gases, and specially constructs the deep learning model, so that the model can dynamically adjust the weights of different sensors according to the classification prediction results, thereby effectively predicting the concentration of the gas.

[0067] The present invention uses a combination of multiple sensor arrays to identify multiple gases. The sensors include methane sensors, carbon monoxide sensors, hydrogen sulfide sensors, Bosch constant temperature sensors, etc. The array is used to identify the target gas. Different target gases have different response sizes on these four sensors, such as Figure 1-Figure 4 As shown in FIG. 1 , the response values ​​of the four target gases on the four sensors under different concentration conditions are shown. It can be seen from the figure that different target gases can be distinguished and identified more accurately according to the corresponding modes of the sensors. Figure 1-Figure 4 In the figure, the horizontal axis 1-10 represents the Bosch constant temperature sensor, 11-20 represents the methane sensor, 21-30 represents the hydrogen sulfide sensor, and 31-40 represents the carbon monoxide sensor.

[0068] The response value is calculated as:

[0069]

[0070] Wherein, Rd represents the response value of the sensor, Ra represents the air baseline resistance, and Rg represents the target gas response resistance.

[0071] Secondly, the present invention explores the attenuation law of the response value of different sensors over time, and simulates the attenuation curve of the sensor over time based on the attenuation law. In the subsequent use of the device, the attenuation coefficient is calculated based on the power-on time statistics of the device, and the response value of the sensor is calibrated based on the attenuation coefficient. Taking the methane sensor as an example, the present invention tracks and explores the attenuation law of the response value of the methane sensor over time. Figure 5 , where the broken line is the attenuation multiple of the methane sensor response value calculated from actual measurements, and the curve is the attenuation law of the methane sensor fitted according to the attenuation multiples at different time points.

[0072] according to Figure 5 , the formula for fitting the attenuation multiple of the methane sensor response value over time is as follows:

[0073]

[0074] Among them, α represents the attenuation multiple, and T represents the number of days the sensor is used. When used, the methane sensor can be calibrated using this formula to obtain the calibrated methane sensor response value:

[0075]

[0076] Where Rd represents the original response value of the methane sensor, represents the time-corrected response value.

[0077] The present invention also explores the influence of temperature and humidity in the environment on the response value of the sensor. The study shows that for every 1% increase in humidity in the environment, the response value of the sensor increases by 1.2%; for every 1% increase in temperature in the environment, the response value of the sensor increases by 1.5%. Based on this rule, the response value of the sensor can be further calibrated by the temperature and humidity change value:

[0078]

[0079] in, represents the time-calibrated response value, represents the response value after environmental calibration, Indicates the percentage change in humidity. Indicates the percentage change in temperature.

[0080] The focus of the present invention is a specially designed DNN network structure called "classification-guided concentration-associated DNN model", which connects the model's classification capability to concentration prediction, so that the model can simultaneously predict gas classification and gas concentration with high accuracy. In order to realize the above concept, the specially designed DNN structure of the present invention is mainly in the output layer. The specific structure is shown in the figure below, where head1 outputs the classification confidence of the four target gases, and head2 outputs the concentrations of the four target gases. Experimental data show that different sensors have different sensitivities to different target gases, that is, different response values. In particular, the carbon monoxide sensor and the Bosch thermostatic sensor have a larger response value to carbon monoxide gas, the methane sensor has a larger response value to methane gas, the carbon monoxide sensor and the Bosch thermostatic sensor have a larger response value to ammonia gas, and the Bosch thermostatic sensor and the hydrogen sulfide sensor have a larger response value to hydrogen sulfide gas. See Figure 1-Figure 4 Based on this, the network connection from classification prediction head1 to concentration prediction head2 is designed, see Figure 6 .

[0081] Among them, the concentration feature is obtained by multiplying Head1 and the response feature through the correlation matrix:

[0082]

[0083] Among them, feature1~feature4 indicate the concentration characteristics of carbon monoxide, methane, ammonia, and hydrogen sulfide respectively, conf1~conf4 represent the confidence of carbon monoxide, methane, ammonia, and hydrogen sulfide predicted by the network respectively, and the matrix in the middle is the correlation matrix. Using this correlation matrix, the classification ability of network prediction can guide the network to predict gas concentration. Except for the correlation matrix, the rest of this network model is a common DNN network structure, which is easy to deploy on edge computing devices.

