Gas detection method and device based on MEMS gas sensor array

Abstract

The present invention relates to the field of gas detection technology, and discloses a gas detection method and device based on a MEMS gas sensor array. The method comprises: grouping a plurality of MEMS gas sensors according to the target gas type and collecting detection data to obtain an original detection data set; performing curve fitting and feature extraction to obtain a feature parameter set and constructing a sensor array feature model; correcting the original detection data set to obtain standardized response data; extracting a plurality of response feature quantities and inputting them into an unknown input observer to perform residual calculation and signal abnormal point identification to obtain abnormal response data; performing piecewise linear regression and dynamic calibration to obtain a target detection data set; inputting the target detection data set into a sensor network model to perform array deployment optimization to obtain an array deployment optimization scheme. The present invention realizes adaptive deployment optimization of the sensor array and improves the spatial coverage efficiency of the detection network.

Description

Technical Field

[0001] The present invention relates to the technical field of gas detection, and in particular to a gas detection method and device based on a MEMS gas sensor array. Background Art

[0002] With the growing demand for gas detection in industrial production and environmental monitoring, MEMS gas sensors are widely used due to their miniaturization, high integration, low power consumption, etc. However, due to the complexity of the industrial environment and the diversity of gas detection objects, a single sensor is difficult to meet the requirements of accurate detection of multi-component gases, which has prompted the development of sensor array technology.

[0003] At present, MEMS gas sensor arrays face multiple challenges in practical applications: first, there is the problem of interference from environmental factors. Changes in temperature and humidity can cause baseline drift. Second, there is the problem of cross-sensitivity between gas molecules. When multiple gases coexist, they will affect the detection accuracy of the sensor. Finally, there is the problem of sensor deployment optimization. Traditional fixed deployment is difficult to adapt to dynamically changing gas diffusion scenarios. Traditional gas detection methods often use a single data processing model, lack a systematic compensation mechanism for multi-source interference, and have deficiencies in abnormal data processing and sensor layout optimization. Summary of the invention

[0004] The present invention provides a gas detection method and device based on a MEMS gas sensor array, which realizes adaptive deployment optimization of the sensor array and improves the spatial coverage efficiency of the detection network.

[0005] In a first aspect, the present invention provides a gas detection method based on a MEMS gas sensor array, the gas detection method based on a MEMS gas sensor array comprising:

[0006] Grouping multiple MEMS gas sensors according to target gas types and collecting detection data to obtain original detection data sets;

[0007] Performing curve fitting and feature extraction on the original detection data set to obtain a feature parameter set, and substituting the feature parameter set into a second-order state space model to construct a sensor array feature model;

[0008] Calculating a temperature correction coefficient and a humidity correction coefficient using the sensor array characteristic model, and correcting the original detection data set to obtain standardized response data;

[0009] Extracting multiple response feature quantities from the standardized response data, and inputting the multiple response feature quantities into an unknown input observer to perform residual calculation and signal abnormal point identification, so as to obtain abnormal response data;

[0010] Selecting a time window length according to the abnormal response data, and performing piecewise linear regression and dynamic calibration on the abnormal data segments to obtain a target detection data set;

[0011] The target detection data set is input into the sensor network model to perform array deployment optimization, and an array deployment optimization solution is obtained.

[0012] In a second aspect, the present invention provides a gas detection device based on a MEMS gas sensor array, the gas detection device based on a MEMS gas sensor array comprising:

[0013] An acquisition module is used to group multiple MEMS gas sensors according to target gas types and collect detection data to obtain an original detection data set;

[0014] A construction module is used to perform curve fitting and feature extraction on the original detection data set to obtain a feature parameter set, and substitute the feature parameter set into a second-order state space model to construct a sensor array feature model;

[0015] A correction module, used to calculate a temperature correction coefficient and a humidity correction coefficient using the sensor array characteristic model, and to correct the original detection data set to obtain standardized response data;

[0016] A calculation module, used for extracting a plurality of response feature quantities from the standardized response data, and inputting the plurality of response feature quantities into an unknown input observer for residual calculation and signal abnormal point identification to obtain abnormal response data;

[0017] A calibration module, used to select a time window length according to the abnormal response data, and perform piecewise linear regression and dynamic calibration on the abnormal data segment to obtain a target detection data set;

[0018] The optimization module is used to input the target detection data set into the sensor network model to perform array deployment optimization and obtain an array deployment optimization solution.

[0019] The third aspect of the present invention provides a gas detection device based on a MEMS gas sensor array, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the gas detection device based on the MEMS gas sensor array executes the above-mentioned gas detection method based on the MEMS gas sensor array.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned gas detection method based on the MEMS gas sensor array.

[0021] In the technical solution provided by the present invention, by establishing a sensor array characteristic model and introducing a second-order state space equation, an accurate description of the dynamic response characteristics of the sensor is achieved, and the accuracy of system modeling is improved; a method combining temperature and humidity correction coefficients with a cross-sensitivity compensation matrix is ​​adopted to effectively eliminate the influence of environmental factors and gas cross-interference, and enhance the reliability of detection data; an unknown input observer is introduced to identify abnormal signals, and combined with a dynamic threshold judgment mechanism, the system's ability to identify and process abnormal data is improved; a method combining piecewise linear regression and dynamic calibration is used to accurately correct abnormal data, and the continuity and accuracy of detection data are guaranteed; based on the gas diffusion law and the Delaunay triangulation algorithm, adaptive deployment optimization of the sensor array is achieved, and the spatial coverage efficiency of the detection network is improved; a multi-level data processing architecture is adopted to optimize the entire process from data acquisition, feature extraction to abnormal processing, and the overall performance and reliability of the system are improved; and the real-time and reliability of data transmission are ensured through the optimized design of the network communication topology. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram of a gas detection method based on a MEMS gas sensor array provided in an embodiment of the present application;

[0024] Figure 2 A schematic block diagram of the structure of a gas detection device based on a MEMS gas sensor array provided in an embodiment of the present application;

[0025] Figure 3 A schematic block diagram of the structure of a gas detection device based on a MEMS gas sensor array provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0028] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0031] See also Figure 1 , Figure 1 A schematic diagram of a gas detection method based on a MEMS gas sensor array provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the gas detection method based on the MEMS gas sensor array provided in the embodiment of the present application includes steps S100 to S600.

[0032] Step S100, grouping multiple MEMS gas sensors according to target gas types and collecting detection data to obtain an original detection data set;

[0033] It is understandable that the execution subject of the present invention may be a gas detection device based on a MEMS gas sensor array, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0034] Specifically, a target gas of standard concentration is introduced into the sensitive layer of the MEMS gas sensor, and a response test is performed to determine the gas detection capability. In this process, the rate of change of the resistance value of the sensitive layer is used as an important parameter to characterize the sensor's response to the target gas, and then the gas-sensitive characteristic parameters are calculated. These parameters form the characteristic data matrix of the sensor. Based on the characteristic data matrix of the sensor, the MEMS gas sensor is divided into target gas types. The response characteristics of each sensor are different, and they are classified according to the slope of the sensor's response curve. For the sensor with the largest slope of the response curve, it is set as the main detection sensor. Through the division method, the responsibilities of each sensor in the array are determined, and a preliminary sensor grouping scheme is obtained. In order to verify the rationality and effectiveness of the sensor grouping, cross-validation is performed. The performance stability of the sensor configuration is checked through cross-validation, especially the impact of the response time. The sensor with the shortest response time is set as the cross-validation sensor to ensure that it can quickly respond to changes in gas concentration in practical applications. Thus, multiple sensor groups are obtained, and the main detection sensor and the cross-validation sensor in each group work together to improve the accuracy and reliability of gas detection. The sampling period of the main detection sensor in each sensor group is set to 10ms to ensure that the changes in gas concentration can be collected at a high frequency. At the same time, the cross-validation sensor sampling period is set to 20ms so that the verification can be effectively performed even when the amount of data is small. Through the setting, the data collection of the sensor response signal can obtain time series data with high timeliness, reflecting the changing trend of the gas concentration. The time series sampling data is converted into digital-to-analog through a high-precision AD conversion circuit. The analog signal is converted into a digital signal through the AD conversion circuit and transmitted to the data processing unit through the I2C bus. In this process, the digitized response data is processed by polynomial fitting to extract the key feature points of the rising section and the recovery section of the response curve. These feature points provide an effective basis for analyzing the sensor response characteristics and judging the gas concentration, and generate the sensor response feature sequence. The sampling period of the temperature and humidity sensor is set to 100ms, and the ambient temperature and relative humidity data are collected. The data is smoothed by the FIR filter to eliminate high-frequency noise and ensure the stability of the data, and the environmental parameter sequence is obtained. The sensor response feature sequence and the environmental parameter sequence are synchronized and matched according to the timestamp. By aligning the sensor response feature sequence and the environmental parameter sequence according to the timestamp, the original detection data set containing the response value, temperature value and humidity value is constructed.

