Deep learning based vocs pollution feature analysis and tracing method
By constructing a deep learning model, the limitations of linear assumptions and weak meteorological coupling capabilities of VOCs source tracing methods are solved, achieving high-accuracy and adaptive pollution source analysis that adapts to environmental changes and data gaps, and the output results have physical meaning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- FUJIAN JINHUANG ENVIRONMENTAL PROTECTION TECH CO LTD
- Filing Date
- 2026-02-12
- Publication Date
- 2026-06-26
AI Technical Summary
Existing VOCs source tracing methods have limitations due to linear assumptions, weak meteorological coupling capabilities, reliance on human experience, and a lack of adaptability, resulting in large analytical errors and low automation.
A deep learning-based approach is adopted to construct a neural network model that includes a spatiotemporal feature extraction module and a source resolution regression module. Nonlinear chemical reaction features and meteorological correlation features of VOCs are extracted through a one-dimensional convolutional neural network and a bidirectional long short-term memory network. Feature fusion is performed by combining meteorological parameters, and decoding is performed using physical constraints and prior knowledge of chemical fingerprints. Transfer learning and online updates are supported.
It significantly improves the accuracy and adaptability of VOCs pollution source analysis, can maintain stable operation when data is missing or the environment changes, and the output results have clear environmental physical meaning.
Smart Images
Figure CN122286469A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of atmospheric environmental monitoring, and in particular to a method for VOCs pollution characteristic analysis and source tracing based on deep learning. Background Technology
[0002] Volatile organic compounds (VOCs) are important precursors to ozone and fine particulate matter. Accurately identifying the sources of VOC pollution is crucial for developing scientific control measures. Current VOC source tracing methods mainly rely on chemical receptor models, such as the positive definite matrix factorization (PMF) model, the chemical mass balance (CMB) model, and principal component analysis (PCA).
[0003] However, the existing technology has the following main drawbacks: Limitations of the linear assumption: Traditional receptor models typically assume that the chemical composition ratio of VOCs emitted from pollution sources remains constant during transport (i.e., linear superposition), neglecting the complex photochemical reactions and secondary transformations of VOCs in the atmosphere, leading to larger resolution errors for long-distance transport or aging air masses.
[0004] Weak meteorological coupling capability: Existing models have difficulty deeply integrating meteorological factors such as wind speed, wind direction, temperature and humidity, and cannot accurately quantify the nonlinear impact of meteorological conditions on the diffusion and accumulation of pollutants.
[0005] Reliance on human experience: Models such as PMF rely heavily on the subjective experience of experts when determining the number of factors and identifying the physical meaning of factors, and are sensitive to outliers and missing values, with low automation.
[0006] Lack of adaptability: Existing models have fixed parameters after training and cannot adapt to the "concept drift" phenomenon caused by seasonal changes or the emergence of new pollution sources.
[0007] Therefore, there is an urgent need for an intelligent source tracing method that can capture the nonlinear chemical characteristics of VOCs, deeply integrate meteorological data, and possess physical interpretability. Summary of the Invention
[0008] To address the aforementioned technical problems in the existing technology, this invention proposes a deep learning-based method for VOCs pollution feature analysis and source tracing.
[0009] This invention proposes a deep learning-based method for VOCs pollution feature analysis and source tracing, including: S1: Acquire online VOCs monitoring time series data and synchronous meteorological parameter data of the target area, and preprocess them to construct a multidimensional input tensor; S2: Construct a deep neural network model that includes a spatiotemporal feature extraction module and a source resolution regression module; S3: Input the multidimensional input tensor into the deep neural network model, and extract the nonlinear coupling features and meteorological correlation features between VOCs components through the spatiotemporal feature extraction module; S4: Use the source analysis regression module to decode the extracted features and output the source component spectrum and time contribution rate sequence of each pollution source category; S5: Based on the source component spectrum and time contribution rate sequence, generate a VOCs pollution feature map and determine the main pollution sources.
[0010] In some specific embodiments, the preprocessing includes: removing outliers from the online monitoring time series data of VOCs and filling in missing data using linear interpolation; replacing data below the detection limit with half of the detection limit; and constructing a two-dimensional feature matrix by using a sliding window to extract VOCs concentration sequences and meteorological parameter sequences at a preset time step, which serves as the input to a deep neural network model.