[0084] By using this DNN model, classification and concentration prediction can be made more accurate, with a classification accuracy of 97.1% and a concentration prediction error of less than 5%; and it can be easily deployed on devices that support edge computing.

[0085] The actual classification confusion matrix and concentration error after the model training are shown in the following table:

[0086] Table 1

[0087]

[0088] The actual test of the four gases resulted in the following classification and concentration prediction results:

[0089] Table 2

[0090]

[0091] The labeled concentrations and predicted concentrations of the four gases are shown in Figure 7-10 shown.

[0092] The above gas identification method is based on edge computing, which relies on the gas identification system.

[0093] The gas identification system of the present invention includes a sensor array module, a data acquisition module, a data preprocessing module, a deep learning model module, a dynamic weight adjustment module, an attenuation law analysis module, an edge computing processing module, a data storage and log module, and a self-check module.

[0094] The sensor array module is used to collect concentration data of various gases, such as methane, carbon monoxide, hydrogen sulfide, etc.

[0095] The sensor array module is composed of multiple sensors, including methane sensors, carbon monoxide sensors, hydrogen sulfide sensors, and Bosch constant temperature sensors. These sensors are combined into an array that can detect multiple target gases at the same time. Each sensor responds differently to specific gases. This combination of multi-sensor arrays can effectively solve the problem of poor selectivity of a single sensor. The module supports flexible configuration of sensors to meet different detection needs, and has the characteristics of high sensitivity and high resolution to ensure accurate detection of changes in trace gas concentrations.

[0096] The data acquisition module is used to receive the sensor output signal in real time and form a gas concentration matrix.

[0097] The data acquisition module receives the signal output by the sensor array in real time and forms matrix data containing the concentration of each gas. The module uses a high-precision analog-to-digital converter to ensure the accuracy and consistency of the collected data. The module design takes into account the multi-channel parallel data acquisition capability to cope with the data output by multiple sensors at the same time. In addition, the module has strong anti-interference ability and can work stably in complex industrial environments to ensure data reliability.

[0098] The data preprocessing module is used to calibrate the collected data in terms of time and environment, such as temperature and humidity, to ensure data accuracy.

[0099] The data preprocessing module performs time calibration, temperature and humidity calibration and other processing on the collected matrix data, solving the impact of sensor response values ​​over time and environmental changes (such as temperature and humidity). Time calibration ensures that the response data of different sensors at different time points can be effectively compared on the same time axis. Temperature and humidity calibration improves the accuracy of measurement results by compensating for the impact of environmental conditions on sensor output. This module uses an adaptive algorithm to continuously optimize preprocessing parameters based on historical data and real-time feedback.

[0100] A deep learning model module for training and deploying classification models to improve the accuracy of gas identification.

[0101] The deep learning model module is the core of the present invention and is responsible for training and deploying the neural network model for gas identification. The model structure combines the differences in the response values ​​of different sensors to different gases, and the model is specially designed based on these differences so that it can effectively capture the characteristic signals of each gas. The model uses a large amount of labeled data during the training process to ensure the high accuracy of the model in different environments. The deployed model runs in real time on the edge device to achieve efficient gas identification. The module also supports dynamic updates and optimizations to ensure that the model is always in the best state.

[0102] The dynamic weight adjustment module is used to dynamically adjust the weight of each sensor in the model according to the response characteristics of the sensor and the classification prediction results.

[0103] The dynamic weight adjustment module dynamically adjusts the weight of the sensor in the model according to the classification prediction results of the deep learning model. This function enables the model to flexibly respond to the detection needs of different gases, thereby improving the accuracy of recognition. For example, when a sensor has a weak response to the target gas, the module can automatically reduce its weight and increase the weight of the sensor with a stronger response, thereby optimizing the overall detection effect. This module combines the sensor response characteristics and classification results, and continuously adjusts the weight parameters through a feedback mechanism to improve the generalization ability of the model.

