Temperature modulation parameter optimization method for gas recognition based on gas sensor

Abstract

The invention provides a temperature modulation parameter optimization method for gas recognition based on a gas sensor, and belongs to the technical field of sensors. The method comprises the following steps: step S11, determining a plurality of classified gases; s12, building a data collection device; step S13, determining a search range and a search step length of modulation parameters; step S14, traversing the modulation parameter combination, and collecting sample data; step S21, projecting sample data of all gases of all modulation parameter combinations to a high-dimensional metric space; step S22, in a high-dimensional measurement space, measuring the similarity degree between the two samples by using Euclidean distance; s23, calculating a sample contour coefficient based on the cohesion degree and the separation degree of the sample; and S24, calculating a contour coefficient, and obtaining an optimization parameter according to the contour coefficient. By using the method, parameter tuning is performed, the required data is less, and the collection speed is high. Scientific basis and technical support are provided for parameter optimization.

Description

Technical Field

[0001] The present invention belongs to the technical field of sensors, and particularly relates to a method for optimizing temperature modulation parameters for gas identification based on gas sensors. Background Art

[0002] In the 1980s, the research on machine olfaction led to the definition of an electronic nose, which is an instrument composed of chemical gas sensors and a pattern recognition system. Nowadays, the term "electronic nose" has been widely used to describe the technology of using gas sensors to identify gas types and concentrations. It has important application values in the fields of food engineering, environmental monitoring, medical diagnosis, safety protection, and agriculture and animal husbandry.

[0003] When gas molecules contact a gas sensor, the resistance on the sensor surface changes, generating specific signals. Classification means of machine learning are used to analyze these signals to determine the type and concentration of the odor.

[0004] Gas sensors usually activate their sensitive layers by heating temperature to detect target gases. Different heating temperatures will affect the reaction rate and degree between gas molecules and the sensitive layer of the sensor, thereby changing the response characteristics of the sensor. In traditional applications, gas sensors usually remain at a fixed heating temperature to provide a stable detection environment.

[0005] However, this way of fixing the heating temperature limits the response characteristics that the sensor can extract, resulting in limited detection ability when facing multiple complex gases or specific gas components.

[0006] By introducing temperature modulation technology, the heating temperature of the gas sensor can be dynamically adjusted to trigger different gas response characteristics at different temperatures. Specifically, at different temperatures, the reaction mechanism between gas molecules and the surface of the sensitive layer may change, such as differences in adsorption and desorption rates, and the diversity of chemical reaction paths. These temperature-related reaction characteristics can significantly enrich the response data of the sensor, providing more dimensional information for feature extraction. As Figure 1 shown, the original resistance response data is scaled to the range of [0,1] to obtain normalized resistance response data. Focusing on the morphological characteristics of the response data, it can be observed that temperature modulation distinguishes different gases through high-frequency oscillations of different amplitudes, thereby providing more identification information.

[0007] Through the above mechanism, temperature modulation technology can greatly enrich the feature dimensions of a single sensor, making it have important potential in the precise detection of specific gases.

[0008] The most common temperature modulation technique is periodic square-wave temperature modulation, which mainly enhances the selectivity and sensitivity to target gases by periodically changing the operating temperature of the sensor (such as alternating high and low temperatures). This method obtains a specific response signal pattern by adjusting the sensor temperature to change the reactivity of its surface with different gas molecules.

[0009] Currently, regarding the parameter selection of periodic square-wave temperature modulation, the relevant research is not yet mature. At present, there is no perfect algorithm to clarify the optimization scheme for parameter selection such as high temperature, low temperature, and modulation period, so as to maximize the improvement effect of periodic square-wave modulation in feature extraction. Summary of the Invention

[0010] The purpose of the present invention is to provide a method for optimizing the parameters of periodic square-wave temperature modulation based on a gas sensor, providing a scientific basis and technical support for parameter optimization to solve the technical problem that there is no perfect algorithm in the prior art to clarify the optimization scheme for parameter selection such as high temperature, low temperature, and modulation period.