[0011] In some specific embodiments, the spatiotemporal feature extraction module includes: A one-dimensional convolutional neural network, with convolutional kernels sliding along the species dimension, is used to extract co-variable chemical reaction features among different VOCs species; A bidirectional long short-term memory network was used to capture the long- and short-term dependencies and forward and backward temporal features of VOCs concentration evolution over time. The outputs of one-dimensional convolutional neural networks and bidirectional long short-term memory networks are fused through a concatenation operation.
[0012] In some specific embodiments, the one-dimensional convolutional neural network is configured with at least three convolutional kernels of different sizes, each corresponding to VOCs species components with different reactivity; convolution operations are performed on kernels of different sizes, and the output feature maps of each convolution operation are stacked in the channel dimension to capture the VOCs component variation features under different reaction rates.
[0013] In some specific embodiments, feature fusion specifically includes: mapping meteorological parameter data into meteorological weight vectors using a fully connected layer; and multiplying the meteorological weight vectors element-wise with the concatenated feature vectors to enhance the feature response values during periods when meteorological conditions have a significant impact on VOCs transmission and diffusion.
[0014] In some specific embodiments, the source apportionment regression module includes a Softmax activation function or a normalized exponential function to force the sum of the contribution rates of all pollution sources to equal 1; it also includes a non-negativity constraint to ensure that the concentration of each species and the contribution rate of each pollution source in the output source component spectrum are non-negative.
[0015] In some specific embodiments, the training process of the deep neural network model adopts a loss function based on physical information. The loss function consists of a reconstruction error term, a chemical fingerprint constraint term, and a sparsity penalty term. The chemical fingerprint constraint term is used to minimize the cosine distance between the source component spectrum output by the model and the preset industry standard source spectrum. The sparsity penalty term adopts L1 regularization to constrain the number of dominant pollution sources at each time point so that it conforms to the sparsity characteristics of environmental pollution source emissions.
[0016] In some specific embodiments, the training of the deep neural network model adopts a transfer learning strategy: the analytical results obtained by processing historical long-term monitoring data using a positive definite matrix factorization model are used as pseudo-labels to initialize the deep neural network model; and the pre-trained model parameters are fine-tuned and updated using manually labeled high-confidence pollution event sample data.
[0017] In some specific embodiments, the method also includes the step of updating the model online: periodically acquiring the latest VOCs monitoring data and meteorological data; determining the deviation between the distribution characteristics of the current monitoring data and the distribution of the training set data; and when the deviation exceeds a preset threshold, using the latest monitoring data to trigger the incremental learning process of the deep neural network model and update the model parameters.
[0018] In some specific embodiments, the VOCs online monitoring time series data includes alkanes, alkenes, aromatic hydrocarbons, halogenated hydrocarbons and oxygen-containing volatile organic compounds; meteorological parameters include: temperature, relative humidity, wind speed, wind direction and atmospheric pressure.
[0019] Compared with the prior art, the above-mentioned technical solution of the present invention has at least the following significant beneficial effects: By using deep neural networks, the nonlinear chemical reaction characteristics and complex meteorological coupling characteristics between VOCs were effectively extracted, overcoming the fundamental defects of traditional linear models in handling atmospheric photochemical transformation processes and significantly improving the accuracy of source apportionment.
[0020] By embedding physical constraints (non-negative, normalized) and prior knowledge of chemical fingerprints into the network structure and loss function, the output of the deep learning model has a clear environmental physical meaning, making it easy for environmental protection departments to apply directly.
[0021] By combining data cleaning, transfer learning, and online incremental update mechanisms, the system can maintain stable operation and accurate parsing capabilities even when faced with missing data, small sample sizes, and dynamic changes in environmental conditions, making it highly valuable for engineering applications. Attached Figure Description
[0022] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many anticipated advantages of the embodiments will be readily recognized as they become better understood through reference to the following detailed description. Other features, objects, and advantages of this application will become more apparent from reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of a deep learning-based VOCs pollution feature analysis and source tracing method according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer system used to implement the electronic device of the present application. Detailed Implementation
[0023] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0024] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0025] Figure 1 An exemplary system architecture 100 is shown that can be applied to the deep learning-based VOCs pollution feature analysis and source tracing method of the embodiments of this application.
[0026] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.
[0027] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various applications can be installed on terminal devices 101, 102, and 103, such as data processing applications, data visualization applications, and web browser applications.
[0028] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules (e.g., software programs or software modules used to provide distributed services) or as a single software program or software module. No specific limitations are imposed here.
[0029] Server 105 can be a server that provides various services, such as a background information processing server that supports the mapping table data displayed on terminal devices 101, 102, and 103. The background information processing server can process the obtained logical address and generate processing results.
