An environmental pollution risk assessment method based on multi-source data fusion

Abstract

The application discloses an environmental pollution risk assessment method based on multi-source data fusion, obtains remote sensing data, ground monitoring data, meteorological data, pollution source emission data and artificial inspection data, and constructs an environmental pollution observation data set; pollution risk evidence is extracted and fused by using an evidence reasoning method, a risk grade fusion support degree and a fusion conflict intensity are obtained; a pollution risk event and an event marker are generated according to the fusion result; the pollution risk event is input into a marker Hox process model, and the risk trigger intensity in a future time window is calculated in combination with the fusion conflict intensity; the pollution risk propagation direction, the propagation range and the risk duration are determined according to the risk trigger intensity; the environmental pollution risk grade is determined in combination with the risk trigger intensity, the risk duration and the pollution risk event marker, and early warning information is output when the early warning condition is met. The application realizes multi-source environmental data fusion analysis, risk evolution prediction, propagation path identification and dynamic early warning.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and environmental risk early warning technology, and in particular to an environmental pollution risk assessment method based on multi-source data fusion. Background Technology

[0002] With the continuous development of refined ecological and environmental supervision, dynamic control of pollution sources, and regional environmental risk early warning systems, environmental pollution risk assessment has become an important technical means for environmental governance and emergency management. Currently, environmental pollution risk assessment typically acquires environmental data through remote sensing monitoring, ground monitoring stations, meteorological monitoring, pollution source emission monitoring, and manual inspections. It then uses statistical analysis, threshold determination, or model calculation methods to identify and assess pollution risks, thereby providing a basis for decision-making in environmental supervision and pollution prevention and control.

[0003] Existing environmental pollution risk assessment technologies still have certain limitations in the collaborative utilization of multi-source heterogeneous data and the dynamic evolution analysis of risks. On the one hand, environmental data from different sources vary in terms of collection time, spatial resolution, data integrity, and reliability. Existing methods often employ simple weighted fusion or single-rule judgment for risk analysis, which is insufficient to effectively handle information conflicts and boundary uncertainties among multi-source data. This can easily lead to significant deviations in risk level assessment results, affecting the reliability of the assessment. On the other hand, most existing technologies focus on static judgments of the pollution state at the current moment, lacking continuous analysis of the formation process and subsequent evolution of pollution risk events. They struggle to combine pollution source characteristics, meteorological diffusion conditions, topographical influences, and spatial correlations to effectively predict the direction, scope, and duration of pollution risk propagation. Consequently, the timeliness and accuracy of risk warnings are limited, making it difficult to meet the application needs of dynamic environmental pollution risk assessment and early warning in complex environmental scenarios.

[0004] Therefore, how to provide an environmental pollution risk assessment method based on multi-source data fusion is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an environmental pollution risk assessment method based on multi-source data fusion. This method utilizes remote sensing monitoring, ground monitoring, meteorological monitoring, pollution source emission monitoring, evidence fusion reasoning, labeled Hawkes process modeling, and risk propagation analysis techniques to fuse multi-source environmental data, thereby completing pollution risk identification, risk event generation, risk propagation prediction, and dynamic early warning, and achieving a full-process assessment of environmental pollution risks.

[0006] An environmental pollution risk assessment method based on multi-source data fusion according to an embodiment of the present invention includes the following steps:

[0007] S1. Construct an environmental pollution observation dataset according to a unified spatial grid and a unified time window;

[0008] S2. Extract pollutant concentration, spatial grid number, collection time, data source credibility and risk level support from the environmental pollution observation dataset to generate pollution risk evidence units;

[0009] S3. Construct the DSmT evidence framework based on pollution risk evidence units to obtain the risk level fusion support and fusion conflict intensity of the corresponding spatial grid and time window.

[0010] S4. Spatial grids whose risk level fusion support exceeds the event generation threshold are identified as pollution risk events.

[0011] S5. Input the pollution risk event and pollution risk event label into the label Hawkes process model, and calculate the risk triggering intensity within the future time window based on the fusion conflict intensity correction event triggering gain and triggering attenuation coefficient.

[0012] S6. Determine the direction and range of pollution risk propagation based on the gradient change of risk trigger intensity in adjacent spatial grids, and determine the duration of risk based on the duration of risk trigger intensity exceeding the duration threshold within a continuous time window.

[0013] S7. Determine the environmental pollution risk level based on the risk trigger intensity, risk duration, and pollution risk event markers, and output early warning information when the risk trigger intensity exceeds the early warning threshold and the intensity of fusion conflict increases within at least two consecutive time windows.

[0014] Optionally, S1 includes:

[0015] S11. Acquire remote sensing data, ground monitoring data, meteorological data, pollution source emission data and manual inspection data of the target area within the monitoring period, and extract spatial coordinates, acquisition time and observation parameters respectively;

[0016] S12. Establish an initial spatial grid based on the boundary of the target area, and calculate the risk sensitivity weight of each initial spatial grid according to the location of the pollution source, the emission intensity of the pollution source, the location of the sensitive receptor, and the type of the sensitive receptor.

[0017] S13. Adjust the scale of the initial spatial grid according to the risk sensitivity weight, and divide the initial spatial grid with a higher risk sensitivity weight into smaller spatial grid units to obtain a unified spatial grid.

[0018] S14. Determine the uniform time window length based on the frequency of pollutant concentration changes, the frequency of pollutant source emission changes, and the frequency of meteorological diffusion changes, and divide the monitoring period into continuous time windows according to the uniform time window length.

