Multi-gas accurate detection method and system

Abstract

The invention relates to the technical field of gas detection, in particular to a multi-gas accurate detection method and system.The method comprises the following steps that gas to be detected flows through a miniature gas enrichment device prepared through an MEMS technology under driving of a miniature gas pump, and target gas is adsorbed and enriched through graphene and metal oxide composite sensitive materials on the surface of a spiral gas inlet flow channel in the miniature gas enrichment device; the desorption concentrated gas is heated to the multi-gas composite detector, and the sensor array reacts with the gas to generate an electric signal; the data acquisition unit performs noise reduction, amplification and analog-to-digital conversion preprocessing on the electric signal; the data processing unit extracts characteristic vectors such as a response curve peak value and a rising slope; inputting the feature vector into a pre-trained machine learning classification-regression combined model, and outputting a gas type identifier and an initial concentration; error compensation is carried out in combination with environment temperature and humidity and a cross sensitivity interference mode library, and finally accurate information of at least nine gases is output. The device has the advantages of high sensitivity, strong anti-interference performance and miniaturization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gas detection, in particular to a multi-gas accurate detection method and system. BACKGROUND

[0002] In the fields of industrial safety, environmental monitoring, public health, etc., the complex scene of coexistence of multiple gases puts forward strict requirements on detection technology: not only at least 9 target gases need to be identified synchronously, but also high sensitivity needs to be maintained for low-concentration gases, and the influence of environmental temperature and humidity fluctuations and multi-gas cross interference needs to be resisted; traditional gas detection technology has exposed significant limitations in such scenes; the existing technology has the following defects: Traditional sensors rely on the direct contact of gas molecules with sensitive materials, and the adsorption amount of low-concentration gases is limited, resulting in a high lower limit of detection, which cannot meet the early warning demand; The response of a sensor array to a mixed gas has "non-specificity", for example, SnO2 sensors exhibit resistance drop for both CO and H2, and it is difficult to distinguish them only by relying on a single feature, resulting in a concentration calculation deviation of more than 30%; Temperature and humidity changes directly affect the adsorption / desorption kinetics of sensors, and traditional methods lack a dynamic compensation mechanism, so when the temperature and humidity fluctuation is >5℃ or the humidity is >60%RH, the detection error significantly increases; Existing multi-gas detection equipment often uses a "single gas sensor stacking" mode, which is bulky and has rough gas path design, making it difficult to adapt to portable or embedded scenarios.

[0003] Therefore, a multi-gas accurate detection method and system are proposed to solve the above problems. SUMMARY

[0004] The purpose of the present application is to provide a multi-gas accurate detection method and system to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides the following technical solutions: A multi-gas accurate detection method, comprising the following steps: Step S1, gas enrichment: the target gas in the measured gas flows through a micro gas enrichment device under the driving of a micro gas pump; the micro gas enrichment device is prepared by MEMS technology, and a spiral-shaped gas inlet channel is arranged inside the micro gas enrichment device; the surface of the spiral-shaped gas inlet channel is covered with a sensitive material layer; the sensitive material layer is a composite material layer of graphene material and metal oxide; The target gas molecules in the measured gas are adsorbed and enriched by the sensitive material layer when flowing through the spiral-shaped gas inlet channel; Step S2, sensitive response: heating desorption is performed on the micro gas concentrator enriched with the target gas, and the released concentrated gas is transported to a multi-gas composite detector; the multi-gas composite detector comprises a sensor array sensitive to multiple gases, which is in contact with the sensitive material layer and reacts to generate corresponding electrical signals; Step S3, signal acquisition and preprocessing: the original electrical signals generated by the sensor array are acquired in real time by using a data acquisition unit, and the original electrical signals are preprocessed, including noise reduction, amplification and analog-to-digital conversion, to obtain a digital signal sequence; Step S4, feature extraction: the digital signal sequence is received by using a data processing unit, and a feature vector used for gas identification is extracted therefrom; the feature vector includes but is not limited to: peak value, response rising slope, falling slope, integral area of the sensor response curve, and response value at a specific time point; Step S5, gas identification and concentration calculation: the feature vector is input into a pre-trained gas identification model, and the gas identification model outputs the type identification and concentration value of the gas; the gas identification model is a classification and regression model based on a machine learning algorithm, which is obtained by training multiple standard gases of known types and concentrations by using the multi-gas composite detector; Step S6, error compensation and result output: an error compensation algorithm is used to correct the concentration value obtained in step S5, so as to reduce the identification error caused by cross-sensitivity and environmental interference; finally, the accurate types and concentrations of multiple gases after compensation are output, so as to realize the detection of at least 9 kinds of gases.

[0006] As a preferred solution, in step S1, the structural parameters of the spiral-shaped gas inlet flow channel are optimized and designed according to the adsorption kinetics model of the target gas, so as to realize the maximum adsorption efficiency of the target gas under a given flow channel volume; the structural parameters include the width, depth, pitch and total length of the flow channel; the metal oxide is one or more selected from SnO2, ZnO, WO3 and CuO; the graphene material is compounded with the metal oxide by a hydrothermal method or an in-situ growth method to form a three-dimensional porous nano structure, so as to increase the specific surface area of the sensitive material layer.

[0007] As a preferred solution, the original electrical signals generated by the sensor array are acquired in real time by using a data acquisition unit, and the original electrical signals are preprocessed, including noise reduction, amplification and analog-to-digital conversion, to obtain a digital signal sequence, including: The collected original electrical signals are subjected to noise reduction processing, and the noise interference is removed by filtering to obtain the noise-reduced signals; The noise-reduced signals are subjected to amplification processing, and the signal amplitude is enhanced by a signal amplifier to obtain the amplified signals; Analog-to-digital conversion is performed on the amplified signal, and an analog signal is converted into a digital signal sequence by an analog-to-digital converter.

[0008] As a preferred solution, the data processing unit receives the digital signal sequence and extracts a feature vector for gas identification therefrom, including: According to the digital signal sequence, a dynamic response curve of each sensor is reconstructed; The peak value of the response process is identified and extracted from the dynamic response curve to obtain a peak value feature; Based on the dynamic response curve, the rising slope of the curve in the response stage is calculated to obtain a rising slope feature; Based on the dynamic response curve, the falling slope of the curve in the recovery stage is calculated to obtain a falling slope feature; The integral area of the response curve is calculated by integrating the dynamic response curve within a set time window to obtain an integral area feature; The response value of the dynamic response curve at a pre-set specific time point is read to obtain a specific time point response value feature; The extracted peak value feature, rising slope feature, falling slope feature, integral area feature, and specific time point response value feature are combined to form a feature vector for representing the type and concentration of the gas.

[0009] As a preferred solution, the feature vector is input into a pre-trained gas identification model, and the gas identification model outputs the type identification and concentration value of the gas, including: The feature vector is input into a classification module of the gas identification model, which identifies the type of the gas based on the pre-processed feature vector and outputs the type identification of the gas; The type identification and the feature vector are input into a regression module of the gas identification model, which calculates and outputs the concentration value of the gas based on the concentration regression model corresponding to the type identification; The gas identification model is obtained through the following training process: The response signal of a standard gas with known type and concentration is collected using a multi-gas composite detector, and the training feature vector is obtained after steps S3 and S4. The feature vector and its corresponding known type and concentration label are used to supervise the training of the model based on the machine learning algorithm, and a joint model capable of simultaneously completing gas type classification and concentration regression is obtained.

[0010] As a preferred solution, in step S6, an error compensation algorithm is used to correct the concentration value obtained in step S5 to reduce the identification error caused by cross-sensitivity and environmental interference, including: Obtain the current environmental parameters, including the environmental temperature and humidity; The environmental parameter, the category identification and the concentration value output by the gas identification model are input into a pre-stored error compensation model; The error compensation model calls a pre-established cross-sensitivity interference mode library for the gas according to the input category identification; Based on the environmental parameter and the interference mode library, a concentration compensation coefficient in the current environment is calculated; The initial concentration value obtained in step S5 is corrected by using the concentration compensation coefficient to obtain an accurate concentration value after compensation; The accurate category identification and the accurate concentration value of each gas after compensation are integrated and output to form a final multi-gas detection result, and simultaneous accurate detection of at least nine kinds of gas is realized.

[0011] A multi-gas accurate detection system, which performs a multi-gas accurate detection method when detecting multiple gases.