[0030] It should be noted that the method provided in this application embodiment can be executed by server 105 or by terminal devices 101, 102, and 103. The corresponding device is generally set in server 105 or can be set in terminal devices 101, 102, and 103.
[0031] It should be noted that a server can be either hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software programs or software modules (such as software programs or software modules used to provide distributed services), or as a single software program or software module. No specific limitations are made here.
[0032] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.
[0033] Figure 2 A flowchart illustrating a deep learning-based method for VOCs pollution feature analysis and source tracing according to an embodiment of this application is shown. Figure 2 As shown, the method includes the following steps: S1: Obtain online VOCs monitoring time series data and synchronous meteorological parameter data of the target area, and perform preprocessing to construct a multidimensional input tensor.
[0034] In a specific embodiment, the VOCs online monitoring time-series data includes alkanes, alkenes, aromatic hydrocarbons, halogenated hydrocarbons, and oxygen-containing volatile organic compounds; meteorological parameters include temperature, relative humidity, wind speed, wind direction, and atmospheric pressure. This clarifies the physical dimensions of the input data, covering the main precursor species and key meteorological factors, ensuring the comprehensiveness and scientific rigor of the source tracing analysis.
[0035] In a specific embodiment, preprocessing includes: removing outliers from the online VOCs monitoring time-series data and filling in missing data using linear interpolation; for filling in missing values, if the concentration of a certain species at time t... Missing, use linear interpolation formula: Replace data below the detection limit with half of the detection limit; construct a two-dimensional feature matrix by using a sliding window to extract VOCs concentration sequences and meteorological parameter sequences at preset time steps. As input to the deep neural network model, T is the length of the time sliding window (e.g., T = 24 hours), N is the number of VOCs, and M is the number of meteorological parameters. Addressing common problems with dirty data in environmental monitoring (missing data, low detection rates, anomalies), interpolation and detection limit replacement strategies ensure data integrity; constructing a two-dimensional feature matrix through a sliding window preserves the temporal locality of the data, providing a standardized input for the deep learning model.
[0036] S2: Construct a deep neural network model that includes a spatiotemporal feature extraction module and a source resolution regression module.
[0037] In a specific embodiment, the spatiotemporal feature extraction module includes: A one-dimensional convolutional neural network, with convolutional kernels sliding along the species dimension, is used to extract co-variable chemical reaction features among different VOCs species; A bidirectional long short-term memory network was used to capture the long- and short-term dependencies and forward and backward temporal features of VOCs concentration evolution over time. The outputs of the one-dimensional convolutional neural network and the bidirectional long short-term memory network are fused through a concatenation operation. By using the aforementioned one-dimensional convolutional neural network and bidirectional long short-term memory network, two problems are solved: the one-dimensional convolutional neural network extracts chemical covariance features (i.e., source fingerprints) among different VOCs species, and the bidirectional long short-term memory network captures the dynamic evolution of concentration over time (i.e., time dependence), significantly improving the richness of feature extraction.
[0038] In a specific embodiment, the one-dimensional convolutional neural network is configured with at least three convolutional kernels of different sizes, each corresponding to VOCs species with different reactivity. Convolution operations are performed on kernels of different sizes, and the output feature maps of each convolution operation are stacked along the channel dimension to capture the VOCs component variation characteristics under different reaction rates. Considering the significant differences in the reactivity (lifetime) of different VOCs species in the atmosphere, the use of convolutional kernels of different sizes to capture the characteristics of fast-reacting species (such as isoprene) and slow-reacting species (such as ethane) improves the model's ability to characterize photochemical reaction processes.
[0039] In a specific embodiment, feature fusion includes: mapping meteorological parameter data into a meteorological weight vector using a fully connected layer; and multiplying the meteorological weight vector element-wise with the concatenated feature vector to enhance the feature response value during periods when meteorological conditions have a significant impact on VOCs transport and diffusion. This application innovatively introduces a meteorological weight vector to simulate the nonlinear modulation effect of meteorological conditions (such as wind dilution and static accumulation) on VOCs concentration in a real atmospheric environment, enabling the model to distinguish between concentration increases caused by emissions and data fluctuations caused by meteorological factors.
[0040] S3: Input the multidimensional input tensor into the deep neural network model, and extract the nonlinear coupling features and meteorological correlation features between VOCs components through the spatiotemporal feature extraction module.