[0019] S15. Map the remote sensing data to the corresponding spatial grid and time window according to the pixel center coordinates, and map the ground monitoring data to the corresponding spatial grid and time window according to the monitoring station coordinates and collection time.

[0020] S16. Map meteorological data to corresponding spatial grids and time windows according to meteorological monitoring location and collection time; map pollution source emission data to corresponding spatial grids and time windows according to emission source location and emission time; and map manual inspection data to corresponding spatial grids and time windows according to inspection location and inspection time.

[0021] S17. Using spatial grids and time windows as spatiotemporal indexes, remote sensing data, ground monitoring data, meteorological data, pollution source emission data, and manual inspection data mapped to the same spatiotemporal index are associated, and spatiotemporal integrity identifiers are generated based on the missing status of various types of data under the current spatiotemporal index.

[0022] S18. Combine the spatiotemporal index, remote sensing observation parameters, ground monitoring parameters, meteorological diffusion parameters, pollution source emission parameters, manual inspection parameters, and spatiotemporal integrity identifier into environmental pollution observation records, and form an environmental pollution observation dataset from all environmental pollution observation records.

[0023] Optionally, S2 includes:

[0024] S21. Read all observation records of the corresponding spatial grid and time window in the environmental pollution observation dataset;

[0025] S22. Calculate the weighted pollution value based on pollutant concentration and pollution source emission data, and determine the spread potential of pollutants within the current spatial grid by combining meteorological diffusion parameters, thus forming a risk-weighted concentration.

[0026] S23. Calculate the credibility of each data source based on the spatiotemporal integrity identifier, data source type, and the deviation between the collection time and the current time window, and associate it with the risk weighted concentration.

[0027] S24. Map the risk-weighted concentration to low-risk, medium-risk, and high-risk threshold ranges to generate preliminary risk level support.

[0028] S25. Adjust the initial risk level support based on manual inspection records and pollution source emission parameters to form the final risk level support.

[0029] S26. Combine the credibility of the data source with the support of the final risk level to form a risk evidence vector;

[0030] S27. For multiple data sources within the same spatial grid and time window, summarize the risk evidence vectors to generate a pollution risk evidence unit.

[0031] Optionally, S3 includes:

[0032] S31. For the same spatial grid number and the same time window number, read all risk evidence vectors within the pollution risk evidence unit, obtain the low-risk level support, medium-risk level support, high-risk level support and data source credibility corresponding to each data source, and multiply each risk level support with the data source credibility to obtain the credibility constraint support. Then, sum the credibility constraint support according to the three risk levels and normalize it to obtain the three-level normalized support of the data source.

[0033] S32. Establish a risk identification framework, assuming the set of basic risk propositions is... Low risk, medium risk, high risk The DSmT proposition space is defined by a superpower set that allows intersection and union operations; within this proposition space, two overlapping boundary propositions are fixedly generated, namely low-risk propositions. Medium risk and medium risk High risk, and low risk Medium risk is defined as only accepting low / medium boundary uncertainty. High-risk is limited to accepting only medium / high boundary uncertainties;

[0034] S33. Perform boundary attribution determination for each data source: Read the interval position of the risk-weighted concentration corresponding to the data source near the preset low / medium threshold and medium / high threshold.

[0035] If the risk-weighted concentration falls within the low / medium boundary buffer, then the boundary share of the medium-risk level normalized support of the data source is transferred to the low-risk level. Medium risk;

[0036] If the risk-weighted concentration falls within the medium / high boundary buffer, then the boundary share of the medium-risk level normalized support of this data source is transferred to the medium-risk level. High risk; the determination of the boundary share adopts the deterministic distance ratio rule, that is, the boundary share is equal to the inverse normalized result of the distance from the risk-weighted concentration to the two adjacent thresholds, and the support of each proposition is renormalized after the transfer;

[0037] S34. Write the basic trust allocation formed by each data source into a unified structured BBA table. The BBA table contains: spatial grid number, time window number, data source identifier, proposition identifier, proposition support, and data source credibility. Using the spatial grid number and time window number as keys, merge the BBA tables of multiple data sources under the same key into a set of evidence to be fused.

[0038] S35. For the evidence set to be fused, sequential conjunction and combination calculation is used: taking the BBA from any data source as the initial synthesis result, it is combined with the BBA from the remaining data sources one by one. During conjunction and combination, only the support of items with non-empty propositional intersection is accumulated, and the support of items with empty propositional intersection is accumulated to the conflict quantity. In the middle; among which the amount of conflict Accumulate according to the following formula:

[0039] ;

[0040] in, and Two data sources are respectively in the proposition Proposition Support level;

[0041] S36. In calculating the amount of conflict During the process, conflict attribution marking is performed on each conflicting product term, and the conflict is decomposed into boundary conflict quantities according to the following rules. Conflict quantity with abnormality :

[0042] If at least one of the conflicting product terms is of low risk Medium risk or medium risk If the risk is high, then the product term should include the boundary conflict amount. ;

[0043] If both propositions in a conflicting product term are basic risk propositions, and the credibility of the data source corresponding to either party is lower than a preset credibility threshold, then the product term is included in the abnormal conflict quantity. ;

[0044] The remaining conflict product terms are included in the boundary conflict quantity. ;

[0045] S37, Regarding the amount of border conflict Perform boundary-oriented reallocation:

[0046] Reassignment occurs only among fundamental risk propositions adjacent to the boundary-overlapping propositions, where when The source involves low risk When the risk level is medium, the risk level is redistributed only to low and medium risk levels.

