Pollution tracing system based on brown carbon optical characteristics and machine learning

Abstract

A pollution traceability system based on brown carbon optical characteristics and machine learning comprises an optical detection unit, a data preprocessing unit, a feature extraction unit, a machine learning analysis unit and a result output unit which are in data connection in sequence. The optical detection unit is used for carrying out on-line sampling on particulate matters in ambient air and acquiring brown carbon optical absorption and scattering data under a multi-wavelength condition; the data preprocessing unit performs background correction, noise suppression and environmental influence correction on the original data; the feature extraction unit constructs a multi-wavelength absorption feature, a spectrum slope feature and a time change feature based on the effective data; the machine learning analysis unit inputs the feature data into a pre-trained pollution source identification model, and outputs a pollution source type and a contribution analysis result; and the result output unit performs visual display, storage and external output. The system can realize automation, quantification and real-time traceability of the particulate matter pollution source, and is suitable for monitoring and management of a complex multi-source pollution environment.

Description

Technical Field

[0001] This invention belongs to the field of air pollution control technology, specifically referring to a pollution source tracing system based on the optical properties of brown carbon and machine learning. Background Technology

[0002] Atmospheric particulate matter pollution is a key concern in environmental science and pollution control. Its sources are complex, including anthropogenic sources such as industrial emissions and fossil fuel combustion, as well as natural sources such as biomass combustion and dust. Accurate identification and source tracing of particulate matter pollution sources are crucial prerequisites for developing targeted pollution control measures and implementing refined environmental management. However, existing pollution source tracing methods largely rely on chemical composition analysis, isotope analysis, or receptor model inversion, which typically require offline laboratory testing of collected samples. This process is time-consuming, costly, and fails to meet the practical needs of real-time monitoring and rapid decision-making.

[0003] In recent years, with the development of atmospheric optical monitoring technology, analytical methods based on the optical properties of particulate matter have gradually attracted attention. Brown carbon, as an organic aerosol component with significant light absorption characteristics, mainly originates from processes such as biomass combustion and secondary organic aerosol generation. It exhibits optical characteristics distinct from black carbon and inorganic particulate matter in the short-wave visible and near-ultraviolet bands. Related studies have shown that brown carbon from different pollution sources exhibits certain differences in spectral absorption intensity, absorption index, and temporal evolution characteristics, providing a potential information basis for pollution source identification. However, current applications of the optical properties of brown carbon are mostly limited to single-index analysis or qualitative judgment, making it difficult to achieve stable and accurate pollution source tracing in complex, multi-source environments.

[0004] Furthermore, with the continuous growth of environmental monitoring data, traditional analytical methods based on empirical rules or linear models are insufficiently adaptable when processing high-dimensional, multivariate optical data, and are easily affected by environmental interference factors, leading to unstable source tracing results. Therefore, how to fully explore the pollution source information contained in the optical properties of brown carbon, and combine it with advanced data analysis methods to improve the automation, accuracy, and real-time performance of pollution source tracing, remains a pressing technical problem to be solved in the current technology. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the present invention provides a pollution source tracing system based on the optical properties of brown carbon and machine learning, so as to at least partially solve the above-mentioned technical problems.

[0006] The technical solution adopted in this invention is as follows: This invention proposes a pollution source tracing system based on the optical properties of brown carbon and machine learning, including an optical detection unit, a data preprocessing unit, a feature extraction unit, a machine learning analysis unit, and a result output unit, with each unit sequentially connected to the other. The optical detection unit is configured to perform online sampling of particulate matter in ambient air and acquire optical absorption and optical scattering data of brown carbon in the particulate matter under different wavelength conditions to form raw detection data reflecting the optical properties of brown carbon. The data preprocessing unit is configured to perform at least background correction, noise suppression, environmental impact correction, and data labeling on the raw detection data. The system includes standardization and other processing to obtain effective optical data that meets the modeling requirements. The feature extraction unit is configured to construct a set of optical feature parameters for brown carbon based on the effective optical data. The set of optical feature parameters includes at least multi-wavelength absorption features, spectral slope features, and time-varying features. The machine learning analysis unit is configured to input the set of optical feature parameters into a pre-trained pollution source identification model and output pollution source type identification results and pollution source contribution analysis results corresponding to the current monitoring data, thereby achieving pollution source tracing. The result output unit is configured to visualize, store, and output the pollution source identification results and pollution source contribution analysis results.

