Regional allergy risk real-time early warning method based on environment micromolecule fingerprint spectrum

Abstract

The invention relates to a regional allergy risk real-time early warning method and device based on an environment micromolecule fingerprint spectrum, and the method comprises the steps: carrying out the real-time sampling of a microvolatile organic compound related to a predefined allergen in the air of a target region, and obtaining an original multi-dimensional characteristic spectrum; extracting a chemical fingerprint spectrum of each known allergen in the original multi-dimensional characteristic spectrum through a VOCs fingerprint spectrum model; dynamically matching the chemical fingerprint spectrum of each known allergen with a preset allergen chemical fingerprint spectrum database by using a Siamese network model, and calculating a comprehensive regional allergy risk factor; and according to the regional allergy risk factor and a preset multi-level allergy risk threshold, determining an allergy risk level of the current target region. According to the method, the purification signal characteristics corresponding to each known allergen are extracted through the VOCs fingerprint spectrum model, and the similarity, relevance and chemical affinity among different allergens are obtained through the Siamese network model, so that the method is closer to the real situation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental trace molecule monitoring, and in particular to a regional allergic risk real-time early warning method and device based on environmental trace molecule fingerprint and computer equipment. BACKGROUND

[0002] With the changes in the environment, industrialization process and changes in lifestyle, the types and concentrations of allergens are increasing, and allergic reactions affect the quality of life of patients. Common allergens include pollen, mold spores, dust mites, pet dander, and some plant volatiles.

[0003] Traditional allergen monitoring methods are mostly limited to pollen allergens and cannot cover other types of allergens (such as mold, mite components, and plant volatiles). They are time-consuming and labor-intensive, have poor real-time performance, and the results are easily limited by sampling points and lack regional representation. Electronic noses have been used in food quality control, environmental pollution monitoring, and disease diagnosis, but are mostly used to identify a single or a few target odors. It is difficult to accurately extract the chemical fingerprints of each allergen for the weak and interfering mixtures produced by multiple trace allergens in the environment. In particular, allergens are usually biological macromolecules that do not directly volatilize. In addition, existing electronic nose systems cannot calculate or infer the synergistic effects between multiple allergens, the modulation effects of environmental factors, and the spatiotemporal dynamics.

[0004] To solve the above problems, the present application discloses a regional allergic risk real-time early warning method based on environmental trace molecule fingerprint, which extracts the purified signal characteristics corresponding to each known allergen through a VOCs fingerprint model, and obtains the similarity, correlation and chemical affinity between different allergens through a Siamese network model, so as to make the comprehensive regional allergic risk factors more close to the real situation. SUMMARY

[0005] In view of the above problems, the present application provides a regional allergic risk real-time early warning method and device based on environmental trace molecule fingerprint and computer equipment.

[0006] According to one aspect of the present application, a regional allergic risk real-time early warning method based on environmental trace molecule fingerprint is provided, comprising:

[0007] A regional allergic risk real-time early warning method based on environmental trace molecule fingerprint, characterized in that it comprises:

[0008] The biomimetic olfactory electronic nose including a multi-array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor and / or a surface acoustic wave sensor is used to sample in real time the trace volatile organic compounds related to predefined allergens present in the air of a target area, to obtain the amplitude, response time, recovery time, impedance change and frequency drift of the sensing response curve to construct an original multi-dimensional feature spectrum; wherein the predefined allergens include pollen, mold spores, plant volatiles and mite components;

[0009] The chemical fingerprint spectrum of each known allergen in the original multi-dimensional feature spectrum is extracted by a VOCs fingerprint spectrum model;

[0010] The chemical fingerprint spectrum of each known allergen is dynamically matched with a preset allergen chemical fingerprint spectrum library using a Siamese network model, and a comprehensive regional allergic risk factor is calculated;

[0011] According to the regional allergic risk factor and the preset multi-level allergic risk threshold, the allergic risk level of the current target area is determined.

[0012] In an alternative way, the chemical fingerprint spectrum of each known allergen in the original multi-dimensional feature spectrum is further extracted by a VOCs fingerprint spectrum model, which comprises:

[0013] The original multi-dimensional feature spectrum is input into the VOCs fingerprint spectrum model for blind source separation to extract an independent component vector corresponding to each known allergen; wherein the VOCs fingerprint spectrum model is trained in advance based on standard VOCs samples of known allergens, and each independent component vector is a purified signal feature of a known allergen;

[0014] The time domain statistical features and frequency domain transform features are obtained by performing multi-domain feature extraction on each independent component vector; and the chemical fingerprint spectrum of each known allergen is constructed according to the time domain statistical features and frequency domain transform features.

[0015] In an alternative way, the VOCs fingerprint spectrum model comprises:

[0016] The variational independent component analysis network layer includes a multi-head self-attention layer and a reversible neural network layer, which is used for probabilistic blind source separation to extract an independent component vector corresponding to each known allergen;

[0017] The time domain statistical feature extraction layer includes a time domain feature branch layer, a time-frequency analysis branch and a non-stationary signal processing branch layer, which is used to extract mean, variance, kurtosis, skewness, zero-crossing rate, autocorrelation coefficient and empirical mode decomposition intrinsic mode function from each independent component vector;

[0018] The spectrum analysis feature extraction layer includes a bottleneck structure layer and a differentiable L2 normalization layer, and is used for adaptively decomposing a time domain signal into an optimal wavelet packet basis and generating a time-frequency energy distribution spectrum.