[0011] To solve the above technical problems, the temperature modulation parameter optimization method for gas identification based on a gas sensor of the present invention includes two stages. The first stage is the data collection stage for parameter optimization, and the second stage is the parameter optimization stage;

[0012] The data collection stage for parameter optimization includes the following steps:

[0013] Step S11: Determine a variety of classified gases;

[0014] Step S12: Build a data collection device;

[0015] Step S13: Determine the search range and search step size of the modulation parameters;

[0016] Step S14: Traverse the modulation parameter combinations and collect sample data;

[0017] The parameter optimization stage includes the following steps:

[0018] Step S21: Project the sample data of all gases of all modulation parameter combinations into a high-dimensional metric space;

[0019] Step S22: In the high-dimensional metric space, use the Euclidean distance to measure the similarity between two samples;

[0020] Step S23: Calculate the sample silhouette coefficient based on the sample cohesion and separation;

[0021] Step S24: Calculate the silhouette coefficient and obtain the optimized parameters according to the silhouette coefficient.

[0022] Further, the data collection device built in step S12 includes a gas chamber, a gas sensor, a microcontroller, and a host computer; the gas sensor is placed in the gas chamber, and the gas sensor is connected to the microcontroller through pins; the microcontroller is connected to the host computer; corresponding gases are placed in the gas chamber according to the needs of data collection; the host computer controls the gas sensor through the microcontroller, collects gas response data and transmits it to the host computer through the serial port protocol.

[0023] Further, the multiple classification gases in step S11 are represented as C1, …, C i , …, C j …, C N , where C i represents the i-th gas, C j represents the j-th gas, and N represents the total number of gases to be classified;

[0024] The sample data collected in step S14 is represented as follows: the number of samples of the i-th gas C i is n i , and all the sample data of the i-th gas C i is represented as where represents the p-th sample of the i-th gas C i ; similarly, represents the q-th sample of the i-th gas C i .

[0025] Further, when the sampling frequency is 2 Hz and the modulation period is 10 seconds, each set of modulation parameter combinations contains 20 sampling points; when the modulation period is less than 10 seconds, the response values at the high-temperature end and the low-temperature end are set to zero, so that each set of modulation parameter combinations also contains 20 sampling points. The response data of the 20 sampling points is corresponded to the coordinate axes of the 20-dimensional metric space one by one, and the response data of a single sample is mapped into a point in the high-dimensional metric space, and the dimension of the high-dimensional metric space is 20.

[0026] Further, the Euclidean distance in step S22 is calculated in the following manner:

[0027]

[0028] where represents the Euclidean distance between sample and sample ; represents the relative resistance value of the m-th sampling point of sample , represents the relative resistance value of the m-th sampling point of sample , and n is the dimension of the high-dimensional metric space.

[0029] Further, in step S23:

[0030] Cohesion degree a p is calculated as follows:

[0031]

[0032] Separation degree b p is calculated as follows:

[0033]

[0034] where n j is the sample quantity of the j-th gas C j sample;

[0035] Sample silhouette coefficient s p is calculated as follows:

[0036]

[0037] where a p is the average distance between the sample and the samples of the same category, b p is the average distance between the sample and the samples of the nearest different gas category, max(a p , b p ) is to take the maximum value of a p and b p .

[0038] Further, in step S24:

[0039] The calculation formula of the silhouette coefficient S is as follows:

[0040]

[0041] where M is the total number of samples of all gases; select the modulation parameter combination with the silhouette coefficient closest to 1 as the optimization parameter.

[0042] Compared with the prior art, the present invention has the following beneficial technical effects: The temperature modulation parameter optimization method for gas recognition based on gas sensors proposed by the present invention first builds a data acquisition device, then uses the data acquisition device to collect sample data of multiple gases, then projects the sample data into a high-dimensional metric space, calculates the similarity and silhouette coefficient between samples, and finally obtains the optimization parameter. Using this method for parameter tuning requires less data and has a fast collection speed, providing a scientific basis and technical support for the parameter optimization of the most commonly used periodic square wave temperature modulation. It is suitable for the preliminary work of the electronic nose. Description of the Drawings

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0044] Figure 1 It is a schematic diagram of the gas response characteristics of the present invention at different temperatures.

