阅读说明：本技术 一种mems薄膜半导体气敏传感器阵列识别方法 (MEMS film semiconductor gas sensor array identification method ) 是由 余岑 李如意 李东风 黄志华 刘红恩 于 2020-12-02 设计创作，主要内容包括：本发明公开了一种MEMS薄膜半导体气敏传感器阵列识别方法,包括,已知物质建模：使用传感器阵列对第一类已知物质进行采样,在所述第一类已知物质的挥发气体浓度逐步上升的过程中时刻,得到n个时刻的电导率变化向量集合；对所述电导率变化向量集合进行标准化得到标准化电导率变化向量集合,对所述标准化电导率变化向量集合进行主元分析,获取第一主成分向量；对所述k类已知物质第一主成分向量进行线性判别式分析,计算得到特征变换矩阵,获得样本类,存入特征数据库中；未知物质检测：重复上述步骤,得到投影到特征空间的一个点t～d,计算t～d与所述特征数据库中的样本类均值的距离是否小于已有样本聚类的分布直径,判断更接近哪一类气体。(The invention discloses a method for identifying an MEMS film semiconductor gas sensor array, which comprises the following steps of modeling by known substances: sampling a first type of known substances by using a sensor array, and obtaining a conductivity change vector set at n moments in the process that the concentration of volatile gas of the first type of known substances gradually rises; normalizing the conductivity change vector set to obtain a normalized conductivity change vector set, and performing principal component analysis on the normalized conductivity change vector set to obtain a first principal component vector; vector feeding the first principal component of the k-type known substancesPerforming line linear discriminant analysis, calculating to obtain a feature transformation matrix, obtaining a sample class, and storing the sample class in a feature database; detection of unknown substances: repeating the steps to obtain a point t projected to the characteristic space d Calculating t d And judging which type of gas is closer to if the distance between the sample type mean value in the characteristic database is smaller than the distribution diameter of the existing sample cluster.)

1. A MEMS film semiconductor gas sensor array identification method is characterized by comprising the following steps,

s1: modeling of known substances: sampling a first type of known substances by using a sensor array, putting the sensor array into a gas-sensitive detection closed box, and gradually injecting volatile gas of the first type of known substances into the gas-sensitive detection closed box to ensure that the concentration of the volatile gas of the first type of known substances in the gas-sensitive detection closed box gradually rises;

s2: t during the gradual increase of the concentration of the volatile gas of the first known substance in step S1_iObtaining conductivity change vectors combined by m sensors at moments, and repeating the steps to obtain a conductivity change vector set at n moments;

s3: normalizing the conductivity change vector set to obtain a normalized conductivity change vector set, performing principal component analysis on the normalized conductivity change vector set to obtain a first principal component vector as data sampled by a single type of known substances, and acquiring the first type of known substances for x times to obtain the first principal component set;

s4: repeating the steps S1, S2 and S3 on the k types of known substances to obtain first principal component vectors of the k types of known substances, carrying out linear discriminant analysis on the first principal component vectors of the k types of known substances, and calculating to obtain a feature transformation matrix W_otpObtaining a sample class and storing the sample class in a characteristic database;

s5: detection of unknown substances: repeating the steps S1, S2 and S3 to obtain a conductivity change vector test set of the unknown substance at n moments, standardizing the conductivity change vector test set, and obtaining a first principal component test vector;

s6: passing the first principal component test vector by W_otpTransforming to feature space, projecting to a point t of the feature space^d；

S7: will t^dComparing with sample classes in the feature database: calculating t^dJudging which type of gas is closer to whether the distance between the sample type mean value in the characteristic database is smaller than the distribution diameter of the existing sample cluster;

s8: and if the first principal component test vector belongs to the category needing alarming in the feature database and exceeds an alarming threshold value, an alarm is given out.

2. The MEMS thin film semiconductor gas sensor array identification method of claim 1, wherein the acquiring is x times, x > 5.

3. The MEMS thin-film semiconductor gas sensor array identification method of claim 1, wherein the k-type known substances are classified into low-risk substances and high-risk substances.

4. The MEMS thin-film semiconductor gas sensor array identification method of claim 3, wherein the low-risk substance comprises alcohol, perfume, toilet water, orange peel, and shaddock peel.

5. The MEMS thin-film semiconductor gas sensor array identification method as claimed in claim 3, wherein the high-risk substances comprise gasoline, rosin water and banana oil.

6. The MEMS thin-film semiconductor gas sensor array identification method of claim 1, wherein one point t of the feature space^dThe corresponding vector dimension is the feature space dimension.