[0019] In an optional mode, the Siamese network model comprises:

[0020] The multi-scale feature pyramid encoding layer comprises a dilated convolution path layer, an adaptive pooling path layer and a residual connection path layer.

[0021] The cross-attention matching layer comprises a query generator, a key-value memory construction layer and a multi-head cross-attention layer, wherein the query generator generates a query vector by linear projection from the to-be-tested chemical fingerprint, and the key-value memory construction layer dynamically constructs a key vector and a value vector from a preset allergen chemical fingerprint library.

[0022] The dynamic relationship reasoning layer comprises a node initialization layer, an edge weight calculation layer, a graph convolution network layer and a readout function.

[0023] The multi-task metric learning head comprises a contrastive learning branch layer, a classification identification branch layer and a gradient balancing layer.

[0024] The risk factor regressor comprises an attention pooling layer, an uncertainty calibration layer and a probability output layer.

[0025] In an optional mode, the graph convolution network layer in the dynamic relationship reasoning layer adopts a multi-hop propagation algorithm.

[0026] The node feature update formula of the graph convolution network layer is as follows:

[0027]

[0028] wherein, , are the feature vectors of the target node i and its neighbor node j in the i-th layer graph convolution, respectively; is the set of all neighbor nodes of the node i; is the importance coefficient of the neighbor node j to the target node i; is the i-th layer learnable weight matrix; is the i-th layer learnable self-connection weight; is a LeakyReLU activation function; The attention coefficient of the graph convolution network layer is as follows:

[0029]

[0030]

[0031] ​​​wherein, , , are learnable balancing factors respectively for adjusting the contribution proportion of the similarity factor, the spatial correlation factor and the chemical affinity factor in the final attention weight; is the similarity factor of the feature vector , ; is the correlation factor of the feature vector , ; is the chemical affinity factor of the feature vector , .

[0032] In an optional manner, the similarity factor is:

[0033]

[0034] wherein, is the cosine similarity of the feature vector , ; is the Gaussian kernel similarity based on the feature space distance ; is the weight; is the control decay rate;

[0035] The correlation factor is:

[0036]

[0037] wherein, is the overall weight of the factor; are the direction angles of the nodes and in the sensor space layout; , are the timestamps of the nodes and being detected; is the time decay constant;

[0038] The chemical affinity factor is:

[0039]

[0040] wherein, are the concentrations of the first chemical component of the allergen and ; is the contribution weight of the first chemical component to the affinity; ​​for concentration scaling parameters.

[0041] In an alternative way, the formula for the integrated regional allergic risk factor is:

[0042]

[0043] wherein, , , are used to balance the contribution of spatiotemporal correlation component synergistic effect component and interaction terms in the total risk, respectively; is the baseline risk component; is the environmental modulation factor;

[0044]

[0045]

[0046]

[0047]

[0048] wherein, is the number of known allergens identified; is the risk weight of the i-th allergen; is the chemical fingerprint feature vector of the i-th allergen detected at present; is the reference standard feature vector of the i-th allergen in the preset library; is the standard deviation of the i-th allergen feature; is the sigmoid function; is the time integration window; is the time decay kernel function; is the time variable; is the gradient of the baseline risk component at time ; is the activation function; is the synergistic effect coefficient between allergens i and j; are the baseline risk components of allergens i and j, respectively; is the distance between allergens i and j in the feature space; is the synergistic effect scale parameter; is the number of environmental parameter categories; is the actual measured value of the k-th environmental parameter at time t; is the optimal value of the k-th environmental parameter; is the tolerance range of the k-th environmental parameter; is a modulation intensity coefficient of the kth environmental parameter; is a Gaussian kernel function.

[0049] In an optional manner, the preset multi-level allergic risk threshold is dynamically adjusted according to historical allergic event data and meteorological parameters;

[0050] wherein the occurrence rate, severity and associated meteorological parameters of historical allergic events are taken as training data and input into a time series model; the meteorological parameters include temperature, humidity, wind speed, rainfall and atmospheric pressure; and the time series model outputs a corresponding allergic risk probability distribution under different meteorological parameters;

[0051] Upon receiving current meteorological forecast data, the dynamic threshold of each level of allergic risk in a future period of time is predicted in real time according to the allergic risk probability distribution, as the preset multi-level allergic risk threshold.

[0052] According to another aspect of the present application, there is provided a device for real-time early warning of regional allergic risk based on environmental trace molecular fingerprint, comprising:

[0053] an environmental sampling and feature spectrum construction module for a biomimetic olfactory electronic nose of an array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor and / or a surface acoustic wave sensor to perform real-time sampling on trace volatile organic compounds related to predefined allergens in the air of a target region, and to obtain amplitudes, response times, recovery times, impedance changes and frequency drifts of sensing response curves to construct original multi-dimensional feature spectra; wherein the predefined allergens include pollen, mold spores, plant volatiles and mite components;

[0054] a chemical fingerprint extraction module for extracting chemical fingerprints of each known allergen in the original multi-dimensional feature spectra through a VOCs fingerprint model;

[0055] a dynamic matching and risk factor calculation module for dynamically matching the chemical fingerprints of each known allergen with a preset allergen chemical fingerprint library by using a Siamese network model, and calculating a comprehensive regional allergic risk factor;

[0056] a risk level determination and early warning module for determining the allergic risk level of a current target region according to the regional allergic risk factor and a preset multi-level allergic risk threshold.