[0035] Step S200, performing curve fitting and feature extraction on the original detection data set to obtain a feature parameter set, and substituting the feature parameter set into a second-order state space model to construct a sensor array feature model;

[0036] Specifically, the original detection data set is resampled and low-pass filtered at equal time intervals Δt=0.1s. Resampling helps convert the data into an equally spaced time series for subsequent analysis, while low-pass filtering can remove high-frequency noise and ensure the stability of the data. After the preprocessing process, a smooth preprocessed data sequence is obtained. According to the characteristics of the response process, the data sequence is segmented and divided into multiple response process data segments. These data segments contain sensor response information at different time stages. The response process data segment is fitted with the least squares curve, where the fitting function is set to f(t)=a·exp(-t / τ1)+b·exp(-t / τ2)+c. The function consists of three parts, corresponding to the fast response time constant τ1, the slow response time constant τ2, and the constant term c, and the fitting coefficients a, b, and c are obtained by the least squares method. Through this fitting process, the response characteristic function of each response process is obtained, reflecting the dynamic response characteristics of the sensor. Through these characteristic functions, key characteristic parameters such as response time tr, recovery time td, response amplitude ΔR and baseline drift R0 are extracted. Response time tr indicates the time required for the sensor response value to reach 90% of the maximum value, and recovery time td indicates the time required for the sensor response value to drop to 10% of the initial value. Response amplitude ΔR reflects the sensitivity of the sensor, while baseline drift R0 indicates the drift change of the sensor when the gas is not present. These parameters constitute the characteristic parameter set of the sensor. Based on the characteristic parameter set, a second-order state space model is established. The state space model of the system is constructed by establishing the state equation dx(t) / dt=Ax(t)+Bu(t) and the output equation y(t)=Cx(t), where x(t) is the state variable, u(t) is the input, y(t) is the output, A is the state matrix, B is the input matrix, and C is the output matrix. This equation describes the dynamic behavior of the sensor array and its relationship with the input signal. In this way, the state change of the sensor array is accurately represented, and the output response corresponding to the change in gas concentration is derived. Discretize the second-order state space model. Use a fixed sampling period for data processing to convert the continuous-time state space equation into a discrete-time equation. After using a sampling period of Δt=0.1 seconds, the continuous-time equation is converted into a discrete state equation x(k+1)=Φx(k)+Γu(k), where Φ is the discrete state transfer matrix, Γ is the discrete input matrix, k is the discrete time sequence number, and x(k+1) represents the state variable at the next moment. The discretized equation can be calculated at each sampling moment to ensure the continuity and stability of the system in the discrete time domain. Use the least squares estimation method to optimize the model parameters. Construct the least squares estimation function J(θ)=Σ(y(k)-ŷ(k))², where θ is the model parameter vector to be estimated, y(k) is the actual output, and ŷ(k) is the output predicted by the model.Through iterative calculation, the error function J(θ) is minimized to obtain the optimized model parameters. The optimization process adjusts the model parameters to make the error between the model output and the actual collected data as small as possible, ensuring that the model can accurately describe the behavior of the sensor array. The optimized model parameters are substituted into the second-order state space equation, and the root mean square error (RMSE) between the model output and the actual data is calculated. When the root mean square error is less than or equal to the preset target value, it is considered that the model has achieved the expected accuracy and the final sensor array characteristic model is obtained. This characteristic model can accurately reflect the dynamic response characteristics of the gas sensor array. Through a systematic modeling and optimization process, the accuracy and reliability of the gas detection method based on the MEMS gas sensor array are effectively improved.

[0037] Step S300, using the sensor array characteristic model to calculate the temperature correction coefficient and the humidity correction coefficient, and correcting the original detection data set to obtain standardized response data;

[0038] Specifically, the response data of the sensor array characteristic model is collected every 5℃ in the temperature range of 5℃-45℃, and the response deviation rate of each temperature point is calculated to obtain the temperature response deviation sequence. By analyzing the temperature response deviation sequence, the sensitivity change of the sensor at different temperatures is revealed. In order to eliminate the influence of temperature, a temperature correction function is established based on the temperature response deviation sequence. The correction function is MT(T)=1 / (1+kT·(T-25)), where kT is the temperature correction coefficient, T is the actual measured temperature, and MT(T) is the temperature correction function. The temperature response deviation sequence is fitted by the least squares method to determine the temperature correction coefficient kT, so that the sensor array can output an accurate response when the temperature changes. Humidity correction is performed. In the relative humidity range of 20% to 80%, the response data of the sensor array is collected every 10%, and the response deviation rate αH=(RH-R50) / R50 of each humidity point is calculated, where RH is the response value at humidity H and R50 is the response value at 50% humidity. By calculating these humidity response deviation rates, the humidity response deviation sequence is obtained. In order to correct the influence of humidity, a humidity correction function MH(H)=1 / (1+kH·(H-50)) is established, where kH is the humidity correction coefficient, H is the actual measured humidity, and MH(H) is the humidity correction function. The humidity response deviation sequence is fitted by the least squares method to obtain the humidity correction coefficient kH, thereby eliminating the interference of humidity on the sensor response. The cross-sensitivity problem of gas is considered, that is, the interaction between the target gas and the interfering gas. The target gas and the interfering gas are mixed in different concentration ratios of 0.2:1, 0.5:1, 1:1, 2:1 and 5:1, and a gas mixing experiment is carried out, and the response data of the sensor array is collected. By analyzing these data, the cross-sensitivity data is calculated to reflect the influence of the interfering gas on the detection of the target gas. Based on the cross-sensitivity data, an interference compensation matrix is ​​constructed to correct the influence of the interfering gas on the detection results. In this way, a standardized interference compensation matrix is ​​obtained, which can effectively eliminate the errors caused by the interfering gas. The temperature and humidity correction and interference compensation results are applied to the original detection data set, and the temperature and humidity compensation calculation is performed on it. After temperature and humidity compensation, the sensor response data will reflect more accurate environmental conditions. Based on the response value after environmental compensation, the concentration of the target gas is calculated to eliminate the influence of cross interference. In order to ensure the comparability and stability of the gas concentration data, normalization processing is performed to obtain standardized response data.

[0039] Step S400, extracting multiple response feature quantities from the standardized response data, and inputting the multiple response feature quantities into the unknown input observer to perform residual calculation and signal abnormal point identification, so as to obtain abnormal response data;

[0040] Specifically, the standardized response data is subjected to dynamic response feature extraction to identify and analyze abnormal situations in the sensor response. For the response curve of each sensor, the dynamic slope and curvature are calculated in the rising section and the recovery section respectively. The dynamic slope can reflect the speed of the sensor response change, while the curvature reflects the acceleration of the response change. The calculated feature quantity helps to identify the different stages of the sensor response. The multiple response feature quantities obtained by calculation provide a dynamic description of the behavior of each sensor. Based on the extracted multiple response feature quantities, a sliding time window is set to analyze the response data. The sliding window is a method of gradually processing the data in the window by continuously moving a window of a fixed time length forward in the time series. In each sliding window, the data is statistically analyzed, the mean and standard deviation in the window are calculated, and the three-times standard deviation principle is applied to mark the abnormal points. When the deviation of a data point exceeds three times the standard deviation of the mean, it is marked as an abnormal point. This method can effectively identify abnormal fluctuations in the response data and obtain a preliminary set of abnormal points. For the initial set of abnormal points, local waveform information is extracted to more accurately identify the specific features of the abnormality. Local waveform extraction includes calculating the amplitude, duration, and rate of change of the waveform near the abnormal point. These features can help the system better understand the specific manifestations of the abnormal phenomenon. By constructing these feature data into a feature vector matrix, the abnormal feature vector is obtained. Based on the abnormal feature vector, the state matrix of the unknown input observer is constructed, and the gain matrix of the observer is calculated by QR decomposition. QR decomposition is a matrix decomposition method in linear algebra that can help optimize the gain matrix in the calculation process, thereby improving the response accuracy of the observer. After the gain matrix is ​​calculated, the pole configuration is performed to ensure the stability and performance of the observer, and a set of observer parameters is obtained. The observer parameters and standardized response data are input into the unknown input observer to generate state estimates and output estimates. By comparing the difference between the actual output and the estimated output, the residual sequence is obtained. The residual sequence can reflect the deviation between the actual system and the model. In order to analyze the abnormal situation in the residual sequence, adaptive threshold calculation is performed. The residual mean and variance in the sliding window are substituted into the statistical discriminant function to calculate the dynamic threshold range. The dynamic threshold range is adaptively adjusted according to the change of data, so that it can flexibly adapt to different situations and enhance the accuracy of anomaly detection. According to the dynamic threshold range, the residual sequence is evaluated in segments. When the residual exceeds the threshold range, it means that the system has abnormal behavior. The data points that continuously exceed the threshold range are marked as abnormal intervals to obtain abnormal data segments. These abnormal data segments contain faults or abnormal conditions. For abnormal data segments, type identification and classification marking are performed. By combining judgment rules, the type and degree of the abnormality are determined according to the characteristics and behavior patterns of the abnormality. Abnormal response data is obtained.