[0047] when The source involves medium risk In high-risk situations, redistribution is only made to medium- and high-risk areas;

[0048] The redistribution ratio adopts the proportional rule of the support of the corresponding basic risk propositions in the synthesis result, and the support of the three basic risk propositions is normalized after redistribution;

[0049] S38. Regarding abnormal conflict quantities Perform abnormal isolation and reallocation: The entire process is transferred to unknown risk propositions, without assigning them to any basic risk propositions, and the unknown risk propositions are used as retained terms in the fusion result for normalization constraints;

[0050] S39. The support of the three basic risk propositions of low risk, medium risk and high risk after normalization is determined as the risk level fusion support of the spatial grid and time window, and the fusion conflict intensity is defined as the weighted combination result of boundary conflict quantity and abnormal conflict quantity, wherein the weight of boundary conflict quantity is fixed to be greater than the weight of abnormal conflict quantity.

[0051] S310 outputs the risk level fusion support and fusion conflict intensity of the corresponding spatial grid and time window.

[0052] Optionally, S4 includes:

[0053] S41. Read the fusion support and fusion conflict intensity of low-risk, medium-risk and high-risk levels corresponding to each spatial grid and time window, and determine the risk dominance level. The risk dominance level is the risk level with the highest fusion support.

[0054] S42. Determine the basic event generation threshold based on the risk-dominant level. When the risk-dominant level fusion support meets the following conditions: At that time, the corresponding spatial grid and time window are identified as support-triggered candidate pollution risk events;

[0055] S43. In addition to support-triggered candidate contamination risk events, perform conflict trigger determination: when the fusion conflict intensity meets the conflict threshold. Furthermore, when the fusion conflict intensity of the spatial grid in the previous time window and the fusion conflict intensity in the current time window satisfy a monotonically increasing relationship, the corresponding spatial grid and time window are identified as conflict-triggered candidate pollution risk events.

[0056] S44. Perform event confirmation for support-triggered candidate pollution risk events and conflict-triggered candidate pollution risk events: When both types of candidate events are established simultaneously, set the event type identifier to composite triggering type; when only one type of candidate event is established, set the event type identifier to the corresponding triggering type; and write the risk dominance level as the triggering risk level into the event record.

[0057] S45. Read the pollutant type information in the environmental pollution observation dataset that is consistent with the spatial grid and time window corresponding to the event, and generate pollutant type identifiers according to the pollutant category coding rules.

[0058] S46. Read the pollution source emission data and generate pollution source category identifiers according to the emission source industry category and emission process type. At the same time, generate emission intensity level identifiers according to the emission intensity range division results.

[0059] S47. Read meteorological diffusion parameters and generate meteorological diffusion condition identifiers according to preset diffusion classification rules; read target area topographic data and generate topographic barrier coefficients according to topographic barrier determination rules.

[0060] S48. Calculate the distance from the center point of the spatial grid to the nearest sensitive receptor and generate a sensitive receptor distance level label according to the distance interval division rule;

[0061] S49. Combine the spatial grid number, time window number, event type identifier, trigger risk level, risk dominance level fusion support, fusion conflict intensity, pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient, and sensitive receptor distance level identifier in a unified field order to form a pollution risk event marking structure, and bind it with the corresponding spatial grid and time window to generate pollution risk event records.

[0062] Optionally, S5 includes:

[0063] S51. Read the pollution risk event records, group them by spatial grid number, and arrange them in ascending order by time window number within each spatial grid to form an event sequence. ;

[0064] S52, For each event Constructing tag vectors The marker vector is generated by concatenating fixed fields in the following order: event type identifier, trigger risk level, pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient level identifier, and sensitive receptor distance level identifier. The marker vector is then used as the key for subsequent trigger gain lookup tables.

[0065] S53. Mapping the intensity of fusion conflicts to conflict levels. The conflict level is divided into three segments based on a fixed threshold: when the fusion conflict intensity is less than... Time to take When the fusion conflict intensity is greater than or equal to and less than Time to take When the fusion conflict intensity is greater than or equal to Time to take and will Write event ;

[0066] S54. Establish a dual-channel Hawkes structure, with a fixed channel set including support trigger channels. Conflict triggering channel Channel attribution follows a deterministic rule: only channels with an event type identifier indicating support-triggered behavior are included. When the event type is identified as conflict-triggered, only [the event] enters [the relevant area]. When the event type is identified as composite triggering, it is copied into two channel events and entered separately. and ;

[0067] S55. Define the risk trigger intensity for each spatial grid. The total intensity is obtained by summing the contributions from both channels, and the intensity expression is fixed as follows:

[0068] ;

[0069] in, Background intensity; For the event The ownership of the passage; The trigger gain is determined jointly by the marker vector and the channel; The trigger attenuation coefficient is determined by the conflict level;

[0070] S56, Constructing Trigger Gain The determination process adopts a deterministic rule of segmented table lookup + product binding:

[0071] Determine the base value of the level based on the trigger risk level. The base value of the level is derived from the preset discrete set. Values;

[0072] By tag vector The pollutant type identification and pollution source category identification determine the category coefficient. The category coefficients are read directly from the preset mapping table;

[0073] By channel Determine the channel coefficient The channel coefficients are from a preset discrete set. Values;

[0074] Determine the trigger gain as and to Apply upper bound Constraints, truncation occurs when the upper bound is exceeded. ;

[0075] S57, Constructing the trigger decay coefficient The determination process adopts a deterministic rule of conflict level-attenuation level table: a preset attenuation level table. And satisfy ,when Pick ,when Pick ,when Pick For compound-triggered events, the attenuation level is fixed down by one level within the conflict triggering channel, i.e., when... Pick ,when Pick ,when Still take

[0076] S58. Construct a parameter setting process, employing a deterministic solution using finite candidate set enumeration + log-likelihood selection: Preset candidate set... For Preset candidate set For Preset candidate set For Traverse within each spatial grid. The combination of events is used to calculate the log-likelihood value of the event sequence under the combination, and the combination with the largest log-likelihood value is selected as the final parameter value of the spatial grid.