[0057] According to still another aspect of the present application, there is provided a computer device, comprising a processor, a memory, a communication interface and a communication bus, the processor, the memory and the communication interface being in communication with each other through the communication bus;

[0058] The memory is used to store at least one executable instruction, and the executable instruction causes the processor to execute the operation corresponding to the above-mentioned real-time early warning method of regional allergic risk based on environmental trace molecular fingerprint.

[0059] According to the scheme provided by the application, the biomimetic olfactory electronic nose including a multi-array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor and / or a surface acoustic wave sensor is used to perform real-time sampling on trace volatile organic compounds related to predefined allergens existing in the air of a target region, to obtain the amplitude, response time, recovery time, impedance change and frequency drift of a sensing response curve to construct an original multi-dimensional feature spectrum; wherein the predefined allergens include pollen, mold spores, plant volatiles and mite components; a VOCs fingerprint model is used to extract the chemical fingerprint of each known allergen in the original multi-dimensional feature spectrum; a Siamese network model is used to dynamically match the chemical fingerprint of each known allergen with a preset allergen chemical fingerprint library, to calculate a comprehensive regional allergic risk factor; and according to the regional allergic risk factor and a preset multi-level allergic risk threshold, the allergic risk level of the current target region is determined. The application extracts the purified signal characteristics corresponding to each known allergen from the complex mixed VOCs signal through the VOCs fingerprint model, solves the problem that the multiple-source trace VOCs signals interfere with each other and are difficult to distinguish, and obtains the similarity, correlation and chemical affinity between different allergens through the Siamese network model, so that the comprehensive regional allergic risk factor is closer to the real situation.

[0060] The above description is only a summary of the technical scheme of the application, in order to enable the technical means of the application to be more clearly understood, and to be implemented according to the content of the specification, and in order to enable the above and other purposes, features and advantages of the application to be more obvious and easy to understand, the following specific embodiments of the application are described. BRIEF DESCRIPTION OF DRAWINGS

[0061] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the present application. Furthermore, the same reference numerals are used throughout the several drawings to represent similar components. In the drawings:

[0062] Figure 1 A flowchart of a real-time early warning method of regional allergic risk based on environmental trace molecular fingerprint of an embodiment of the application is shown;

[0063] Figure 2 A framework diagram of a real-time early warning device of regional allergic risk based on environmental trace molecular fingerprint of an embodiment of the application is shown;

[0064] Figure 3 FIG. 1 shows a structural schematic diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0065] Exemplary embodiments of the present application will be described in detail with reference to the drawings. Although exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.

[0066] Figure 1 FIG. 1 shows a structural schematic diagram of a computer device according to an embodiment of the present application. Figure 1 As shown in FIG. 1, the computer device includes the following components:

[0067] In step S101, a biomimetic olfactory electronic nose including a multi-array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor, and / or a surface acoustic wave sensor is used to perform real-time sampling of trace volatile organic compounds (VOCs) related to predefined allergens present in the air of a target area, and to obtain the amplitude, response time, recovery time, impedance change, and frequency drift of the sensor response curve to construct an original multi-dimensional feature spectrum. The predefined allergens include pollen, mold spores, plant volatiles, and mite components.

[0068] In this embodiment, the semiconductor metal oxide gas sensor (MOS), the quartz crystal microbalance sensor (QCM), and the surface acoustic wave sensor (SAW) are used to detect trace molecules in the environment, which can capture allergen-related VOCs (trace volatile organic compounds) with extremely low concentrations in the air, and achieve subtle allergic risk assessment and early warning. The electronic nose simulates the synergistic effect of biological olfactory receptors through array sensors, has different response modes to different VOCs, forms a unique fingerprint, and can distinguish complex VOC mixtures. The sensor response parameters such as amplitude, response time, recovery time, impedance change, and frequency drift not only contain the concentration information (amplitude) of VOCs, but also contain the kinetic information (response / recovery time) of their interaction with the sensor surface and the changes in physical and chemical properties (impedance / frequency drift).

