A multi-component gas detection method and system based on artificial olfaction

Abstract

The present invention discloses a multi-component gas detection method and system based on artificial olfaction. The system includes an olfactory sensor array, a data acquisition module, a host computer control center, a pattern recognition algorithm module, and a visual interface. The olfactory sensor array is composed of a variety of metal oxide semiconductor sensors with different response characteristics, which can collect multi-dimensional gas information of the test environment. The data acquisition module collects the sensor information and transmits it to the host computer control center, which is then analyzed by an independent pattern recognition algorithm module that fully utilizes the cross-response of the sensors. This invention can not only realize the qualitative and quantitative identification of pure component gases, but also demix and identify mixed gases. It has a wide range of applications and can be widely used in safety detection scenarios such as the chemical industry, pharmaceutical industry, tunnel operations, and municipal gas.

Description

Technical Field

[0001] The present invention relates to a gas detection technology in a complex environment in the chemical industry, and in particular to a multi-component gas detection method and system based on artificial olfaction. Background Art

[0002] At present, gas detection methods can be mainly divided into electrochemical detection method, catalytic combustion method, gas chromatography, infrared spectroscopy absorption method, and gas sensitive sensor detection method.

[0003] Electrochemical detection method: Electrochemical sensors are primarily composed of electrodes, a breathable membrane, and an electrolyte. The gas to be measured passes through the breathable membrane and reaches the working electrode, where a chemical reaction occurs, ultimately converting chemical energy into electrical energy, forming a current signal that is linearly related to the concentration of the gas to be measured. Electrochemical gas sensors are characterized by low cost, high detection sensitivity, and a certain degree of selectivity for certain types of gas to be measured; however, their detection range is relatively small, and since the gas requires an electrolyte as a carrier for transmission, this has a certain impact on the response time. Most electrochemical sensors require the use of liquid electrolytes, which will cause some corrosion to the sensor, affecting its performance and lifespan.

[0004] Catalytic combustion method: A catalytic combustion gas sensor consists of a detection module that reacts with the gas being measured and a compensation module that does not. When a catalytic combustion sensor comes into contact with a flammable gas, the detection module combusts, causing its temperature to rise, which in turn changes the resistance of the thermistor. The compensation module, on the other hand, does not combust, maintaining a constant temperature and its resistance unchanged. The difference in resistance between the detection and compensation modules can be detected as a bias voltage using a Wheatstone bridge circuit. This voltage is proportional to the gas concentration, and the offset voltage between the two modules determines the concentration of the gas being measured. The catalytic combustion method is characterized by its fast response, stable and reliable operation, and low price. However, it is limited to combustible gas detection and has certain limitations. Furthermore, the sensor's sensitivity is easily affected by chemical substances. In low oxygen environments, the gas being measured may not fully combust, resulting in large measurement errors.

[0005] Gas chromatography: A gas chromatography detection system usually includes a carrier gas source for introducing samples into the system, a channel for samples to enter the detection system - an inlet, a chromatographic column for separating mixed gases, and a detection module and a data acquisition and processing module for generating chromatograms. The gas to be tested is carried by the carrier gas into the gas chromatograph. When passing through the chromatographic column, the gas to be tested can be separated due to the different adsorption properties of the chromatographic column to different gases. Finally, a chromatogram can be obtained based on the test results. The different peaks in the chromatogram represent different gas components in the sample. By measuring the peak height or peak area, the concentration information of each gas can be obtained. The characteristic of gas chromatography is high monitoring accuracy, but this method needs to be analyzed in the laboratory, the response time is very long, and real-time detection cannot be performed. It is not suitable for use in home or industrial field environmental monitoring.

[0006] Infrared absorption: Infrared light of a fixed wavelength and intensity is passed through a chamber filled with the gas to be measured. The gas's ability to absorb infrared light of a specific wavelength is exploited. When experimental conditions such as the chamber's optical pathlength remain unchanged, the concentration of the gas to be measured can be inferred by analyzing the change in the intensity of the emitted light after absorption by the substance. This method utilizes the characteristic spectrum of the gas molecules to detect them, offering very high resolution and excellent selectivity for different gas molecules. However, the equipment is expensive and complex, and requires high ambient lighting conditions for the test.