[0012] As can be seen from the technical solutions provided by the above-mentioned application, the multi-gas accurate detection method and system provided by the application have the beneficial effects that: The combination of the micro gas concentrator and the high-performance sensitive material effectively solves the problem that low-concentration target gas is difficult to be accurately captured: the micro gas concentrator is prepared by using the MEMS process, the internal spiral-shaped gas inlet flow channel can prolong the contact time of the to-be-detected gas and the sensitive material, and the flow channel parameters are optimized based on a target gas adsorption kinetics model, so that high-efficiency adsorption of target gas molecules can be realized in a limited volume; the sensitive material layer is a composite material of graphene and SnO2, ZnO and other metal oxides, and a three-dimensional porous structure is formed by using a hydrothermal method or an in-situ growth method, so that the specific surface area is greatly increased and the adsorption capacity of target gas molecules is improved; the synergistic effect of the two can concentrate low-concentration (such as 0.1 ppm level) target gas to a range in which the detector can accurately respond, so that the application can detect trace gas that is difficult to be identified by traditional methods, and the lower limit of the concentration of gas detection is expanded; The technical combination of “sensor array + multi-feature extraction + machine learning model” breaks the limitations of traditional single-gas detection equipment: the multi-gas composite detector is configured with a sensor array sensitive to different types of gas (such as a graphene-SnO2 sensor for detecting reducing gas, a graphene-WO3 sensor for detecting oxidizing gas, and an electrochemical sensor for detecting polar gas), which can simultaneously respond to multiple gas signals; step S4 extracts multi-dimensional features such as the peak value, the rising slope and the integral area of the sensor response curve, which can accurately distinguish the signal differences of different gases; the gas identification model of step S5 is a “classification-regression combined model”, which can output the category identification and the initial concentration value of multiple gases after training by standard gas, and finally realizes simultaneous accurate detection of at least nine kinds of gas by combining the error compensation of step S6, which is suitable for complex scenes such as industrial environment and indoor air where multiple gases coexist. This invention effectively eliminates the impact of various interferences on the detection results through dual optimization of "signal preprocessing + error compensation": the signal preprocessing process in step S3 (noise reduction, amplification, analog-to-digital conversion) can remove industrial electromagnetic interference, sensor thermal noise, etc., ensuring the purity of the original electrical signal; the error compensation algorithm in step S6 specifically addresses two core interferences: first, it corrects the mutual interference between different gases through the "cross-sensitivity interference mode library"; second, it adjusts the concentration compensation coefficient in combination with environmental temperature and humidity parameters to eliminate errors caused by environmental fluctuations. The dual optimization significantly reduces the error of the final output gas concentration value, and the detection accuracy is much higher than that of traditional uncompensated schemes, ensuring the reliability of the results. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall structure of a multi-gas precision detection method and system according to the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0015] To better understand the above technical solutions, the following will provide a detailed description of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0016] like Figure 1 As shown, this embodiment of the invention provides a method for precise detection of multiple gases, including the following steps: Step S1, Gas enrichment: The gas to be tested flows through a micro gas enricher driven by a micro gas pump. The micro gas enricher is fabricated using MEMS technology and has a spiral inlet channel inside. The surface of the spiral inlet channel is covered with a sensitive material layer. The sensitive material layer is a composite material layer of graphene and metal oxide. When the target gas molecules in the gas to be tested flow through the spiral inlet channel, they are adsorbed and enriched by the sensitive material layer; Step S2, Sensitive Response: The micro gas enrichment device containing the target gas is heated and desorbed, and the released concentrated gas is delivered to the multi-gas composite detector; the multi-gas composite detector contains a sensor array sensitive to multiple gases, which contacts and reacts with the sensitive material layer to generate corresponding electrical signals. Step S3, Signal Acquisition and Preprocessing: The raw electrical signals generated by the sensor array are acquired in real time using the data acquisition unit, and the raw electrical signals are preprocessed, including noise reduction, amplification and analog-to-digital conversion, to obtain a digital signal sequence; Step S4, feature extraction: receiving the digital signal sequence by the data processing unit, and extracting the feature vector for gas identification therefrom; the feature vector includes but is not limited to: the peak value of the sensor response curve, the rising slope, the falling slope, the integral area, and the response value at a specific time point; Step S5, gas identification and concentration calculation: inputting the feature vector into the pre-trained gas identification model, and the gas identification model outputs the type identification and concentration value of the gas; the gas identification model is a classification and regression model based on a machine learning algorithm, which is obtained by training a plurality of standard gases of known types and concentrations by a multi-gas composite detector; Step S6, error compensation and result output: correcting the concentration value obtained in step S5 by using an error compensation algorithm to reduce the identification error caused by cross-sensitivity and environmental interference; finally, outputting the accurate types and concentrations of a plurality of gases after compensation, to realize the detection of at least 9 kinds of gases.