[0077] S59. Based on the final parameter values, recursively calculate for each spatial grid within the future time window according to the time window step size. And the corresponding future time window As the risk trigger intensity output, the output result shall include at least the spatial grid number, the future time window number, and the risk trigger intensity value.

[0078] Optionally, S6 includes:

[0079] S61. Read the risk triggering intensity value of each spatial grid within each time window. Let the spatial grid number be... The time window number is Spatial grid In the time window The risk trigger intensity within is denoted as Simultaneously read the spatial grid In the time window Internal fusion conflict intensity value ;

[0080] S62, for each spatial grid Establish adjacency set Adjacency set To be compatible with spatial grids The set of all adjacent spatial grids sharing a boundary;

[0081] S63, in the same time window Inside, for spatial grids Each adjacent spatial grid in its adjacent set Calculate the coupling propagation discrimination value: First, calculate the difference in risk triggering intensity between the adjacent spatial grid and the current spatial grid, i.e., using minus Obtain the difference; then use the fusion conflict intensity value of adjacent spatial grids. The difference is amplified by multiplying the difference by one and adding the fusion conflict strength value to obtain the coupled propagation discrimination value for propagation determination.

[0082] S64. Among adjacent spatial grids with positive coupling propagation discriminant values, select the adjacent spatial grid with the largest coupling propagation discriminant value as the main propagation pointing grid, and then... The direction pointing to the grid of the main propagation is determined as the main propagation direction of pollution risk.

[0083] S65, using spatial grids Starting with a grid, propagation chain expansion is performed along the main propagation direction: when the risk triggering strength of the current grid is less than the propagation termination threshold. The propagation terminates when there are no adjacent grids with positive coupling propagation criteria. The entire set of spatial grids covered by the propagation chain is determined as the range of pollution risk propagation, where the propagation termination threshold is... This is the preset minimum propagation strength threshold;

[0084] S66, Set the duration threshold as The duration threshold To determine the risk trigger intensity threshold for the sustained risk status; within a continuous time window sequence, each window of the propagation chain starting grid and the propagation chain covering grid is checked to see if the risk trigger intensity is not less than the sustained threshold and the main propagation direction remains consistent in adjacent time windows. Time windows that simultaneously meet both conditions are marked as stable propagation time windows.

[0085] S67. Calculate the length of the longest continuous time window segment of the stable propagation time window, and determine the time length corresponding to this continuous time window segment as the risk duration.

[0086] S68. Output the main propagation direction of pollution risk, the propagation range of pollution risk, and the duration of risk corresponding to the output spatial grid number and time window number.

[0087] The beneficial effects of this invention are:

[0088] This invention constructs an environmental pollution observation dataset with a unified spatiotemporal index by acquiring remote sensing data, ground monitoring data, meteorological data, pollution source emission data, and manual inspection data. It then employs the DSmT evidence fusion method to jointly analyze the multi-source environmental data, achieving collaborative fusion and conflict identification of risk information from different data sources. Compared to traditional weighted averaging or single threshold determination methods, this approach effectively addresses the issues of credibility differences and information conflicts among multi-source data, improving the accuracy and reliability of environmental pollution risk assessment results.

[0089] This invention constructs a pollution risk event generation mechanism and incorporates conflict intensity into a labeled Hawkes process model to model and analyze the triggering relationships and evolution patterns of pollution risk events, enabling the prediction of risk trigger intensity within future time windows. By integrating information such as pollutant type, pollution source category, meteorological diffusion conditions, topographic barrier coefficient, and sensitive receptor distance into a unified modeling process, the ability to characterize the formation process and development trend of pollution risks is enhanced, improving the timeliness and accuracy of environmental pollution risk prediction. Attached Figure Description

[0090] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0091] Figure 1 This is a flowchart of an environmental pollution risk assessment method based on multi-source data fusion proposed in this invention;

[0092] Figure 2 This is a schematic diagram of the risk triggering intensity calculation process based on the marked Hawkes process proposed in this invention. Detailed Implementation

[0093] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0094] refer to Figure 1 - Figure 2 An environmental pollution risk assessment method based on multi-source data fusion includes the following steps:

[0095] S1. Acquire remote sensing data, ground monitoring data, meteorological data, pollution source emission data and manual inspection data of the target area, and construct an environmental pollution observation dataset according to a unified spatial grid and a unified time window;

[0096] S2. Extract pollutant concentration, spatial grid number, collection time, data source credibility and risk level support from the environmental pollution observation dataset to generate pollution risk evidence units;

[0097] S3. Construct the DSmT evidence framework based on pollution risk evidence units, identify conflicts and redistribute proportional conflicts for the risk level support corresponding to different data sources, and obtain the risk level fusion support and fusion conflict intensity of the corresponding spatial grid and time window.