[0007] Gas sensor method: Gas sensors are made of metal oxide semiconductor (MOS) gas-sensitive materials. When the gas being measured reaches the semiconductor surface, it undergoes a redox reaction with the sensor, causing a change in the semiconductor's resistance. By measuring this change in sensor resistance, the gas concentration can be inferred. The gas sensor method is characterized by good response, low cost, and simple operation. However, the gas sensor has poor selectivity and is susceptible to cross-response.

[0008] Methods such as infrared spectroscopy and gas chromatography have a wide detection range and high sensitivity, but require manual sampling before measurement and analysis. The entire process and cycle are cumbersome and have high latency, making them unsuitable for industrial detection applications with high real-time requirements. Moreover, the equipment required for these methods is complex and expensive, making them unsuitable for large-scale deployment. Currently, the most widely used method is the gas sensor detection method based on MOS sensors. Summary of the Invention

[0009] To address the problems of the prior art, the present invention provides a multi-component gas detection method and system based on artificial olfaction. This invention can not only achieve qualitative and quantitative identification of pure component gases, but also demix and identify mixed gases.

[0010] In order to achieve the above objectives, the present invention provides the following technical solutions.

[0011] A multi-component gas detection system based on artificial olfaction, comprising: an array sensor, a data acquisition module, a pattern recognition algorithm module, a host computer control center and a visual interface;

[0012] The array sensor includes a plurality of sensors with different response characteristics, and is used to obtain multi-dimensional gas information of the environment to be measured;

[0013] The data acquisition module is used to collect information from the array sensor and send it to the host computer control center after filtering and amplification processing;

[0014] The host computer control center is used to send data acquisition instructions to the data acquisition module to control data acquisition, receive data sent back by the data acquisition module and send it to the pattern recognition algorithm module for analysis and control the display output of the visualization interface;

[0015] The pattern recognition algorithm module is used to pre-process the data transmitted from the host computer control center and then use an independent neural network algorithm to predict the concentration of pure gas components and unmix and identify mixed gases;

[0016] The visual interface provides human-computer interaction functions and outputs various data of the system operation process.

[0017] As a further improvement of the present invention, the pattern recognition algorithm module includes a pure component qualitative and quantitative recognition algorithm module and a mixed component separation recognition algorithm module;

[0018] The pure component qualitative and quantitative identification algorithm module uses a self-attention mechanism to calculate the degree of correlation between sensors;

[0019] The mixed component separation and identification algorithm module uses the correlation between the same components and the independence between different components to construct the objective function, maps the original data to a space with strong correlation between the same components and weak correlation between different components, and uses correlation matching to unmix.

[0020] As a further improvement of the present invention, the pure component qualitative and quantitative identification algorithm module is used to input the preprocessed data into the backbone network, use the self-attention mechanism to calculate the relevant features between sensors, then use the fully connected network to integrate the features, and finally use the multi-label regression node to output the component concentration information of the gas to be measured.

[0021] As a further improvement of the present invention, the pure component qualitative and quantitative identification algorithm module uses a multi-label regression model to integrate gas classification and regression problems into one model; finally, using a multi-label regression node to output the component concentration information of the gas to be measured refers to outputting a set of multi-label regression nodes, each node corresponding to a gas component. When the pure component gas is input into the model, the node corresponding to the gas outputs its concentration value, and the output of other nodes is 0, indicating that there is no other gas.

[0022] As a further improvement of the present invention, the output of the mixed component separation and identification algorithm module is a feature sequence, and the objective function of the model is to maximize the correlation between feature sequences of the same component and the independence between feature sequences of different components.