[0017] In this embodiment, the structural parameters of the spiral-shaped gas inlet flow channel are optimized and designed according to the adsorption kinetics model of the target gas, so as to realize the maximum adsorption efficiency of the target gas under a given flow channel volume; the structural parameters include the width, depth, pitch and total length of the flow channel; the metal oxide is one or more selected from SnO2, ZnO, WO3 and CuO; the graphene material and the metal oxide are compounded by a hydrothermal method or an in-situ growth method to form a three-dimensional porous nano structure, so as to increase the specific surface area of the sensitive material layer; Further, by miniaturizing the gas path design, compounding high-performance sensitive materials, and optimizing the structured flow channel, directional adsorption and efficient concentration of target gas molecules in the to-be-detected gas are realized, the concentration of the low-concentration target gas is improved to the range that can be accurately responded by the subsequent detector, and the sensitivity and accuracy of multi-gas detection are laid a foundation; the following are the detailed steps: Step S1-1: gas path driving system building and to-be-detected gas conveying: Miniature air pump selection and parameter matching: according to the flow rate requirement of the target gas and the concentrator volume, a piezoelectric or electromagnetic miniature air pump is selected to ensure that the air pump output flow is stable at 50-200 mL / min and the working pressure is controlled at 0.1-0.3 MPa, so as to avoid insufficient adsorption due to excessive air flow or low detection efficiency due to weak air flow; for example, for low-concentration (0.1-1 ppm) volatile organic compounds (VOCs), a piezoelectric air pump with a flow rate of 100 mL / min is selected to balance the adsorption efficiency and detection time; Gas path sealing and impurity filtration: Fluorine rubber sealing ring is used to seal the interfaces of gas pump, concentrator and connecting pipeline. The interfaces are reinforced by laser welding process to ensure that the gas path leakage rate is less than 0.1% / h. A metal mesh filter with a pore size of 5-10 μm is connected in series between the gas pump and the concentrator to remove dust, liquid droplets and other impurities in the test gas, avoiding clogging the flow channel or contaminating the sensitive material layer. Directional delivery control of test gas: The working time sequence of the gas pump is set by a programmable logic controller (PLC) to control the one-way flow of the test gas along the path "gas pump→filter→spiral inlet flow channel→concentrator outlet", avoiding backflow of gas flow causing the desorption of adsorbed target gas molecules, ensuring the stable residence time of gas flow in the flow channel (initially set to 2-5 s, and then fine-tuned according to the adsorption kinetics model); Step S1-2: MEMS process preparation of micro gas concentrator and flow channel optimization: MEMS process selection and substrate preparation: Single crystal silicon or sapphire is selected as the concentrator substrate (thickness 200-500 μm). The embryonic form of the spiral flow channel is constructed on the substrate surface by photolithography and dry etching (ICP-RIE) process. The advantage of MEMS process is that the flow channel size accuracy can reach ±5 μm, meeting the requirements of miniaturization and mass production, while ensuring the smoothness of the inner wall of the flow channel (roughness Ra≤0.1 μm), reducing the gas flow resistance and dead volume; Optimization of spiral inlet flow channel parameters: According to the adsorption kinetics model of the target gas (such as Langmuir monolayer adsorption model or BET multilayer adsorption model), the core parameters of the flow channel are optimized: Flow channel width: designed to be 200-300 μm, too wide will result in too fast gas flow speed and insufficient contact time with sensitive materials, too narrow will easily block, for small molecule gases such as CO, NO2, etc., the width can be appropriately narrowed to 150 μm to improve the contact frequency; Flow channel depth: set to 150-200 μm, increasing the depth can increase the amount of sensitive material attached (0.5-1 μg of sensitive material per millimeter of flow channel), but it needs to match the substrate thickness to avoid structural weakness; Flow channel pitch: controlled at 500-800 μm, to avoid short circuiting between adjacent flow channels, ensuring that the gas in each flow channel can fully contact the sensitive material; Total length of flow channel: according to the adsorption equilibrium time calculation, usually designed to be 5-8 cm, to ensure that the adsorption efficiency of target gas molecules in the flow channel is ≥85% (for example, for formaldehyde gas, an 8 cm flow channel can achieve an adsorption efficiency of 92%); Flow channel surface modification treatment: remove organic residues on the inner wall of the flow channel by plasma cleaning (Ar atmosphere, power 100-200W, time 5-10min), then coat a 10-20nm thick Al2O3 film using atomic layer deposition (ALD) process to improve the adhesion between the flow channel and the sensitive material layer, and prevent the sensitive material from falling off under the impact of the gas flow; Step S1-3: composite preparation and performance optimization of the sensitive material layer: Composite sensitive material component selection: according to the adsorption characteristics of the target gas, select graphene and a specific metal oxide for compounding: For reducing gases (such as CO, H2S), SnO2 (band gap 3.6eV, sensitive to reducing gases) is selected for compounding with graphene; For oxidizing gases (such as NO2, O3), WO3 (band gap 2.7eV, easily changes resistance in the presence of oxidizing gases) is selected for compounding with graphene; If multiple gases need to be detected simultaneously, SnO2-ZnO (mass ratio 3:1) can be used to compound with graphene to expand the detection range by utilizing the synergistic sensitive effect of different metal oxides; Preparation of three-dimensional porous structure (hydrothermal method / in-situ growth method): Hydrothermal method (take graphene / SnO2 as an example): 0.5g of graphene oxide is dispersed in 50mL of deionized water, ultrasonic treatment for 30min (power 300W) to obtain a uniform dispersion; 1mmol of SnCl4・5H2O is added, and after magnetic stirring for 1h, the pH is adjusted to 8-9 with ammonia water; the mixed solution is transferred to a polytetrafluoroethylene reaction kettle, and heated at 180℃ for 6h; after the reaction is completed, centrifugal separation is performed (speed 8000r / min, time 10min), and vacuum drying is performed at 60℃ for 12h to obtain a three-dimensional porous composite material, which has a specific surface area of 150-200m² / g and a porosity of 60%-70%, significantly improving the gas adsorption capacity; In-situ growth method (take graphene / WO3 as an example): graphene film is laid on the inner wall of the flow channel, and soaked in WO3 precursor solution (0.1mol / L Na2WO4 solution, pH adjusted to 2), and heated in a water bath at 60℃ for 3h to allow WO3 nanoparticles to grow in-situ on the surface of graphene, forming a three-dimensional structure of "graphene substrate-WO3 nanoneedle" to enhance the gas molecule capture ability; Sensitive material layer coating and curing: use microdroplet jet printing technology (resolution 5-10μm) to uniformly coat the composite sensitive material on the inner wall of the spiral flow channel, and control the coating thickness to be 5-10μm; after coating, perform annealing treatment in a nitrogen atmosphere (temperature 200-300℃, time 2h) to tightly bond the sensitive material layer with the flow channel substrate, and at the same time remove the residual solvents in the material to ensure stable adsorption performance; Step S1-4: Control of adsorption and enrichment process of target gas molecules: Dynamic adjustment of adsorption flow rate: Real-time monitoring of gas flow rate by flow controller, fine-tuning of flow rate according to initial concentration of target gas: When the concentration of target gas is 0.1-1 ppm (low concentration), the flow rate is reduced to 80-100 mL / min, and the residence time of gas flow in the flow channel is extended to 4-5 s to ensure sufficient adsorption; When the concentration is 1-10 ppm (medium concentration), the flow rate is set to 120-150 mL / min, and the residence time is 2-3 s to balance the adsorption efficiency and detection speed; Flow rate adjustment based on the "concentration-flow rate-adsorption efficiency" curve drawn from pre-experiments to ensure that the enrichment factor at different concentrations is stable at 10-50 times; Monitoring and termination of adsorption saturation: A micro auxiliary sensor (such as a semiconductor gas sensor) is installed at the outlet of the enricher to monitor the concentration of target gas in the outlet gas in real time; When the outlet concentration remains stable for 30 s (fluctuation ≤5%), it is determined that the adsorption has reached equilibrium, triggering the gas pump to stop working to avoid excessive adsorption leading to saturation of the sensitive material (the desorption efficiency will decrease by 20%-30% after saturation); Environmental control during enrichment process: Wrap the enricher with a constant temperature heating sheet (temperature controlled at 25-30°C) to avoid environmental temperature fluctuations (such as ±5°C) causing changes in the adsorption performance of the sensitive material; At the same time, monitor the humidity in the gas path through a humidity sensor, and when the humidity >60%RH, start a micro dehumidification module (using molecular sieve dehumidification) to reduce the interference of water vapor on the adsorption of polar gases (such as formaldehyde, NH3), ensuring the stability of the enrichment process.