[0045] Figure 2 It is a schematic diagram of the structure of the gas sensor data collection device of the present invention.

[0046] Figure 3 It is a schematic diagram of the periodic square wave temperature modulation parameters of the present invention.

[0047] Figure 4 For the embodiment T of the present invention base = 290 °C, ΔT = 20 °C, t cycle = 10 s, the relative resistance response changes of each gas.

[0048] Figure 5 For the embodiment T of the present invention base = 320 °C, ΔT = 20 °C, t cycle = 10 s, the relative resistance response changes of each gas.

[0049] Figure 6 For the embodiment T of the present invention base = 290 °C, ΔT = 20 °C, t cycle = 10 s, the distribution of the relative resistance response reduced to two-dimensional space using principal component analysis.

[0050] Figure 7 For the embodiment T of the present invention base = 320 °C, ΔT = 20 °C, t cycle = 10 s, the distribution of the relative resistance response reduced to two-dimensional space using principal component analysis.

[0051] Reference numerals in the figure: 1 - gas chamber; 2 - gas sensor; 3 - microcontroller; 4 - host computer. Detailed implementation manners

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0053] The temperature modulation parameter optimization method for gas recognition based on gas sensors proposed by the present invention mainly includes two stages. The first stage is the data collection stage for parameter optimization, and the second stage is the parameter optimization stage.

[0054] Specifically, the data collection stage for parameter optimization includes the following steps:

[0055] Step S11: Determine multiple classification gases.

[0056] The periodic square-wave temperature modulation parameter optimization method for gas sensors of the present invention can be used for classifying and recognizing multiple gases. The gas categories of the classification gases are represented as C1, …, C i , …, C j …, C N , where C i represents the i-th gas, C j represents the j-th gas, and N represents the total number of gases to be classified.

[0057] Step S12: Build a data collection device.

[0058] As Figure 2 shown, the built data collection device includes a gas chamber 1, a gas sensor 2, a microcontroller 3, and a host computer 4. The gas sensor 2 is placed inside the gas chamber 1, and the gas sensor 2 is connected to the microcontroller 3 through pins; the microcontroller 3 is connected to the host computer 4. Corresponding gases are placed inside the gas chamber 1 according to the need for data collection. The host computer 4 controls the gas sensor 2 through the microcontroller 3, collects gas response data, and transmits it to the host computer through the serial port protocol.

[0059] The gas sensor is a BME680 multi-functional sensor. The gas sensor supports programmatically adjusting the temperature of its heater through the IIC or SPI protocol to achieve the temperature modulation function. The microcontroller is selected as ESP32S3, which can control the heating temperature of the gas sensor through the IIC or SPI protocol, simultaneously collect the resistance value of the sensitive layer of the gas sensor, and transmit the collected data to the host computer for processing through the serial port protocol.

[0060] The data sampling frequency of the data collection device is set to 2 Hz. The gas in the gas chamber is replaced according to the need for data collection to collect the relative resistance responses of different gases.

[0061] Step S13: Determine the search range and search step size of the modulation parameters. Figure 3 shows the parameters to be modulated for periodic square-wave temperature modulation. The modulation parameters include the low-temperature temperature parameter T base , the temperature change amount ΔT of the periodic square-wave heating, and the modulation period t cycle . T baseThe search range of is between 200 degrees Celsius and 390 degrees Celsius, with a step size of 10 degrees Celsius; the search range of ΔT is 10 degrees Celsius, 20 degrees Celsius, 40 degrees Celsius, t cycle The search range of is between 1 second and 10 seconds, with a step size of 1 second.

[0062] Step S14: Traverse the modulation parameter combinations and collect sample data.