[0104] The attenuation law analysis module is used to study and model the attenuation law of sensor response values ​​to improve the accuracy of long-term monitoring.

[0105] The attenuation law analysis module specifically studies the attenuation law of sensor response values ​​over time. This module establishes attenuation models for various sensors through long-term monitoring data, and simulates the attenuation curve of the sensor over time. Taking the methane sensor as an example, the module tracks the attenuation trend of the response value of the methane sensor over time in detail, and generates an attenuation multiple curve to predict future sensor performance. In actual applications, the module will automatically calculate the attenuation coefficient of the sensor according to the power-on time of the device, and calibrate the real-time data of the sensor based on the coefficient to ensure the accuracy of long-term monitoring.

[0106] Edge computing processing module, used to process data and run deep learning models in real time on edge devices, improving response speed and processing capabilities.

[0107] The edge computing processing module is responsible for real-time data processing on edge devices and running deep learning models. The present invention optimizes the model structure so that the deep learning model can run efficiently on edge devices with limited resources, thereby achieving rapid response and real-time gas identification. The module adopts a distributed computing architecture and can dynamically adjust the allocation of computing resources on edge devices to meet different data processing requirements. In addition, the edge computing processing module supports collaborative work with cloud systems, and achieves optimal resource configuration through seamless connection between the edge and the cloud.

[0108] The data storage and log module is used to save historical data and test records for subsequent analysis and review.

[0109] The data storage and log module is used to save the system's historical data and detection records, including sensor response values, preprocessing results, model outputs, and dynamically adjusted weight parameters. This module uses highly reliable storage media to ensure data security and persistence. The log function records the process of each detection and processing in detail, providing an important basis for subsequent analysis and troubleshooting. In addition, the data storage and log module also supports data compression and backup functions to improve storage efficiency and data security.

[0110] The self-test module is used to adjust the environment of the sensor array module to standard temperature and humidity conditions, and compare the actual detected data under these conditions with the attenuation predicted data to achieve the effect of sensor calibration and update the sensor attenuation model of the system.

[0111] The self-test module relies on the gas identification device to form a closed and adjustable environment, so that the sensor array module is in a standard environment.

[0112] The gas identification device of the present invention is equipped with a gas identification system. The device includes an identifier body 1, a lifting seat 2 is provided on the identifier body 1 through a through groove, and a gas sensor 3 is provided on the lifting seat 2, a slider 5 is provided on the lifting seat 2, and the slider 5 is slidably installed on a spring guide rod 4, a magnetic sheet 6 is provided on the slider 5, and an electromagnetic seat 7 is installed in the identifier body 1, and the lifting seat 2 is received into the identifier body 1 by energizing the electromagnetic seat 7, a sealing plate 10 is rotatably installed on the identifier body 1 through a side seat 9, and a transmission structure driven by the slider 5 is provided in the side seat 9, and when the lifting seat 2 and the gas sensor 3 are received into the identifier through the transmission structure, the sealing plate 10 closes the through groove, and a temperature and humidity control device is installed in the identifier body 1.

[0113] The identification device mainly cooperates with the self-test module to calibrate the model of the identification system. It can control the sensor array under a standard temperature and humidity environment, thereby calculating its actual attenuation degree and updating the attenuation formula, thereby completing the gas identification work more accurately.

[0114] The through groove is arranged on the top surface of the identifier body 1 , and the gas sensor 3 includes a methane sensor, a carbon monoxide sensor, a hydrogen sulfide sensor, and a Bosch constant temperature sensor. The gas sensor 3 array is arranged on the lifting seat 2 .

[0115] The sensor array composed of gas sensors 3 is arranged on the lifting seat 2. When the lifting seat 2 rises, the gas sensors 3 pop out from the through slots of the identifier body 1, so that gas identification can be performed. When self-test is required, the lifting seat 2 is lowered so that the gas sensors 3 enter the cavity of the identifier body 1, thereby using the temperature and humidity control device in the identifier body 1 to create a standard temperature and humidity environment to complete the self-test.