[0023] As a further improvement of the present invention, the mixed component separation and identification algorithm module is used to use the pure component data to construct a benchmark feature library through the trained model, and the features between different components in the benchmark feature library are unrelated. The data of the mixed component to be measured is then input into the model to obtain the features of the data to be measured, and the correlation between the features of the data to be measured and the features in the benchmark feature library is calculated one by one. The benchmark feature with a greater correlation with the feature to be measured is selected according to the number of mixed gases in the mixed component, and the pure component gas corresponding to the benchmark feature is the component of the gas to be measured.

[0024] As a further improvement of the present invention, the data acquisition module includes a lower computer control module, an A / D conversion module, a filtering circuit and an amplifying circuit; the lower computer control module, the A / D conversion module, the filtering circuit and the amplifying circuit are electrically connected in sequence, the lower computer control module is electrically connected to the upper computer control center, and the amplifying circuit is electrically connected.

[0025] As a further improvement of the present invention, the array sensor is provided with sensor power supplies and signal interfaces on both sides, and the plurality of sensor arrays are arranged between the sensor power supplies and signal interfaces on both sides.

[0026] As a further improvement of the present invention, the sensor is a metal oxide semiconductor sensor.

[0027] A multi-component gas detection method based on artificial olfaction comprises the following steps:

[0028] The array sensor acquires multi-dimensional gas information of the environment to be tested; the data acquisition module collects the information of the array sensor and transmits it to the host computer control center, which then transmits the data to the pattern recognition algorithm module for analysis to obtain the recognition result.

[0029] As a further improvement of the present invention, after pre-processing the data transmitted from the host computer control center, the pattern recognition algorithm module uses a blind source separation mixed gas unmixing recognition algorithm to perform multi-component recognition according to the detection requirements, and simultaneously uses an independent neural network algorithm to predict and identify the concentration of pure gas components, including:

[0030] Use the self-attention mechanism to calculate the correlation between different sensors;

[0031] The objective function is constructed by using the correlation between the same components and the independence between different components. The original data is mapped into a space with strong correlation between the same components and weak correlation between different components, and correlation matching is used for unmixing.

[0032] The preprocessed data is input into the backbone network, and the self-attention mechanism is used to calculate the relevant features between sensors. The fully connected network is then used to integrate the features, and finally a multi-label regression node is used to output the component concentration information of the gas to be measured.

[0033] A multi-label regression model is used to integrate gas classification and regression problems into one model. Finally, a multi-label regression node is used to output the component concentration information of the gas to be tested. This means outputting a set of multi-label regression nodes, each corresponding to a gas component. When a pure component gas is input into the model, the node corresponding to the gas outputs its concentration value, and the output of other nodes is 0, indicating the absence of other gases.

[0034] A benchmark feature library is constructed using the trained model using pure component data. The features of different components in the benchmark feature library are uncorrelated. The data of the mixed component to be measured is then input into the model to obtain the features of the data to be measured. The correlation between the features of the data to be measured and the features in the benchmark feature library is calculated one by one. The benchmark feature with a greater correlation with the feature to be measured is selected based on the number of mixed gases in the mixed component. The pure component gas corresponding to the benchmark feature is the component of the gas to be measured.

[0035] Compared with existing gas detection methods, the advantages of this method are:

[0036] The present invention utilizes multiple sensors to create a sensor array, inheriting the advantages of simple sensor structure and fast response. Furthermore, it utilizes a pattern recognition algorithm module for high-precision qualitative and quantitative identification of gases. It also overcomes the cross-response drawbacks of traditional single-sensor gas detection and enables the separation and identification of mixed gases. The array sensor utilizes a variety of metal oxide semiconductor sensor components with different response characteristics, enabling the collection of multi-dimensional gas information from the test environment. The data acquisition module transmits the collected sensor information to a host computer control center, where it is then analyzed by an independent pattern recognition algorithm module that fully utilizes the cross-response of the sensors. This invention not only enables the qualitative and quantitative identification of pure component gases, but also enables the demixing and identification of mixed gases, making it widely applicable in safety monitoring scenarios such as the chemical industry, pharmaceutical industry, tunnel operations, and municipal gas. The gas detection method proposed in this invention boasts high response speed, high precision, and high accuracy, enabling the identification of mixed gas components. It is widely applicable for gas component analysis and identification in scenarios such as the chemical industry, pharmaceutical industry, tunnel operations, and municipal gas, demonstrating its high application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present invention in any way. In addition, the shapes and proportional dimensions of the components in the drawings are only schematic and are used to help understand the present invention, and are not intended to specifically limit the shapes and proportional dimensions of the components of the present invention. In the drawings:

[0038] Figure 1 This is the overall structure diagram of the present invention: including artificial olfactory sensor module 1, data acquisition module 2, pattern recognition algorithm module 3, host computer control center 4, and visualization interface 5;

[0039] Figure 2 is a diagram of an array of artificial olfactory sensor modules;

[0040] Figure 3 It is a pure component gas qualitative and quantitative identification algorithm module;

[0041] Figure 4 It is an algorithm for unmixing and identifying components of mixed gases. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the embodiments described 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 efforts should fall within the scope of protection of the present invention.

[0043] It should be noted that when an element is referred to as being "disposed on" another element, it may be directly on the other element or there may be an element centered thereon. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an element centered thereon. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only embodiments.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0045] This invention proposes a method and system for soft identification of gas components based on an artificial olfactory sensor module 1. This gas detection system requires sensor detection, data acquisition and processing, and human-computer interaction. Therefore, the system can be divided into four parts: the artificial olfactory sensor module 1, a lower-level computer data acquisition and processing module, a gas soft identification algorithm module, and a host computer visualization interface 5. This improves gas detection accuracy and enables demixing and identification of mixed gases.

[0046] The first object of the present invention is to provide an artificial olfactory sensor module gas detection system comprising an artificial olfactory sensor module 1, a data acquisition module 2, a host computer control center 4, a pattern recognition algorithm module 3, and a visualization interface 5. During detection, the host computer control center 4 sends instructions to the slave computer. The slave computer acquires the sensor's response value through the A / D conversion module 202 and sends it to the host computer. The host computer uses a specific pattern recognition algorithm module 3 to process the acquired data, ultimately achieving qualitative and quantitative identification of the gas, and finally outputs the results on the visualization interface 5. A detailed description of each component is as follows:

[0047] 1) Structure of artificial olfactory sensor module 1 composed of metal oxide semiconductor sensors

[0048] The sensor 1 is composed of metal oxide semiconductor sensors with different response characteristics, which can collect multi-dimensional information of the gas.

[0049] To overcome the problem of a single sensor's cross-response to different gases, which can lead to inaccurate identification of the sample being measured, a design of an artificial olfactory sensor module 1 is proposed. The artificial olfactory sensor module 1 is composed of different sensors with different response characteristics, which can simultaneously detect gases and obtain multi-dimensional data.

[0050] 2) Acquisition module 2

[0051] Receive instructions from the host computer, collect and condition the signals output by the artificial olfactory sensor module 1, and transmit them to the host computer;

[0052] The acquisition module 2 controls the data collection, receives the data sent back by the data acquisition module 2 and sends it to the pattern recognition algorithm module 3 for analysis, and controls the display output of the visualization interface 5;

[0053] 3) Pattern recognition algorithm module 3

[0054] Based on the data collected by the artificial olfactory sensor module 1, a pattern recognition algorithm module 3 is proposed, which mainly includes a pure component gas qualitative and quantitative identification algorithm module 302 and a mixed gas component separation and identification algorithm module 303. For the identification of pure component gases, a multi-label regression model is proposed to unify qualitative and quantitative identification, realizing a single model to predict the concentrations of different gases. For mixed component data, a decorrelation model is proposed, which can achieve the separation and identification of mixed gas components using only the prior knowledge of pure gas components;

[0055] The pattern recognition algorithm module 3 is used to analyze and process the data transmitted from the host computer control center 4, and then use the general denoising and dimensionality reduction algorithm to predict the concentration of pure gas components and unmixing and identifying mixed gases using an independent neural network algorithm;

[0056] 4) Visual interface for human-computer interaction 5

[0057] Display of raw sensor data, display of data feature extraction results, and output of recognition results.