[0041] In a specific embodiment, the training process of the deep neural network model employs a loss function based on physical information. This loss function consists of a reconstruction error term, a chemical fingerprint constraint term, and a sparsity penalty term. The chemical fingerprint constraint term minimizes the cosine distance between the source component spectrum output by the model and a preset industry-standard source spectrum. The sparsity penalty term uses L1 regularization to constrain the number of dominant pollution sources at each time point, ensuring it conforms to the sparsity characteristics of environmental pollution source emissions. The total loss function... , and These are the weight hyperparameters for the chemical fingerprint constraint term and the sparsity penalty term, respectively; the reconstruction error term. Where T represents the total number of time steps in the input time series, and t is the index of the current time step. For the observation data matrix, Let S be the contribution rate vector of each pollution source at time t, and let S be the source component spectrum matrix. The square of the Frobenius norm; chemical fingerprint constraint term K is the preset total number of pollution source categories; k represents the pollution source category index; The source component spectral vector of the k-th type of pollution source predicted by the model; The industry standard reference source spectrum vector representing the pre-defined k-th type of pollution source represents the cosine similarity. This represents the cosine similarity function; the smaller the value of this term, the more similar the predicted source spectrum is to the standard source spectrum; sparsity penalty term. , The L1 norm is represented by the contribution rate of the source. Applying L1 regularization induces a sparse contribution rate distribution in the model output, meaning that only a few dominant pollution sources are at play at a given time. The chemical fingerprint constraint ensures that the resolved source spectrum does not deviate from the actual industry emission characteristics; the sparsity penalty term conforms to the objective law that there are usually few dominant pollution sources in the real environment at the same time, reducing mathematical ambiguity.
[0042] In a specific embodiment, the training of the deep neural network model employs a transfer learning strategy: the analytical results obtained from processing historical long-term monitoring data using a positive definite matrix factorization model are used as pseudo-labels to initialize the deep neural network model; the pre-trained model parameters are then fine-tuned using manually labeled high-confidence pollution event sample data. By pre-training with long-accumulated data from traditional PMF models and then fine-tuning with a small amount of high-precision manually labeled data, the stability of traditional methods is inherited while leveraging the high-precision advantages of deep learning, significantly shortening the training time.
[0043] S4: Use the source analysis regression module to decode the extracted features and output the source component spectrum and time contribution rate sequence of each pollution source category.
[0044] In specific embodiments, the source apportionment regression module includes a Softmax activation function or a normalized exponential function to force the sum of the contribution rates of all pollution sources to equal 1; it also includes a non-negativity constraint to ensure that the concentrations of each species and the contribution rates of each pollution source in the output source composition spectrum are non-negative. Through Softmax (normalization) and non-negativity constraints, the model output is forced to conform to the basic physical laws of environmental source apportionment (i.e., the sum of contribution rates is 1 and the concentration is not negative), enhancing the reliability of the results.
[0045] S5: Based on the source component spectrum and time contribution rate sequence, generate a VOCs pollution feature map and determine the main pollution sources.
[0046] In a specific embodiment, the method further includes an online update step for the model: periodically acquiring the latest VOCs monitoring data and meteorological data; determining the deviation between the distribution characteristics of the current monitoring data and the distribution of the training set data; and when the deviation exceeds a preset threshold, triggering the incremental learning process of the deep neural network model using the latest monitoring data to update the model parameters. By automatically triggering incremental learning based on the deviation of the monitoring data distribution, the system can adapt to seasonal emission changes or sudden pollution events, extending the model's lifespan.
[0047] The following is for reference. Figure 3 It shows a schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application. Figure 3 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.
[0048] like Figure 3 As shown, the computer system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 302 or programs loaded from storage section 308 into random access memory (RAM) 303. RAM 303 also stores various programs and data required for the operation of system 300. CPU 301, ROM 302, and RAM 303 are interconnected via bus 304. Input / output (I / O) interface 305 is also connected to bus 304.
[0049] The following components are connected to I / O interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a liquid crystal display (LCD) and speakers, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card and a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to I / O interface 305 as needed. A removable medium 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 310 as needed so that computer programs read from it can be installed into storage section 308 as needed.
[0050] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the functions defined in the methods of this application. It should be noted that the computer-readable storage medium of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable storage medium other than a computer-readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0051] Computer program code for performing the operations of this application can be written in one or more programming languages or a combination thereof. Programming languages include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0052] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0053] The modules described in the embodiments of this application can be implemented in software or in hardware.