[0116] The spring on the spring guide rod 4 provides elastic force to make the lifting seat 2 extend from the through slot, the magnetic sheet 6 is embedded in the side of the slider 5, and the electromagnetic seat 7 is installed at the bottom of the identifier body 1.

[0117] By energizing the electromagnetic seat 7 , a magnetic force is generated to act on the magnetic sheet 6 of the slider 5 , so that the lifting seat 2 overcomes the elastic force of the spring guide rod 4 and moves downward.

[0118] A torsion spring shaft 12 is provided on the side seat 9, and the sealing plate 10 is connected to the torsion spring shaft 12, a rotating wheel 13 is installed on the torsion spring shaft 12, and a winding rope 14 is wound and connected on the rotating wheel 13, a sleeve 15 is vertically installed at the bottom of the side seat 9, and the winding rope 14 is arranged through the sleeve 15, a base 16 is installed at the bottom of the winding rope 14, and a side pressure block 8 is installed on the slider 5, and after the slider 5 descends, the base 16 is pushed down by the side pressure block 8 to drive the sealing plate 10 to rotate, and a sealing gasket 11 is provided on the sealing plate 10.

[0119] A sealing plate 10 with a sealing gasket 11 is also provided for the through-slot structure on the identifier body 1. When the lifting seat 2 descends, the side pressure block 8 carried by the upper slider 5 can push the base 16 downward, thereby pulling the winding rope 14, so that the rotating wheel 13 drives the torsion spring shaft 12 to rotate. Through the structural setting, the sealing plate 10 can rotate 180 degrees to close the through-slot and seal the gas sensor 3 in the cavity of the identifier body 1, so as to create a standard temperature and humidity environment.

[0120] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A gas identification method based on edge computing, characterized in that: The gas identification method comprises the following steps: Step 1: constructing a sensor array, using a plurality of sensors of different types to form a sensor array, through which matrix data of gas is collected; Step 2: Data collection and response value calculation, collecting the response data of the target gas sample through the sensor and calculating the response value of each sensor; Step 3: Time calibration, study the attenuation law of the sensor response value over time, and fit the time attenuation model. During the use of the device, calculate the attenuation coefficient based on the power-on time statistics of the device, and calibrate the sensor response value based on the attenuation coefficient; Step 4: Temperature and humidity calibration, consider the impact of ambient temperature and humidity on the sensor response value, and calibrate the temperature and humidity separately. The impact of ambient temperature and humidity on the sensor response value is that for every 1% increase in ambient humidity, the sensor response value increases by 1.2%; for every 1% increase in ambient temperature, the sensor response value increases by 1.5%. Based on this rule, the sensor response value is further calibrated by the temperature and humidity change value; Step 5: Deep learning model training, building a classification-guided concentration-associated DNN model, and using the calibrated sensor response values ​​to train the model to ensure that the model has the ability to classify gases and predict concentrations; Step 6: Model output and association. In the output layer of the DNN model, the classification confidence of the target gas is output through Head1, the concentration of the target gas is output through Head2, and the classification features are associated with the concentration features through the association matrix; Step 7: Model deployment and edge computing: deploy the trained DNN model on edge computing devices to achieve real-time gas identification and concentration prediction; The sensors include methane sensor, carbon monoxide sensor, hydrogen sulfide sensor, and Bosch constant temperature sensor. The formula for the attenuation multiple of the methane sensor response value over time is as follows: Among them, α represents the attenuation multiple, T represents the number of days the sensor is used, and the methane sensor can be calibrated using this formula to obtain the calibrated methane sensor response value: Where Rd represents the original response value of the methane sensor, represents the response value after time calibration; The response value of the sensor is calibrated by the temperature and humidity change value as follows: in, represents the time-calibrated response value, represents the response value after environmental calibration, Indicates the percentage change in humidity. Indicates the percentage of temperature change; The concentration feature is obtained by multiplying Head1 and the response feature through the correlation matrix: Among them, feature1~feature4 indicate the concentration characteristics of carbon monoxide, methane, ammonia, and hydrogen sulfide respectively, conf1~conf4 represent the confidence of carbon monoxide, methane, ammonia, and hydrogen sulfide gases predicted by the network respectively, and the matrix in the middle is the correlation matrix.