[0058] As a preferred embodiment, the visualization interface 5 provides a human-computer interaction function and outputs various data of the system operation process, including sensor information, original data information, data dimension reduction results, feature extraction results and final recognition results.

[0059] As a preferred embodiment, the system adopts a hardware and software separation design, that is, the system can be divided into a lower computer hardware system and a host computer software system. The lower computer hardware system is composed of an artificial olfactory sensor module 1 and a data acquisition module 2, and the host computer software system is composed of a pattern recognition algorithm module 3, a host computer control center 4, and a visual interface 5. The hardware and software separation design can realize module reuse, that is, using the same module with different pattern recognition algorithm modules 3 can realize different functions;

[0060] Further preferably, the pure component qualitative and quantitative identification algorithm module 302 uses a self-attention mechanism to calculate the correlation between sensors, making full use of the cross-response characteristics that traditional gas sensors try to avoid during detection; the pure component qualitative and quantitative identification algorithm module 302 uses a multi-label regression model to integrate gas classification and regression problems into one model;

[0061] Further preferably, the mixed component separation and identification algorithm module 303 uses the correlation between the same components and the independence between different components to construct the objective function, maps the original data to a space with strong correlation between the same components and weak correlation between different components, and uses correlation matching to unmix;

[0062] The second purpose of the present invention is to provide a multi-component gas detection method based on artificial olfaction, comprising the following steps: before measurement, the sensor 1 is placed in the gas environment to be measured, the data acquisition module 2 collects the sensor information and transmits it to the host computer control center 4, and the host computer control center 4 then transmits the data to the pattern recognition algorithm module 3 for analysis to obtain the recognition result.

[0063] After the pattern recognition algorithm module 3 pre-processes the data transmitted from the host computer control center 4, it uses the blind source separation mixed gas unmixing recognition algorithm to perform multi-component recognition according to the detection requirements, and uses the independent neural network algorithm to predict and identify the concentration of pure gas components, specifically including:

[0064] Use the self-attention mechanism to calculate the correlation between different sensors;

[0065] The objective function is constructed by using the correlation between the same components and the independence between different components. The original data is mapped into a space with strong correlation between the same components and weak correlation between different components, and correlation matching is used for unmixing.

[0066] The preprocessed data is input into the backbone network, and the self-attention mechanism is used to calculate the relevant features between sensors. The fully connected network is then used to integrate the features, and finally a multi-label regression node is used to output the component concentration information of the gas to be measured.

[0067] A multi-label regression model is used to integrate gas classification and regression problems into one model. Finally, a multi-label regression node is used to output the component concentration information of the gas to be tested. This means outputting a set of multi-label regression nodes, each corresponding to a gas component. When a pure component gas is input into the model, the node corresponding to the gas outputs its concentration value, and the output of other nodes is 0, indicating the absence of other gases.

[0068] A benchmark feature library is constructed using the trained model using pure component data. The features of different components in the benchmark feature library are uncorrelated. The data of the mixed component to be measured is then input into the model to obtain the features of the data to be measured. The correlation between the features of the data to be measured and the features in the benchmark feature library is calculated one by one. The benchmark feature with a greater correlation with the feature to be measured is selected based on the number of mixed gases in the mixed component. The pure component gas corresponding to the benchmark feature is the component of the gas to be measured.

[0069] Therefore, the present invention has the following advantages:

[0070] 1) Using metal oxide semiconductor sensors to build sensor arrays has rapid response, low cost, stable structure and wide application range;

[0071] 2) Propose an identification algorithm module, use a model to unify the gas classification problem and concentration prediction problem, and can effectively demix and identify mixed gases;

[0072] 3) Separate hardware and software design: the lower computer sensor part and the upper computer visual interface 5 are separated, and only data and instructions are exchanged between the two. This design allows for the reuse of hardware modules, that is, the same sensor array can play different functions when equipped with different pattern recognition algorithm modules 3;

[0073] 4) Design a visual interactive interface to facilitate the use of the system.