[0007] Furthermore, the optical detection unit includes an air sampling component, a multi-wavelength light source component, and an optical signal detection component. The multi-wavelength light source component covers the ultraviolet to visible light band and is configured to emit incident light of different wavelengths to the sampled particles.

[0008] Furthermore, the optical signal detection component is configured to detect the absorption and scattering signals of particulate matter to incident light of different wavelengths to obtain the absorption and scattering coefficients of brown carbon under multiple wavelength conditions.

[0009] Furthermore, the data preprocessing unit includes a background correction module, a noise filtering module, a humidity correction module, and an abnormal data removal module. The data preprocessing unit is configured to eliminate the influence of environmental factors on optical detection data.

[0010] Furthermore, the feature extraction unit is configured to calculate the absorption index of brown carbon based on the absorption coefficient at different wavelengths and construct multi-wavelength absorption ratio features and spectral slope features.

[0011] Furthermore, the feature extraction unit is configured to extract dynamic change features from the time series of optical data, the dynamic change features including rate of change features and periodicity features.

[0012] Furthermore, the machine learning analysis unit incorporates at least one pollution source identification model, which is trained based on historical pollution sample data, and the historical pollution sample data is pre-labeled with pollution source types.

[0013] Furthermore, the pollution source identification model is a random forest model, a support vector machine model, a neural network model, or a combination of the above models.

[0014] Furthermore, the machine learning analysis unit is configured to output the category probability and pollution contribution ratio of each pollution source to achieve source tracing analysis of mixed pollution situations.

[0015] Furthermore, the result output unit includes a visualization display module and a data interface module. The visualization display module is configured to display the pollution source type and its changing trend, and the data interface module is configured to output the pollution source tracing results to an external environmental management system.

[0016] Compared with the prior art, the present invention has the following advantages: Based on the specific optical absorption and scattering characteristics of brown carbon under multi-wavelength conditions, a stable and repeatable acquisition of optical information of brown carbon in environmental particulate matter was achieved by constructing a multi-dimensional set of optical feature parameters and combining data preprocessing and feature extraction workflows. Compared with traditional pollution source tracing methods that rely on offline analysis of chemical components, this embodiment can complete continuous monitoring without damaging the sample structure, improving the timeliness and completeness of pollution source tracing data and reducing the impact of human intervention and experimental errors on the source tracing results.

[0017] By introducing a machine learning analysis unit and combining the set of optical feature parameters of brown carbon with a pre-trained pollution source identification model, this embodiment can automatically complete the identification of pollution source types and their contribution analysis, avoiding the subjective problems caused by traditional experience-based judgment and inference from a single indicator. It maintains high identification accuracy and robustness in complex, multi-source pollution scenarios, and is particularly suitable for pollution situations with high levels of brown carbon, such as biomass combustion and fossil fuel combustion, effectively improving the scientific rigor and reliability of pollution source tracing.

[0018] The system adopts a modular structure design with clear data interfaces between functional units, facilitating system expansion and practical deployment. The results output unit enables the visualization and external output of pollution source identification results, directly providing data support for environmental management, pollution early warning, and governance decisions. This reduces reliance on high-cost laboratory analysis conditions, demonstrating high engineering practical value and promising prospects for widespread application. Attached Figure Description

[0019] Figure 1This is a schematic diagram illustrating the working principle of the pollution source tracing system based on the optical properties of brown carbon and machine learning proposed in an embodiment of the present invention.

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a pollution source tracing system based on the optical properties of brown carbon and machine learning. This system is used to detect the optical properties of brown carbon carried by particulate matter in ambient air and, combined with a machine learning model, to identify pollution sources and analyze their pollution contribution. The system includes an optical detection unit, a data preprocessing unit, a feature extraction unit, a machine learning analysis unit, and a result output unit. These units are sequentially connected via wired or wireless means to form a complete data acquisition, processing, and analysis chain.

[0024] The optical detection unit is used to sample particulate matter in ambient air online and acquire optical property data of brown carbon. It includes an air sampling component, a multi-wavelength light source component, and an optical signal detection component.

