Olfactory chip-based outdoor environment health monitoring method

Abstract

The invention discloses an outdoor environment health monitoring method based on an olfactory chip, and relates to the technical field of environment health monitoring. A traditional sensor and an olfactory chip are adopted at a monitoring node at the same time, so that synchronous acquisition of temperature, humidity, gas concentration and complex smell fingerprints is realized; the limitation of identifying multi-component pollutants in a complex environment of an industrial area by a single data source is fundamentally made up; the traditional method is difficult to distinguish cross interference of various gases, however, in the invention, a deep neural network is combined with an attention mechanism to extract implicit features of sensing data, and data cross correction is realized through space-time correlation analysis, so that the accuracy of pollutant identification is remarkably improved; further introducing a graph neural network to carry out modeling on spatio-temporal data of each monitoring node, capturing space structure information of the whole monitoring network, and forming a fusion feature vector with relatively high distinguishing capability; the problem of signal cross interference when various malodorous gases are mixed in an industrial area is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental health monitoring, and in particular to an outdoor environmental health monitoring method based on an olfactory chip. Background Art

[0002] At present, outdoor environmental health monitoring generally uses gas sensors and spectrometers to conduct fixed-point detection of harmful gases in the air, and reflects environmental quality through regular sampling. This type of method can effectively capture temperature, humidity and concentration data of specific pollutants in stable areas or when monitoring fixed emission sources, but in actual applications, the monitoring effect on industrial areas is not good.

[0003] There are many types of gases in industrial environments. The odorous substances emitted include not only conventional gases such as sulfur dioxide and hydrogen sulfide, but also various volatile organic compounds, which form cross-interference. A single sensor array has difficulty distinguishing the subtle differences between these gases during detection, resulting in data confusion, which in turn affects the accurate assessment of the degree of pollution. Secondly, the real-time nature of environmental changes is high, the equipment in industrial areas operates frequently, and the process fluctuates greatly. Traditional monitoring methods are difficult to respond quickly to sudden pollution incidents, and early warnings lag behind the actual spread of pollution.

[0004] To make up for this deficiency, some traditional solutions introduce a multi-sensor combination to pre-process and linearly fuse data from different types of sensors. However, this method still has difficulty in effectively extracting the characteristic signals of each pollutant in a complex mixed gas. Another solution uses gas chromatography-mass spectrometry technology. Although it can improve detection accuracy, the equipment is large and costly, making it difficult to achieve wide-area real-time monitoring and rapid on-site response. In this context, how to more accurately distinguish complex gas mixtures has become a key issue that needs to be addressed. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an outdoor environmental health monitoring method based on an olfactory chip to solve the problem that there are many kinds of odorous substances emitted from industrial areas, and cross-responses often occur, making it difficult for traditional single sensor arrays to distinguish complex gas mixtures.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] The embodiment of the present invention provides an outdoor environment health monitoring method based on an olfactory chip, which comprises:

[0009] Step S1, in the area to be monitored, the area is divided according to the preset grid, and monitoring nodes are arranged in each area, each monitoring node includes a sensor unit and an olfactory chip sensor module;

[0010] Step S2, filtering, standardizing and denoising the raw data collected by each monitoring node to generate pre-processed data;

[0011] Step S3, integrating the preprocessed data in step S2 using a multimodal data fusion algorithm to generate a fused feature vector;

[0012] Step S4, based on historical monitoring data, using a supervised learning method to perform model training on the fused feature vector of step S3 to establish a multi-component gas recognition model;

[0013] Step S5: deploy the trained recognition model on the edge computing platform and analyze the fused feature vector collected in real time:

[0014] When the identification result shows that the pollution index of a monitoring node exceeds the preset threshold, the local alarm device is immediately activated;

[0015] Step S6, after the local alarm device is activated, the real-time data and alarm information of each monitoring node are transmitted to the control console via the wireless network, and the control console stores, displays and compares the real-time data and alarm information with historical data.

[0016] As a preferred solution of the outdoor environment health monitoring method based on olfactory chip described in the present invention, wherein: the sensor unit is used to collect temperature, humidity and gas concentration data;

[0017] The olfactory chip sensor module is an electronic nose device composed of multiple miniature odor sensor elements, which is used to capture multi-component odor fingerprints in the air in real time.

[0018] As a preferred solution of the outdoor environment health monitoring method based on olfactory chip described in the present invention, the preprocessed data includes traditional sensor data and transient odor response data output by the olfactory chip module.