7. The MEMS thin film semiconductor gas sensor array identification method of claim 1, wherein the linear discriminant analysis comprises setting a feature library dimension j, wherein the feature library dimension j < m.

8. The MEMS thin film semiconductor gas sensor array identification method of claim 1, wherein the conductivity of the sensor is changed to,where ρ isⁱIs t_iThe conductivity of the sensor at the moment.

9. The MEMS thin-film semiconductor gas sensor array identification method as claimed in claim 1, wherein the eigenvectors corresponding to the first few eigenvalues with the cumulative contribution rate of the principal component analysis reaching 85% are principal components.

10. The MEMS thin film semiconductor gas sensor array identification method of claim 1, wherein the normalization is,wherein

Technical Field

The invention relates to the technical field of combustible substance detection, in particular to an MEMS (micro-electromechanical systems) film semiconductor gas sensor array identification method.

Background

The dangerousness of different kinds of flammable liquids varies, for example, gasoline burns rapidly and deflagration phenomena may occur, while alcohol of the same volume burns relatively slowly and dangerously and alcohol, as a main component in perfume, floral water, may cause some false alarm phenomena. Aromatic hydrocarbon substances contained in orange peel and shaddock peel also belong to flammable liquid, but the content of the aromatic hydrocarbon substances is low, and unnecessary alarm can be caused in a real scene.

The patent publication No. CN108107086A discloses a gas detection method based on an array gas sensor and the gas sensor, and belongs to the field of gas identification. The method comprises the following steps: collecting the resistance value of the array gas sensor in real time; calculating the resistance value change rate of the target sensor, and calculating the resistance value change rate of the gas sensor once every time a new resistance value is acquired; judging whether M of N continuous resistance value change rates of at least one gas sensor in the target sensor is greater than a resistance value change rate threshold value; if so, judging whether the sensitivity of most of the gas sensors to all the target gases is greater than a sensitivity threshold value; if so, taking the time point of the last resistance value acquisition for calculating the first resistance value change rate in the N continuous resistance value change rates as a gas sampling point.

The invention of publication number CN107478687A discloses a multi-component gas sensor, which comprises an electric control unit, wherein the electric control unit is connected with a heating unit, a Pt10 heater and a metal oxide sensitive element, the heating unit is connected with the Pt10 heater, the metal oxide sensitive element is laid on a ceramic heating substrate, the ceramic heating substrate is attached to the Pt10 heater, and the number of the metal sensitive elements is multiple. In this structure, Pt10 is used as the main component of the heater, which is used as a thermistor to feed back the heating temperature value in real time through the resistance value, and then various components of the gas to be detected are detected by the metal oxide sensitive element.

In order to distinguish the danger degree of inflammable substances, characteristic data with high burning danger such as gasoline, rosin water, banana oil and the like needs to be collected, characteristic data with relatively low danger such as alcohol, orange peel, shaddock peel and the like needs to be collected, and the characteristic data are labeled according to the substance types to construct a characteristic database. When alarm monitoring is carried out, the collected data is compared with entries in a feature database after features are extracted, and whether the collected data belongs to a known type in the database or a type which is not included in the database is judged.

Disclosure of Invention

The embodiment of the invention provides an MEMS film semiconductor gas sensor array identification method. The following presents a simplified summary in order to provide a basic understanding of some aspects of the disclosed embodiments. This summary is not an extensive overview and is intended to neither identify key/critical elements nor delineate the scope of such embodiments. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is presented later.

According to an embodiment of the present invention, there is provided a MEMS thin film semiconductor gas sensor array recognition method, including,

s1: modeling of known substances: sampling a first type of known substances by using a sensor array, putting the sensor array into a gas-sensitive detection closed box, and gradually injecting volatile gas of the first type of known substances into the gas-sensitive detection closed box to ensure that the concentration of the volatile gas of the first type of known substances in the gas-sensitive detection closed box gradually rises;

s2: t during the gradual increase of the concentration of the volatile gas of the first known substance in step S1_iObtaining conductivity change vectors combined by m sensors at moments, and repeating the steps to obtain a conductivity change vector set at n moments;

s3: normalizing the conductivity change vector set to obtain a normalized conductivity change vector set, performing principal component analysis on the normalized conductivity change vector set to obtain a first principal component vector as data sampled by a single type of known substances, and acquiring the first type of known substances for x times to obtain the first principal component set;

s4: repeating the steps S1, S2 and S3 on the k types of known substances to obtain first principal component vectors of the k types of known substances, carrying out linear discriminant analysis on the first principal component vectors of the k types of known substances, and calculating to obtain a feature transformation matrix W_otpObtaining a sample class and storing the sample class in a characteristic database;

s5: detection of unknown substances: repeating the steps S1, S2 and S3 to obtain a conductivity change vector test set of the unknown substance at n moments, standardizing the conductivity change vector test set, and obtaining a first principal component test vector;

s6: passing the first principal component test vector by W_otpTransforming to feature space, projecting to a point t of the feature space^d；