[0025] The air sampling assembly includes a sampling pump, an air intake pipe, and a particulate matter sieving structure. It continuously introduces ambient air into the detection chamber and controls the particle size of the particles entering the chamber, ensuring that the target object is fine particulate matter suspended in the air. A multi-wavelength light source assembly is located on one side of the detection chamber. Its spectrum covers the ultraviolet to visible light range. The multi-wavelength light source assembly can emit multiple incident lights of different wavelengths sequentially or simultaneously. These wavelengths can be selected based on the significant absorption characteristics of brown carbon in the ultraviolet and short-wave visible light regions.

[0026] An optical signal detection component is positioned within or relative to the detection cavity to collect the absorption and scattering signals generated by particulate matter with incident light of different wavelengths. By detecting the absorption and scattering signals under different wavelength conditions, the absorption and scattering coefficients of brown carbon under multiple wavelength conditions can be obtained, thus forming raw detection data reflecting the optical properties of brown carbon.

[0027] The data preprocessing unit is connected to the optical detection unit and is used to process the raw detection data to obtain effective optical data that meets the requirements of subsequent modeling. The data preprocessing unit includes a background correction module, a noise filtering module, a humidity correction module, and an outlier removal module.

[0028] The system comprises several modules: a background correction module to eliminate the influence of the light source background signal and detector dark current on the optical data; a noise filtering module to filter out random noise introduced during the acquisition process; a humidity correction module to correct optical deviations caused by particulate matter moisture content under high humidity conditions based on synchronously acquired ambient humidity information; and an outlier removal module to identify and remove data points that significantly deviate from the normal range of variation. Through these processes, stable, continuous, and effective optical data suitable for modeling and analysis is obtained.

[0029] The feature extraction unit is used to construct a set of optical feature parameters for brown carbon based on the effective optical data. This feature extraction unit calculates the absorption index of brown carbon according to the absorption coefficient at different wavelengths, and further constructs multi-wavelength absorption ratio features and spectral slope features to characterize the absorption differences of brown carbon in different wavelength ranges.

[0030] Furthermore, the feature extraction unit analyzes the time series of optical data, extracting dynamic change features from continuously sampled data. These dynamic change features include the rate of change of the absorption coefficient over time and periodic change features. The set of optical feature parameters for brown carbon constructed in this way can comprehensively reflect the spectral characteristics of brown carbon and its behavior over time.

[0031] The machine learning analysis unit is connected to the feature extraction unit and has at least one built-in pollution source identification model. The pollution source identification model is trained based on historical pollution sample data, which is pre-labeled with corresponding pollution source types. The pollution source identification model can be a random forest model, a support vector machine model, a neural network model, or a combination of these models.

[0032] During system operation, the machine learning analysis unit inputs the real-time acquired set of brown carbon optical feature parameters into the pollution source identification model for inference calculations, and outputs the pollution source type identification result corresponding to the current monitoring data. Simultaneously, the machine learning analysis unit further outputs the category probability and pollution contribution ratio of each pollution source, thereby achieving source tracing analysis of multiple pollution sources in mixed pollution scenarios.

[0033] The result output unit is connected to the machine learning analysis unit and is used to visualize, store, and output the pollution source tracing results. The result output unit includes a visualization display module and a data interface module.

[0034] The visualization module is used to display the pollution source type, the proportion of pollution source contribution, and the trend information over time in a graphical manner; the data interface module is used to output the pollution source tracing results to the external environmental management system, monitoring platform, or decision support system in a standard data format to realize pollution monitoring and management applications.

[0035] During system operation, the optical detection unit continuously samples particulate matter in the ambient air online and acquires optical absorption and scattering data of brown carbon under different wavelength conditions; the data preprocessing unit corrects and modifies the raw detection data; the feature extraction unit constructs a set of optical feature parameters of brown carbon; the machine learning analysis unit completes pollution source identification and pollution contribution analysis based on the feature parameters; and finally, the result output unit displays, stores, and outputs the source tracing results, thereby completing pollution source tracing based on the optical properties of brown carbon and machine learning.

[0036] Example 2 This embodiment, based on specific embodiment 1, provides an implementation method for a pollution source tracing system based on the collaboration of mobile monitoring and edge computing. The overall system structure of this embodiment still includes an optical detection unit, a data preprocessing unit, a feature extraction unit, a machine learning analysis unit, and a result output unit. The data connection relationships and functional divisions between the units are consistent with those in embodiment 1. The difference is that the system uses a mobile monitoring platform as a carrier and sets up an edge computing module locally to achieve rapid pollution source tracing in complex environments.