[0019] As a preferred solution of the outdoor environmental health monitoring method based on olfactory chip described in the present invention, wherein: in step S3, the multimodal data fusion algorithm includes a deep neural network structure, which is used to extract the characteristics of the data of each sensor module, and use spatiotemporal correlation analysis to achieve data cross-correction and separate the mutual interference information between multi-component gases.

[0020] As a preferred solution of the outdoor environmental health monitoring method based on olfactory chip described in the present invention, in the data fusion step of step S3, the graph neural network is also used to model and analyze the spatiotemporal data of each monitoring node to identify multi-component cross-interference signals.

[0021] As a preferred solution of the outdoor environment health monitoring method based on olfactory chip described in the present invention, wherein: in step S3, the step of integrating using multimodal data fusion algorithm to generate fusion feature vectors is:

[0022] A deep neural network is used to process the raw data, and the processing formula is:

[0023] F i =σ(W2(σ(W1X i +b1))+b2),

[0024] Among them, X i represents the raw data collected by the i-th sensor module, W1 represents the first-layer weight matrix, b1 represents the first-layer bias vector, W2 represents the second-layer weight matrix, b2 represents the second-layer bias vector, σ represents the activation function, and F i Represents the extracted deep features;

[0025] The similarity function is defined to capture the spatiotemporal correlation between data. The function formula is:

[0026]

[0027] Among them, F i,t represents the feature of node i at time t, F j,t represents the characteristics of node j at time t, W a represents the similarity weight matrix, ReLU represents the rectified linear unit function, φ(F i,t ,F j,t ) indicates the similarity between the two;

[0028] Based on the similarity, the attention weight is calculated using the following formula:

[0029]

[0030] Among them, α ij represents the attention weight of node i to neighboring node j, represents the neighborhood set of node i, exp represents the exponential function, F l,t Represents the characteristics of node l at time t;

[0031] Weighted aggregation of neighborhood features yields G i,t :

[0032]

[0033] Among them, G i,t Represents the weighted aggregation feature of node i at time t.

[0034] As a preferred solution of the outdoor environmental health monitoring method based on an olfactory chip described in the present invention, in step S3, the step of integrating using a multimodal data fusion algorithm to generate a fusion feature vector also includes:

[0035] The graph neural network is used to perform graph convolution processing on each monitoring node. The convolution formula is:

[0036]

[0037] Among them, H i Represents the features of node i after graph convolution processing, represents node i and its neighborhood, c ij represents the normalization factor,

[0038] W g represents the graph convolution weight matrix, σ represents the activation function as before,

[0039] The weighted aggregation features and graph convolution features are cascaded to form a fused feature vector, which can be expressed as:

[0040] Z i =[G i,t ;H i ],

[0041] Among them, Z i represents the fused feature vector of node i, [;] represents G i,t With H i Cascade by column.

[0042] As a preferred solution of the outdoor environmental health monitoring method based on the olfactory chip described in the present invention, the recognition model is used to automatically determine the concentration levels of various malodorous gases and other harmful components;

[0043] The recognition model is trained using back propagation and gradient descent algorithms, and dynamic parameter adjustment is performed according to the actual monitoring environment.

[0044] As a preferred solution of the outdoor environmental health monitoring method based on an olfactory chip described in the present invention, in step S4, the step of using a supervised learning method to perform model training on the fused feature vector of step S3 to establish a multi-component gas recognition model is as follows:

[0045] Using the fused feature vector Z generated in step S3 i A multi-component gas identification model is established, and its output is defined as:

[0046]

[0047] in, represents the model prediction output of node i, g(·;ω) represents the recognition model function, ω represents the model parameter,

[0048] Construct the loss function, the function formula is:

[0049]

[0050] Among them, L represents the total loss, N represents the total number of samples, and y i represents the true label of node i, l(·) represents the cross entropy loss function;

[0051] Gradient descent is used to update the parameters, and the update formula is:

[0052]

[0053] Among them, η represents the learning rate, represents the gradient of the loss function with respect to the parameter ω.

[0054] As a preferred solution of the outdoor environmental health monitoring method based on olfactory chip described in the present invention, in step S5, the trained recognition model is deployed on the edge computing platform, and the step of analyzing the fusion feature vector collected in real time is as follows:

[0055] The trained recognition model is deployed on the edge computing platform to process the fused feature vector collected in real time. Its real-time output is expressed as:

[0056]

[0057] Set the alarm conditions as:

[0058] like Then the alarm device of node i is triggered;

[0059] Wherein, τ represents the preset alarm threshold.