[0069] Specifically, according to the type and concentration range of VOCs that can be released by the predefined allergens (pollen, mold spores, plant volatiles, mite components), MOS sensors, QCM sensors (coated with polymer membranes selectively adsorbing specific VOCs) and SAW sensors (coated with selective adsorption membranes) of different coatings or materials are selected. The selected multiple sensors are arranged in an array, each sensor showing different response curves to different VOCs or VOC mixtures, forming a unique olfactory fingerprint. The sensor array, gas flow path, heating element (for MOS sensors), oscillation circuit (for QCM / SAW sensors), signal conditioning circuit and data acquisition module are integrated in an electronic nose device. The electronic nose is equipped with a miniature pump for accurately sucking air samples from the target area into the sensor cavity. In addition, to remove large particles such as dust in the air, a pre-filter is provided at the air inlet. The sampling period (e.g. every minute, every five minutes or every hour) is set for air sampling and data reading to monitor in real time. When the air sample enters the sensor cavity, the VOCs therein interact with the sensor surface, causing changes in the physical or electrical properties of the sensor. The maximum change value of the sensor response is recorded, such as the resistance change of the MOS sensor, the frequency drift of the QCM / SAW sensor. The time required from the start of VOCs exposure to the sensor response reaching a stable value (or a preset percentage, such as 90%). The time required from the removal of VOCs to the sensor response returning to the baseline value (or a preset percentage). For QCM and SAW sensors, the change in their resonant frequency is recorded. The collected data is stored and transmitted to the server through wireless (such as Wi-Fi, LoRa). As shown in Table 1, the electronic nose records various parameters of each sensor during air sample exposure in real time.

[0070] Table 1

[0071]

[0072] Step S102, extracting the chemical fingerprint of each known allergen in the original multi-dimensional feature spectrum through the VOCs fingerprint model.

[0073] In an alternative way, extracting the chemical fingerprint of each known allergen in the original multi-dimensional feature spectrum through the VOCs fingerprint model further comprises:

[0074] Inputting the original multi-dimensional feature spectrum into the VOCs fingerprint model for blind source separation to extract an independent component vector corresponding to each known allergen; wherein the VOCs fingerprint model is trained in advance based on standard VOCs samples of known allergens, and each independent component vector is a purified signal feature of a known allergen;

[0075] The multi-domain feature extraction is performed on each independent component vector to obtain time-domain statistical features and frequency-domain transform features; and a chemical fingerprint of each known allergen is constructed according to the time-domain statistical features and the frequency-domain transform features.

[0076] In the embodiment, the ambient air is usually mixed with multiple VOCs, and the sensing response of the VOCs on the sensor array of the electronic nose is expressed as a superimposed signal. It is difficult to accurately distinguish the contribution of different allergens by using a traditional method. The present application can separate the pure signal feature of each allergen (independent component vector shown in Table 2) from the mixed signal by using a blind source separation method, significantly improve the recognition accuracy of a single allergen (for example, the cypress pollen is taken as an example, as shown in Table 3), and effectively work even in a complex mixture. Each independent component vector is defined as the purified signal feature of the known allergen, can strip the background noise and other interference, focus on the unique signal of a specific allergen, and thus reduce false positives and false negatives. The VOCs fingerprint model is trained in advance based on the standard VOCs sample of the known allergen, learns the typical signal pattern of each allergen in a controlled environment, and can better identify and match the known allergen when facing complex environmental data in the real world.

[0077] Table 2

[0078]

[0079] Table 3

[0080]

[0081] In an optional manner, the VOCs fingerprint model comprises:

[0082] A variational independent component analysis network layer comprising a multi-head self-attention layer and a reversible neural network layer is used for probabilistic blind source separation and extraction of an independent component vector corresponding to each known allergen;

[0083] A time-domain statistical feature extraction layer comprising a time-domain feature branch layer, a time-frequency analysis branch, and a non-stationary signal processing branch layer is used for extracting mean, variance, kurtosis, skewness, zero-crossing rate, autocorrelation coefficient, and empirical mode decomposition intrinsic mode function from each independent component vector;

[0084] A spectrogram analysis feature extraction layer comprising a bottleneck structure layer and a differentiable L2 normalization layer is used for adaptively decomposing the time-domain signal into an optimal wavelet packet basis and generating a time-frequency energy distribution spectrogram.

[0085] In this embodiment, the variational independent component analysis network layer can extract independent source signals (i.e. VOCs fingerprints of single allergens) from the mixed signals in a probabilistic manner. Through source separation by the variational independent component analysis network layer, and then extracting rich features from the time domain and the frequency domain respectively, the complementary features are finally fused to form a more discriminative chemical fingerprint.

[0086] In step S103, the Siamese network model is used to dynamically match the chemical fingerprint of each known allergen with the preset allergen chemical fingerprint library, and calculate the comprehensive regional allergic risk factor.

[0087] In this embodiment, the Siamese network can accurately identify similar chemical fingerprint characteristics in the presence of noise, signal drift or insufficient sample size through metric learning, overcoming the problems of large environmental sensing data fluctuations and strong nonlinearity. At the same time, it learns from the similarity of sample pairs and does not rely on large-scale labeled data, which is especially suitable for situations where allergen samples are scarce in real environments.

[0088] In an optional manner, the Siamese network model comprises:

[0089] The multi-scale feature pyramid encoding layer comprises a dilated convolution path layer, an adaptive pooling path layer and a residual connection path layer.

[0090] The cross-attention matching layer comprises a query generator, a key-value memory construction layer and a multi-head cross-attention layer, wherein the query generator generates a query vector through linear projection from the to-be-tested chemical fingerprint, and the key-value memory construction layer dynamically constructs a key vector and a value vector from the preset allergen chemical fingerprint library.