[0074] Example

[0075] The following is a detailed description of the invention with reference to the accompanying drawings:

[0076] like Figure 1 As shown, the present invention mainly consists of five parts: a sensor 11, a data acquisition module 22, a pattern recognition algorithm module 3, a host computer control center 4, and a visualization interface 5.

[0077] The main functions of each part are as follows: the sensor 11 is responsible for converting the gas information of the environment to be tested into electrical signal information; the data acquisition module 2 is composed of a lower computer control module 201, an A / D conversion module 202, a filter circuit 203 and an amplifier circuit 204. The lower computer control module 201, the A / D conversion module 202, the filter circuit 203 and the amplifier circuit 204 are electrically connected in sequence, the lower computer control module 201 is electrically connected to the upper computer control center 4, and the amplifier circuit 204 is electrically connected to the sensor 1.

[0078] The lower computer control module 201 is responsible for receiving the sampling rate, sampling time and other parameters of the upper computer and collecting data according to the parameters and transmitting the collected data to the upper computer. The A / D conversion module 202, the filter circuit 203, and the amplifier circuit 204 are responsible for amplifying and removing noise from the tiny electrical signals output by the sensor and converting them into digital quantities; the pattern recognition algorithm module 3 is responsible for processing the signals sent back by the lower computer and obtaining the composition and concentration information of the unknown gas to be measured. The data preprocessing uses the traditional data preprocessing methods of normalization, dimensionality reduction, and whitening. The qualitative and quantitative analysis and unmixing of the data use a neural network model designed for the cross-response of the sensor. The process is as follows: Figure 3 、 Figure 4 As shown; the host computer control center 4 is responsible for the operation scheduling of the entire system, including instruction transmission, data reception, data analysis and processing, result display, etc.; the visual interface 5 is responsible for displaying the data processing results.

[0079] like Figure 2 As shown, the sensor array is composed of various metal oxide semiconductor sensors with different response characteristics.

[0080] like Figure 3As shown, the backbone of this neural network model is built using the self-attention mechanism in the Transformer. This backbone network extracts features to calculate the correlation between each sensor's data, fully leveraging the cross-response information of the sensors. The model's final output is a set of multi-label regression nodes, each corresponding to a gas component. When a pure component gas is input into the model, the node corresponding to that gas outputs its concentration value, while the output of other nodes is 0, indicating the absence of other gases. In summary, the overall process of the algorithm is as follows: preprocessed data is input into the backbone network, the self-attention mechanism is used to calculate the correlation features between sensors, and then a fully connected network is used to integrate the features. Finally, a multi-label regression node outputs the component concentration information of the gas being measured.

[0081] like Figure 4 As shown in the figure, in the training process of the network model, only pure component data is used for training, and its feature extraction network is Figure 3 The output of the model is a feature sequence. The specific structure is shown in Table 2. The objective function of the model is to maximize the correlation between feature sequences of the same component and the independence between feature sequences of different components.

[0082] Table 1. Structure of the pure component qualitative and quantitative identification model (where x represents the number of sensors, dim represents the data length of each sensor, n represents the total number of gases output, and @ represents the dimension separation symbol)

[0083] name Input Dimension Output dimension enter x@dim x@dim Self-attention x@dim x@dim Self-attention x@dim x@dim Self-attention x@dim x@dim Self-attention x@dim x@dim Reshape 1@x*dim x*dim Linear x*dim 2048 ReLU 2048 2048 Linear 2048 n Sigmoid n n

[0084] Table 2 Mixture component separation and identification model structure (where x represents the number of sensors, dim represents the data length of each sensor, n represents the total number of output gases, and @ represents the dimension separation symbol)

[0085]

[0086]

[0087] The network structure is shown in Table 1. Its goal is to minimize the gap between the network output and the target. The objective function and loss function are shown in Equation 1 and Equation 2. In Equation 1, f is the mean square error calculation function.