[0037] In this embodiment, the optical detection unit is integrated inside a mobile monitoring platform, which can be a vehicle-mounted monitoring device, a drone monitoring device, or a portable mobile monitoring terminal. An air sampling component is located at the platform's air inlet for continuous ambient air sampling during movement. A multi-wavelength light source component and an optical signal detection component are housed within a sealed detection cavity, enabling stable multi-wavelength optical detection of particulate matter during movement. This allows for the acquisition of optical absorption and scattering data of brown carbon at different spatial locations and under different time conditions, forming corresponding raw detection data.

[0038] The data preprocessing unit and feature extraction unit are preferably deployed in the edge computing module within the mobile monitoring platform. The data preprocessing unit performs background correction, noise suppression, humidity correction, and outlier removal on the raw detection data to reduce the impact of vibration and airflow changes in the mobile environment on the optical detection results. The feature extraction unit calculates the absorption index of brown carbon based on the processed effective optical data and constructs multi-wavelength absorption ratio features, spectral slope features, and time-varying features. The time-varying features are further combined with the mobile trajectory information to form a set of dynamic feature parameters with spatial identification.

[0039] The machine learning analysis unit is also deployed in an edge computing module or a host computing platform connected to it. Its built-in pollution source identification model is trained based on historical pollution sample data. This historical pollution sample data not only includes optical feature parameters of brown carbon but also associates corresponding spatial locations and time labels. After receiving the set of feature parameters, the machine learning analysis unit outputs the pollution source type identification result corresponding to the current monitoring location, as well as the category probability and pollution contribution ratio of each pollution source, thereby realizing the source tracing analysis of the distribution of pollution sources within the mobile monitoring path.

[0040] The result output unit includes a local visualization display module and a remote data interface module. The local visualization display module is used to display the pollution source type, pollution contribution ratio, and their trend information with spatial location in real time on the mobile monitoring platform; the data interface module is used to upload the pollution source tracing results and location information to the external environmental management system to realize regional-scale analysis of the spatial distribution of pollution sources and assessment of pollution diffusion paths.

[0041] During the operation of this embodiment, the system conducts patrol monitoring in the target area through a mobile monitoring platform, collects real-time data on the optical characteristics of brown carbon, and completes data preprocessing, feature extraction and pollution source identification locally, ultimately achieving rapid and precise source tracing of mixed emissions from multiple pollution sources in complex areas.

[0042] Example 3 This embodiment provides an implementation method for a pollution source tracing system based on multi-site networking and cloud-based collaborative learning. The system as a whole still includes an optical detection unit, a data preprocessing unit, a feature extraction unit, a machine learning analysis unit, and a result output unit. The basic structure and function of each unit are consistent with those in Embodiment 1 and will not be described again. The difference in this embodiment is that the system adopts a multi-monitoring site network deployment and uses a cloud platform to uniformly manage and jointly model the data from multiple sites.

[0043] In this embodiment, the optical detection units are deployed at multiple fixed environmental monitoring stations, each located in a different functional area or pollution-characteristic area, including but not limited to industrial areas, areas around major traffic arteries, residential areas, and background stations. Each monitoring station is independently equipped with an air sampling component, a multi-wavelength light source component, and an optical signal detection component to continuously collect brown carbon optical absorption data and optical scattering data of particulate matter in the ambient air at that station, forming raw detection data with timestamps and station identifiers.

[0044] The data preprocessing unit and feature extraction unit are preferably located in the local computing module of each monitoring station. The data preprocessing unit performs background correction, noise filtering, humidity correction, and outlier removal on the raw detection data to eliminate the influence of different environmental conditions at different stations on the optical detection data. The feature extraction unit constructs a set of brown carbon optical feature parameters based on the preprocessed effective optical data. The set of feature parameters includes multi-wavelength absorption features, absorption index, spectral slope features, and time-varying features, and is uniformly converted into a standardized feature format.

[0045] The machine learning analysis unit is divided into a site-side analysis module and a cloud-based joint analysis module. The site-side analysis module is used to perform preliminary pollution source identification on locally collected optical feature parameters and output the pollution source type and its probability distribution for the corresponding time period at the site. The cloud-based joint analysis module is deployed on a remote server or cloud computing platform to aggregate feature parameters and preliminary identification results from multiple monitoring sites and to train and update the pollution source identification model based on historical pollution sample data from multiple sites.