[0091] The dynamic relationship inference layer comprises a node initialization layer, an edge weight calculation layer, a graph convolution network layer and a readout function.

[0092] The multi-task metric learning head comprises a contrastive learning branch layer, a classification identification branch layer and a gradient balancing layer.

[0093] The risk factor regressor comprises an attention pooling layer, an uncertainty calibration layer and a probability output layer.

[0094] In this embodiment, the multi-scale feature pyramid encoding layer fuses dilated convolution (enlarges the receptive field), adaptive pooling (adapts to different input lengths), and residual connection (preserves original information), which can effectively deal with the variability of sensor signals in the time dimension. The cross-attention matching layer enables end-to-end similarity calculation between the test fingerprint spectrum and all known allergens in the spectrum library through query-key-value, supports dynamic updating of the spectrum library (such as adding new pollen or mold), and avoids the rigid problem of traditional template matching. The dynamic relationship reasoning layer uses a graph convolution network to fuse feature similarity, spatial co-occurrence, and chemical affinity to more accurately evaluate the synergistic sensitization effect under complex exposure (such as pollen and ozone jointly exacerbating respiratory reactions). The multi-task metric learning head simultaneously performs contrastive learning (pulling similar ones and pushing away dissimilar ones), classification identification (determining the type of allergen), and gradient balancing (relieving class imbalance), significantly enhancing the recognition ability in rare or newly emerging allergen scenarios. The risk factor regressor automatically reduces the prediction confidence when the signal-to-noise ratio is low or the sensor drifts.

[0095] In an alternative way, the graph convolution network layer in the dynamic relationship reasoning layer adopts a multi-hop propagation algorithm;

[0096] wherein the node feature update formula of the graph convolution network layer is:

[0097]

[0098] wherein, , are the feature vectors of the target node i and its neighbor node j in the layer graph convolution; is the set of all neighbor nodes of node i; is the importance coefficient of neighbor node j to target node i; is the layer learnable weight matrix; is the layer learnable self-connection weight; is the LeakyReLU activation function;

[0099] The attention coefficient of the graph convolution network layer is:

[0100]

[0101] wherein, , , is a learnable balance factor used to adjust the contribution proportion of the similarity factor, the spatial correlation factor, and the chemical affinity factor in the final attention weight; is the feature vector ,a similarity factor of a feature vector of , a correlation factor of a feature vector of , a chemical affinity factor of

[0102] In this embodiment, the multi-hop propagation algorithm allows information to propagate multiple layers in the graph, thereby modeling indirect association relationships (pollen→ozone→mold spore activity enhancement) and more comprehensively reflecting the synergistic or antagonistic effects of allergens in real environments. Among them, the similarity factor adopts cosine similarity or Euclidean distance normalized value to measure the feature proximity of , the spatial correlation factor is based on historical data of deployed sensors to count the co-occurrence frequency of the two types of allergens in the same period; and the chemical affinity factor gives higher affinity scores to structural similar categories according to the pre-defined chemical category knowledge base (pollen volatiles are mostly monoterpenes, and mold metabolites are mostly aldehydes and ketones).

[0103] In an optional manner, the similarity factor is:

[0104]

[0105] wherein, a cosine similarity of a feature vector of , a Gaussian kernel similarity based on the feature space distance a weight; a control decay rate;

[0106] The correlation factor is:

[0107]

[0108] wherein, a total weight of the factor; a directional angle of nodes and in the sensor space layout; , a time stamp of nodes and being detected; a time decay constant;

[0109] The chemical affinity factor is:

[0110]

[0111] wherein,​​ are the concentrations of the chemical components of the allergens and respectively. are the concentrations of the chemical components of the allergens are the contribution weights of the chemical components to the affinity. are the contribution weights of the chemical components to the affinity. are the concentration scaling parameters.

[0112] In this embodiment, the model is able to adapt to environmental changes (e.g. sensor position shift, allergen concentration fluctuation) and time dependence (e.g. seasonal regularity of pollen dissemination) by Gaussian kernel similarity (weight adjusted by distance decay) and time decay function (correlation strength controlled by timestamp difference).

[0113] In an alternative way, the formula of the comprehensive regional allergy risk factor is:

[0114]

[0115] wherein, , , are used to balance the contribution of the spatio-temporal correlation component , the synergistic effect component and the interaction term in the total risk respectively. is the base risk component. is the environmental modulation factor.

[0116]

[0117]

[0118]

[0119]

[0120] wherein, is the number of the identified known allergen species; is the risk weight of the i-th allergen species; is the chemical fingerprint feature vector of the i-th allergen species detected at present; is the reference standard feature vector of the i-th allergen species in the preset library; is the standard deviation of the i-th allergen feature; is the sigmoid function. is the time integration window. is the time decay kernel function. is the time variable. is the base risk component at time is the base risk component at time gradient; is an activation function; is a synergy effect coefficient between allergens i and j; is a base risk component of allergen i and j, respectively; is a distance of allergen i and j in feature space; is a synergy effect scale parameter; is the number of environmental parameter categories; is the actual measurement value of the kth environmental parameter at time t; is the optimal value of the kth environmental parameter; is the tolerance range of the kth environmental parameter; is the modulation intensity coefficient of the kth environmental parameter; is a Gaussian kernel function.