[0088] min f(Y,Y0) Formula 1

[0089]

[0090] max f(X)+min g(X,Y) Equation 3

[0091]

[0092] The objective function and loss function of the model are shown in Equations 3 and 4. In both equations, X represents a set of data of the same type, and Y represents a set of data of different types from the data of type X. In Equation 3, f represents the correlation calculation function, and g represents the independence calculation function. In Equation 4, the COR function calculates the correlation coefficient between two variables, and the COV function calculates the covariance between two variables. α and β are hyperparameters used to adjust the weight of each loss.

[0093] In the prediction process of the network model, the pure component data is first used to build a benchmark feature library through the trained model. The features of different components in this feature library are uncorrelated. Then the data of the mixed component to be tested is input into the model to obtain the features of the data to be tested. The correlation between the features to be tested and the features in the benchmark feature library is calculated one by one. According to the number of mixed gases in the mixed component, the benchmark feature with a greater correlation with the features to be tested is selected. The pure component gas corresponding to the benchmark feature is the composition of the gas to be tested.

[0094] Many embodiments and applications beyond the examples provided will be apparent to those skilled in the art upon reading the foregoing description. Therefore, the scope of the present teachings should be determined not with reference to the foregoing description, but rather with reference to the preceding claims, along with the full scope of equivalents to which such claims are entitled. For the purpose of completeness, all articles and references, including the disclosures of patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein from the preceding claims is not a disclaimer of such subject matter, nor should it be interpreted that the applicants did not consider such subject matter to be part of the disclosed inventive subject matter.

[0095] The above content is a further detailed description of the present invention, and it cannot be considered that the specific implementation methods of the present invention are limited to these. For ordinary technicians in the technical field to which the present invention belongs, they can make several simple deductions or substitutions without departing from the concept of the present invention, which should be regarded as belonging to the scope of protection of the present invention determined by the submitted claims.

Claims

1. A multi-component gas detection system based on artificial olfaction, characterized in that: include: Artificial olfactory sensor module (1), data acquisition module (2), pattern recognition algorithm module (3), host computer control center (4) and visualization interface (5); The artificial olfactory sensor module (1) comprises a plurality of sensors with different response characteristics, and is used to obtain multi-dimensional gas information of the environment to be measured; The data acquisition module (2) is used to collect information from the artificial olfactory sensor module (1) and send it to the host computer control center (4) after filtering and amplification processing; The host computer control center (4) is used to send data acquisition instructions to the data acquisition module (2) to control data acquisition, receive data sent back by the data acquisition module (2) and send it to the pattern recognition algorithm module (3) for analysis and control the display output of the visualization interface (5); The pattern recognition algorithm module (3) is used to pre-process the data transmitted from the host computer control center (4), and then use the blind source separation mixed gas unmixing recognition algorithm to perform multi-component recognition according to the detection requirements, and use the independent neural network algorithm to predict and identify the concentration of pure gas components; The visual interface (5) provides a human-computer interaction function and outputs various data of the system operation process; The pattern recognition algorithm module (3) includes a pure component qualitative and quantitative recognition algorithm module (302) and a mixed component separation recognition algorithm module (303); The pure component qualitative and quantitative identification algorithm module (302) uses a self-attention mechanism to calculate the degree of correlation between different sensors; The mixed component separation and identification algorithm module (303) uses the correlation between the same components and the independence between different components to construct the objective function, maps the original data to a space with strong correlation between the same components and weak correlation between different components, and uses correlation matching to unmix; The output of the mixed component separation and identification algorithm module (303) is a feature sequence, and the objective function of the model is to maximize the correlation between feature sequences of the same component and the independence between feature sequences of different components; The pure component qualitative and quantitative identification algorithm module (302) is used to input the pre-processed data into the backbone network, use the self-attention mechanism to calculate the relevant features between sensors, then use the fully connected network to integrate the features, and finally use the multi-label regression node to output the component concentration information of the gas to be measured; The pure component qualitative and quantitative identification algorithm module (302) uses a multi-label regression model to integrate gas classification and regression problems into one model; finally, using a multi-label regression node to output component concentration information of the gas to be measured means outputting a set of multi-label regression nodes, each node corresponding to a gas component, and when a pure component gas is input into the model, the node corresponding to the gas outputs its concentration value, and the output of other nodes is 0, indicating that no other gas exists; The mixed component separation and identification algorithm module (303) is used to construct a reference feature library using pure component data through a trained model, wherein the features of different components in the reference feature library are uncorrelated, and then the data of the mixed component to be measured is input into the model to obtain the features of the data to be measured, and the correlation between the features of the data to be measured and the features in the reference feature library is calculated one by one, and the reference feature with a greater correlation with the feature to be measured is selected according to the amount of mixed gas in the mixed component, and the pure component gas corresponding to the reference feature is the component of the gas to be measured.