[0063] When t cycle takes the larger value, the corresponding response data already contains the key information when t cycle takes the smaller value. Therefore, it is only necessary to collect data when t cycle takes the maximum value (i.e., 10 seconds), and generate the corresponding data for the smaller t by padding zeros at the end. cycle corresponding data.

[0064] During the data collection process, the response data of each complete temperature modulation cycle is regarded as an independent sample. Fix t cycle to 10 seconds, traverse all T basc , ΔT parameter combinations, collect multiple sample data of different gases for each group of parameter combinations, and collect the response data of different gases, that is, the relative resistance value, under the T base , ΔT, t cycle modulation parameter combinations within a single temperature modulation cycle as the data for parameter optimization.

[0065] The number of samples of the i-th gas C i is n i , and all the sample data of the i-th gas C i is expressed as where represents the p-th sample of the i-th gas C i ; similarly, represents the q-th sample of the i-th gas C i .

[0066] The parameter optimization stage includes the following steps:

[0067] Step S21: Project the sample data of all gases with all T base , ΔT, t cycle modulation parameter combinations into a high-dimensional metric space.

[0068] Taking a group of T base , ΔT, t cycle modulation parameter combinations as an example, for t cycleFor the case of taking 10 seconds, when the sampling frequency is 2 Hz, each temperature modulation window contains 20 sampling points. The response data of the 20 sampling points are corresponded to the coordinate axes of the 20-dimensional metric space one by one, and the response data of a single sample are mapped to a point in the high-dimensional metric space, where the dimension of the high-dimensional metric space is 20.

[0069] Because it is stipulated that t cycle takes 10 seconds when collecting data, so for t cycle <10, for the optimization of t cycle , the method of setting the response values to zero at the high-temperature end and the low-temperature end is adopted to achieve an approximation of the case of t cycle <10. (For example, 9 sampling points are set to zero at the high-temperature end and the low-temperature end respectively, and a total of 18 sampling points are set to zero, then it corresponds to the case of t cycle taking 1 second.) Similarly, the response data of the 20 sampling points are corresponded to the coordinate axes of the 20-dimensional metric space one by one, and the response data of a single sample are mapped to a point in the high-dimensional metric space.

[0070] Step S22: In the high-dimensional metric space, the Euclidean distance is used to measure the similarity between two samples.

[0071] In the high-dimensional metric space, the formula for the Euclidean distance is:

[0072]

[0073] Among them, represents the Euclidean distance between sample and sample ; represents the relative resistance value of the m-th sampling point of sample , represents the relative resistance value of the m-th sampling point of sample , n is the dimension of the high-dimensional metric space, that is, the number of sampling points of the sample. In the present invention, the dimension of the high-dimensional metric space is equal to the number of sampling points contained in each temperature modulation window when t cycle takes 10 seconds and the sampling frequency is 2 Hz.

[0074] The goal of parameter optimization is that the distances between the same gas samples are as close as possible (cohesion), and the distances between different gas samples are as far as possible (separation).

[0075] Step S23: Calculate the silhouette coefficient of the sample based on the sample cohesion and separation.

[0076] For gas sample , the cohesion is defined to represent the average distance between this sample and other samples under the same gas category. Specifically, the cohesion a p is calculated as follows:

[0077]

[0078] Cohesion degree a p Measures the similarity or cohesion between this sample and gas samples of the same category. The higher the cohesion degree, the closer the samples under the same gas category are in the feature space, indicating stronger similarity.

[0079] In the classification task, in addition to measuring the similarity of samples of the same category, it is also necessary to pay attention to the differences between gas samples of different categories. For gas samples Define the separation degree to represent the average distance between this sample and the samples of the nearest different gas category. Specifically, the separation degree b p The calculation formula is as follows:

[0080]

[0081] Where n j Is the number of samples of the j-th gas C j The larger the separation degree, the farther the samples between different gas categories are in the feature space, indicating stronger discrimination.