[0098] S4. Spatial grids whose risk level fusion support exceeds the event generation threshold are identified as pollution risk events, and the pollutant type, pollution source category, meteorological diffusion conditions, topographic barrier coefficient, sensitive receptor distance, and fusion conflict intensity are written into the pollution risk event tag.

[0099] S5. Input the pollution risk event and pollution risk event label into the label Hawkes process model, and calculate the risk triggering intensity within the future time window based on the fusion conflict intensity correction event triggering gain and triggering attenuation coefficient.

[0100] S6. Determine the direction and range of pollution risk propagation based on the gradient change of risk trigger intensity in adjacent spatial grids, and determine the duration of risk based on the duration of risk trigger intensity exceeding the duration threshold within a continuous time window.

[0101] S7. Determine the environmental pollution risk level based on the risk trigger intensity, risk duration, and pollution risk event markers, and output early warning information when the risk trigger intensity exceeds the early warning threshold and the intensity of fusion conflict increases within at least two consecutive time windows.

[0102] In this embodiment, S1 includes:

[0103] S11. Acquire remote sensing data, ground monitoring data, meteorological data, pollution source emission data and manual inspection data of the target area within the monitoring period, and extract spatial coordinates, acquisition time and observation parameters respectively;

[0104] S12. Establish an initial spatial grid based on the boundary of the target area, and calculate the risk sensitivity weight of each initial spatial grid according to the location of the pollution source, the emission intensity of the pollution source, the location of the sensitive receptor, and the type of the sensitive receptor.

[0105] S13. Adjust the scale of the initial spatial grid according to the risk sensitivity weight, so that the initial spatial grid with a higher risk sensitivity weight is divided into smaller spatial grid units to obtain a unified spatial grid.

[0106] S14. Determine the uniform time window length based on the frequency of pollutant concentration changes, the frequency of pollutant source emission changes, and the frequency of meteorological diffusion changes, and divide the monitoring period into continuous time windows according to the uniform time window length.

[0107] S15. Map the remote sensing data to the corresponding spatial grid and time window according to the pixel center coordinates, and map the ground monitoring data to the corresponding spatial grid and time window according to the monitoring station coordinates and collection time.

[0108] S16. Map meteorological data to corresponding spatial grids and time windows according to meteorological monitoring location and collection time; map pollution source emission data to corresponding spatial grids and time windows according to emission source location and emission time; and map manual inspection data to corresponding spatial grids and time windows according to inspection location and inspection time.

[0109] S17. Using spatial grids and time windows as spatiotemporal indexes, remote sensing data, ground monitoring data, meteorological data, pollution source emission data, and manual inspection data mapped to the same spatiotemporal index are associated, and spatiotemporal integrity identifiers are generated based on the missing status of various types of data under the current spatiotemporal index.

[0110] S18. Combine the spatiotemporal index, remote sensing observation parameters, ground monitoring parameters, meteorological diffusion parameters, pollution source emission parameters, manual inspection parameters, and spatiotemporal integrity identifier into environmental pollution observation records, and form an environmental pollution observation dataset from all environmental pollution observation records.

[0111] In this embodiment, S2 includes:

[0112] S21. Read all observation records of the corresponding spatial grid and time window in the environmental pollution observation dataset, and extract the spatial grid number, time window number, pollutant concentration, meteorological diffusion parameters, pollution source emission parameters, manual inspection records, and spatiotemporal integrity identifier.

[0113] S22. Calculate the weighted pollution value based on pollutant concentration and pollution source emission data, and determine the spread potential of pollutants within the current spatial grid by combining meteorological diffusion parameters, thus forming a risk-weighted concentration.

[0114] S23. Calculate the credibility of each data source based on the spatiotemporal integrity identifier, data source type, and the deviation between the collection time and the current time window, and associate it with the risk weighted concentration.

[0115] S24. Map the risk-weighted concentration to low-risk, medium-risk, and high-risk threshold ranges to generate preliminary risk level support.

[0116] S25. Adjust the initial risk level support based on manual inspection records and pollution source emission parameters to form the final risk level support, which is used to reflect the risk judgment after the integration of multi-source data.

[0117] S26. Combine the data source credibility with the final risk level support to form a risk evidence vector. Each vector includes: spatial grid number, time window number, risk weighted concentration, data source credibility, and support for each risk level.

[0118] S27. For multiple data sources within the same spatial grid and time window, summarize the risk evidence vectors to generate a pollution risk evidence unit.

[0119] In this embodiment, S3 includes:

[0120] S31. For the same spatial grid number and the same time window number, read all risk evidence vectors within the pollution risk evidence unit, obtain the low-risk level support, medium-risk level support, high-risk level support and data source credibility corresponding to each data source, and multiply each risk level support with the data source credibility to obtain the credibility constraint support. Then, sum the credibility constraint support according to the three risk levels and normalize it to obtain the three-level normalized support of the data source.

[0121] S32. Establish a risk identification framework, assuming the set of basic risk propositions is... Low risk, medium risk, high risk The DSmT proposition space is defined by a superpower set that allows intersection and union operations; within this proposition space, two overlapping boundary propositions are fixedly generated, namely "low risk". "Medium risk" and "Medium risk" "High risk", and "low risk" "Medium risk" is limited to accepting only low / medium boundary uncertainties. "High risk" is limited to accepting only those with medium / high risk boundaries;

[0122] S33. Perform boundary attribution determination for each data source: Read the range of risk-weighted concentration corresponding to the data source near the preset low / medium threshold and medium / high threshold. If the risk-weighted concentration falls into the low / medium boundary buffer, transfer the boundary share in the medium-risk level normalized support of the data source to "low risk". "Medium risk"; if the risk-weighted concentration falls within the medium / high boundary buffer, then the boundary share of the normalized support for the medium risk level of this data source is transferred to "medium risk". "High risk"; the determination of the boundary share adopts the deterministic distance ratio rule, that is, the boundary share is equal to the inverse normalized result of the distance from the risk-weighted concentration to the two adjacent thresholds. After the transfer, the support of each proposition is renormalized so that each data source forms a unique basic trust allocation in the proposition space.