[0060] The beneficial effects of the present invention are as follows: the present invention adopts both traditional sensors and olfactory chips at the monitoring nodes to realize the synchronous collection of temperature, humidity, gas concentration and complex odor fingerprints, fundamentally making up for the limitation of a single data source in identifying multi-component pollutants in complex environments of industrial areas; traditional methods are difficult to distinguish cross-interference of various gases, while the present invention utilizes deep neural networks combined with attention mechanisms to extract implicit features of sensor data, and realizes data cross-correction through spatiotemporal correlation analysis, which significantly improves the accuracy of pollutant identification.

[0061] The present invention further introduces a graph neural network to model the spatiotemporal data of each monitoring node, captures the spatial structure information of the entire monitoring network, and forms a fused feature vector with high distinguishing ability; this multimodal data fusion method effectively solves the problem of signal cross-interference when multiple odorous gases are mixed in industrial areas, enables the data of each node to complement each other, and greatly improves the overall monitoring effect.

[0062] In addition, the supervised learning model constructed by the present invention adopts cross entropy loss and gradient descent method to achieve efficient training of fused feature vectors. The trained model has the ability to automatically determine the concentration of various pollutants, and can dynamically adjust parameters according to the actual environment to adapt to complex and changeable industrial scenarios; after being deployed on the edge computing platform, it can analyze and collect data in real time, quickly respond to abnormal situations, trigger local alarms, and transmit abnormal information to the central console for centralized management and display.

[0063] In summary, the present invention forms a closed-loop control system from data collection, feature extraction to model training and real-time early warning, which not only improves the monitoring accuracy and response speed, but also effectively overcomes the shortcomings of traditional environmental monitoring methods in dealing with multi-component cross-interference and real-time data integration, and provides reliable technical support and data guarantee for industrial area environmental governance. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] 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 only 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.

[0065] Figure 1 It is a schematic diagram of the flow of the outdoor environment health monitoring method based on the olfactory chip of the present invention. DETAILED DESCRIPTION

[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0067] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0068] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0069] Example 1, reference Figure 1 This embodiment provides an outdoor environment health monitoring method based on an olfactory chip, comprising:

[0070] Step S1, in the area to be monitored, the area is divided according to the preset grid, and monitoring nodes are arranged in each area, each monitoring node includes a sensor unit and an olfactory chip sensor module;

[0071] The sensor unit is used to collect temperature, humidity and gas concentration data;

[0072] The olfactory chip sensor module is an electronic nose device composed of multiple micro odor sensor elements, which is used to capture multi-component odor fingerprints in the air in real time;

[0073] Step S2, filtering, standardizing and denoising the raw data collected by each monitoring node to generate pre-processed data;

[0074] The preprocessed data includes traditional sensor data and transient odor response data output by the olfactory chip module;

[0075] Step S3, integrating the pre-processed data in step S2 using a multimodal data fusion algorithm to generate a fused feature vector;

[0076] In step S3, the multimodal data fusion algorithm includes a deep neural network structure, which is used to extract the characteristics of the data of each sensor module, and use spatiotemporal correlation analysis to achieve data cross-correction and separate the mutual interference information between multi-component gases;

[0077] In the data fusion step of step S3, the graph neural network is also used to model and analyze the spatiotemporal data of each monitoring node to identify multi-component cross-interference signals;

[0078] In step S3, a multimodal data fusion algorithm is used for integration, and the steps of generating a fusion feature vector are as follows:

[0079] A deep neural network is used to process the raw data, and the processing formula is:

[0080] F i =σ(W2(σ(W1X i +b1))+b2),

[0081] Among them, X irepresents the raw data collected by the i-th sensor module, W1 represents the first-layer weight matrix, b1 represents the first-layer bias vector, W2 represents the second-layer weight matrix, b2 represents the second-layer bias vector, σ represents the activation function, and F i Represents the extracted deep features;

[0082] The similarity function is defined to capture the spatiotemporal correlation between data. The function formula is:

[0083]

[0084] Among them, F i,t represents the feature of node i at time t, F j,t represents the characteristics of node j at time t, W a represents the similarity weight matrix, ReLU represents the rectified linear unit function, φ(F i,t ,F j,t ) indicates the similarity between the two;

[0085] Based on the similarity, the attention weight is calculated using the following formula:

[0086]

[0087] Among them, α ij represents the attention weight of node i to neighboring node j, represents the neighborhood set of node i, exp represents the exponential function, F l,t Represents the characteristics of node l at time t;

[0088] Weighted aggregation of neighborhood features yields G i,t :

[0089]

[0090] Among them, G i,t Represents the features of node i after weighted aggregation at time t;

[0091] In step S3, the multimodal data fusion algorithm is used for integration, and the step of generating a fusion feature vector also includes:

[0092] The graph neural network is used to perform graph convolution processing on each monitoring node. The convolution formula is:

[0093]

[0094] Among them, H i Represents the features of node i after graph convolution processing, represents node i and its neighborhood, c ij represents the normalization factor,

[0095] Wg represents the graph convolution weight matrix, σ represents the activation function as before,

[0096] The weighted aggregation features and graph convolution features are cascaded to form a fused feature vector, which can be expressed as:

[0097] Z i =[G i,t ;H i ],

[0098] Among them, Z i represents the fused feature vector of node i, [;] represents G i,t With H i Cascade by column;

[0099] Specifically, deep neural networks are used here to extract implicit features from the original data, and the attention mechanism is used to perform weighted aggregation on the node data at different times. Graph convolution is used to capture the spatial correlation between the monitoring nodes. The fused feature vector formed by the cascade of the two not only retains the local response information, but also reflects the overall spatiotemporal structure.

[0100] Step S4, based on historical monitoring data, using a supervised learning method to perform model training on the fused feature vector of step S3 to establish a multi-component gas recognition model;

[0101] The recognition model is used to automatically determine the concentration levels of various malodorous gases and other harmful components;

[0102] The recognition model is trained using back propagation and gradient descent algorithms, and dynamic parameter adjustments are made according to the actual monitoring environment;

[0103] In step S4, the fusion feature vector of step S3 is trained using a supervised learning method, and the steps of establishing a multi-component gas recognition model are as follows:

[0104] Using the fused feature vector Z generated in step S3 i A multi-component gas identification model is established, and its output is defined as:

[0105]

[0106] in, represents the model prediction output of node i, g(·;ω) represents the recognition model function, ω represents the model parameter,

[0107] Construct the loss function, the function formula is:

[0108]

[0109] Among them, L represents the total loss, N represents the total number of samples, and y irepresents the true label of node i, l(·) represents the cross entropy loss function;

[0110] Gradient descent is used to update the parameters, and the update formula is:

[0111]

[0112] Among them, η represents the learning rate, Represents the gradient of the loss function with respect to the parameter ω;

[0113] Specifically, a supervised learning framework is constructed here, the fused feature vector is input into the recognition model, and the loss function is defined to measure the difference between the model output and the true label. The parameters are updated using the gradient descent algorithm so that the model can continuously adapt to the actual monitoring data.

[0114] Step S5: deploy the trained recognition model on the edge computing platform and analyze the fused feature vector collected in real time:

[0115] When the identification result shows that the pollution index of a monitoring node exceeds the preset threshold, the local alarm device is immediately activated;

[0116] In step S5, the trained recognition model is deployed on the edge computing platform, and the step of analyzing the fused feature vector collected in real time is as follows:

[0117] The trained recognition model is deployed on the edge computing platform to process the fused feature vector collected in real time. Its real-time output is expressed as:

[0118]

[0119] Set the alarm conditions as:

[0120] like Then the alarm device of node i is triggered;

[0121] Among them, τ represents the preset alarm threshold;

[0122] Specifically, here, by deploying the trained recognition model on the edge platform, the real-time collected data is quickly analyzed, and the model prediction output is compared with the preset threshold. If the threshold is exceeded, the alarm device is automatically triggered. At the same time, the abnormal information is transmitted to the console through the wireless network for storage and display, which helps to respond to environmental abnormalities in a timely manner;

[0123] Step S6, after the local alarm device is activated, the real-time data and alarm information of each monitoring node are transmitted to the control console via the wireless network, and the control console stores, displays and compares the real-time data and alarm information with historical data.

[0124] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for outdoor environmental health monitoring based on an olfactory chip, characterized in that: include, Step S1, in the area to be monitored, the area is divided according to the preset grid, and monitoring nodes are arranged in each area, each monitoring node includes a sensor unit and an olfactory chip sensor module; Step S2, filtering, standardizing and denoising the raw data collected by each monitoring node to generate pre-processed data; Step S3, integrating the preprocessed data in step S2 using a multimodal data fusion algorithm to generate a fused feature vector; Step S4, based on historical monitoring data, using a supervised learning method to perform model training on the fused feature vector of step S3 to establish a multi-component gas recognition model; Step S5: deploy the trained recognition model on the edge computing platform and analyze the fused feature vector collected in real time: When the identification result shows that the pollution index of a monitoring node exceeds the preset threshold, the local alarm device is immediately activated; Step S6, after the local alarm device is activated, the real-time data and alarm information of each monitoring node are transmitted to the control console via the wireless network, and the control console stores, displays and compares the real-time data and alarm information with historical data.