[0082] Based on the cohesion degree and separation degree defined above, further calculate the silhouette coefficient of each sample. The silhouette coefficient of the sample is used to measure the selectivity of the sample. The silhouette coefficient s p The calculation formula is as follows:

[0083]

[0084] Where a p Is the average distance between the sample And samples of the same category, b p Is the average distance between the sample And the samples of the nearest different gas category, max(a p , b p ) Is to take the maximum value of a p And b p . The value range of the silhouette coefficient s p Is [-1, 1]. Among them, the closer the value of the silhouette coefficient is to 1, the higher the similarity of the sample to the gas of the same category, and at the same time, the greater the difference from the gas of other categories. On the contrary, the value close to -1 indicates that the sample is more likely to belong to other categories.

[0085] Step S24: Calculate the silhouette coefficient and obtain the optimization parameter according to the silhouette coefficient.

[0086] To comprehensively evaluate the selectivity of the entire high-dimensional measurement space, the silhouette coefficient is introduced as a measurement index. The silhouette coefficient is the average of the silhouette coefficients of all samples, and the calculation formula for the silhouette coefficient S is as follows:

[0087]

[0088] Among them, M is the total number of samples of all gases. The silhouette coefficient is used to quantify the clustering quality of a single sample in the high-dimensional measurement space, and the value range is [-1, 1]. The closer the value is to 1, the higher the similarity of the sample to samples in the same category and the greater the difference from samples in different categories. Therefore, the closer the silhouette coefficient is to 1, the better the discrimination between the overall gas categories and the more ideal the clustering effect among samples.

[0089] Select the parameter combination T base , ΔT, t cycle as the optimization parameters. These optimization parameters can maximize the feature extraction ability of a single gas sensor, ensuring that the clustering effect not only has high cohesion and high separation but also maintains consistency among different samples, avoiding affecting the classification accuracy due to unstable clustering.

[0090] To demonstrate the effectiveness of the silhouette coefficient in measuring the feature extraction ability, in this embodiment, gases of three categories, cooking wine, vinegar, and coffee, are identified. The fixed parameters are ΔT = 20°C, t cycle = 10, and T base = 320°C and T base = 290°C are taken as examples for comparison. The main difference between these two cases is that the low-temperature temperature parameter T base differs by 30°C. Under these two benchmark temperature conditions, the relative resistance response changes of each gas are as shown in Figure 4 and Figure 5 .

[0091] When T base = 290°C, the resistance change ranges of cooking wine and vinegar are similar, and the response patterns of the three volatile gases are relatively similar. This makes it difficult to distinguish the volatile gases of cooking wine and vinegar in subsequent work. In contrast, under the condition of T base = 320°C, the resistance change ranges of cooking wine, vinegar, and coffee are quite different. Although the resistance response patterns of vinegar and coffee are similar, the difference in their change ranges is sufficient to distinguish the two.

[0092] Principal Component Analysis (PCA) is a common dimensionality reduction method. It aims to project high-dimensional data onto new coordinate axes with the maximum variance through orthogonal transformation, thereby retaining the main features of the data and reducing the data dimension. This method can effectively improve the visualization effect and analysis efficiency of the data. In this embodiment, principal component analysis is performed on the original 20-dimensional data and projected it into a two-dimensional space to more intuitively analyze the distribution characteristics of three volatile gases under different experimental conditions. The data projection results after dimensionality reduction are as Figure 6 and Figure 7 shown.

[0093] From the dimensionality reduction results, it can be observed that at T base = 290 °C, the relative resistance changes of cooking wine and vinegar are relatively similar, resulting in an overlap in their distributions in the two-dimensional space. When the reference temperature is increased to T base = 320 °C, the relative resistance changes of the three gases can be clearly distinguished in the dimensionality-reduced space.

[0094] To quantitatively evaluate the influence of different parameter combinations on the feature separation effect, the silhouette coefficient is used for analysis. The calculation results show that when T base = 290 °C, the silhouette coefficient is S = 0.6788; when T base = 320 °C, the silhouette coefficient increases to S = 0.9244. This result indicates that the silhouette coefficient can effectively measure the influence of different experimental parameters on the feature extraction ability.