2. A gas identification method based on edge computing according to claim 1, characterized in that: The response value of the sensor is calculated by the following formula: Wherein, Rd represents the response value of the sensor, Ra represents the air baseline resistance, and Rg represents the target gas response resistance.

3. A gas identification system, applied to the gas identification method based on edge computing as claimed in claim 1, characterized in that: The gas identification method is based on edge computing, which relies on a gas identification system. The gas identification system includes a sensor array module, a data acquisition module, a data preprocessing module, a deep learning model module, a dynamic weight adjustment module, an attenuation law analysis module, an edge computing processing module, a data storage and log module, and a self-check module. A sensor array module, constructed by step 1, for collecting concentration data of multiple gases; A data acquisition module, used in step 2, to receive the sensor output signal in real time and form a gas concentration matrix; A data preprocessing module, used in step 3 and step 4 to perform time calibration and environment calibration on the collected data; A deep learning model module, used in step 5 to train and deploy a classification model to improve the accuracy of gas identification; A dynamic weight adjustment module, used in step 6 to dynamically adjust the weight of each sensor in the model according to the response characteristics of the sensor and the classification prediction results; An attenuation law analysis module, used in steps 3-5 to study and model the attenuation law of the sensor response value; An edge computing processing module, used in step 7 to process data and run deep learning models in real time on edge devices; A data storage and log module, used in the gas identification method to store historical data and detection records; The self-check module is used in the gas identification method to adjust the environment of the sensor array module to standard temperature and humidity conditions, and compare the actual detected data under these conditions with the attenuation predicted data to calibrate the sensor and update the sensor attenuation model.

4. A gas identification device equipped with the gas identification system according to claim 3, characterized in that: The device comprises a recognition device body (1), a lifting seat (2) is arranged on the recognition device body (1) through a through slot, and a gas sensor (3) is arranged on the lifting seat (2), a slider (5) is arranged on the lifting seat (2), and the slider (5) is slidably mounted on a spring guide rod (4), a magnetic sheet (6) is arranged on the slider (5), and an electromagnetic seat (7) is installed in the recognition device body (1), and the lifting seat (2) is received into the recognition device body (1) by energizing the electromagnetic seat (7), a sealing plate (10) is rotatably mounted on the recognition device body (1) through a side seat (9), and a transmission structure driven by the slider (5) is arranged in the side seat (9), and when the lifting seat (2) and the gas sensor (3) are received into the recognition device through the transmission structure, the sealing plate (10) closes the through slot, and a temperature and humidity control device is installed in the recognition device body (1).

5. A gas identification device according to claim 4, characterized in that: The through groove is arranged on the top surface of the identifier body (1), and the gas sensor (3) comprises a methane sensor, a carbon monoxide sensor, a hydrogen sulfide sensor, and a Bosch constant temperature sensor, and the gas sensor (3) array is arranged on the lifting seat (2).

6. A gas identification device according to claim 4, characterized in that: The spring on the spring guide rod (4) provides elastic force to make the lifting seat (2) extend out of the through slot, the magnetic sheet (6) is embedded in the side of the slider (5), and the electromagnetic seat (7) is installed at the bottom of the identifier body (1).

7. A gas identification device according to claim 4, characterized in that: The side seat (9) is provided with a torsion spring shaft (12), and the sealing plate (10) is connected to the torsion spring shaft (12), a rotating wheel (13) is installed on the torsion spring shaft (12), and a winding rope (14) is wound and connected to the rotating wheel (13), a sleeve (15) is vertically installed at the bottom of the side seat (9), and the winding rope (14) passes through the sleeve (15) for installation, a base (16) is installed at the bottom end of the winding rope (14), and a side pressure block (8) is installed on the slider (5), and after the slider (5) is lowered, the side pressure block (8) pushes the base (16) downward to drive the sealing plate (10) to rotate, and a sealing pad (11) is provided on the sealing plate (10).

Images