2. The outdoor environment health monitoring method based on an olfactory chip according to claim 1, characterized in that: The sensor unit is used to collect temperature, humidity and gas concentration data; The olfactory chip sensor module is an electronic nose device composed of multiple miniature odor sensor elements, which is used to capture multi-component odor fingerprints in the air in real time.

3. The outdoor environment health monitoring method based on an olfactory chip as claimed in claim 2, characterized in that: The preprocessed data includes traditional sensor data and transient odor response data output by the olfactory chip module.

4. The outdoor environment health monitoring method based on an olfactory chip as claimed in claim 3, characterized in that: In step S3, the multimodal data fusion algorithm includes a deep neural network structure, which is used to extract the characteristics of the data of each sensor module, and use spatiotemporal correlation analysis to achieve data cross-correction and separate the mutual interference information between multi-component gases.

5. The outdoor environment health monitoring method based on an olfactory chip according to claim 4, characterized in that: In the data fusion step of step S3, the graph neural network is also used to model and analyze the spatiotemporal data of each monitoring node to identify multi-component cross-interference signals.

6. The outdoor environment health monitoring method based on an olfactory chip according to claim 5, characterized in that: In step S3, the step of integrating by using a multimodal data fusion algorithm to generate a fusion feature vector is as follows: A deep neural network is used to process the raw data, and the processing formula is: F i =σ(W2(σ(W1X i +b1))+b2), Among them, X i represents the raw data collected by the i-th sensor module, W1 represents the first-layer weight matrix, b1 represents the first-layer bias vector, W2 represents the second-layer weight matrix, b2 represents the second-layer bias vector, σ represents the activation function, and F i Represents the extracted deep features; The similarity function is defined to capture the spatiotemporal correlation between data. The function formula is: Among them, F i,t represents the feature of node i at time t, F j,t represents the characteristics of node j at time t, W a represents the similarity weight matrix, ReLU represents the rectified linear unit function, φ(F i,t ,F j,t ) indicates the similarity between the two; Based on the similarity, the attention weight is calculated using the following formula: Among them, α ij represents the attention weight of node i to neighboring node j, represents the neighborhood set of node i, exp represents the exponential function, F l,t Represents the characteristics of node l at time t; Weighted aggregation of neighborhood features yields G i,t : Among them, G i,t Represents the weighted aggregation feature of node i at time t.

7. The outdoor environment health monitoring method based on an olfactory chip according to claim 6, characterized in that: In step S3, the step of integrating using a multimodal data fusion algorithm to generate a fused feature vector also includes: The graph neural network is used to perform graph convolution processing on each monitoring node. The convolution formula is: Among them, H i Represents the features of node i after graph convolution processing, represents node i and its neighborhood, c ij represents the normalization factor, W g represents the graph convolution weight matrix, σ represents the activation function as before, The weighted aggregation features and graph convolution features are cascaded to form a fused feature vector, which can be expressed as: Z i =[G i,t ;H i ], Among them, Z i represents the fused feature vector of node i, [;] represents G i,t With H i Cascade by column.

8. The outdoor environment health monitoring method based on an olfactory chip according to claim 7, characterized in that: The identification model is used to automatically determine the concentration levels of various malodorous gases and other harmful components; The recognition model is trained using back propagation and gradient descent algorithms, and dynamic parameter adjustment is performed according to the actual monitoring environment.

9. The outdoor environment health monitoring method based on an olfactory chip as claimed in claim 8, characterized in that: In step S4, the supervised learning method is used to perform model training on the fused feature vector of step S3 to establish a multi-component gas recognition model. Using the fused feature vector Z generated in step S3 i A multi-component gas identification model is established, and its output is defined as: in, represents the model prediction output of node i, g(·;ω) represents the recognition model function, ω represents the model parameter, Construct the loss function, the function formula is: Among them, L represents the total loss, N represents the total number of samples, and y i represents the true label of node i, l(·) represents the cross entropy loss function; Gradient descent is used to update the parameters, and the update formula is: Among them, η represents the learning rate, represents the gradient of the loss function with respect to the parameter ω.

10. The outdoor environment health monitoring method based on an olfactory chip according to claim 9, characterized in that: In step S5, the trained recognition model is deployed on the edge computing platform, and the step of analyzing the fused feature vector collected in real time is as follows: The trained recognition model is deployed on the edge computing platform to process the fused feature vector collected in real time. Its real-time output is expressed as: Set the alarm conditions as: like Then the alarm device of node i is triggered; Wherein, τ represents the preset alarm threshold.

Images