[0095] It can be understood that the present invention is described through some embodiments. Those skilled in the art know that without departing from the spirit and scope of the present invention, various changes or equivalent replacements can be made to these features and embodiments. Additionally, under the teaching of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the scope protected by the present invention.

Claims

1. A temperature modulation parameter optimization method for gas identification based on a gas sensor, characterized in that: The method comprises two stages, the first stage is a data collection stage for parameter optimization, and the second stage is a parameter optimization stage; The data collection phase for parameter optimization consists of the following steps: Step S11: determining multiple types of gases; Step S12: Building a data collection device; Step S13: Determine the search range and search step of the modulation parameters; Step S14: traverse the modulation parameter combinations and collect sample data; The parameter optimization phase includes the following steps: Step S21: projecting sample data of all gases of all modulation parameter combinations into a high-dimensional quantitative space; Step S22: In the high-dimensional measurement space, the Euclidean distance is used to measure the similarity between two samples; Step S23: Calculate the sample silhouette coefficient based on the sample cohesion and separation; Step S24: Calculate the silhouette coefficient and obtain the optimization parameters according to the silhouette coefficient.

2. The temperature modulation parameter optimization method for gas identification based on a gas sensor according to claim 1, characterized in that: The data collection device constructed in step S12 includes a gas chamber, a gas sensor, a microcontroller, and a host computer; the gas sensor is placed in the gas chamber, and the gas sensor is connected to the microcontroller through pins; the microcontroller is connected to the host computer; corresponding gas is placed in the gas chamber according to the needs of data collection; the host computer controls the gas sensor through the microcontroller, collects gas response data and transmits it to the host computer through the serial port protocol.

3. The temperature modulation parameter optimization method for gas identification based on a gas sensor according to claim 1, characterized in that: In step S11, the various classified gases are represented by C1, ..., C i ,…,C j …,C N , where C i represents the i-th gas, C j represents the jth gas, and N represents the total number of gases to be classified; The sample data collected in step S14 are expressed as follows: i The sample size is n i , the i-th gas C i All sample data are expressed as in, represents the i-th gas C i The pth sample of represents the i-th gas C i The qth sample of .

4. The temperature modulation parameter optimization method for gas identification based on a gas sensor according to claim 3 is characterized in that: In step S21, When the sampling frequency is 2 Hz and the modulation period is 10 seconds, each group of modulation parameter combinations contains 20 sampling points; when the modulation period is less than 10 seconds, the high-temperature end and low-temperature end response values ​​are set to zero so that the modulation parameter combination also contains 20 sampling points. The response data of the 20 sampling points are corresponded one by one to the coordinate axes of the 20-dimensional mass space, and the response data of a single sample is mapped to a point in the high-dimensional mass space, and the dimension of the high-dimensional mass space is 20.

5. The method for optimizing temperature modulation parameters for gas identification based on a gas sensor according to claim 4, characterized in that: Step S22: The Euclidean distance is calculated as follows: in, Representation sample With sample The Euclidean distance between Representation sample The relative resistance value of the mth sampling point, Representation sample The relative resistance value of the mth sampling point, where n is the dimension of the high-dimensional measurement space.

6. The method for optimizing temperature modulation parameters for gas identification based on a gas sensor according to claim 5, characterized in that: In step S23: Cohesion p The calculation formula is as follows: Separation b p The calculation formula is as follows: Among them, n j is the jth gas C j Sample size; Sample silhouette coefficient s p The calculation formula is as follows: Among them, a p It is a sample The average distance to samples of the same category, b p It is a sample The average distance to the nearest sample of different gas categories, max(a p ,b p ) is to take a p With b p The maximum value in .

7. The method for optimizing temperature modulation parameters for gas identification based on a gas sensor according to claim 6, characterized in that: In step S24: The calculation formula of the silhouette coefficient S is as follows: Where M is the total number of samples of all gases; the modulation parameter combination with the silhouette coefficient closest to 1 is selected as the optimization parameter.

Images