[0018] In this embodiment, precise heating desorption is used to achieve efficient release of target gas in the enricher, and the concentrated target gas is directed to the multi-gas composite detector, and the sensor array and sensitive material layer of the detector react specifically to convert the gas component signal into an electrical signal that can be collected, providing raw response data for subsequent signal processing and gas identification. The following are the detailed steps: Step S2-1: Construction and parameter control of micro gas enricher heating desorption system: Design and installation of MEMS integrated heating module: Use MEMS technology to integrate ultra-thin heating sheets (thickness 5-10 μm, material Ti / Pt alloy, resistance 50-100 Ω) on the back of the micro gas enricher substrate, and precisely align the heating sheets with the flow channel area (covering more than 90% of the total flow channel area) to ensure uniform heat transfer to the sensitive material layer in the flow channel. The heating module is connected to the external driving circuit through gold wire bonding to achieve precise current control and avoid local overheating damage to the sensitive material; Dynamic optimization of heating desorption parameters: according to the desorption energy of target gas (such as CO desorption energy about 20-30 kJ / mol, formaldehyde desorption energy about 35-45 kJ / mol) and the thermal stability of sensitive material (graphene-metal oxide composite material resistant temperature ≤300℃), the heating parameters are set by PID (proportional-integral-derivative) controller: Heating rate: control for 5-10℃ / s, too fast will easily lead to the collapse of the structure of sensitive material, too slow will prolong the desorption time (total desorption time control in 30-60s); Desorption temperature: set different temperature for different target gas, such as set for 150-180℃ when detecting CO, H2S, set for 180-220℃ when detecting NO2, O3, to ensure the desorption efficiency of target gas ≥95%, and avoid the crystal transformation of metal oxide (such as SnO2, WO3); Soaking time: after reaching the set desorption temperature, soak for 5-10s to ensure that the target gas molecules adsorbed in the flow channel are completely released, the flow channel temperature is monitored in real time by micro thermocouple sensor (accuracy ±0.5℃), and feedback to PID controller to realize temperature fluctuation ≤±2℃; Desorption process protection mechanism: series overcurrent protection resistor (100-200Ω) in heating circuit, when the heating sheet current exceeds the rated value (such as 50mA), the circuit is automatically cut off; at the same time, temperature fuse sheet (fusing temperature 250℃) is attached to the surface of the concentrator, to prevent temperature runaway damage to the concentrator in extreme cases, and ensure the safety and stability of the desorption process; Step S2-2: Concentrated gas directional delivery system construction and loss control: Transportation path planning and component selection: design "micro gas concentrator outlet→ micro one-way valve→ heat preservation pipeline→ multi-gas composite detector inlet" directional delivery path, component selection needs to match the miniaturization demand: Micro one-way valve: select silicone valve core (diameter 1-2mm), working pressure 0.05-0.2MPa, to prevent concentrated gas backflow to the concentrator (backflow will lead to secondary adsorption of undesorbed gas, reducing the concentration of concentration); Heat preservation pipeline: use polytetrafluoroethylene material pipeline (inner diameter 2-3mm, length 5-10cm), pipe outside wrapped with aluminum foil heat insulation layer with thickness of 10-15μm, to avoid condensation of concentrated gas in the delivery process due to temperature drop (such as VOCs gas condensation will lead to concentration loss ≥10%); Auxiliary delivery micro gas pump: series micro gas pump (flow 50-100mL / min) between one-way valve and detector, to provide stable delivery power, ensure the flow rate of concentrated gas in the pipeline is controlled at 0.5-1m / s, avoid too fast flow rate leading to insufficient response of detector, or too slow flow rate leading to gas diffusion loss; Concentration stability monitoring during delivery: A miniature optical concentration sensor (detection accuracy 0.1 ppm) is installed in the middle section of the heat preservation pipeline to monitor the concentration changes of the concentrated gas in real time. When the concentration fluctuation exceeds ±5%, the feedback is sent to the auxiliary gas pump controller to adjust the flow rate (e.g., if the concentration decreases, the flow rate is appropriately increased to 1.2 m / s to reduce the diffusion time), ensuring that the concentration deviation of the concentrated gas delivered to the sensor array is ≤±3%, providing stable gas samples for the sensor array; Step S2-3: Multi-gas composite probe structure design and sensor array configuration: Probe cavity design and flow field optimization: The probe cavity is made of stainless steel (volume 5-10 mL), with an internal "diffusion type" flow channel (inlet → buffer chamber → sensor array area → outlet). The buffer chamber has a volume of 1-2 mL, which can reduce the gas flow impact when the concentrated gas enters, avoiding fluctuations in sensor response. The flow channel structure is optimized through fluid simulation (e.g., CFD) to ensure uniform flow velocity (deviation ≤10%) in the sensor array area, and each sensor can contact an equal amount of concentrated gas; Sensor array selection and arrangement: According to the requirement of "detecting at least 9 kinds of gas", a multi-type sensor array (total number 8-12) is configured, and the sensor types and detection targets are matched as follows: Graphene-SnO2 composite sensor (2-3): For reducing gases (CO, H2, methane), the semiconductor properties of SnO2 are utilized, and the sensor resistance decreases after target gas adsorption; Graphene-WO3 composite sensor (2-3): For oxidizing gases (NO2, O3, Cl2), the sensor resistance rises after target gas adsorption; Graphene-ZnO composite sensor (1-2): For sulfides (H2S, SO2), it has specific response; Electrochemical sensor (2-3): For polar gases (formaldehyde, NH3, ethanol), an electric current signal proportional to the concentration is generated through electrochemical reaction; The sensor array is installed in the middle of the probe cavity in a "ring-shaped uniform arrangement" manner, with the sensor sensitive surface facing the gas flow direction, and the distance between adjacent sensors is 5-8 mm to avoid mutual interference (e.g., the electrolyte evaporation of electrochemical sensors affecting semiconductor sensors); Sensor array preheating and baseline calibration: Before the concentrated gas is delivered, the sensor array is preheated for 10-15 min (temperature 25-30°C, consistent with the ambient temperature), and the initial signals of each sensor (such as the initial resistance of the semiconductor sensor and the initial current of the electrochemical sensor) are monitored by the data acquisition unit. The "baseline threshold range" is set (such as an initial resistance fluctuation of ≤2% and an initial current of ≤0.1 μA), and if it exceeds the range, the calibration program is triggered (by passing pure nitrogen to adjust the baseline to the standard value) to ensure that the initial state of the sensor is consistent and to reduce subsequent response errors; Step S2-4: Sensor array reaction with sensitive material layer and generation of electrical signals: Concentrated gas contact reaction with sensor sensitive material: After the concentrated gas enters the detector cavity, it reacts with the sensitive material layer on the surface of the sensor array (the same as the "graphene material and metal oxide composite material layer" in S1, the sensor sensitive surface is pre-coated with this material with a thickness of 3-5 μm) to produce specific interactions: For semiconductor sensors (graphene-SnO2 / WO3 / ZnO): target gas molecules are adsorbed on the surface of the sensitive material, and the charge transfer changes the carrier concentration of the material (such as CO molecules transferring electrons to SnO2, increasing the carrier concentration and reducing the resistance); For electrochemical sensors: target gas molecules undergo redox reactions between the working electrode and the counter electrode of the sensor (such as formaldehyde being oxidized to CO2 and H2O at the working electrode, producing electron transfer and forming an electric current); The reaction process lasts for 5-10 s until the sensor signal reaches a stable response state (signal fluctuation ≤1% / s); Real-time monitoring and preliminary determination of electrical signals: the electrical signals (resistance change signals for semiconductor sensors and current signals for electrochemical sensors) are transmitted to the pre-signal conditioning circuit through the sensor array leads to monitor the signal change trend in real time: If the change in the electrical signal of a certain sensor is ≥5% (relative to the baseline value), it is determined that the sensor has an effective response to the target gas, and the response start time and signal rise rate are recorded; If the change in the electrical signal is <5%, it is determined that the sensor has not detected the corresponding target gas, and it is marked as "no response" to avoid invalid signals entering the subsequent processing flow; At the same time, an electromagnetic shielding layer (material: copper foil, thickness: 5-10 μm) is added to the circuit to reduce signal noise caused by external electromagnetic interference (such as industrial equipment electromagnetic radiation) and ensure that the signal-to-noise ratio of the original electrical signal is ≥30 dB, providing high-quality data sources for the signal preprocessing of S3.

[0019] In this embodiment, the data acquisition unit is used to collect the original electric signal generated by the sensor array in real time, and the original electric signal is preprocessed, including noise reduction, amplification and analog-digital conversion, to obtain a digital signal sequence, including: The collected original electric signal is subjected to noise reduction processing, and noise interference is removed through filtering to obtain a noise-reduced signal. The noise-reduced signal is subjected to amplification processing, and the signal amplitude is enhanced through a signal amplifier to obtain an amplified signal. The amplified signal is subjected to analog-digital conversion processing, and the analog signal is converted into a digital signal sequence through an analog-digital converter. Further, the data acquisition unit is used to capture the original electric signal output by the sensor array in the multi-gas composite detector in real time, and then a three-stage preprocessing procedure of “noise reduction-amplification-analog-digital conversion” is used to eliminate signal interference, enhance weak signal strength, and convert continuous analog signals into calculable discrete digital signal sequences, thereby providing a standardized and high-quality data source for the subsequent feature extraction in step S4, and ensuring the accuracy of subsequent gas recognition and concentration calculation from the signal level. The following are the detailed steps: Step S3-1: The data acquisition unit is built and the original electric signal is captured in real time: Hardware adaptation and function positioning of the acquisition unit: The data acquisition unit needs to be precisely matched with the sensor array of the multi-gas composite detector (including sensors sensitive to multiple gases, such as graphene-metal oxide composite semiconductor sensors and electrochemical sensors), and the core function is to capture the original electric signal output by each sensor in real time without omission. According to the signal characteristics of the sensor array (such as the voltage signal corresponding to the change of the output resistance of the semiconductor sensor and the current signal output by the electrochemical sensor), the acquisition unit needs to have multi-channel synchronous acquisition capability to ensure that the signals of different sensors are aligned in time dimension (to avoid timestamp misalignment of the subsequent response curve due to acquisition delay), and the acquisition frequency needs to match the dynamic response speed of the sensor (such as being able to capture the whole process of the signal from rising to stable and then recovering without losing key features such as peak value and slope). Original electric signal capture and temporary storage: After starting the acquisition unit, it will receive the original electric signal transmitted by the sensor array in real time (the signal amplitude may differ, such as weak mV-level voltage signal or nA-level current signal), and temporarily store it in the local cache module (such as high-speed RAM) in the format of “sensor channel number-acquisition timestamp-original signal value”. The temporary storage process needs to ensure data integrity to avoid loss of original signal due to transmission interruption or storage delay, and to provide a complete original data basis for subsequent preprocessing. Step S3-2: Noise reduction processing of the original electric signal (first stage of preprocessing): Noise source and filtering strategy matching: There are two main types of interference that can easily mix into the original electrical signal: one is electromagnetic interference in industrial environments (such as power frequency 50Hz / 60Hz interference), equipment vibration interference; the second is the thermal noise of the sensor itself, circuit noise (high-frequency random noise); For different types of noise, use the corresponding filtering method to remove interference: For power frequency electromagnetic interference, use notch filtering (center frequency matches power frequency, stopband attenuation needs to ensure that the interference signal amplitude is reduced to less than 5% of the original value) to accurately filter out periodic interference in this frequency band, avoiding regular fluctuations in the signal caused by it; For high-frequency random noise (frequencies are usually higher than the effective signal bandwidth of the sensor), use low-pass filtering (cutoff frequency is set according to the effective signal range of the sensor, such as not more than 100Hz) to retain the effective components of the signal related to the gas response, while eliminating the masking of high-frequency noise on signal details; Noise reduction effect verification: After filtering, the signal waveform before and after noise reduction needs to be compared to ensure that the effective signal (such as the rising edge of the sensor response, peak value) has not been distorted, and the noise amplitude has been reduced to less than 10% of the original signal. Finally, get a smooth and pure noise-reduced signal to exclude the interference of subsequent amplification processing; Step S3-3: Signal amplification processing after noise reduction (second stage of preprocessing): Amplification needs and amplifier selection: The signal after noise reduction may still be weak (such as the signal output by the electrochemical sensor after noise reduction is still 0.1-1mV level voltage), if directly into analog-to-digital conversion, easy to cause precision loss due to signal amplitude below the conversion threshold; Therefore, the signal amplitude needs to be enhanced through a signal amplifier: The amplifier is preferably a programmable gain amplifier (PGA), which can dynamically adjust the amplification factor according to the signal amplitude difference of different sensors (such as semiconductor sensor signals may be 0.5-2V, no high gain is needed; Electrochemical sensor signal needs 100-1000 times gain), to ensure that the signal amplitude of all sensors after amplification is uniformly in the range of 0.5-5V (adapted to the best input range of subsequent analog-to-digital conversion), and the amplification process needs to be linear (non-linear distortion ≤2%), to avoid introducing false signal characteristics due to amplification; Amplified signal stability detection: After amplification is completed, the signal amplitude fluctuation needs to be monitored in real time, if the fluctuation amplitude exceeds ±3% (relative to the mean value of the signal after amplification), it is determined to be abnormal (may be amplifier parameter drift or temporary sensor failure), trigger the re-amplification process to ensure that the output amplified signal is stable and reliable; Step S3-4: Analog-to-digital conversion of the amplified signal (third stage of preprocessing): Analog-to-digital converter (ADC) parameter configuration: In order to convert continuous analog signals into discrete digital signals that can be recognized by the data processing unit, an appropriate ADC needs to be selected, and the core parameter configuration needs to meet: Resolution: Not less than 12 bits, to ensure that the amplified signal can be distinguished from the small changes (such as 0.5-5V, 12-bit ADC can achieve about 1.2mV resolution, enough to capture the signal difference corresponding to the change of gas concentration); Conversion rate: Synchronized with the acquisition frequency of the data acquisition unit, to avoid data accumulation due to insufficient conversion speed, and to ensure that each amplified analog signal sampling point can be converted in time; Reference voltage: Use high-precision reference voltage source (such as 2.5V±0.01%), to avoid reference voltage fluctuations leading to different digital values corresponding to the same analog signal, and to ensure conversion accuracy; Digital signal sequence generation and output: After the ADC completes the conversion, the output digital signal is organized in a structured format of "acquisition timestamp-sensor channel number-digital signal value", forming a digital signal sequence; This sequence needs to be transmitted to the data processing unit (such as embedded processor, cloud processing module) for storage, while it needs to be verified (such as data length verification) to ensure that the sequence is complete and error-free, and finally provides a standardized digital data source for feature extraction in step S4.