[0123] S34. Write the basic trust allocation formed by each data source into a unified structured BBA table. The BBA table shall contain at least: spatial grid number, time window number, data source identifier, proposition identifier, proposition support, and data source credibility. Using the spatial grid number and time window number as keys, merge the BBA tables of multiple data sources under the same key into a set of evidence to be fused.

[0124] S35. For the evidence set to be fused, sequential conjunction and combination calculation is used: taking the BBA from any data source as the initial synthesis result, it is combined with the BBA from the remaining data sources one by one. During conjunction and combination, only the support of items with non-empty propositional intersection is accumulated, and the support of items with empty propositional intersection is accumulated to the conflict quantity. In the middle; among which the amount of conflict Accumulate according to the following formula:

[0125] ;

[0126] in, and Two data sources are respectively in the proposition Proposition Support level;

[0127] S36. In calculating the amount of conflict During the process, conflict attribution marking is performed on each conflicting product term, and the conflict is decomposed into boundary conflict quantities according to the following rules. Conflict quantity with abnormality :

[0128] If at least one of the conflicting product terms states "low risk" Medium risk or medium risk If the risk is "high", then the product term will be included in the boundary conflict quantity. ;

[0129] If both propositions in a conflicting product term are basic risk propositions, and the credibility of the data source corresponding to either party is lower than a preset credibility threshold, then the product term is included in the abnormal conflict quantity. ;

[0130] The remaining conflict product terms are included in the boundary conflict quantity. ;

[0131] S37, Regarding the amount of border conflict Perform boundary-oriented reallocation: reallocate only between fundamental risk propositions adjacent to the boundary overlapping propositions, where when The source involves "low risk" When the risk level is "medium", it is only redistributed to low and medium risk levels; when The source involves "medium risk" When the risk level is "high", the risk level is redistributed only to medium and high risk levels. The redistribution ratio is based on the proportion of the support of the corresponding basic risk propositions in the composite result, and the support of the three basic risk propositions is normalized after redistribution.

[0132] S38. Regarding abnormal conflict quantities Perform abnormal isolation and reallocation: The entire process is transferred to unknown risk propositions, without assigning them to any basic risk propositions, and the unknown risk propositions are used as retained terms in the fusion result for normalization constraints;

[0133] S39. The support of the three basic risk propositions of low risk, medium risk and high risk after completion is determined as the risk level fusion support of the spatial grid and time window. The fusion conflict intensity is defined as the weighted combination result of boundary conflict quantity and abnormal conflict quantity. The weight of boundary conflict quantity is fixed to be greater than the weight of abnormal conflict quantity, so that the boundary uncertain conflict retains a higher influence in the subsequent risk evolution.

[0134] S310 outputs the risk level fusion support and fusion conflict intensity of the corresponding spatial grid and time window.

[0135] In this embodiment, S4 includes:

[0136] S41. Read the fusion support and fusion conflict intensity of low-risk, medium-risk and high-risk levels corresponding to each spatial grid and time window, and determine the risk dominance level. The risk dominance level is the risk level with the highest fusion support.

[0137] S42. Determine the basic event generation threshold based on the risk-dominant level. When the risk-dominant level fusion support meets the following conditions: At that time, the corresponding spatial grid and time window are identified as support-triggered candidate pollution risk events;

[0138] S43. In addition to support-triggered candidate contamination risk events, perform conflict trigger determination: when the fusion conflict intensity meets the conflict threshold. Furthermore, when the fusion conflict intensity of the spatial grid in the previous time window and the fusion conflict intensity in the current time window satisfy a monotonically increasing relationship, the corresponding spatial grid and time window are identified as conflict-triggered candidate pollution risk events.

[0139] S44. Perform event confirmation for support-triggered candidate pollution risk events and conflict-triggered candidate pollution risk events: When both types of candidate events are established simultaneously, set the event type identifier to composite triggering type; when only one type of candidate event is established, set the event type identifier to the corresponding triggering type; and write the risk dominance level as the triggering risk level into the event record.

[0140] S45. Read the pollutant type information in the environmental pollution observation dataset that is consistent with the spatial grid and time window corresponding to the event, and generate pollutant type identifiers according to the pollutant category coding rules.

[0141] S46. Read the pollution source emission data and generate pollution source category identifiers according to the emission source industry category and emission process type. At the same time, generate emission intensity level identifiers according to the emission intensity range division results.

[0142] S47. Read meteorological diffusion parameters and generate meteorological diffusion condition identifiers according to preset diffusion classification rules; read target area topographic data and generate topographic barrier coefficients according to topographic barrier determination rules.

[0143] S48. Calculate the distance from the center point of the spatial grid to the nearest sensitive receptor and generate a sensitive receptor distance level label according to the distance interval division rule;

[0144] S49. Combine the spatial grid number, time window number, event type identifier, trigger risk level, risk dominance level fusion support, fusion conflict intensity, pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient, and sensitive receptor distance level identifier in a unified field order to form a pollution risk event marking structure, and bind it with the corresponding spatial grid and time window to generate pollution risk event records.