[0041] Step S500, selecting a time window length according to the abnormal response data, and performing piecewise linear regression and dynamic calibration on the abnormal data segment to obtain a target detection data set;

[0042] Specifically, the abnormal response data is subjected to time series statistics to determine the appropriate time window length. By performing time series analysis on the abnormal response data, the interval distribution and duration between abnormal points are calculated to understand the time characteristics of the abnormal data. The reference time window length is set according to the length of the normal response cycle. The reference time window serves as a reference for subsequent abnormal data segment processing to reasonably divide the time period of the abnormal response data. The abnormal response data is segmented according to the reference time window length. The continuously occurring abnormal points are merged into abnormal data segments, and the start and end positions are marked for each data segment to obtain a sequence containing multiple abnormal data segments, which helps to identify the specific time period when the abnormal behavior occurs. For each abnormal data segment, the boundary is extended, that is, normal data of a time window length is added before and after the abnormal data segment. By adding context data, the processing deviation caused by data mutation is alleviated, and the processing of the abnormal data segment is made more stable and effective. The extended data segment is subjected to data normalization processing, the mean and variance of the normalized data are calculated, and the data points that deviate from the mean by more than three times the variance are eliminated, effectively removing noise and irrelevant data to obtain the target data segment. Linear regression analysis is performed based on the target data segment. By using a sliding window to calculate the regression coefficient segment by segment, a linear fit is performed on each abnormal data segment. The purpose of linear regression is to find a fitting straight line that best reflects the data trend, and then obtain regression correction data. Regression correction data can help the system adjust the abnormal data segment to make it conform to the expected response mode. In order to improve the accuracy of data correction, dynamic weight allocation is performed on the regression correction data. The correction coefficient is set based on the credibility of the data, that is, a higher weight is given to more reliable data, so as to optimize the correction process, make the corrected data more accurate, and avoid affecting the overall result due to the poor quality of some data. The weighted calibration data is obtained by weighted averaging the corrected data. The weighted calibration data is compared and analyzed with the normal response data. The accuracy and reliability of the calibration results are evaluated by calculating the correlation coefficient and the mean square error. If the correlation between the weighted calibration data and the normal response data is high and the mean square error is low, it means that the calibration process is successful and the calibrated data is closer to the normal response state. The verified data is used to verify the validity of the model. The verified data is reconstructed in time series, and the calibrated abnormal segment data replaces the original abnormal data and is rearranged in chronological order. In this way, the abnormal response data can be successfully corrected without affecting the overall data structure to obtain the target detection data set.

[0043] Step S600: input the target detection data set into the sensor network model to perform array deployment optimization and obtain an array deployment optimization solution.

[0044] Specifically, the target detection data set is input into the sensor network model for spatial distribution analysis. The behavior of the gas is analyzed according to the gas concentration gradient and diffusion rate in the detection area, and the gas diffusion parameters are obtained. These parameters help to describe the propagation characteristics of the gas in space. Based on the gas diffusion parameters, a gas diffusion model is constructed, which can simulate the diffusion process of the gas in space. Based on the gas diffusion model, the detection area is divided into multiple small grids, and the gas concentration distribution of each grid point is calculated to obtain the concentration distribution field, which reflects the spatial variation of the gas concentration in the detection area. The concentration distribution field is analyzed by contour lines. The optimal range of gas diffusion is determined by calculating the diffusion radius at the maximum concentration gradient. The diffusion radius reflects the propagation range of the gas and helps to determine the effective coverage range of each detection unit. Through this analysis, the optimal detection radius is obtained. Based on the optimal detection radius, the detection area is Delaunay triangulated, and the adjacent sensor nodes are connected through triangular grids to construct a preliminary sensor deployment network. Delaunay triangulation is a spatial grid division method that can ensure the connectivity between adjacent nodes and minimize redundant and invalid node connections. Node optimization is performed on the initial deployment network. Calculate the coverage overlap and connectivity of each sensor node. Coverage overlap refers to the overlap of detection ranges between multiple sensor nodes, while connectivity refers to whether the connection between sensor nodes is stable and efficient. In the optimization process, redundant nodes are deleted. These nodes will neither effectively expand the detection range nor increase the connectivity of the network. Through the optimization process, a streamlined and efficient sensor network deployment plan is obtained. Based on the optimized deployment plan, the spatial correlation coefficient between sensor nodes is calculated, which measures the spatial correlation between different nodes. For nodes with spatial correlation coefficients lower than the preset threshold, the positions are adjusted so that these nodes can better cooperate with other nodes to improve the coverage efficiency and connectivity of the entire network. Through the optimization process, the spatial correlation matrix is ​​obtained. The spatial correlation matrix is ​​input into the network communication model to calculate the data transmission path and communication load between nodes. The data transmission path reflects the information flow mode between sensor nodes, while the communication load represents the transmission pressure between nodes and the overall load of the network. Through calculation, the communication topology is obtained, and then the reliability of the topology is verified. Reliability verification includes calculating the connectivity of the network and the data transmission delay. Connectivity measures whether the communication between nodes in the network is unimpeded, while data transmission delay reflects the efficiency of information transmission in the network. If the connectivity is low or the transmission delay is long, it means that the network deployment needs further optimization. Through this process, the overall structure of the network is verified and adjusted to ensure its reliability and efficiency in practical applications. The optimized array deployment scheme is obtained, which can maximize the detection capability of the sensor network and ensure its stability and efficiency in complex environments.

[0045] In the embodiment of the present invention, by establishing a sensor array characteristic model and introducing a second-order state space equation, an accurate description of the dynamic response characteristics of the sensor is achieved, and the accuracy of system modeling is improved; a method combining temperature and humidity correction coefficients with a cross-sensitivity compensation matrix is ​​adopted to effectively eliminate the influence of environmental factors and gas cross-interference, and enhance the reliability of detection data; an unknown input observer is introduced to identify abnormal signals, and combined with a dynamic threshold judgment mechanism, the system's ability to identify and process abnormal data is improved; by combining a piecewise linear regression with a dynamic calibration method, accurate correction of abnormal data is achieved, and the continuity and accuracy of detection data are ensured; based on the gas diffusion law and the Delaunay triangulation algorithm, adaptive deployment optimization of the sensor array is achieved, and the spatial coverage efficiency of the detection network is improved; a multi-level data processing architecture is adopted to optimize the entire process from data acquisition, feature extraction to abnormal processing, thereby improving the overall performance and reliability of the system; and the real-time and reliability of data transmission are ensured through the optimized design of the network communication topology.

[0046] In a specific embodiment, the process of executing step S100 may specifically include the following steps:

[0047] The sensitive layer of the MEMS gas sensor is subjected to a response test by passing a target gas of standard concentration, and the gas-sensing characteristic parameters are calculated according to the change rate of the resistance value of the sensitive layer to obtain the sensor characteristic data matrix;

[0048] The target gas types of MEMS gas sensors are divided according to the sensor characteristic data matrix, and the sensor with the largest slope of the response curve is set as the main detection sensor to obtain the initial sensor grouping scheme;

[0049] Cross-validating the sensor configuration of the initial sensor grouping scheme, setting the sensor with the shortest response time as the cross-validation sensor, and obtaining multiple sensor groups;

[0050] The sampling period of the main detection sensor in each sensor group is set to 10ms, and the sampling period of the cross-validation sensor is set to 20ms, and the sensor response signal is collected to obtain the time series sampling data;

[0051] The time series sampling data is converted into digital-to-analog data by a high-precision AD conversion circuit, and transmitted to the data processing unit through the I2C bus to obtain digital response data. The digital response data is then subjected to polynomial fitting processing to extract the characteristic points of the rising and recovery segments of the response curve to obtain the sensor response characteristic sequence.

[0052] The sampling period of the temperature and humidity sensor is set to 100ms, the ambient temperature and relative humidity data are collected, and the data is smoothed through the FIR filter to obtain the environmental parameter sequence;

[0053] The sensor response feature sequence and the environmental parameter sequence are synchronously matched according to the timestamp to construct the original detection data set containing the response value, temperature value and humidity value.