[0121] In this embodiment, Capture the evolution trend of risk over time (such as sudden increase in pollen concentration) through time integration window and gradient. Quantify the nonlinear interaction between allergens (such as the synergistic sensitization effect of pollen and mold spores) based on feature space distance and synergy effect coefficient. Exponential decay of the tolerance of environmental parameters deviating from the optimal value through Gaussian kernel function, to adapt to the risk fluctuation under different environmental conditions.

[0122] Step S104, according to the regional allergic risk factor and the preset multi-level allergic risk threshold, determine the allergic risk level of the current target region.

[0123] In an optional manner, the preset multi-level allergic risk threshold is dynamically adjusted according to historical allergic event data and meteorological parameters;

[0124] Among them, the incidence and severity of historical allergic events and the meteorological parameters associated with them are used as training data to input a time series model; the meteorological parameters include temperature, humidity, wind speed, rainfall and atmospheric pressure, etc.; the time series model outputs the corresponding allergic risk probability distribution under different meteorological parameters;

[0125] When receiving the current meteorological forecast data, the dynamic threshold of each level of allergic risk in a future period of time is predicted in real time according to the allergic risk probability distribution, as the preset multi-level allergic risk threshold.

[0126] In this embodiment, a time series model is constructed according to real historical allergic events and synchronous meteorological data (as shown in Table 4), the threshold setting has an environmental physical basis, and avoids the deviation of subjective experience setting. Meanwhile, the interaction of temperature, humidity, wind speed, rainfall, atmospheric pressure and other meteorological factors is considered to obtain the amplification effect of the combination of "high temperature + low humidity + strong wind" on pollen diffusion, or the triggering effect of "rapid humidity rise after rainfall" on mold spore release.

[0127] Table 4

[0128]

[0129] According to the scheme provided by the present application, the biomimetic olfactory electronic nose including a multi-array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor and / or a surface acoustic wave sensor is used to sample the trace volatile organic compounds related to the predefined allergens in the air of the target area in real time, the amplitude, response time, recovery time, impedance change and frequency drift of the sensing response curve are obtained to construct the original multi-dimensional feature spectrum; wherein the predefined allergens include pollen, mold spores, plant volatiles and mite components; the chemical fingerprint of each known allergen in the original multi-dimensional feature spectrum is extracted through the VOCs fingerprint model; the chemical fingerprint of each known allergen is dynamically matched with the preset allergen chemical fingerprint library by using the Siamese network model, and the comprehensive regional allergic risk factor is calculated; and the allergic risk level of the current target area is determined according to the regional allergic risk factor and the preset multi-level allergic risk threshold. The present application extracts the purified signal characteristics corresponding to each known allergen from the complex mixed VOCs signal through the VOCs fingerprint model, solves the problem that the multiple source trace VOCs signals interfere with each other and are difficult to distinguish, and obtains the similarity, correlation and chemical affinity between different allergens through the Siamese network model, so that the comprehensive regional allergic risk factor is closer to the real situation.

[0130] Figure 2 The framework schematic diagram of the regional allergic risk real-time early warning device based on environmental trace molecular fingerprint according to the embodiments of the present application is shown. The regional allergic risk real-time early warning device based on environmental trace molecular fingerprint includes:

[0131] The environmental sampling and feature spectrum construction module 210 is a biomimetic olfactory electronic nose including a multi-array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor and / or a surface acoustic wave sensor, which is used to sample the trace volatile organic compounds related to the predefined allergens in the air of the target area in real time, and the amplitude, response time, recovery time, impedance change and frequency drift of the sensing response curve are obtained to construct the original multi-dimensional feature spectrum; wherein the predefined allergens include pollen, mold spores, plant volatiles and mite components;

[0132] The chemical fingerprint extraction module 220 extracts the chemical fingerprint of each known allergen in the original multi-dimensional characteristic spectrum through a VOCs fingerprint model;

[0133] The dynamic matching and risk factor calculation module 230 dynamically matches the chemical fingerprint of each known allergen with a preset allergen chemical fingerprint library by using a Siamese network model, and calculates a comprehensive regional allergic risk factor;

[0134] The risk level determination and early warning module 240 determines the allergic risk level of the current target region according to the regional allergic risk factor and a preset multi-level allergic risk threshold.

[0135] Figure 3 The specific embodiments of the present application do not limit the specific implementation of the computer device.

[0136] As Figure 3 shown, the computer device can include a processor 302, a communications interface 304, a memory 306, and a communications bus 308.

[0137] The processor 302, the communications interface 304, and the memory 306 can communicate with each other through the communications bus 308. The communications interface 304 is configured to communicate with network elements such as clients or other servers. The processor 302 is configured to execute the program 310, and specifically can execute the related steps in the above-mentioned regional allergic risk real-time early warning method based on environmental trace molecule fingerprint.

[0138] Specifically, the program 310 can include program code, which includes computer operation instructions.

[0139] The processor 302 can be a central processing unit CPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present application. The one or more processors included in the computer device can be the same type of processor, such as one or more CPUs; or can be different types of processors, such as one or more CPUs and one or more ASICs.