[0145] In this embodiment, S5 includes:

[0146] S51. Read the pollution risk event records, group them by spatial grid number, and arrange them in ascending order by time window number within each spatial grid to form an event sequence. Each event At least include the time window number Event type identifier, trigger risk level, risk dominance level fusion support, fusion conflict intensity, pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient, and sensitive receptor distance level identifier;

[0147] S52, For each event Constructing tag vectors The marker vector is generated by concatenating fixed fields in the following order: event type identifier, trigger risk level, pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient level identifier, and sensitive receptor distance level identifier. The marker vector is then used as the key for subsequent trigger gain lookup tables.

[0148] S53. Mapping the intensity of fusion conflicts to conflict levels. The conflict level is divided into three segments based on a fixed threshold: when the fusion conflict intensity is less than... Time to take When the fusion conflict intensity is greater than or equal to and less than Time to take When the fusion conflict intensity is greater than or equal to Time to take and will Write event ;

[0149] S54. Establish a dual-channel Hawkes structure, with a fixed channel set including support trigger channels. Conflict triggering channel Channel attribution follows a deterministic rule: only channels with an event type identifier indicating support-triggered behavior are included. When the event type is identified as conflict-triggered, only [the event] enters [the relevant area]. When the event type is identified as composite triggering, it is copied into two channel events and entered separately. and ;

[0150] S55. Define the risk trigger intensity for each spatial grid. The total intensity is obtained by summing the contributions from both channels, and the intensity expression is fixed as follows:

[0151] ;

[0152] in, Background intensity; For the event The ownership of the passage; The trigger gain is determined jointly by the marker vector and the channel; The trigger attenuation coefficient is determined by the conflict level;

[0153] S56, Constructing Trigger Gain The determination process adopts a deterministic rule of "segmented table lookup + product binding":

[0154] Determine the base value of the level based on the trigger risk level. The base value of the level is derived from the preset discrete set. Values;

[0155] By tag vector The pollutant type identification and pollution source category identification determine the category coefficient. The category coefficients are read directly from the preset mapping table;

[0156] By channel Determine the channel coefficient The channel coefficients are from a preset discrete set. Values;

[0157] Determine the trigger gain as and to Apply upper bound Constraints, truncation occurs when the upper bound is exceeded. ;

[0158] S57, Constructing the trigger decay coefficient The determination process adopts a deterministic rule of "conflict level - attenuation level table": a preset attenuation level table. And satisfy ,when Pick ,when Pick ,when Pick For compound-triggered events, the attenuation level is fixed down by one level within the conflict triggering channel, i.e., when... Pick ,when Pick ,when Still take ;

[0159] S58. Construct a parameter setting process, using a deterministic solution of "finite candidate set enumeration + log-likelihood selection": preset candidate set For Preset candidate set For Preset candidate set For Traverse within each spatial grid. The combination of events is used to calculate the log-likelihood value of the event sequence under the combination, and the combination with the largest log-likelihood value is selected as the final parameter value of the spatial grid.

[0160] S59. Based on the final parameter values, recursively calculate for each spatial grid within the future time window according to the time window step size. And the corresponding future time window As the risk trigger intensity output, the output result shall include at least the spatial grid number, the future time window number, and the risk trigger intensity value.

[0161] In this embodiment, S6 includes:

[0162] S61. Read the risk triggering intensity value of each spatial grid within each time window. Let the spatial grid number be... The time window number is Spatial grid In the time window The risk trigger intensity within is denoted as Simultaneously read the spatial grid In the time window Internal fusion conflict intensity value The fusion conflict intensity value is used to characterize the degree of inconsistency among multi-source evidence;

[0163] S62, for each spatial grid Establish adjacency set Adjacency set To be compatible with spatial grids The set of all adjacent spatial grids sharing a boundary;

[0164] S63, in the same time window Inside, for spatial grids Each adjacent spatial grid in its adjacent set Calculate the coupling propagation discrimination value: First, calculate the difference in risk triggering intensity between the adjacent spatial grid and the current spatial grid, i.e., using minus Obtain the difference; then use the fusion conflict intensity value of adjacent spatial grids. The difference is amplified by multiplying the difference by one and adding the fusion conflict strength value to obtain the coupled propagation discrimination value for propagation-oriented decision-making. The formula for calculating the coupled propagation discrimination value is as follows:

[0165] ;

[0166] in, Representing a time window Inner grid With adjacent grid The discriminant value for coupling propagation between them;

[0167] S64. Among adjacent spatial grids with positive coupling propagation discriminant values, select the adjacent spatial grid with the largest coupling propagation discriminant value as the main propagation pointing grid, and then... The direction pointing to the grid of the main propagation is determined as the main propagation direction of pollution risk.

[0168] S65, using spatial grids Starting with a grid, propagation chain expansion is performed along the main propagation direction: when the risk triggering strength of the current grid is less than the propagation termination threshold. The propagation terminates when there are no adjacent grids with positive coupling propagation criteria. The entire set of spatial grids covered by the propagation chain is determined as the range of pollution risk propagation, where the propagation termination threshold is... This is the preset minimum propagation strength threshold;

[0169] S66, Set the duration threshold as The duration threshold The risk trigger intensity threshold is used to determine the duration of the risk. Within a continuous time window sequence, the starting grid and the covered grid of the propagation chain are checked window by window to determine whether the "risk trigger intensity is not less than the duration threshold" and the "main propagation direction is consistent in adjacent time windows". Time windows that meet both conditions are marked as stable propagation time windows.