[0054] Specifically, the response test is performed on the sensitive layer of the MEMS gas sensor. A target gas of standard concentration is passed through the sensitive layer of the sensor, and the change in its resistance value is monitored and recorded. When the target gas passes through the sensitive layer, the chemical reaction between the gas molecules and the sensor surface will cause a change in the sensor resistance, which in turn reflects the change in gas concentration. By testing the relationship between gas concentration and resistance change, the gas-sensitive characteristic parameters are calculated to reveal the sensor's response ability to different gases. For example, assuming the resistance change rate is:

[0055] ;

[0056] in, is the change in resistance, is the sensor at time The resistance value is is the reference resistance (i.e. the resistance when there is no gas). By calculating the resistance change rate, the gas-sensitive characteristic parameters of each sensor are obtained, including sensitivity, response time, recovery time, etc. These characteristic parameters are summarized into a sensor characteristic data matrix, which reflects the response characteristics of each sensor to the target gas. The MEMS gas sensors are divided into target gas types according to the sensor characteristic data matrix. By analyzing the slope of the response curve of each sensor, the sensor with the largest slope of the response curve is selected as the main detection sensor. The slope of the response curve is obtained by performing differential or derivative calculations on the sensor's resistance response curve. For example, given a sensor response curve of , whose slope is expressed as:

[0057] ;

[0058] in, is the resistance change, is the time change. The sensor with the largest slope is selected as the main detection sensor, because this sensor responds most sensitively to changes in gas concentration and can most effectively detect changes in the concentration of the target gas. In this way, a preliminary sensor grouping scheme is obtained, in which the main detection sensor of each group has the maximum response sensitivity. Cross-validation is performed on the initial sensor grouping scheme to ensure the effectiveness and reliability of the sensor configuration. The sensor with the shortest response time is set as the cross-validation sensor. The response time refers to the time required for the sensor to reach a steady state from the initial state. By testing the response time of different sensors, the sensor with a shorter response time is selected for cross-validation. The calculation of the response time depends on the time from the beginning of the sensor reaction to reaching 90% of the response value, which is specifically expressed as:

[0059] ;

[0060] in, is the response time, is the maximum resistance response value of the sensor. Through cross-validation, it is ensured that the sensors in each sensor group have a good balance in response speed and sensitivity, so as to obtain more accurate detection results. Through multiple sensor groups after cross-validation, multiple sensor configuration schemes are obtained, and each scheme can work independently. For the sensors in each sensor group, the sampling period of the main detection sensor is set to 10ms, while the sampling period of the cross-validation sensor is set to 20ms. The setting of the sampling period affects the accuracy and timeliness of data acquisition. A shorter sampling period can provide higher frequency response data, which helps to accurately capture subtle changes in the sensor. After collecting the response signal of the sensor, a high-precision analog-to-digital conversion circuit (ADC) is used for digital-to-analog conversion to convert the analog signal into a digital signal, and the data is transmitted to the data processing unit through the I2C bus. In the data processing unit, the digitized response data is subjected to polynomial fitting processing. Polynomial fitting is used to smooth the response curve and reduce the impact of noise on the data. For example, assuming that the sensor response curve is fitted by a quadratic polynomial:

[0061] ;

[0062] in, , ,and are the fitting coefficients, is a time variable. By fitting, the characteristic points of the rising and recovery segments of the response curve are extracted to obtain the response characteristic sequence of the sensor. By polynomial fitting, the response data is made smoother and more stable, which is convenient for subsequent analysis and processing. The ambient temperature and humidity data are collected through the temperature and humidity sensor. The sampling period of the temperature and humidity sensor is set to 100ms, and the temperature and relative humidity data are collected every 100 milliseconds. Since environmental parameters (such as temperature and humidity) will affect the response of the sensor, the temperature and humidity data are smoothed. Smoothing is performed through the FIR filter (finite impulse response filter) to reduce the high-frequency noise in the temperature and humidity data and obtain a more stable environmental parameter sequence. The output of the FIR filter is expressed as:

[0063] ;

[0064] in, is the output after filtering, is the input signal, are the filter coefficients, is the order of the filter. The smoothed environmental parameter sequence contains the temperature and humidity values ​​at each time point. The sensor response feature sequence and the environmental parameter sequence are synchronized and matched according to the timestamp. The timestamp refers to the time point recorded during data collection, usually in seconds or milliseconds. Through timestamp synchronization, the sensor response data is accurately matched with the environmental parameter data, ensuring that the response value, temperature value and humidity value of each sensor match at the same time, and constructing an original detection data set containing the response value, temperature value and humidity value.

[0065] In a specific embodiment, the process of executing step S200 may specifically include the following steps:

[0066] The original detection data set is resampled and low-pass filtered at equal time intervals of Δt=0.1s to obtain a preprocessed data sequence, and the preprocessed data sequence is segmented to obtain a response process data segment;

[0067] The least square curve fitting is performed on the response process data segment, and the fitting function is f(t)=a·exp(-t / τ1)+b·exp(-t / τ2)+c, where f(t) is the fitting function, t is the time variable, τ1 is the fast response time constant, τ2 is the slow response time constant, a, b, c are the fitting coefficients, and the response characteristic function is obtained;

[0068] According to the response characteristic function, the sensor response time tr, recovery time td, response amplitude ΔR and baseline drift R0 are extracted, where tr is the time when the response value reaches 90% of the maximum value, and td is the time when the response value drops to 10% of the initial value, and the characteristic parameter set is obtained;

[0069] Based on the characteristic parameter set, the state equation dx(t) / dt=Ax(t)+Bu(t) and the output equation y(t)=Cx(t) are established, where x(t) is the state variable, u(t) is the system input, y(t) is the system output, A is the state matrix, B is the input matrix, and C is the output matrix, and the second-order state space equation is obtained;

[0070] Discretize the second-order state space equation, use a sampling period of Δt=0.1s, and convert the continuous time equation into a discrete state equation x(k+1)=Φx(k)+Γu(k), where Φ is the discrete state transfer matrix, Γ is the discrete input matrix, k is the discrete time sequence number, and x(k+1) is the discrete state equation;

[0071] Based on the discrete state equation, the least squares estimation function J(θ)=Σ(y(k)-ŷ(k))2 is constructed, where θ is the model parameter vector to be estimated, y(k) is the actual output, and ŷ(k) is the model predicted output. J(θ) is minimized through iterative calculation to obtain the optimized model parameters.

[0072] The optimized model parameters are substituted into the second-order state-space equation, and the root mean square error between the model output and the actual data is calculated. When the root mean square error is less than or equal to the preset target value, the sensor array characteristic model is obtained.

[0073] Specifically, the original detection data is resampled using equal time intervals. Resample every 10 seconds. Evenly distribute the data at fixed time intervals to avoid difficulties in subsequent analysis due to irregular data points. After data resampling, low-pass filter the data. The low-pass filter removes high-frequency noise in the data, making the response curve smoother and easier to process later. The data after low-pass filtering is the preprocessed data sequence. The preprocessed data is segmented to divide the entire data sequence into different response process data segments. A long response signal is divided into multiple small segments with certain characteristics, so that each data segment can be analyzed separately to better extract useful information. Fit the response process data segment. The common fitting method is least squares fitting. A double exponential function is used for fitting. The function has two time constants, which are used to describe the fast response and slow response characteristics of the sensor. The fitting function is:

[0074] ;

[0075] in, is the fitted response value, and are the fitting coefficients, is the fast response time constant, is the slow response time constant, is the baseline drift, is a time variable. By fitting, the characteristic function of each response process is obtained, which can accurately describe the law of the change of resistance of the gas sensor over time during the response process. After fitting the response data, the key characteristic parameters are extracted. These characteristic parameters include response time , recovery time , response amplitude and baseline drift Response time It refers to the time when the sensor response value reaches 90% of the maximum value, calculated by the following formula:

[0076] ;

[0077] in, is the maximum response value of the sensor, is the response value at the current moment, is the reference resistance. Recovery time It refers to the time it takes for the sensor's response value to drop to 10% of its initial value and is calculated in a similar way. It represents the maximum response change of the sensor, that is:

[0078] ;

[0079] Baseline drift It represents the resistance value of the sensor in the absence of gas and is used as a reference resistor. These characteristic parameters constitute the characteristic data set of the sensor, which can effectively describe the dynamic characteristics of the sensor during the response process. A state space model is established based on these characteristic parameters. The state space model is a mathematical model used to describe the dynamic behavior of the system. A second-order state space model is used, which includes a state equation and an output equation. The state equation describes the dynamic changes of the system and is usually expressed as:

[0080] ;

[0081] in, is a state variable, is the input variable, is the state matrix, is the input matrix. The state variables Usually contains all the necessary information to describe the dynamic behavior of the system, while the input variables is the external control signal or input data. The output equation represents the output value of the system, expressed as:

[0082] ;

[0083] in, is the output of the system, is the output matrix. The state space model can describe the response changes of the sensor under different gas concentrations. In order to perform numerical calculations, the continuous state space equation is discretized and The sampling period is 1 second. The continuous dynamic system is converted into a discrete system that can be processed by the computer. The discretized state equation is expressed as:

[0084] ;

[0085] in, is the discrete state transfer matrix, is the discrete input matrix, is a discrete time sequence number, is a discrete time point The state variables at the moment. Through discretization, the continuous time system is converted into a discrete time system, so that it can be numerically solved in the computer. Based on the discretized state space equation, a least squares estimation function is constructed to optimize the model parameters. The least squares method is a method of fitting a model by minimizing the sum of squared errors. Using the least squares estimation function:

[0086] ;

[0087] in, is the model parameter vector to be estimated, is the actual output, is the model prediction output. Through iterative calculation, the model parameters are continuously adjusted so that the error function Minimize and get the optimized model parameters Substitute the optimized model parameters into the second-order state space equation and calculate the root mean square error (RMSE) between the model output and the actual data. The root mean square error is calculated using the following formula:

[0088] ;

[0089] in, is the number of samples, is the actual output, is the model prediction output. If the RMS error is less than or equal to the preset target value, it means that the model fitting effect is good, and the sensor array characteristic model obtained at this time can be used for further gas detection and analysis.