[0020] In this embodiment, the data processing unit receives the digital signal sequence and extracts the feature vector for gas identification, including: According to the digital signal sequence, the dynamic response curve of each sensor is reconstructed; From the dynamic response curve, the peak value of the response process is identified and extracted, obtaining the peak value feature; Based on the dynamic response curve, the rising slope of the curve in the response stage is calculated, obtaining the rising slope feature; Based on the dynamic response curve, the falling slope of the curve in the recovery stage is calculated, obtaining the falling slope feature; Integrate the dynamic response curve within a set time window, calculate the integral area of the response curve, and obtain the integral area feature; Read the response value of the dynamic response curve at a pre-set specific time point, and obtain the specific time point response value feature; Combine the extracted peak value feature, rising slope feature, falling slope feature, integral area feature, and specific time point response value feature to form a feature vector representing the gas type and concentration; Further, step S4 analyzes and transforms the digital signal sequence output by step S3 through the data processing unit, first reconstructs the dynamic response curve of each sensor, and then extracts the key features that can accurately represent the gas type and concentration from the curve. Finally, the standardized feature vector is formed by combining the key features, which provides input parameters with discrimination for the gas recognition model in step S5. It is the core link between "preprocessed digital signal" and "gas recognition calculation", and directly affects the accuracy of subsequent gas type judgment and concentration calculation. The following are the detailed steps: Step S4-1: Digital signal sequence reception and preliminary arrangement: Signal reception and channel classification: the data processing unit first receives the digital signal sequence output by step S3 (format: "timestamp-sensor channel number-digital signal value"), and classifies the signals according to "sensor channel number"; each channel corresponds to a specific sensor in the multi-gas composite detector (such as a graphene-SnO2 composite sensor for detecting reducing gases, a graphene-WO3 composite sensor for detecting oxidizing gases, etc.), ensuring that the signals of different sensors are independently grouped to avoid confusion; Data integrity check and completion: for each group of sensor digital signal sequence, sort by timestamp in ascending order, check for missing timestamps or abnormal signal values (such as extreme values outside the normal response range of the sensor); if there is a small amount of missing data, complete it by "linear interpolation of adjacent signal values", ensuring the continuity of the signal sequence; if there are abnormal values, they are determined as invalid data and are removed, and the mean value of the normal signals before and after the sensor is used to replace them, ensuring the accuracy of the subsequent curve reconstruction; Step S4-2: Sensor dynamic response curve reconstruction: Curve coordinate system setting: take "time" as the horizontal axis (unit: seconds, value range covering the whole period from the start of the sensor response to the signal returning to the baseline, usually 30-60 seconds), and "digital signal value" as the vertical axis (unit: volts, corresponding to the voltage response value of the sensor after preprocessing), to establish the dynamic response curve coordinate system of each sensor; Continuous curve fitting: based on the sorted digital signal sequence, use a smoothing fitting algorithm (such as cubic spline fitting) to connect the discrete signal points into a continuous dynamic response curve; the fitting process needs to strictly preserve the true trend of the signal; including "response phase" (signal rising from baseline), "stable phase" (signal reaching peak value and then stabilizing for a short time), "recovery phase" (signal falling from peak value to baseline), avoiding overfitting that may cause false peaks or sudden changes in slope in the curve, ensuring that the curve can truly reflect the interaction process between the sensor and the target gas; Step S4-3: Core feature extraction (five key features): Step S4-3-1: Response process peak feature extraction: Response stage definition: Firstly, identify the "response stage" from the dynamic response curve; that is, the interval from the first deviation of the signal from the baseline (deviation amplitude ≥ 5% of the baseline value, the baseline value is the initial stable signal value after preheating of the sensor in step S2) to the first continuous downward trend (signal value decreases at 3 consecutive time points) after the signal reaches the maximum value, which is the core range of the peak value; Peak confirmation and recording: In the defined response stage, the maximum value of the signal value is selected as the "initial peak value"; to exclude false peaks caused by noise (such as small sharp points formed by transient interference), the signal values at 5 time points before and after the initial peak value are verified: if the initial peak value is greater than 1.1 times of all verification point signal values, then the value is confirmed as "valid peak value"; if not, reselect the second largest value and repeat the verification, and finally record the peak characteristics (i.e. the specific value of the valid peak value) of the sensor response curve; Step S4-3-2: Response stage rising slope feature extraction: Rising stage division: Taking the "response stage starting point" (the time point at which the signal deviates from the baseline by ≥5%, denoted as t1) and the "peak value occurrence time point" (denoted as t2) as boundaries, the "rising stage" of the curve is defined; this stage reflects the rate at which target gas molecules are adsorbed by the sensor sensitive material, the faster the rate, the shorter the rising stage and the greater the slope; Slope calculation: Linear fitting is performed on the curve of the rising stage, and the curve slope of the rising stage is obtained through the fitting result, which is the "rising slope feature"; if the rising stage curve is highly nonlinear (low fitting goodness), then the "two-point method" is used to calculate: the difference between the peak signal value and the signal value at t1 is divided by the time difference between t2 and t1 to obtain the rising slope, ensuring that this feature accurately reflects the speed of gas adsorption; Step S4-3-3: Recovery stage descending slope feature extraction: Recovery stage division: Taking the "peak value occurrence time point t2" and the "time point at which the signal recovers to 90% of the baseline" (denoted as t3) as boundaries, the "recovery stage" of the curve is defined; this stage reflects the rate at which target gas molecules are desorbed from the sensor sensitive material, the faster the rate, the shorter the recovery stage and the greater the slope (absolute value); Slope calculation: Consistent with the logic of rising slope extraction, linear fitting is performed on the curve of the recovery stage to obtain the curve slope of the descending stage (usually negative, reflecting the signal downward trend), which is the "descending slope feature"; if the curve is highly nonlinear, the "two-point method" is used to calculate: the difference between the signal value at t3 and the peak signal value is divided by the time difference between t3 and t2 to obtain the descending slope, which characterizes the speed of gas desorption; Step S4-3-4: Response curve integral area feature extraction: Integral time window setting: uniformly set the integral time window as "from the starting point of the response phase t1 to the ending point of the recovery phase t3", ensuring to cover the whole cycle of the interaction between the sensor and the target gas, and the integral area size can reflect the total reaction amount of the target gas and the sensitive material of the sensor, which is positively correlated with the gas concentration; Integral area calculation: calculate the area between the dynamic response curve and the baseline (the baseline value is the initial stable signal value of the sensor) in the time window, which is the "integral area feature"; during the calculation process, all the areas enclosed by the curve and the baseline need to be covered to avoid area calculation deviation caused by too small window setting, which affects the judgment of the total reaction amount of the gas; Step S4-3-5: feature extraction of response value at specific time point Specific time point presetting: based on a large amount of standard gas experimental data (basic data provided for gas recognition model training in step S5), 2-3 specific time points with significant discrimination degree are preset, for example: The first time point: 1 second after the starting point of the response phase t1 (reflecting the signal intensity in the initial stage of gas rapid adsorption); The second time point: the middle time point between the peak value occurrence time point t2 and the ending point of the recovery phase t3 (reflecting the signal intensity in the stable desorption stage of the gas); The preset principle is to ensure that these time points can effectively distinguish the response differences of different gases and avoid setting the time points in the area with gentle signal change; Response value reading: on the dynamic response curve, the signal value corresponding to the preset specific time point is accurately read, if the time point does not exactly correspond to the discrete signal point of the fitting curve, the accurate response value is obtained through linear interpolation of the adjacent signal points, and finally the "response value features at specific time points" are recorded, which supplement the dynamic information not covered by the peak value, the rising slope and the falling slope; Step S4-4: feature vector combination and standardization Single sensor feature set construction: for each sensor in the multi-gas composite detector, the "peak value feature, rising slope feature, falling slope feature, integral area feature, and response value feature at specific time point" are combined in a fixed order (such as "peak value → rising slope → falling slope → integral area → response value at the first specific time point → response value at the second specific time point") to form the "single sensor feature set" of the sensor; Multi-sensor feature vector integration: according to the channel number of the sensor array (such as first reducing gas sensor, then oxidizing gas sensor, and finally polar gas sensor), the "single sensor feature sets" of all sensors are sequentially spliced to form the "initial feature vector" covering the entire sensor array; Feature standardization processing: in order to eliminate the influence of different feature magnitude differences on the subsequent gas identification model (for example, the integral area value may be 10-3 order of magnitude, and the peak value may be 10 0 order of magnitude), the initial feature vector is standardized; by calculating the mean and standard deviation of the feature in the historical training data, each feature value is converted into a standardized value of a uniform order of magnitude, and finally a "gas identification feature vector" is obtained, which is transmitted to the gas identification model in step S5.