[0046] In the cloud-based joint analysis module, the pollution source identification model is jointly trained by incorporating feature data from multiple sites, thereby enhancing the model's ability to identify regional pollution transmission, coordinated emissions from pollution sources, and temporal evolution patterns. After the model training is completed, the cloud distributes the updated model parameters or model version to each monitoring station, enabling the pollution source identification model to be periodically updated and optimized online.

[0047] The result output unit includes a site-level result display module and a regional-level result display module. The site-level result display module is used to display the pollution source type, pollution contribution ratio, and its trend over time for a single monitoring site. The regional-level result display module is used to perform spatial fusion analysis on the pollution source tracing results of multiple sites in the cloud platform, generate regional-scale pollution source distribution maps, pollution transmission trend maps, and pollution source contribution assessment results, and provide unified data services to the external environmental management system through a data interface.

[0048] During the operation of this embodiment, the system acquires optical characteristic data of brown carbon through continuous monitoring at multiple sites, and realizes dynamic updates of the pollution source identification model by relying on cloud-based joint modeling, thereby achieving long-term, stable and high-precision source tracing analysis of brown carbon-related pollution sources at the regional scale.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0050] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A pollution source tracing system based on the optical properties of brown carbon and machine learning, characterized in that: It includes an optical detection unit, a data preprocessing unit, a feature extraction unit, a machine learning analysis unit, and a result output unit, with each of these units connected in sequence. The optical detection unit is configured to perform online sampling of particulate matter in ambient air and acquire optical absorption data and optical scattering data of brown carbon in the particulate matter under different wavelength conditions to form raw detection data reflecting the optical properties of brown carbon. The data preprocessing unit is configured to perform at least background correction, noise suppression, environmental impact correction and data standardization on the raw detection data to obtain effective optical data that meets the modeling requirements. The feature extraction unit is configured to construct a set of optical feature parameters for brown carbon based on the effective optical data. The set of optical feature parameters includes at least multi-wavelength absorption features, spectral slope features, and time-varying features. The machine learning analysis unit is configured to input the set of optical feature parameters into a pre-trained pollution source identification model, and output pollution source type identification results and pollution source contribution analysis results corresponding to the current monitoring data, thereby realizing pollution source tracing. The result output unit is configured to visualize, store, and output the pollution source identification results and pollution source contribution analysis results.

2. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 1, characterized in that: The optical detection unit includes an air sampling component, a multi-wavelength light source component, and an optical signal detection component. The multi-wavelength light source component covers the ultraviolet to visible light band and is configured to emit incident light of different wavelengths to the sampled particles.

3. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 2, characterized in that: The optical signal detection component is configured to detect the absorption and scattering signals of particulate matter to incident light of different wavelengths to obtain the absorption and scattering coefficients of brown carbon under multiple wavelength conditions.

4. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 3, characterized in that: The data preprocessing unit includes a background correction module, a noise filtering module, a humidity correction module, and an abnormal data removal module. The data preprocessing unit is configured to eliminate the influence of environmental factors on optical detection data.

5. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 4, characterized in that: The feature extraction unit is configured to calculate the absorption index of brown carbon based on the absorption coefficient at different wavelengths and construct multi-wavelength absorption ratio features and spectral slope features.

6. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 5, characterized in that: The feature extraction unit is configured to extract dynamic change features from the time series of optical data, the dynamic change features including rate of change features and periodic features.

7. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 6, characterized in that: The machine learning analysis unit has at least one pollution source identification model built in. The pollution source identification model is trained based on historical pollution sample data, which is pre-labeled with pollution source types.

8. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 7, characterized in that: The pollution source identification model is a random forest model, a support vector machine model, a neural network model, or a combination of the above models.

9. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 8, characterized in that: The machine learning analysis unit is configured to output the category probability and pollution contribution ratio of each pollution source to achieve source tracing analysis of mixed pollution situations.

10. The pollution source tracing system based on the optical properties of brown carbon and machine learning according to claim 9, characterized in that: The result output unit includes a visualization display module and a data interface module. The visualization display module is configured to display the pollution source type and its changing trend, and the data interface module is configured to output the pollution source tracing results to an external environmental management system.

Images