[0090] In a specific embodiment, the process of executing step S300 may specifically include the following steps:

[0091] The response data of the sensor array characteristic model is collected every 5°C in the temperature range of 5°C-45°C, and the response deviation rate of each temperature point is calculated to obtain the temperature response deviation sequence;

[0092] Based on the temperature response deviation sequence, a temperature correction function MT(T)=1 / (1+kT·(T-25)) is established, where kT is the temperature correction coefficient, T is the measured temperature, and MT(T) is the temperature correction function. The temperature correction coefficient is determined by least squares fitting.

[0093] The sensor array characteristic model collects response data every 10% in the relative humidity range of 20%-80%, and calculates the response deviation rate of each humidity point αH=(RH-R50) / R50, where RH is the response value at humidity H, R50 is the response value at 50% humidity, and αH is the response deviation rate. Finally, the humidity response deviation sequence is obtained;

[0094] According to the humidity response deviation sequence, a humidity correction function MH(H)=1 / (1+kH·(H-50)) is established, where kH is the humidity coefficient, H is the humidity, and MH(H) is the humidity correction function. The humidity correction coefficient is determined by least squares fitting.

[0095] The target gas and the interfering gas were mixed in concentration ratios of 0.2:1, 0.5:1, 1:1, 2:1, and 5:1, the sensor response data was collected, the cross-sensitivity data was calculated, and an interference compensation matrix was constructed based on the cross-sensitivity data to obtain a standardized interference compensation matrix;

[0096] The temperature and humidity compensation calculation is performed on the original detection data set to obtain the response value after environmental compensation, and the gas concentration is calculated on the response value after environmental compensation to obtain the concentration data with cross interference eliminated, and the concentration data with cross interference eliminated is normalized to obtain the standardized response data.

[0097] Specifically, the sensor array characteristic model is tested for response in the temperature range of 5°C to 45°C, and the response data is collected every 5°C. The target gas is input into the sensor array, the response data at different temperatures are recorded, and the response deviation rate at each temperature point is calculated. The response deviation rate refers to the difference between the response of the sensor and the reference response (usually the response at 25°C) at a specific temperature. Temperature response deviation rate Calculated by the following formula:

[0098] ;

[0099] in, The temperature is The sensor response at is the response value at the reference temperature of 25°C. By calculating the response deviation rate of all temperature points, the temperature response deviation sequence is obtained. This sequence is used to describe the impact of temperature changes on the sensor response. Based on the temperature response deviation sequence, a temperature correction function is established. , which is of the form:

[0100] ;

[0101] in, is the temperature correction coefficient, is the measured temperature, is the temperature correction function. In order to determine the temperature correction coefficient , the temperature response deviation series is fitted by the least squares method. The least squares method optimizes the coefficients by minimizing the following objective function:

[0102] ;

[0103] in, It is Temperature points, is the response data at this temperature. By solving this least squares problem, the optimal temperature correction coefficient is obtained In view of the influence of humidity, the sensor array characteristic model is tested for response in the relative humidity range of 20% to 80%, and the response data is collected every 10% humidity. Similar to temperature correction, the humidity response deviation rate The calculation formula is:

[0104] ;

[0105] in, The humidity is The sensor response at It is the response value at 50% humidity. Humidity response deviation rate Describe the change of sensor response under different humidity. The humidity response deviation sequence is used to establish the humidity correction function. The humidity correction function is in the form of:

[0106] ;

[0107] in, is the humidity correction factor, For humidity, is the humidity correction function. The humidity correction coefficient is obtained by fitting the humidity response deviation sequence using the least squares method.

[0108] ;

[0109] in, It is humidity points, is the response data under this humidity. Optimize the humidity correction coefficient by the least squares method , and obtain an accurate humidity correction function. After completing the temperature and humidity correction, consider the cross-sensitivity of the gas sensor. Since multiple gases affect the sensor array, the response of the sensor is not only affected by the target gas, but also by the interfering gas. In order to eliminate this cross-interference, the target gas and the interfering gas are mixed in different concentration ratios through experiments, such as 0.2:1, 0.5:1, 1:1, 2:1 and 5:1, and the response data of the sensor is collected to calculate the cross-sensitivity data. Cross-sensitivity data refers to the change in the response of the sensor to the target gas and the interfering gas under different gas concentration conditions. These data are used to construct an interference compensation matrix, which describes the cross-effect of different gases in the sensor response. Based on these data, a standardized interference compensation matrix is ​​constructed to eliminate the influence of interfering gases by adjusting the response of the sensor. For the original detection data set, the temperature and humidity compensation calculation is applied to obtain the response value after environmental compensation. The process of temperature and humidity compensation is to combine the corrected response data with the actual environmental parameters (such as temperature and humidity) to adjust the sensor output to make it more consistent with the actual concentration of the target gas. Based on the compensated response data, the gas concentration is calculated using a pre-established gas concentration model that is obtained by fitting the experimental data. The calculated concentration data is the concentration data after environmental compensation and cross-interference elimination. In order to ensure that the output data of the sensor can be directly used in practical applications, the concentration data after cross-interference elimination is normalized. The concentration data is mapped to a standard range for subsequent analysis and processing. Normalization is performed using the following formula:

[0110] ;

[0111] in, is the normalized concentration data, is the original concentration data, and are the minimum and maximum values ​​of the concentration data respectively. After normalization, the standardized response data is obtained.

[0112] In a specific embodiment, the process of executing step S400 may specifically include the following steps:

[0113] The dynamic response characteristics are extracted from the standardized response data, and the dynamic slope and curvature of the response curve of each sensor are calculated in the rising section and the recovery section to obtain multiple response characteristic quantities;

[0114] A sliding time window is set based on multiple response characteristics, the statistical mean and standard deviation of the data in the sliding window are calculated, and the abnormal points are marked using the principle of three times the standard deviation to obtain the initial abnormal point set;

[0115] Extract the local waveform of the initial abnormal point set, calculate the amplitude, duration and change rate of the waveform, construct the eigenvector matrix, obtain the abnormal eigenvector, and construct the unknown input observer state matrix based on the abnormal eigenvector. Calculate the observer gain matrix through QR decomposition, and perform pole configuration to obtain the observer parameters.

[0116] Input observer parameters and standardized response data into the unknown input observer to generate state estimation and output estimation, and make a difference between the actual output and the estimated output to obtain the residual sequence;

[0117] Perform adaptive threshold calculation on the residual sequence, substitute the mean and variance of the residual in the sliding window into the statistical discriminant function to obtain the dynamic threshold range;

[0118] The residual sequence is evaluated segmentally according to the dynamic threshold range, and the data points that continuously exceed the threshold range are marked as abnormal intervals to obtain abnormal data segments. The abnormal data segments are then type-identified and classified, and the type and degree of the abnormality are determined by combining judgment rules to obtain abnormal response data.

[0119] Specifically, for the standardized response data of each sensor, dynamic response feature extraction is performed. The important features in the sensor response curve are identified, especially the dynamic changes in the rising section and the recovery section. For each sensor response curve, the dynamic slope and curvature are calculated in the rising section and the recovery section respectively. The slope indicates the rate of change of the curve, while the curvature reflects the degree of curvature of the curve. Set the response curve to ,in is the time variable, is the response value. Dynamic slope and dynamic curvature The calculation is obtained by the following formula:

[0120] ;

[0121] ;

[0122] in, is the first derivative of the response, indicating the rate of change of the response; It is the second derivative of the response, which indicates the acceleration or curvature of the response change. These two characteristic quantities can help the system capture important dynamic information in the response process. By calculating these characteristic quantities, multiple response characteristic quantities are obtained. Based on the dynamic response characteristic quantity, a sliding time window is set to process the response data in segments. The size of the sliding window is selected according to the characteristics of the data, and the common setting is a time interval of fixed length for each slide. For the data in each window, calculate its statistical mean and standard deviation. The mean reflects the central trend of the data, while the standard deviation measures the degree of fluctuation of the data. Based on these statistical characteristics, the principle of three times the standard deviation is used to mark outliers. If a data point exceeds the range of the mean of the data in the window plus or minus three times the standard deviation, the data point is considered to be an outlier. Mathematical expression:

[0123] ;

[0124] in, is the mean of the data in the sliding window, is the standard deviation, It is The response value at each time point, is the number of data in the window. According to the triple standard deviation principle, points that meet the following conditions are marked as abnormal:

[0125] ;

[0126] Extract the local waveform features of the initial abnormal point set and calculate the amplitude, duration and change rate of the waveform. The amplitude represents the maximum deviation of the waveform, the duration is the length of time the waveform changes, and the change rate describes the speed of the waveform change. These features are constructed into a feature vector matrix as a description of each abnormal point. Set the amplitude of the waveform to , duration is , the rate of change is , then the abnormal feature vector is expressed as:

[0127] ;

[0128] The state matrix of the unknown input observer is constructed using the abnormal eigenvector. The observer gain matrix is ​​calculated by QR decomposition, and the pole configuration is performed to obtain the observer parameters. The purpose of the observer model is to dynamically track the behavior of the system through state estimation, so as to identify the difference between the actual output and the estimated output, that is, the residual. Assume that the state equation is:

[0129] ;

[0130] The output equation is:

[0131] ;

[0132] in, is the state vector of the system, is the input of the system, is the output of the system, is the state matrix, is the input matrix, is the output matrix. By discretizing these equations and calculating the gain matrix using QR decomposition, the discrete state equation is obtained:

[0133] ;

[0134] in, is the discrete state transfer matrix, is the input matrix, is a discrete time sequence number. Through the optimization algorithm, the state of the system is estimated, and the state estimate and output estimate are generated. The difference (residual) between the actual output and the estimated output is calculated by the following formula:

[0135] ;

[0136] in, is the residual, is the actual output, is the estimated output. When performing adaptive threshold calculation on the residual sequence, the threshold range is dynamically adjusted according to the mean and variance of the residuals in the sliding window. Calculate the mean of the residuals in the sliding window and standard deviation , and then substitute it into the statistical discriminant function to obtain the dynamic threshold range:

[0137] ;

[0138] According to the dynamic threshold range, the data points that continuously exceed the threshold range are marked as abnormal intervals to obtain abnormal data segments. The abnormal data segments are identified and classified. By combining the judgment rules, the type and degree of the abnormality are determined, such as too fast response, too slow response or abnormal amplitude, and finally the abnormal response data is obtained. According to these marked abnormal intervals, appropriate correction measures are taken to improve the accuracy and reliability of gas detection.