[0021] In this embodiment, the feature vector is input into the pre-trained gas identification model, and the gas identification model outputs the type identification and concentration value of the gas, including: The feature vector is input into the classification module of the gas identification model, which identifies the type of gas based on the preprocessed feature vector, and outputs the type identification of the gas; The type identification and the feature vector are input into the regression module of the gas identification model, which calculates and outputs the concentration value of the gas based on the concentration regression model corresponding to the type identification; The gas identification model is obtained through the following training process: The response signal of the standard gas with known type and concentration is collected using a multi-gas composite detector, and the training feature vector is obtained after steps S3 and S4; the feature vector and its corresponding known type and concentration label are used to supervise the training of the model based on the machine learning algorithm, and a joint model capable of simultaneously completing gas type classification and concentration regression is obtained; Further, step S5 inputs the standardized feature vector output by step S4 into the pre-trained gas identification model, determines the gas type through the classification function of the model, calculates the gas concentration through the regression function, realizes the conversion from "feature data" to "gas information", is the core judgment link of the present application for completing multi-gas accurate detection, directly outputs the gas type identification and initial concentration value, and provides basic data for error compensation in subsequent step S6; the following are the detailed steps: Step S5-1: feature vector input verification and model adaptation: Feature vector integrity and effectiveness verification: the data processing unit first receives the gas identification feature vector output by step S4, checks whether the dimension and feature quantity of the vector are consistent with the input requirements of the gas identification model (for example, if the model is designed for 8 sensors and 6 types of features, the input vector needs to be 48-dimensional and have no missing feature values); at the same time, it is checked whether there are abnormal values (such as extreme values exceeding the feature range of the historical training data) in the feature vector; if there are a small number of abnormal values, they are corrected by "feature value truncation" (adjusting the abnormal values to the maximum / minimum value of the feature in the training data); if the proportion of abnormal values exceeds 10%, it is determined that the feature vector is invalid, and the feature is extracted again in step S4 to ensure that the data input into the model is compliant; Feature vector and model input format adaptation: convert the passed feature vector into a format supported by the gas identification model (such as tensor format, array format), while matching the input interface protocol of the model (such as API interface, local data interaction protocol), to ensure that the feature vector can be transmitted to the calculation module of the model completely and without delay, avoiding data loss or calculation errors due to format incompatibility; Step S5-2: The classification module performs gas type identification: Core function and algorithm logic of the classification module: the classification module of the gas identification model is constructed based on machine learning algorithms (such as random forest, support vector machine, neural network, etc.), and its core function is to use the "feature differences" of different gases in the feature vector (for example, CO and NO2 have significant differences in peak value and rising slope on the graphene-SnO2 sensor) to determine the type of the gas to be measured; the classification module matches the input feature vector through the "feature-type" correspondence learned in the training phase; for example, compare the peak value distribution of each sensor in the feature vector, the response value change trend at a certain time point, and whether it is consistent with the feature pattern of the known gas in the training data; Gas type identification output and confidence judgment: after the classification module completes feature matching, it outputs the corresponding gas type identification (such as "CO-001" for carbon monoxide and "NO2-003" for nitrogen dioxide, which includes the gas name abbreviation and unique code for easy subsequent call of the regression model); at the same time, it outputs the confidence of this type of judgment (such as 98%), if the confidence is lower than the preset threshold (such as 85%), it is determined as "to be confirmed type", which needs to be further verified in combination with the response consistency of multiple sensors (for example, if the sensors detecting oxidizing gases all respond, while the reducing gas sensors do not respond, the confidence of the oxidizing gas type can be improved), and finally determine the unique and high-confidence gas type identification; Differentiation and processing of mixed gas types: if the multi-gas composite detector detects multiple gases at the same time (i.e. the feature vector has the feature patterns of multiple gases), the classification module identifies the type of each gas through "multi-label classification logic"; for example, first determine the "main gas" type with the highest proportion in the feature vector, and then identify the secondary gas type, and output the corresponding type identification (such as "main gas: CO-001, secondary gas: H2S-005"), to ensure that the invention can realize simultaneous identification of multiple gases and meet the requirement of "at least 9 gas detection"; Step S5-3: The regression module calculates the gas concentration value: Linkage logic between regression module and classification module: after the classification module outputs the gas species identification, the identification will be input into the regression module together with the feature vector of step S4; the regression module will automatically call the "concentration regression sub-model" pre-trained for the gas according to the species identification; the correlation between the concentration of different gases and the feature vector is different (for example, the concentration of formaldehyde is more strongly related to the integral area of the electrochemical sensor, while the concentration of CO is more strongly related to the peak value of the graphene-SnO2 sensor), so a regression sub-model needs to be built for each gas to ensure the accuracy of the concentration calculation; Concentration value calculation and preliminary result output: the regression sub-model performs numerical calculation on the input feature vector based on the quantitative relationship between "features and concentration" (for example, the higher the peak value of a sensor in the feature vector, the higher the concentration of the corresponding gas); for example, combined with the "peak value-concentration" variation law fitted in the training stage, the corresponding gas concentration value (in units such as ppm, mg / m³) is calculated according to the peak value of the current feature vector; after the calculation is completed, the initial concentration value of the gas is output, and the feature weight used in the calculation process (such as the peak value of a certain sensor accounts for 60% in the concentration calculation) is recorded, providing a reference for error compensation in subsequent step S6; Step S5-4: training and optimization process of gas identification model: Training data preparation and selection of standard gas: the training of the gas identification model requires the use of standard gases with known species and concentrations (at least covering 9 target detection gases, each gas containing 5-10 concentration gradients such as 0.1 ppm, 0.5 ppm, 1 ppm, 5 ppm, etc., covering the concentration range of actual detection scenarios); the response signals of these standard gases are collected by the multi-gas composite detector of the present application, and a large number of "feature vector-known species-known concentration" training samples are generated through step S3 (signal acquisition and preprocessing) and step S4 (feature extraction), to construct a training data set (the sample amount needs to be ≥1000 groups to ensure sufficient model training); Model construction and supervised training: based on the training data set, a "classification-regression combined model" is constructed by supervised training; first, the classification sub-model is trained with "feature vector" as input and "known gas species" as label, and the model parameters (such as the number of decision trees of random forest and the number of hidden layer nodes of neural network) are adjusted to make the species identification accuracy reach the preset target (such as ≥95%); then, for each gas, the corresponding regression sub-model is trained with "feature vector" as input and "known gas concentration" as label, to control the concentration calculation error within the preset range (such as ≤5%); during the training process, the "cross-validation" method (dividing the training data set into training set and validation set, for example, 8:2) is adopted to avoid model overfitting (such as the difference between the identification accuracy of the validation set and the training set is ≤3%, which is determined as no overfitting); Model iteration optimization and performance verification: after training, use "test standard gas" (covering different concentrations of 9 gases) that does not participate in training to verify the performance of the model; if the accuracy of species identification or the error of concentration calculation does not meet the standard, supplement more training samples (especially low concentration, mixed gas samples), adjust the model parameters and train again until the model performance meets the design requirements; finally, a stable gas identification model is formed and stored in the system for actual gas identification and concentration calculation in step S5.