[0140] Memory 306, for storing program 310. Memory 306 can include high-speed RAM memory and can also include non-volatile memory, such as at least one disk memory.

[0141] According to the scheme provided by the application, the biomimetic olfactory electronic nose including a multi-array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor and / or a surface acoustic wave sensor is used to sample the trace volatile organic compounds related to the predefined allergens in the air of the target area in real time, the amplitude, response time, recovery time, impedance change and frequency drift of the sensing response curve are obtained to construct the original multi-dimensional feature spectrum; wherein the predefined allergens include pollen, mold spores, plant volatiles and mite components; the chemical fingerprint spectrum of each known allergen in the original multi-dimensional feature spectrum is extracted through the VOCs fingerprint spectrum model; the chemical fingerprint spectrum of each known allergen is dynamically matched with the preset allergen chemical fingerprint spectrum library by using the Siamese network model, and the comprehensive regional allergic risk factor is calculated; according to the regional allergic risk factor and the preset multi-level allergic risk threshold, the allergic risk level of the current target area is determined. The application extracts the purified signal characteristics corresponding to each known allergen from the complex mixed VOCs signal through the VOCs fingerprint spectrum model, solves the problem that the multi-source trace VOCs signals interfere with each other and are difficult to distinguish, and obtains the similarity, correlation and chemical affinity between different allergens through the Siamese network model, so that the comprehensive regional allergic risk factor is closer to the real situation.

[0142] Those skilled in the art will appreciate that the modules in the apparatuses in the embodiments can be adapted and placed in one or more apparatuses other than the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and furthermore can be divided into multiple sub-modules or sub-units or sub-components. Any combination of all the features disclosed in the present specification (including the accompanying claims, abstract and drawings), and any method or apparatus so disclosed, can be taken in any combination, except that at least some of such features and / or processes or units are mutually exclusive, unless explicitly stated otherwise. Each feature disclosed in the present specification (including the accompanying claims, abstract and drawings) can be replaced by alternative features serving the same, equivalent or similar purpose, unless explicitly stated otherwise. Furthermore, the skilled person will appreciate that the combination of features of different embodiments implies that the features of the different embodiments are meant to be combined, unless explicitly stated otherwise. For example, in the claims below, any of the embodiments can be used in any combination. The application can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unitary claim, several of the devices mentioned in the embodiments can be implemented by means of one and the same hardware item. The steps of the above-described embodiments, unless explicitly stated otherwise, are not to be understood as having to be carried out in the order in which they are described.

Claims

1. A method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting, characterized in that, include: A biomimetic electronic olfactory nose, comprising a multi-array semiconductor metal oxide gas sensor, a quartz crystal microbalance sensor, and / or a surface acoustic wave sensor, is used to sample trace amounts of volatile organic compounds associated with predefined allergens in the air of a target area in real time. The amplitude, response time, recovery time, impedance change, and frequency drift of the sensing response curve are obtained to construct an original multidimensional feature spectrum. The predefined allergens include pollen, mold spores, plant volatiles, and mite components. The chemical fingerprint of each known allergen in the original multidimensional feature spectrum is extracted using the VOCs fingerprinting model; The Siamese network model is used to dynamically match the chemical fingerprint of each known allergen with a pre-set allergen chemical fingerprint database to calculate the comprehensive regional allergy risk factor. Based on regional allergy risk factors and preset multi-level allergy risk thresholds, the allergy risk level of the current target area is determined.

2. The method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting as described in claim 1, characterized in that, The extraction of the chemical fingerprint of each known allergen from the original multidimensional feature spectrum using the VOCs fingerprinting model further includes: The original multidimensional feature spectrum is input into the VOCs fingerprint model for blind source separation, and independent component vectors corresponding to each known allergen are extracted; wherein, the VOCs fingerprint model is pre-trained based on standard VOCs samples of known allergens, and each independent component vector is a purification signal feature of a known allergen; Multi-domain feature extraction is performed on each independent component vector to obtain time-domain statistical features and frequency-domain transform features; a chemical fingerprint spectrum for each known allergen is constructed based on the time-domain statistical features and frequency-domain transform features.

3. The method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting according to claim 2, characterized in that, The VOCs fingerprint model includes: The variational independent component analysis network layer, including a multi-head self-attention layer and a reversible neural network layer, is used to perform probabilistic blind source separation and extract independent component vectors corresponding to each known allergen. The time-domain statistical feature extraction layer includes a time-domain feature branch layer, a time-frequency analysis branch layer, and a non-stationary signal processing branch layer, which are used to extract the mean, variance, kurtosis, skewness, zero-crossing rate, autocorrelation coefficient, and empirical mode decomposition intrinsic mode function from each independent component vector; The spectral analysis feature extraction layer, including a bottleneck structure layer and a differentiable L2 normalization layer, is used to adaptively decompose the time-domain signal onto the optimal wavelet packet basis and generate a time-frequency energy distribution spectrum.