[0139] In a specific embodiment, the process of executing step S500 may specifically include the following steps:

[0140] Perform time series statistics on abnormal response data, calculate the interval distribution and duration between abnormal points, set the time window according to the length of the normal response cycle, and obtain the benchmark time window length;

[0141] The abnormal response data is segmented according to the length of the reference time window, and the continuously occurring abnormal points are merged into abnormal data segments. The start and end positions of each abnormal data segment are marked to obtain an abnormal data segment sequence;

[0142] The boundary of each data segment in the abnormal data segment sequence is extended, and normal data of a time window length is added before and after the abnormal data segment to obtain the extended data segment. The extended data segment is normalized, and the data mean and variance are calculated. The data points that deviate from the mean by more than three times the variance are removed to obtain the target data segment.

[0143] Perform linear regression analysis on the target data segment, use a sliding window to calculate the regression coefficient segment by segment, perform linear fitting on the abnormal data to obtain regression correction data, and dynamically assign weights to the regression correction data. Set the correction coefficient based on the data credibility, perform weighted averaging on the corrected data, and obtain weighted calibration data.

[0144] The weighted calibration data is compared and analyzed with the normal response data, the correlation coefficient and the mean square error are calculated, the calibration results are verified, the verified data are obtained, and the time series of the verified data is reconstructed. The calibrated abnormal segment data replaces the original abnormal data and is rearranged in chronological order to obtain the target detection data set.

[0145] Specifically, we conduct time series statistics on abnormal response data, calculate the interval distribution and duration between abnormal points, understand the frequency, interval and duration characteristics of abnormal points, and set an appropriate time window accordingly. Set the length of the normal response cycle to , which is obtained by analyzing the statistical characteristics of the normal response data. For example, assuming that the length of the normal response cycle is s, and the time window is set based on this cycle length. Assuming that an abnormal response point appears in a given time period, if the time interval between adjacent abnormal points is , then the interval distribution of these outliers is obtained by calculating the mean and standard deviation:

[0146] ;

[0147] On this basis, by analyzing the interval distribution of abnormal points, the length of the benchmark time window is reasonably selected. , and based on this, the abnormal response data is segmented and the continuously occurring abnormal points are merged into abnormal data segments. The starting and ending points of each abnormal data segment are marked by the following method: the starting point of each abnormal data segment is set to the position of the first abnormal point , the end point is the position of the last outlier point , forming a sequence of abnormal data segments :

[0148] ;

[0149] The boundaries of each abnormal data segment are expanded, and normal data of a time window length is added before and after each abnormal segment to obtain the expanded data segment, which can better capture the context information of the abnormal segment and avoid overly limited abnormal data analysis. After expansion, the expanded data segment is normalized to scale the data to a uniform range. Normalized data It is done by the following formula:

[0150] ;

[0151] in, and are the mean and standard deviation of the extended data segment. The purpose of normalization is to reduce the scale difference of the data and make subsequent analysis more stable. and standard deviation , and according to the three-times standard deviation principle, remove the data points that deviate from the mean by more than three times the standard deviation, that is:

[0152] ;

[0153] After removing extreme values, the target data segment is obtained Perform linear regression analysis on the target data segment and use a sliding window to calculate the regression coefficient segment by segment. Assuming that the target data segment is fitted by a linear model, the form of the linear regression function is:

[0154] ;

[0155] in, is the regression coefficient, which represents the rate of change of the response over time, is the intercept, which indicates the baseline value of the response. The regression coefficient is obtained by the least squares method. and The goal of the least squares method is to minimize the following objective function:

[0156] ;

[0157] in, is the actual response value, is the linear fitting value. By solving the optimization problem, the regression correction data is obtained. Dynamic weights are assigned to the regression correction data, and the correction coefficient is set based on the credibility of the data. The credibility is estimated by calculating the variance or other statistical characteristics of the data. For example, setting the correction coefficient The credibility factor for each data point is calculated as follows:

[0158] ;

[0159] in, It is a data point The weighted average of the corrected data is used to obtain the weighted calibration data.

[0160] ;

[0161] The weighted calibration data is compared with the normal response data, and the correlation coefficient and mean square error (MSE) are calculated. Reflects the degree of linear correlation between weighted calibration data and normal response data. The calculation formula is:

[0162] ;

[0163] in, and are the means of the weighted calibration data and the normal response data, respectively, and is its standard deviation. The calculation formula of mean square error (MSE) is:

[0164] ;

[0165] The validity of the calibration results is verified by comparing the obtained correlation coefficient and mean square error. If the calibration results meet expectations, the calibrated abnormal segment data replaces the original abnormal data. The time series of the verified data is reconstructed, and the replaced data is rearranged in chronological order to obtain the target detection data set.

[0166] In a specific embodiment, the process of executing step S600 may specifically include the following steps:

[0167] The target detection data set is input into the sensor network model for spatial distribution analysis. The gas diffusion parameters are obtained according to the gas concentration gradient and diffusion rate in the detection area. A gas diffusion model is constructed based on the gas diffusion parameters. The detection area is gridded, and the gas concentration distribution of each grid point is calculated to obtain the concentration distribution field.

[0168] Perform contour analysis on the concentration distribution field, calculate the diffusion radius at the maximum concentration gradient, set the diffusion radius as the coverage range of the detection unit, and obtain the optimal detection radius. Based on the optimal detection radius, perform Delaunay triangulation on the detection area, connect adjacent sensor nodes to build a triangular network, and obtain the initial deployment network.

[0169] The nodes of the initial deployment network are optimized, the coverage overlap and connectivity of each sensor node are calculated, and the redundant nodes are deleted to obtain the optimized deployment plan. The spatial correlation coefficient between sensor nodes is calculated based on the optimized deployment plan, and the nodes with spatial correlation coefficients lower than the threshold are adjusted to obtain the spatial correlation matrix.

[0170] The spatial correlation matrix is ​​input into the network communication model, the data transmission path and communication load between nodes are calculated, the communication topology is obtained, and the reliability of the communication topology is verified. The network connectivity and data transmission delay are calculated to determine the array deployment optimization plan.

[0171] Specifically, the target detection data set is input into the sensor network model for spatial distribution analysis. The gas concentration gradient and diffusion rate in the detection area are calculated through the gas diffusion model. The gas diffusion process follows the classical diffusion equation, and the following model is used to describe the change in gas concentration:

[0172] ;

[0173] in, is the gas concentration at location and time The distribution of is the diffusion coefficient of the gas, is the Laplace operator, describing the spatial second-order derivative of gas concentration, is the gas flow rate, is the concentration gradient. By solving this equation, the concentration distribution of the gas in the detection area is obtained. Based on this diffusion equation, the diffusion parameters of the gas, such as the diffusion coefficient, are calculated. and flow rate , and use these parameters to build a mathematical model to describe the gas diffusion process. The detection area is gridded, and the gas concentration value of each grid point is calculated according to the gas concentration distribution model to obtain the concentration distribution field. The discretization of the concentration distribution field is achieved by setting a series of grid points in the area. Assume that the side length of the grid is , then the concentration at each grid point is It is obtained by numerically solving the diffusion equation. The concentration distribution field of the gas is expressed as:

[0174] ;

[0175] in, represents the gas concentration function, represents the position coordinates of the i-th grid point in space, represents the time variable, express It is a point in the discrete grid point set. The grid point set is the set of all the divided grid points in the detection area. According to the concentration distribution field, the contour analysis is performed to determine the diffusion radius at the maximum concentration gradient. This radius represents the diffusion range of the gas in the area. Assume that the maximum concentration gradient occurs at a certain point , the corresponding diffusion radius is calculated by the following formula:

[0176] ;