[0022] In this embodiment, step S6 uses an error compensation algorithm to correct the concentration value obtained in step S5 to reduce the identification error caused by cross-sensitivity and environmental interference, including: Obtain the current environmental parameters, including environmental temperature and humidity; Input the environmental parameters, species identification and concentration value output by the gas identification model into the pre-stored error compensation model; The error compensation model calls the pre-established cross-sensitivity interference mode library for the gas according to the input species identification; Based on the environmental parameters and the interference mode library, calculate the concentration compensation coefficient under the current environment; Use the concentration compensation coefficient to correct the initial concentration value obtained in step S5 to obtain the accurate concentration value after compensation; Integrate the accurate species identification and accurate concentration value of various gases after compensation to form the final multi-gas detection result, realizing the synchronous accurate detection of at least 9 gases; Further, step S6 corrects the initial concentration value output by step S5 through the error compensation algorithm to eliminate the influence of cross-sensitivity (mutual interference between different gases) and environmental interference (temperature and humidity changes) on the detection result, and finally integrates and outputs accurate gas species and concentration information, ensuring that the present application realizes the synchronous accurate detection of at least 9 gases, which is a key optimization link to improve the accuracy of multi-gas detection; the following are the detailed steps: Step S6-1: Current environmental parameter acquisition and preprocessing: Environmental parameter acquisition module deployment and selection: deploy the environmental parameter acquisition module in the surrounding area of the multi-gas composite detector (ensure consistency with the environment of the detection cavity), and acquire two key parameters of "environmental temperature" and "environmental humidity"; the temperature acquisition selects a high-precision platinum resistance temperature sensor (measurement range -20℃ to 60℃, accuracy ±0.2℃), and the humidity acquisition selects a capacitive humidity sensor (measurement range 10%RH to 90%RH, accuracy ±3%RH); the sampling frequency of the two types of sensors is set to 1Hz, ensuring real-time reflection of environmental changes (avoiding compensation deviation due to lagging environmental parameters); Environmental parameter data preprocessing: The collected temperature and humidity raw data are filtered (using sliding average filtering with a window size of 5 sampling points) to eliminate noise caused by transient environmental fluctuations (such as temperature sudden change caused by air flow sweeping over the sensor); at the same time, check if the data is out of the normal measurement range of the sensor, if there are abnormal values (such as temperature > 60℃ or humidity < 10%RH), replace it with the "average value of the previous 5 valid data" to ensure that the input error compensation model is stable and reliable; Step S6-2: Input data integration of error compensation model: Input data type and source confirmation: The error compensation model needs to receive three types of core data, and needs to ensure data integrity and relevance: Step S5 output "gas type identification" (such as "CO-001" "NO2-003", containing unique code to distinguish different gases); Step S5 output "initial concentration value" (unit such as ppm, mg / m³, for uncompensated concentration calculation result); Step S6-1 preprocessed "environmental temperature and humidity parameters" (standardized values such as temperature 25℃, humidity 50%RH); When integrating data, you need to match the environmental parameters of the same timestamp for each "type identification-initial concentration value" to avoid compensation logic failure due to time misalignment (such as using 10-minute-old temperature and humidity to compensate the current concentration value); Input data format adaptation and verification: Convert the integrated three types of data into the format supported by the error compensation model (such as structured array, key-value pair format), and verify the logical rationality of the data; for example, if the gas type identification is "formaldehyde-005" (electrochemical sensor detection), it needs to confirm whether the initial concentration value is within the detection range of the sensor (such as 0.01ppm to 10ppm), if it is out of range, it is determined as invalid input, return to step S5 to recalculate the concentration, ensure that the input model data meets the compensation logic requirements; Step S6-3: Cross-sensitivity interference pattern library call: Core content and basis of interference pattern library: The system pre-stores a "cross-sensitivity interference pattern library" for each target detection gas (at least 9 types), and the data in the library is pre-built through a large number of standard gas experiments; in the experiment, simulate the mixed existence of multiple gases in the actual detection scene (such as CO mixed with H2S, NO2 mixed with O3), record the interference degree of each gas on the detection results of other gases under different mixing ratios (such as 10ppm CO will cause the initial concentration value of H2S to be 5% higher), and record the change rule of the interference degree under different temperature and humidity environments (such as every 5℃ increase in temperature, the interference degree of CO on H2S increases by 2%), forming a structured interference data entry; Library matching call based on gas type identification: the error compensation model automatically matches and calls the corresponding interference mode library entry of the gas according to the input "gas type identification"; for example, if the type identification is "H2S-005", the "interference gas list affecting H2S detection", "concentration-interference rate corresponding relationship of different interference gases", "correction coefficient of temperature and humidity on interference rate" and other data in the library are called to provide interference basis for subsequent calculation of compensation coefficient, ensure that only the actual interference factors of the current gas are compensated, and avoid new errors caused by indiscriminate compensation; Step S6-4: Concentration compensation coefficient calculation: The calculation logic and dimension of the compensation coefficient: based on the called cross-sensitivity interference mode library data and the current environmental temperature and humidity parameters, the concentration compensation coefficient is calculated in two steps: Cross-interference compensation coefficient calculation: according to the "interference gas-interference rate" relationship in the mode library, combined with the possible concentration of the interference gas in the current detection scene (obtained from the identification result of the mixed gas in step S5, such as detecting CO concentration of 8 ppm), the interference rate of the interference gas on the current gas (such as 8 ppm CO corresponding to the interference rate of H2S of 4%) is calculated, and the cross-interference compensation coefficient (such as 0.96, i.e. 1-4%) is obtained; Environmental interference compensation coefficient calculation: according to the "temperature and humidity-interference rate correction coefficient" in the mode library, combined with the current pre-processed temperature and humidity parameters, the correction value of environmental factors on the interference rate (such as the current temperature is 3°C higher than the experimental standard temperature, and the interference rate correction value is 1.2%) is calculated, the cross-interference compensation coefficient is adjusted, and the final "comprehensive concentration compensation coefficient" (such as 0.96x(1-1.2%)≈0.948) is obtained; The value range of the compensation coefficient is controlled between 0.8 and 1.2, if the calculation result exceeds the range, it is determined as an abnormality (may be due to the interference mode library data not covering extreme cases), the default compensation coefficient (such as 1.0) of the gas is used, and overcompensation is avoided; Step S6-5: Initial concentration value correction and accurate concentration confirmation: Initial concentration value correction process: multiply the "initial concentration value" output in step S5 by the "comprehensive concentration compensation coefficient" calculated in step S6-4 to obtain the corrected "accurate concentration value"; for example, the initial concentration value of H2S is 5 ppm, and the comprehensive compensation coefficient is 0.948, then the accurate concentration value is 5x0.948≈4.74 ppm; if there are multiple interference gases (such as CO and NO2 interfering with H2S detection), the "multiple interference coefficient superposition correction" logic is used, and the compensation coefficient corresponding to each interference is multiplied in turn to ensure that all interference factors are covered; Reasonable verification of accurate concentration value: After correction, check whether the accurate concentration value is within the "reasonable detection range" of the current gas (the range is pre-stored in the system, such as the reasonable range of H2S is 0.01 ppm to 20 ppm), if it is out of range, recheck the interference mode library data and compensation coefficient calculation process to exclude calculation errors; if it is within the reasonable range, record the decimal point after the accurate concentration value (retain 2 digits, such as 4.74 ppm), ensure that the data accuracy meets the detection requirements; Step S6-6: multi-gas detection result integration and output: Structured integration of detection results: for each detected gas (including the main gas and secondary gas in the mixed gas), a structured detection result entry is constructed in the format of "gas unique code-gas name-accurate concentration value-detection timestamp-compensation coefficient-confidence"; for example, "code: CO-001; name: carbon monoxide; accurate concentration: 3.25 ppm; detection time: 2024-XX-XXXX:XX:XX; compensation coefficient: 0.98; confidence: 96%", ensuring complete result information and facilitating subsequent data tracing and analysis; Result output form and synchronization requirements: the system supports two output forms to meet different scene requirements: Local display output: real-time display of structured results of all detected gases on the display screen of the detector, in a clear list form, highlighting the gases with concentrations exceeding the safety threshold (such as using red font to mark gases with concentrations exceeding the safety threshold); Data interface output: through RS485, Ethernet and other interfaces, structured results are transmitted to the upper computer system (such as the monitoring platform) in a standardized protocol (such as Modbus protocol), with a transmission frequency consistent with the detection frequency (such as 1 time / minute), ensuring data synchronization; The final output result needs to cover the accurate information of at least 9 target gases, realizing the core goal of multi-gas synchronous accurate detection, and providing data support for subsequent safety warning, environmental regulation, etc.