[0054] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: acquire online monitoring time-series data of Vocs in the target area and synchronous meteorological parameter data, and preprocess them to construct a multidimensional input tensor; construct a deep neural network model including a spatiotemporal feature extraction module and a source apportionment regression module; input the multidimensional input tensor into the deep neural network model, and extract the nonlinear coupling features and meteorological correlation features between Vocs components through the spatiotemporal feature extraction module; decode the extracted features using the source apportionment regression module, and output the source component spectrum and time contribution rate sequence of each pollution source category; based on the source component spectrum and time contribution rate sequence, generate a Vocs pollution feature map and determine the main pollution sources.
[0055] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.
Claims
1. A deep learning-based method for VOCs pollution feature analysis and source tracing, characterized in that, include: S1: Acquire online VOCs monitoring time series data and synchronous meteorological parameter data of the target area, and preprocess them to construct a multidimensional input tensor; S2: Construct a deep neural network model that includes a spatiotemporal feature extraction module and a source resolution regression module; S3: Input the multidimensional input tensor into the deep neural network model, and extract the nonlinear coupling features and meteorological correlation features between VOCs components through the spatiotemporal feature extraction module; S4: Use the source analysis regression module to decode the extracted features and output the source component spectrum and time contribution rate sequence of each pollution source category; S5: Based on the source component spectrum and time contribution rate sequence, generate a VOCs pollution feature map and determine the main pollution sources.
2. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 1, characterized in that, The preprocessing specifically includes: removing outliers from the online VOCs monitoring time series data and filling in missing data using linear interpolation; replacing data below the detection limit with half of the detection limit; and constructing a two-dimensional feature matrix by using a sliding window to extract VOCs concentration sequences and meteorological parameter sequences at a preset time step, which serves as the input to the deep neural network model.
3. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 1, characterized in that, The spatiotemporal feature extraction module includes: A one-dimensional convolutional neural network, with convolutional kernels sliding along the species dimension, is used to extract co-variable chemical reaction features among different VOCs species; A bidirectional long short-term memory network was used to capture the long- and short-term dependencies and forward and backward temporal features of VOCs concentration evolution over time. The outputs of the one-dimensional convolutional neural network and the bidirectional long short-term memory network are fused through a concatenation operation.
4. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 3, characterized in that, The one-dimensional convolutional neural network is configured with at least three convolutional kernels of different sizes, each corresponding to VOCs species components with different reactivity. Convolutional operations with convolutional kernels of different sizes are performed, and the output feature maps of each convolutional operation are stacked in the channel dimension to capture the VOCs component variation features under different reaction rates.
5. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 3, characterized in that, The feature fusion specifically includes: mapping the meteorological parameter data into a meteorological weight vector using a fully connected layer; and multiplying the meteorological weight vector element-wise with the concatenated feature vector to enhance the feature response value during periods when meteorological conditions have a significant impact on VOCs transmission and diffusion.
6. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 1, characterized in that, The source apportionment regression module includes a Softmax activation function or a normalized exponential function to force the sum of the contribution rates of all pollution sources to equal 1; it also includes a non-negativity constraint to ensure that the concentration of each species and the contribution rate of each pollution source in the output source component spectrum are non-negative.
7. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 1, characterized in that, The training process of the deep neural network model adopts a loss function based on physical information. The loss function consists of a reconstruction error term, a chemical fingerprint constraint term, and a sparsity penalty term. The chemical fingerprint constraint term is used to minimize the cosine distance between the source component spectrum output by the model and the preset industry standard source spectrum. The sparsity penalty term adopts L1 regularization to constrain the number of dominant pollution sources at each time point, so that it conforms to the sparsity characteristics of environmental pollution source emissions.
8. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 1, characterized in that, The training of the deep neural network model adopts a transfer learning strategy: the analytical results obtained by processing historical long-term monitoring data using a positive definite matrix factorization model are used as pseudo-labels to initialize the deep neural network model; and the pre-trained model parameters are fine-tuned and updated using manually labeled high-confidence pollution event sample data.
9. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 1, characterized in that, It also includes the step of updating the model online: periodically acquiring the latest VOCs monitoring data and meteorological data; determining the deviation between the distribution characteristics of the current monitoring data and the distribution of the training set data; when the deviation exceeds a preset threshold, using the latest monitoring data to trigger the incremental learning process of the deep neural network model and update the model parameters.
10. The deep learning-based VOCs pollution feature analysis and source tracing method according to claim 1, characterized in that, The online VOCs monitoring time series data includes alkanes, alkenes, aromatic hydrocarbons, halogenated hydrocarbons, and oxygen-containing volatile organic compounds; the meteorological parameters include temperature, relative humidity, wind speed, wind direction, and atmospheric pressure.