4. The method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting according to claim 1, characterized in that, The Siamese network model includes: The multi-scale feature pyramid encoding layer includes dilated convolutional path layers, adaptive pooling path layers, and residual connection path layers; The cross-attention matching layer includes a query generator, a key-value memory construction layer, and a multi-head cross-attention layer. The query generator generates a query vector from the chemical fingerprint spectrum to be tested through linear projection. The key-value memory construction layer dynamically constructs key vectors and value vectors from a preset allergen chemical fingerprint spectrum library. The dynamic relation inference layer includes a node initialization layer, an edge weight calculation layer, a graph convolutional network layer, and a readout function; The multi-task metric learning head includes a contrastive learning branch layer, a classification and recognition branch layer, and a gradient equalization layer; The risk factor regressor includes an attention pooling layer, an uncertainty calibration layer, and a probability output layer.

5. The method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting according to claim 4, characterized in that, The graph convolutional network layer in the dynamic relation inference layer employs a multi-hop propagation algorithm. The node feature update formula for the graph convolutional network layer is as follows: ； in, , They were respectively in the second The feature vectors of the target node i and its neighboring nodes j in the layer graph convolution; Let i be the set of all neighboring nodes of node i; Let j be the importance coefficient of neighbor node j to target node i; For the first Layer-learnable weight matrix; For the first Layer-learnable self-connection weights; Use the LeakyReLU activation function; The attention coefficient of the graph convolutional network layer is: ； in, , , These are learnable balancing factors used to adjust the contribution ratios of similarity factor, spatial association factor, and chemical affinity factor in the final attention weight; For feature vectors , Similarity factor; For feature vectors , The correlation factors; For feature vectors , Chemical affinity factor.

6. The method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting according to claim 5, characterized in that, The similarity factor is: ； in, For feature vectors , Cosine similarity; For feature space distance Gaussian kernel similarity; As weight; To control the decay rate; The correlation factor is: ； in, The overall weight of the factor; For nodes and Orientation angle in sensor spatial layout; , For nodes and The detected timestamp; The time decay constant; The chemical affinity factor is: ； in, Allergens and The The concentration of a chemical component; For the first The contribution weight of each chemical component to affinity; This is the concentration scaling parameter.

7. The method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting according to claim 1, characterized in that, The formula for calculating the comprehensive regional allergy risk factor is as follows: ； in, , , They are used to balance the spatiotemporal correlation components respectively. Synergistic effect component And the contribution of interaction items to the total risk; Basic risk component; Environmental modulation factor; ； ； ； ； in, The number of known allergens identified; Let i be the risk weight of the i-th allergen; This represents the chemical fingerprint feature vector of the i-th allergen currently detected. This is the reference standard feature vector of the i-th allergen in the preset library; Let be the standard deviation of the characteristics of the i-th allergen; It is the sigmoid function; For time integration window; This is the time decay kernel function; It is a time variable; For at any time Basic risk components The gradient; For activation functions; The coefficient representing the synergistic effect between allergens i and j; These are the baseline risk components for allergens i and j, respectively; Let be the distance between allergens i and j in the feature space; For the synergistic effect scale parameter; The number of types of environmental parameters; This represents the actual measured value of the k-th environmental parameter at time t. This represents the optimal value for the k-th environmental parameter. The tolerance range for the kth environmental parameter; Let be the modulation intensity coefficient for the k-th environmental parameter; This is the Gaussian kernel function.

8. The method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting according to claim 1, characterized in that, The preset multi-level allergy risk thresholds are dynamically adjusted based on historical allergy event data and meteorological parameters; The model uses historical allergy event incidence, severity, and associated meteorological parameters as training data, inputting them into a time series model. These meteorological parameters include temperature, humidity, wind speed, rainfall, and atmospheric pressure. The time series model outputs the probability distribution of allergy risk under different meteorological parameters. Upon receiving current weather forecast data, the system predicts dynamic thresholds for allergy risks at various levels in real time based on the allergy risk probability distribution, which are then used as the preset multi-level allergy risk thresholds.

9. A real-time early warning device for regional allergy risk based on environmental trace molecular fingerprinting, characterized in that, include: An environmental sampling and feature spectrum construction module is used in a biomimetic olfactory electronic nose that incorporates array semiconductor metal oxide gas sensors, quartz crystal microbalance sensors, and / or surface acoustic wave sensors. This module performs real-time sampling of trace volatile organic compounds (VOCs) in the air of a target area that are associated with predefined allergens. It acquires the amplitude, response time, recovery time, impedance change, and frequency drift of the sensing response curve to construct an original multidimensional feature spectrum. The predefined allergens include pollen, mold spores, plant volatiles, and mite components. The chemical fingerprint extraction module is used to extract the chemical fingerprint of each known allergen from the original multidimensional feature spectrum using a VOCs fingerprint model. The dynamic matching and risk factor calculation module is used to dynamically match the chemical fingerprint spectrum of each known allergen with a preset allergen chemical fingerprint spectrum library using the Siamese network model, and calculate the comprehensive regional allergy risk factor. The risk level assessment and early warning module is used to determine the allergy risk level of the current target area based on regional allergy risk factors and preset multi-level allergy risk thresholds.

10. A computer device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described method for real-time early warning of regional allergy risk based on environmental trace molecular fingerprinting.

Images