2. The multi-component gas detection system based on artificial olfaction according to claim 1, characterized in that: The data acquisition module (2) comprises a lower computer control module (201), an A / D conversion module (202), a filtering circuit (203) and an amplifying circuit (204); the lower computer control module (201), the A / D conversion module (202), the filtering circuit (203) and the amplifying circuit (204) are electrically connected in sequence, the lower computer control module (201) is electrically connected to the upper computer control center (4), and the amplifying circuit (204) is electrically connected to the artificial olfactory sensor module (1).

3. The multi-component gas detection system based on artificial olfaction according to claim 1 or 2, characterized in that: The artificial olfactory sensor module (1) is composed of metal semiconductor oxide sensors with different response characteristics, the metal semiconductor oxide sensors are arranged in a rectangular shape, and the power supply and signal interfaces of the metal semiconductor oxide sensors are arranged at the edge of the module.

4. A multi-component gas detection method based on artificial olfaction, based on the multi-component gas detection system based on artificial olfaction according to claim 1, characterized in that: The following steps are involved: The artificial olfactory sensor module (1) acquires multi-dimensional gas information of the environment to be tested; the data acquisition module (2) acquires the information of the artificial olfactory sensor module (1) and transmits it to the upper computer control center (4); the upper computer control center (4) then transmits the data to the pattern recognition algorithm module (3) for analysis to obtain a recognition result; After pre-processing the data transmitted from the host computer control center (4), the pattern recognition algorithm module (3) uses a blind source separation mixed gas unmixing recognition algorithm to perform multi-component recognition according to detection requirements, and uses an independent neural network algorithm to predict and recognize the concentration of pure gas components.

5. The multi-component gas detection method based on artificial olfaction according to claim 4, characterized in that: After the pattern recognition algorithm module (3) pre-processes the data transmitted from the host computer control center (4), it uses the mixed component separation and recognition algorithm module (303) to perform multi-component recognition according to the detection requirements, and simultaneously uses an independent neural network algorithm to predict and recognize the concentration of pure gas components, including: Use the self-attention mechanism to calculate the correlation between different sensors; The objective function is constructed by using the correlation between the same components and the independence between different components. The original data is mapped into a space with strong correlation between the same components and weak correlation between different components, and correlation matching is used for unmixing. The preprocessed data is input into the backbone network, and the self-attention mechanism is used to calculate the relevant features between sensors. The fully connected network is then used to integrate the features, and finally a multi-label regression node is used to output the component concentration information of the gas to be measured. A multi-label regression model is used to integrate gas classification and regression problems into one model. Finally, a multi-label regression node is used to output the component concentration information of the gas to be tested. This means outputting a set of multi-label regression nodes, each corresponding to a gas component. When a pure component gas is input into the model, the node corresponding to the gas outputs its concentration value, and the output of other nodes is 0, indicating the absence of other gases. A benchmark feature library is constructed using the trained model using pure component data. The features of different components in the benchmark feature library are uncorrelated. The data of the mixed component to be measured is then input into the model to obtain the features of the data to be measured. The correlation between the features of the data to be measured and the features in the benchmark feature library is calculated one by one. The benchmark feature with a greater correlation with the feature to be measured is selected based on the number of mixed gases in the mixed component. The pure component gas corresponding to the benchmark feature is the component of the gas to be measured.

Images