[0177] in, is the diffusion radius, is the diffusion coefficient, is the flow rate. The diffusion radius reflects the maximum range over which the gas can effectively diffuse. The calculated diffusion radius is set as the coverage of the detection unit to determine the optimal detection radius. , that is, the spatial range of the detection unit. Based on the optimal detection radius , perform Delaunay triangulation on the detection area and generate the initial deployment plan of the sensor network. Delaunay triangulation is a spatial grid division method that effectively divides the spatial area into several non-overlapping triangular units and ensures that the inscribed circle of each triangle does not contain other points. The advantage of this method is that it automatically generates the network topology according to the density and distribution of the sensors and maintains good geometric properties. The initial deployment network is obtained by the following steps: select the location of the sensor and To determine the coverage of each sensor; connect adjacent sensor nodes through the Delaunay triangulation method to form the initial topology of the sensor network. Optimize the nodes of the initial deployment network and calculate the coverage overlap and connectivity of each sensor node. The coverage overlap and connectivity of the sensor nodes are evaluated by calculating the spatial distance and signal strength between the nodes. For example, for any two sensor nodes and , the spatial distance between them is ,like , then the two nodes are considered to overlap each other. The coverage overlap of the nodes is obtained by calculating the number of nodes within the coverage range. For some redundant nodes, they are deleted to optimize the network structure. By calculating the spatial correlation coefficient between sensor nodes , optimize the position of the node. The spatial correlation coefficient reflects the relative position relationship between sensors and is calculated by the following formula:

[0178] ;

[0179] in, For sensor nodes and If the spatial correlation coefficient is lower than a certain threshold , you need to Adjust the position to improve the overall coverage of the network. After node optimization, the spatial association matrix is ​​input into the network communication model to calculate the data transmission path and communication load between nodes. The communication load reflects the communication overhead and bandwidth requirements of each node, and is estimated by calculating the data transmission volume of each link in the network. The network communication model uses the shortest path algorithm in graph theory, such as the Dijkstra algorithm, to calculate the shortest path from one node to another, and then estimate the data transmission delay and load. The communication topology is obtained and its reliability is verified. The reliability of the communication topology is verified by the following two aspects: network connectivity: check whether there are isolated nodes or disconnected parts in the network. Networks with high connectivity usually have stronger robustness; data transmission delay: calculate the transmission delay of data from the source node to the target node in the network to ensure that the delay is within an acceptable range. By verifying the connectivity and transmission delay of the network, the array deployment optimization plan is determined to ensure that the sensor array can efficiently monitor the gas concentration while ensuring the reliability and low latency of network communication.

[0180] See also Figure 2 , Figure 2 A schematic block diagram of the structure of a gas detection device 200 based on a MEMS gas sensor array provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, a gas detection device 200 based on a MEMS gas sensor array includes:

[0181] The acquisition module 210 is used to group multiple MEMS gas sensors according to the target gas type and collect detection data to obtain an original detection data set;

[0182] A construction module 220 is used to perform curve fitting and feature extraction on the original detection data set to obtain a feature parameter set, and substitute the feature parameter set into the second-order state space model to construct a sensor array feature model;

[0183] A correction module 230 is used to calculate a temperature correction coefficient and a humidity correction coefficient using a sensor array characteristic model, and to correct an original detection data set to obtain standardized response data;

[0184] A calculation module 240 is used to extract multiple response feature quantities from the standardized response data, and input the multiple response feature quantities into the unknown input observer to perform residual calculation and signal abnormal point identification to obtain abnormal response data;

[0185] A calibration module 250 is used to select a time window length according to the abnormal response data, and perform piecewise linear regression and dynamic calibration on the abnormal data segment to obtain a target detection data set;

[0186] The optimization module 260 is used to input the target detection data set into the sensor network model to perform array deployment optimization and obtain an array deployment optimization solution.

[0187] Through the coordinated cooperation of the above-mentioned components, by establishing a sensor array characteristic model and introducing a second-order state-space equation, an accurate description of the dynamic response characteristics of the sensor is achieved, thereby improving the accuracy of system modeling; a method combining temperature and humidity correction coefficients with a cross-sensitivity compensation matrix is ​​adopted to effectively eliminate the influence of environmental factors and gas cross-interference, thereby enhancing the reliability of detection data; an unknown input observer is introduced for abnormal signal recognition, and combined with a dynamic threshold judgment mechanism, the system's ability to recognize and process abnormal data is improved; an accurate correction of abnormal data is achieved by combining piecewise linear regression with dynamic calibration, thereby ensuring the continuity and accuracy of detection data; based on the gas diffusion law and the Delaunay triangulation algorithm, adaptive deployment optimization of the sensor array is achieved, thereby improving the spatial coverage efficiency of the detection network; a multi-level data processing architecture is adopted to optimize the entire process from data acquisition, feature extraction to abnormal processing, thereby improving the overall performance and reliability of the system; and the real-time and reliability of data transmission are ensured through the optimized design of the network communication topology.

[0188] See also Figure 3 , Figure 3 A schematic block diagram of the structure of a gas detection device 300 based on a MEMS gas sensor array provided in an embodiment of the present application, wherein the gas detection device 300 based on a MEMS gas sensor array includes a processor 301 and a memory 302, wherein the processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0189] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned gas detection methods based on the MEMS gas sensor array.

[0190] The processor 301 is used to provide computing and control capabilities to support the operation of the entire gas detection device 300 based on the MEMS gas sensor array.

[0191] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned gas detection methods based on the MEMS gas sensor array.

[0192] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation of the gas detection device 300 based on the MEMS gas sensor array involved in the scheme of the present application. The specific gas detection device 300 based on the MEMS gas sensor array may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0193] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0194] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the gas detection device 300 based on the MEMS gas sensor array described above can refer to the corresponding process of the aforementioned gas detection method based on the MEMS gas sensor array, and will not be repeated here.

[0195] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement a gas detection method based on a MEMS gas sensor array as provided in an embodiment of the present application.

[0196] The computer-readable storage medium may be an internal storage unit of the gas detection device 300 based on the MEMS gas sensor array in the aforementioned embodiment, such as a hard disk or memory of the gas detection device 300 based on the MEMS gas sensor array. The computer-readable storage medium may also be an external storage device of the gas detection device 300 based on the MEMS gas sensor array, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped with the gas detection device 300 based on the MEMS gas sensor array.

[0197] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0198] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0199] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A gas detection method based on a MEMS gas sensor array, characterized in that: include: Grouping multiple MEMS gas sensors according to target gas types and collecting detection data to obtain original detection data sets; The original detection data set is subjected to curve fitting and feature extraction to obtain a feature parameter set, and the feature parameter set is substituted into a second-order state space model to construct a sensor array feature model; specifically comprising: resampling and low-pass filtering the original detection data set at an equal time interval Δt=0.1s to obtain a preprocessed data sequence, and segmenting the preprocessed data sequence to obtain a response process data segment; performing least squares curve fitting on the response process data segment, and the fitting function f(t)=a·exp(-t / τ1)+b·exp(-t / τ2)+c, wherein f(t) is a fitting function, t is a time variable, τ1 is a fast response time constant, τ2 is a slow response time constant, and a, b, and c are fitting coefficients to obtain a response characteristic function; according to the response characteristic function, extracting the sensor response time tr, recovery time td, response amplitude ΔR, and baseline drift R0, wherein where tr is the time when the response value reaches 90% of the maximum value, and td is the time when the response value drops to 10% of the initial value, and a characteristic parameter set is obtained; based on the characteristic parameter set, a state equation dx(t) / dt=Ax(t)+Bu(t) and an output equation y(t)=Cx(t) are established, where x(t) is the state variable, u(t) is the system input, y(t) is the system output, A is the state matrix, B is the input matrix, and C is the output matrix, and a second-order state space equation is obtained; the second-order state space equation is discretized, and a sampling period of Δt=0.1s is used to convert the continuous time equation into a discrete state equation x(k+1)=Φx(k)+Γu(k), where Φ is the discrete state transfer matrix, Γ is the discrete input matrix, k is the discrete time sequence number, and x(k+1) is the discrete state equation; based on the discrete state equation, a least squares estimation function J(θ)=Σ(y(k)-ŷ(k)) is constructed. 2 , where θ is the model parameter vector to be estimated, y(k) is the actual output, ŷ(k) is the model predicted output, and J(θ) is minimized by iterative calculation to obtain the optimized model parameters; the optimized model parameters are substituted into the second-order state space equation, and the root mean square error between the model output and the actual data is calculated. When the root mean square error is less than or equal to the preset target value, the sensor array characteristic model is obtained; Calculating a temperature correction coefficient and a humidity correction coefficient using the sensor array characteristic model, and correcting the original detection data set to obtain standardized response data; Extracting multiple response feature quantities from the standardized response data, and inputting the multiple response feature quantities into an unknown input observer to perform residual calculation and signal abnormal point identification, so as to obtain abnormal response data; Selecting a time window length according to the abnormal response data, and performing piecewise linear regression and dynamic calibration on the abnormal data segments to obtain a target detection data set; The target detection data set is input into the sensor network model to perform array deployment optimization, and an array deployment optimization solution is obtained.