S7: will t^dComparing with sample classes in the feature database: calculating t^dJudging which type of gas is closer to whether the distance between the sample type mean value in the characteristic database is smaller than the distribution diameter of the existing sample cluster;

s8: and if the first principal component test vector belongs to the category needing alarming in the feature database and exceeds an alarming threshold value, an alarm is given out.

Preferably, the acquisition is x times, x > 5.

Preferably, the k-class known substances are classified into low-risk substances and high-risk substances.

Preferably, the substance with low risk includes alcohol, perfume, toilet water, orange peel, and shaddock peel.

Preferably, the substances with high risk comprise gasoline, rosin water and banana oil.

Preferably, a point t of said eigenspace^dThe corresponding vector dimension is the feature space dimension.

Preferably, the linear discriminant analysis includes setting a feature library dimension j, the feature library dimension j < m.

Preferably, the conductivity of the sensor is varied by,where ρ isⁱIs t_iThe conductivity of the sensor at the moment.

Preferably, the feature vector corresponding to the first several feature values with the cumulative contribution rate of the principal component analysis reaching 85% is a principal component.

Preferably, the normalization is performed as,wherein

The technical scheme provided by the embodiment of the invention has the following beneficial effects:

according to the MEMS film semiconductor gas sensor array identification method provided by the invention, various flammable and explosive gas and liquid characteristic libraries can be constructed by applying the algorithm; the algorithm can identify the component content of the target to be detected, and compares the component content with various flammable and explosive gas liquid characteristic libraries, so that the objects with different danger degrees can be distinguished, the grading early warning accuracy is realized to the maximum extent, and the false alarm, the false alarm and the missing alarm are avoided. For example, gasoline, alcohol and orange peel can cause the gas sensor to give an alarm in life, but the gasoline and the alcohol contain more alcohols and alkanes, the orange peel contains more hydrocarbons, if the algorithm is adopted, the judgment result is that the gasoline and the alcohol are required to be enhanced and the alarm is highlighted, and if the orange peel is extruded and gives an alarm, the alarm can be weakened or not given.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

FIG. 1 is a flow chart illustrating a MEMS thin film semiconductor gas sensor array identification method according to one exemplary embodiment;

FIG. 2 is a diagram illustrating LDA projection clustering, according to an exemplary embodiment;

Detailed Description

The following description and the drawings sufficiently illustrate specific embodiments of the invention to enable those skilled in the art to practice them. The examples merely typify possible variations. Individual components and functions are optional unless explicitly required, and the sequence of operations may vary. Portions and features of some embodiments may be included in or substituted for those of others. The scope of embodiments of the invention encompasses the full ambit of the claims, as well as all available equivalents of the claims. Embodiments may be referred to herein, individually or collectively, by the term "invention" merely for convenience and without intending to voluntarily limit the scope of this application to any single invention or inventive concept if more than one is in fact disclosed. Herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed. The embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the structures, products and the like disclosed by the embodiments, the description is relatively simple because the structures, the products and the like correspond to the parts disclosed by the embodiments, and the relevant parts can be just described by referring to the method part.

It should be noted that: the relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

It should be noted that although the various steps of the methods of the present disclosure are depicted in the drawings in a particular order, this does not require or imply that these steps must be performed in this particular order, or that all of the depicted steps must be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step execution, and/or one step broken down into multiple step executions, etc.

The invention is further described with reference to the following figures and examples:

in MEMS thin film semiconductor gas sensor array identification, most sensors generate conductivity change along with the change of target gas concentration, and the conductivity change presents an approximate linear relation. Due to the different metal oxides and modification materials used in MEMS sensor arrays, the response capabilities of individual sensors to the same target gas are not the same. Assuming that the gas at each sensor in the sensor array is uniformly distributed, a sensor conductivity change vector set a ═ a in the gas concentration change process can be obtained by sampling at n times¹,A²,…AⁿTherein ofIs t_iM sensor conductivity change vectors at a time.