[0023] A multi-gas accurate detection system, which performs a multi-gas accurate detection method when detecting multiple gases.

[0024] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for precise detection of multiple gases, characterized in that: Includes the following steps: Step S1, Gas enrichment: The gas to be tested flows through a micro gas enricher driven by a micro gas pump. The micro gas enricher is fabricated using MEMS technology and has a spiral inlet channel inside. The surface of the spiral inlet channel is covered with a sensitive material layer. The sensitive material layer is a composite material layer of graphene and metal oxide. When the target gas molecules in the gas to be tested flow through the spiral inlet channel, they are adsorbed and enriched by the sensitive material layer; Step S2, Sensitive Response: The micro gas enrichment device containing the target gas is heated and desorbed, and the released concentrated gas is delivered to the multi-gas composite detector; the multi-gas composite detector contains a sensor array sensitive to multiple gases, which contacts and reacts with the sensitive material layer to generate corresponding electrical signals. Step S3, Signal Acquisition and Preprocessing: The raw electrical signals generated by the sensor array are acquired in real time using the data acquisition unit, and the raw electrical signals are preprocessed, including noise reduction, amplification and analog-to-digital conversion, to obtain a digital signal sequence; Step S4, Feature Extraction: The data processing unit receives the digital signal sequence and extracts the feature vector for gas identification. The feature vector includes, but is not limited to, the peak value, the rising slope, the falling slope, the integral area, and the response value at a specific time point of the sensor response curve. Step S5, Gas Identification and Concentration Calculation: Input the feature vector into the pre-trained gas identification model. The gas identification model outputs the gas type identifier and concentration value. The gas identification model is a classification and regression model based on machine learning algorithms, which is trained on multiple standard gases of known types and concentrations by a multi-gas composite detector. Step S6, Error Compensation and Result Output: The concentration value obtained in step S5 is corrected using an error compensation algorithm to reduce the recognition error caused by cross-sensitivity and environmental interference; finally, the accurate type and concentration information of multiple gases after compensation is output, realizing the detection of at least 9 gases.

2. The multi-gas precise detection method according to claim 1, characterized in that: In step S1, the structural parameters of the spiral inlet channel are optimized based on the adsorption kinetics model of the target gas to achieve the maximum adsorption efficiency of the target gas within a given channel volume. The structural parameters include the width, depth, pitch, and total length of the flow channel; the metal oxide is one or more metal oxides selected from SnO2, ZnO, WO3, and CuO; the graphene material and the metal oxide are combined by hydrothermal method or in-situ growth method to form a three-dimensional porous nanostructure to increase the specific surface area of ​​the sensitive material layer.

3. The multi-gas precise detection method according to claim 1, characterized in that: The raw electrical signals generated by the sensor array are acquired in real time using a data acquisition unit. These raw electrical signals are then preprocessed, including noise reduction, amplification, and analog-to-digital conversion, to obtain a digital signal sequence, including: The acquired raw electrical signal is denoised by filtering to remove noise interference and obtain the denoised signal. The noise-reduced signal is amplified by using a signal amplifier to increase the signal amplitude, resulting in an amplified signal. The amplified signal undergoes analog-to-digital conversion, transforming the analog signal into a digital signal sequence.

4. The multi-gas precise detection method according to claim 1, characterized in that: The data processing unit receives a digital signal sequence and extracts feature vectors for gas identification, including: Based on the digital signal sequence, the dynamic response curves of each sensor are reconstructed; Peak values ​​of the response process are identified and extracted from the dynamic response curve to obtain peak characteristics; Based on the dynamic response curve, the rising slope of the curve during the response stage is calculated to obtain the rising slope characteristics. Based on the dynamic response curve, the descent slope of the curve during the recovery phase is calculated to obtain the descent slope characteristics. Integrate the dynamic response curve over a set time window, calculate the area of ​​integration of the response curve, and obtain the area of ​​integration characteristic. Read the response value of the dynamic response curve at a pre-set specific time point to obtain the response value characteristics at that specific time point; The extracted peak features, rising slope features, falling slope features, integral area features, and response value features at specific time points are combined to form a feature vector for characterizing gas type and concentration.

5. The multi-gas precise detection method according to claim 1, characterized in that: The feature vector is input into a pre-trained gas recognition model, which outputs the gas type identifier and concentration value, including: The feature vector is input into the classification module of the gas identification model. The classification module identifies the gas type based on the preprocessed feature vector and outputs the gas type identifier. The species identifier and feature vector are input into the regression module of the gas identification model. The regression module calculates and outputs the concentration value of the gas based on the concentration regression model corresponding to the species identifier. The gas recognition model is obtained through the following training process: The response signals of standard gases of known types and concentrations are collected using a multi-gas composite detector. After processing in steps S3 and S4, a training feature vector is obtained. The feature vector and its corresponding known type and concentration labels are used to perform supervised training on the model based on the machine learning algorithm, resulting in a joint model that can simultaneously perform gas type classification and concentration regression.

6. The multi-gas precise detection method according to claim 1, characterized in that: In step S6, an error compensation algorithm is used to correct the concentration value obtained in step S5 to reduce the recognition error caused by cross-sensitivity and environmental interference, including: Obtain current environmental parameters, including ambient temperature and humidity; The environmental parameters, along with the type identifier and concentration value output by the gas identification model, are input into the pre-stored error compensation model. The error compensation model calls a pre-established library of cross-sensitivity interference patterns for that gas based on the input type identifier. Based on the environmental parameters and the interference mode library, calculate the concentration compensation coefficient under the current environment; The initial concentration value obtained in step S5 is corrected using the concentration compensation coefficient to obtain the compensated accurate concentration value. The system integrates and outputs the accurate type identification and concentration values ​​of various gases after compensation to form the final multi-gas detection result, achieving simultaneous and accurate detection of at least 9 gases.

7. A multi-gas precision detection system, characterized in that: The system performs multi-gas precise detection method according to any one of claims 1-6 when performing multi-gas detection.

Images