2. The gas detection method based on the MEMS gas sensor array according to claim 1, characterized in that: The method of grouping the multiple MEMS gas sensors according to the target gas type and collecting detection data to obtain the original detection data set includes: The sensitive layer of the MEMS gas sensor is subjected to a response test by passing a target gas of standard concentration, and the gas-sensing characteristic parameters are calculated according to the change rate of the resistance value of the sensitive layer to obtain the sensor characteristic data matrix; Classify the target gas types of the MEMS gas sensors according to the sensor characteristic data matrix, set the sensor with the largest response curve slope as the main detection sensor, and obtain an initial sensor grouping scheme; Cross-validating sensor configurations for the initial sensor grouping scheme, setting sensors with the shortest response time as cross-validated sensors, and obtaining a plurality of sensor groups; The sampling period of the main detection sensor in each sensor group is set to 10ms, and the sampling period of the cross-validation sensor is set to 20ms, and the sensor response signal is collected to obtain the time series sampling data; The time series sampling data is converted into digital-to-analog data by a high-precision AD conversion circuit, and transmitted to a data processing unit through an I2C bus to obtain digital response data, and the digital response data is subjected to polynomial fitting processing to extract the characteristic points of the rising section and the recovery section of the response curve to obtain a sensor response characteristic sequence; The sampling period of the temperature and humidity sensor is set to 100ms, the ambient temperature and relative humidity data are collected, and the data is smoothed through the FIR filter to obtain the environmental parameter sequence; The sensor response feature sequence and the environmental parameter sequence are synchronously matched according to timestamps to construct an original detection data set including response values, temperature values ​​and humidity values.

3. The gas detection method based on MEMS gas sensor array according to claim 1, characterized in that: The method of calculating the temperature correction coefficient and the humidity correction coefficient by using the sensor array characteristic model and correcting the original detection data set to obtain standardized response data includes: The sensor array characteristic model collects response data at intervals of 5°C within a temperature range of 5°C to 45°C, calculates the response deviation rate of each temperature point, and obtains a temperature response deviation sequence; Based on the temperature response deviation sequence, a temperature correction function MT(T)=1 / (1+kT·(T-25)) is established, where kT is the temperature correction coefficient, T is the measured temperature, and MT(T) is the temperature correction function, and the temperature correction coefficient is determined by least squares fitting; The sensor array characteristic model collects response data every 10% in the relative humidity range of 20%-80%, and calculates the response deviation rate αH=(RH-R50) / R50 at each humidity point, where RH is the response value at humidity H, R50 is the response value at 50% humidity, and αH is the response deviation rate, and finally obtains the humidity response deviation sequence; According to the humidity response deviation sequence, a humidity correction function MH(H)=1 / (1+kH·(H-50)) is established, wherein kH is the humidity coefficient, H is the humidity, and MH(H) is the humidity correction function, and the humidity correction coefficient is determined by least squares fitting; The target gas and the interfering gas are mixed in concentration ratios of 0.2:1, 0.5:1, 1:1, 2:1, and 5:1, sensor response data are collected, cross-sensitivity data are calculated, and an interference compensation matrix is ​​constructed based on the cross-sensitivity data to obtain a standardized interference compensation matrix; The original detection data set is subjected to temperature and humidity compensation calculation to obtain the response value after environmental compensation, the response value after environmental compensation is subjected to gas concentration calculation to obtain the concentration data with cross interference eliminated, and the concentration data with cross interference eliminated is normalized to obtain the standardized response data.

4. The gas detection method based on the MEMS gas sensor array according to claim 3, characterized in that: The step of extracting a plurality of response feature quantities from the standardized response data and inputting the plurality of response feature quantities into an unknown input observer for residual calculation and signal abnormal point identification to obtain abnormal response data includes: Extracting dynamic response features from the standardized response data, calculating the dynamic slope and curvature of the response curve of each sensor in the rising section and the recovery section, respectively, to obtain a plurality of response feature quantities; A sliding time window is set based on the multiple response feature quantities, a statistical mean and a standard deviation are calculated for the data in the sliding window, and an abnormal point is marked using a three times standard deviation principle to obtain an initial abnormal point set; Extracting a local waveform from the initial abnormal point set, calculating the amplitude, duration and rate of change of the waveform, constructing a eigenvector matrix, obtaining an abnormal eigenvector, and constructing an unknown input observer state matrix based on the abnormal eigenvector, calculating an observer gain matrix through QR decomposition, and performing pole configuration to obtain observer parameters; Inputting the observer parameters and the standardized response data into the unknown input observer, generating a state estimation value and an output estimation value, and performing a difference between the actual output and the estimated output to obtain a residual sequence; Performing adaptive threshold calculation on the residual sequence, substituting the mean and variance of the residual in the sliding window into the statistical discriminant function to obtain a dynamic threshold range; The residual sequence is evaluated in segments according to the dynamic threshold range, and the data points that continuously exceed the threshold range are marked as abnormal intervals to obtain abnormal data segments. The abnormal data segments are then type-identified and classified, and the type and degree of the abnormality are determined by combining judgment rules to obtain abnormal response data.

5. The gas detection method based on the MEMS gas sensor array according to claim 4, characterized in that: The step of selecting a time window length according to the abnormal response data, and performing piecewise linear regression and dynamic calibration on the abnormal data segments to obtain a target detection data set includes: Performing time series statistics on the abnormal response data, calculating the interval distribution and duration between abnormal points, setting a time window according to the length of a normal response cycle, and obtaining a reference time window length; Segmenting the abnormal response data according to the reference time window length, merging consecutive abnormal points into abnormal data segments, marking the start and end positions of each abnormal data segment, and obtaining an abnormal data segment sequence; Perform boundary expansion on each data segment of the abnormal data segment sequence, add normal data of a time window length before and after the abnormal data segment, respectively, to obtain an extended data segment, perform data normalization processing on the extended data segment, calculate the data mean and variance, remove data points that deviate from the mean by more than three times the variance, and obtain a target data segment; Performing linear regression analysis on the target data segment, calculating the regression coefficient segment by segment using a sliding window, performing linear fitting on the abnormal data to obtain regression correction data, dynamically assigning weights to the regression correction data, setting correction coefficients based on data credibility, and performing weighted averaging on the corrected data to obtain weighted calibration data; The weighted calibration data is compared and analyzed with the normal response data, the correlation coefficient and the mean square error are calculated, the calibration result is verified, the verified data is obtained, and the time series of the verified data is reconstructed, the calibrated abnormal segment data replaces the original abnormal data, and the data is rearranged in chronological order to obtain the target detection data set.

6. The gas detection method based on the MEMS gas sensor array according to claim 5, characterized in that: The step of inputting the target detection data set into the sensor network model to perform array deployment optimization and obtain an array deployment optimization solution includes: The target detection data set is input into the sensor network model for spatial distribution analysis, and the gas diffusion parameters are obtained according to the gas concentration gradient and diffusion rate in the detection area. A gas diffusion model is constructed based on the gas diffusion parameters, the detection area is gridded, and the gas concentration distribution of each grid point is calculated to obtain the concentration distribution field; Performing contour analysis on the concentration distribution field, calculating the diffusion radius at the maximum concentration gradient, setting the diffusion radius as the coverage range of the detection unit, obtaining the optimal detection radius, and performing Delaunay triangulation on the detection area based on the optimal detection radius, connecting adjacent sensor nodes to construct a triangular network, and obtaining an initial deployment network; Optimizing nodes of the initial deployment network, calculating coverage overlap and connectivity of each sensor node, deleting redundant nodes, obtaining an optimized deployment plan, and calculating spatial correlation coefficients between sensor nodes based on the optimized deployment plan, adjusting the positions of nodes with spatial correlation coefficients lower than a threshold, and obtaining a spatial correlation matrix; The spatial association matrix is ​​input into the network communication model, the data transmission path and communication load between nodes are calculated, the communication topology is obtained, and the reliability of the communication topology is verified. The network connectivity and data transmission delay are calculated to determine the array deployment optimization plan.

7. A gas detection device based on a MEMS gas sensor array, characterized in that: A gas detection method based on a MEMS gas sensor array according to any one of claims 1 to 6, comprising: An acquisition module is used to group multiple MEMS gas sensors according to target gas types and collect detection data to obtain an original detection data set; A construction module is used to perform curve fitting and feature extraction on the original detection data set to obtain a feature parameter set, and substitute the feature parameter set into a second-order state space model to construct a sensor array feature model; A correction module, used to calculate a temperature correction coefficient and a humidity correction coefficient using the sensor array characteristic model, and to correct the original detection data set to obtain standardized response data; A calculation module, used for extracting a plurality of response feature quantities from the standardized response data, and inputting the plurality of response feature quantities into an unknown input observer for residual calculation and signal abnormal point identification to obtain abnormal response data; A calibration module, used to select a time window length according to the abnormal response data, and perform piecewise linear regression and dynamic calibration on the abnormal data segment to obtain a target detection data set; The optimization module is used to input the target detection data set into the sensor network model to perform array deployment optimization and obtain an array deployment optimization solution.

8. A gas detection device based on a MEMS gas sensor array, characterized in that: The gas detection device based on the MEMS gas sensor array includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the gas detection device based on the MEMS gas sensor array to execute the gas detection method based on the MEMS gas sensor array according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the gas detection method based on the MEMS gas sensor array as described in any one of claims 1 to 6 is implemented.

Images