A MEMS thin film semiconductor gas sensor array recognition method as shown in fig. 1, includes,

s1: modeling of known substances: the method comprises the steps of sampling a first type of known substances by using a sensor array, putting the sensor array into a gas-sensitive detection closed box, and gradually injecting volatile gas of the first type of known substances into the gas-sensitive detection closed box, so that the concentration of the volatile gas of the first type of known substances in the gas-sensitive detection closed box is gradually increased.

S2: t during the gradual increase of the concentration of the volatile gas of the first known substance in step S1_iObtaining the conductivity change vector formed by combining m sensors,repeating the steps to obtain a conductivity change vector set A ═ A at n moments¹,A²,…,Aⁿ}; the change in the electrical conductivity of the sensor is,where ρ isⁱIs t_iThe conductivity of the sensor at the moment.

S3: normalizing the conductivity change vector set to obtain a normalized conductivity change vector set B ═ { B ═ B¹,B²,…,Bⁿ-the normalization is carried out in the order of,

wherein

B＝{B¹,B²,…BⁿThe sensor obtained for n sampling points has a set of vectors responding to the change of the concentration of the target gas, and ideally, the vectors in the set are nearly equal, but factors such as interfering gas, measurement error and the degree of nonlinearity of the sensor in real detection can cause { B }¹,B²,…BⁿThere is a certain deviation between them. In MEMS film semiconductor gas sensor array recognition, it is difficult to realize that each gas sensor only responds to specific gas through an adsorption mechanism of organic molecules and semiconductor film materials, response characteristics of semiconductor films made of different materials and processes are different to a certain extent, a response mode of current main gas components can be obtained by using a PCA method, and the response mode is compared with a pre-collected characteristic database to obtain possible categories of the main gas components. For the MEMS film semiconductor gas sensor array recognition algorithm, the normalization processing result B is { B ═ B by PCA method¹,B²,…BⁿAnalyzing to obtain one or a few of the samples with higher contribution rateAnd the characteristic vector corresponding to each principal component is compared with the characteristic vector of each target gas in the database, so that one or a few target gas types which have larger influence on the current measurement are determined.

Performing principal component analysis on the standardized conductivity change vector set to obtain a first principal component vectorAs one data of a single type of known substance, x times are acquired for a first type of known substance, said x times>5, obtaining the first principal component vector setThe pivot analysis calculation process is as follows,

step 1. calculating correlation matrix

Calculating B ═ B¹,B²,…BⁿThe correlation matrix of } is:

wherein r is_ijIs BⁱAnd B^jThe correlation coefficient of (a) is calculated,

step 2. calculating principal Components

Calculating a correlation matrix R_n×nCharacteristic value λ of_iAnd feature vectorThe characteristic values are arranged from large to small, and the contribution rate of the characteristic values is calculated

And taking the eigenvectors corresponding to the first eigenvalues with the cumulative contribution rate of 85% as principal components.

Step 3, explain the principal ingredients

Feature vector of principal componentAnd performing clustering comparison with known target gases in the characteristic database as input to determine the main component as the corresponding target gas in the characteristic database. If no vector with larger correlation coefficient can be found in the feature database, the gas is indicated as the gas which is not recorded in the feature database.

Step 4, calculating the score of the principal component

Will { B¹,B²,…BⁿSubstituting the obtained value into a principal component formula formed by principal component feature vectors,

is t_iThe principal component k at a time is scored,

the law of the change of the principal component k with time, namely the target gas concentration change curve corresponding to the principal component, is described. And (4) comparing the collected and marked records in the special library to determine the substances corresponding to the main components. The goal of PCA is to keep the most information in the reduced dimension data, but it does not mean that the data can be easily distinguished after the dimension reduction, and feature database construction and type discrimination need to be completed by LDA (linear discriminant analysis) method based on PCA method.

LDA focuses more on the distribution state of samples in space and the distance analysis among the samples, and the basic idea is to project a high-dimensional pattern sample to an optimal identification vector space so as to achieve the effects of extracting classification information and compressing the dimension of a feature space, and after projection, the pattern sample is ensured to have the maximum inter-class distance and the minimum intra-class distance in a new subspace, namely, the pattern has the optimal separability in the space. Therefore, it is an effective feature extraction method. Using this approach, the inter-class scatter matrix of the post-projection mode pattern can be maximized, while the intra-class scatter matrix is minimized. That is, it can ensure that the pattern samples after projection have the minimum intra-class distance and the maximum inter-class distance in the new space, i.e., the pattern has the best separability in the space.

As shown in fig. 2, the principle of LDA is that labeled data (points) are projected into a space with lower dimensions by a projection method, so that the projected points are classified into different categories, that is, points of the same category are closer in the projected space, and points of different categories are farther in the projected space. The goal of LDA analysis is to project a data point onto one or more straight lines, where a is a sample vector, w is a feature vector or feature vector matrix corresponding to the straight line, and y is the projected sample point.

S4: repeating the steps S1, S2 and S3 on the k types of known substances to obtain a first principal component vector of the k types of known substances, carrying out linear discriminant analysis on the first principal component vector of the k types of known substances, and setting a feature library dimension j, wherein the feature library dimension j is<m, calculating to obtain a characteristic transformation matrix W_optObtaining a sample class and storing the sample class in a characteristic database; the k-type known substances are divided into low-risk substances and high-risk substances, the low-risk substances comprise alcohol, perfume, floral water, orange peel and shaddock peel, and the high-risk substances comprise gasoline, rosin water and banana oil; the specific calculation steps of the LDA analysis are as follows,

step 1, data standardization processing

Using m sensors p-type known substances to carry out multiple sampling, and carrying out standardization processing on sampling data, wherein the raw data a ═ a₁,a₂,…,a_m]The value representing single sampling by using m sensors is obtained after normalization processingWherein

The sample is expressed as the jth sample in the ith class after normalization.

Assuming that C-class samples are shared, the number of each class of samples is M_i，Is the total number of training samples.

Step 2, for each category, calculating m-dimensional mean vector

Representing the mean vector of the training samples of the ith class.

Representing the mean vector of all samples.

Step 3. construct interspecies scatter matrix S_bAnd intra-class scatter matrix S_w

Inter-class scatter matrixThe sum of covariance matrixes of the samples and the population is calculated by each sample according to the class to which the sample belongs, and the discrete redundancy degree between all the classes and the population is described.

Intra-class scatter matrixIs the sum of covariance matrixes between each sample in the class and the class to which the sample belongs, and describes the dispersion between each sample and the class in the class from the whole.

For better discrimination between species, it is desirable that the degree of coupling between species be low and the degree of intra-species polymerization be high, i.e., S_bMiddle value is larger, S_wThe numerical value in (1) is small, so that the classification effect is good.

Step 4, calculating the eigenvalue of the matrix and the corresponding eigenvector

Introducing Fisher discrimination criterion expression:

whereinIs any m-dimensional column vector.

The Fisher linear discriminant analysis is selected so thatVector to maximumThe physical meaning of the projection direction is that the projected sample has the largest inter-class dispersion and the smallest intra-class dispersion.

Substituting the result in the step 3 into the Fisher identification criterion expression:

hypothesis matrixIs represented by (u)_i-u) is inThe square of the geometric distance in space, so the molecules of Fisher linear identification form are the projection space of the sampleThe sum of squares of the geometric distances between classes can be deduced by the same principle that the denominator is the sample in the projection spaceAnd the classification problem is converted into a low-dimensional space, so that after the sample is projected to the space, the ratio of the sum of squares of the distances between the classes to the sum of squares of the distances in the classes is maximum, namely the optimal classification effect.

To find a projection matrix W formed by an optimal set of discrimination vectors_opt

Wherein the column vector w_iAs a generalized characteristic equationD largest eigenvalues of (a) are associated with the eigenvector (matrix S)_w ^-1S_bThe feature vector of (d) and the number of optimal projection axes d is less than or equal to c-1.

Step 5, using the conversion matrix W_optMapping samples to be determined onto a new feature subspace

After obtaining a new sample to be determined, the matrix W is transformed_optIt is mapped to a feature subspace and then compared to the clusters marked in the database to determine to which class the sample belongs.

S5: detection of unknown substances: repeating the steps S1, S2 and S3 to obtain a test set of conductivity change vectors of unknown substances at n momentsNormalized conductivity change vector test setObtaining a first principal component test vector

S6: testing the first principal component to obtain a vectorThrough W_optTransforming to feature space, projecting to a point t of the feature space^dA point t of said eigenspace^dThe corresponding vector dimension is a feature space dimension;

s7: will t^dComparing with sample classes in the feature database: calculating t^dJudging which type of gas is closer to whether the distance between the sample type mean value in the characteristic database is smaller than the distribution diameter of the existing sample cluster;

s8: if the first principal component tests the vectorThe alarm belongs to the category needing alarm in the characteristic database, and the alarm is sent out when the alarm threshold value is exceeded.

It is to be understood that the present invention is not limited to the procedures and structures described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.