[0170] S67. Calculate the length of the longest continuous time window segment of the stable propagation time window, and determine the time length corresponding to this continuous time window segment as the risk duration.

[0171] S68. Output the main propagation direction of pollution risk, the scope of pollution risk propagation, and the duration of risk corresponding to the spatial grid number and time window number.

[0172] In this embodiment, S7 includes:

[0173] S71, Read Spatial Grid Number In the current time window Risk trigger intensity within Duration of risk And the generated pollution risk event markers;

[0174] S72. Construct a baseline value for risk level determination. Based on the risk trigger intensity and risk duration, form a comprehensive risk value. The comprehensive risk value is calculated by multiplying the risk trigger intensity by a duration weighting coefficient. The duration weighting coefficient is determined by the interval to which the risk duration belongs. The calculation result is used as the input value for risk level classification. The comprehensive risk value is denoted as [missing value]. The duration weighting coefficient is The calculation formula is:

[0175] ;

[0176] S73, Setting a set of risk level classification thresholds When the overall risk value is less than When it is determined to be a low-risk level; when the comprehensive risk value is greater than or equal to and less than When it is determined to be at a medium risk level; when the comprehensive risk value is greater than or equal to and less than When it is determined to be a higher risk level; when the comprehensive risk value is greater than or equal to It was determined to be at a high-risk level at that time;

[0177] S74. Read the pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient, and sensitive receptor distance level identifier from the pollution risk event marker, and execute the risk correction rule on the determined risk level: when the pollutant type belongs to the high-risk pollutant category or the sensitive receptor distance level is in the closest range, the risk level is raised by one level.

[0178] S75, Set the warning threshold as follows When the risk trigger intensity within the current time window Greater than or equal to the warning threshold When the intensity of the fusion conflict shows a monotonically increasing state for at least two consecutive time windows, it is determined that the warning trigger condition is met.

[0179] S76. When the warning triggering conditions are met, a warning message is generated. The warning message shall include at least the spatial grid number, time window number, current risk level, risk trigger intensity value, risk duration, main propagation direction, and pollution risk event marker content.

[0180] S77. Output the generated early warning information to the early warning management module and record the early warning trigger time window and corresponding risk level for subsequent risk evolution tracking.

[0181] Example:

[0182] To verify the feasibility of the proposed environmental pollution risk assessment method based on multi-source data fusion, this invention was applied to an environmental pollution risk monitoring scenario in an industrial park. This industrial park houses various enterprises engaged in chemical production, surface treatment, warehousing and logistics, energy supply, and wastewater pretreatment. Pollution sources include stationary emission sources, intermittent emission sources, fugitive emission sources, and mobile emission sources, involving multiple types of pollutants such as volatile organic compounds, particulate matter, nitrogen oxides, sulfur dioxide, and nitrogen-containing pollutants. Sensitive receptors such as residential areas, schools, waterways, farmland, and ecological buffer zones are located around the park, requiring continuous monitoring of pollution emissions, meteorological diffusion, pollution transmission, and early warning responses by environmental management departments. Traditional monitoring methods mainly rely on fixed monitoring stations, online monitoring devices at enterprises, and manual inspection records. When pollution anomalies occur in grids with insufficient coverage of fixed stations, or when discrepancies exist between remote sensing monitoring, ground monitoring, pollution source emission data, and manual inspection records, problems such as delayed risk detection, unstable risk level assessment, and mismatch between the early warning range and the actual diffusion range easily arise.

[0183] In this scenario, the system first integrates remote sensing data, ground monitoring data, meteorological data, pollution source emission data, and manual inspection data, extracting spatial coordinates, acquisition time, and observation parameters for each type of data. The system establishes an initial spatial grid based on the boundaries of the monitored area and calculates risk sensitivity weights according to the location of pollution sources, emission intensity, location of sensitive receptors, and type of sensitive receptors. Smaller-scale grids are used for densely populated enterprise areas, areas adjacent to sensitive receptors, and areas with historically high anomalies, while larger-scale grids are used for ordinary production auxiliary areas, thus obtaining a unified spatial grid. The system then determines a unified time window based on the frequency of pollutant concentration changes, the frequency of pollution source emission changes, and the frequency of meteorological diffusion changes, mapping data from different sources to the corresponding spatial grids and time windows, forming environmental pollution observation records indexed by the spatial grids and time windows. Each environmental pollution observation record includes remote sensing observation parameters, ground monitoring parameters, meteorological diffusion parameters, pollution source emission parameters, manual inspection parameters, and a spatiotemporal integrity identifier.

[0184] The system reads all observation records within each spatial grid and time window, extracting pollutant concentrations, meteorological diffusion parameters, pollution source emission parameters, manual inspection records, and spatiotemporal integrity indicators. It calculates a weighted pollution value based on pollutant concentrations and pollution source emission data, and then combines this with parameters such as wind speed, wind direction, humidity, air pressure, and diffusion stability to determine the pollutant's propagation potential within the current grid, forming a risk-weighted concentration. For data sources with collection times close to the current time window, normal equipment status, and high historical stability, the system calculates a higher data source credibility. For data sources with missing data, collection delays, incomplete inspection descriptions, or abnormal equipment calibration, the system lowers the corresponding credibility. The system maps the risk-weighted concentration to low-risk, medium-risk, and high-risk threshold ranges, and adjusts the risk level support based on manual inspection records and pollution source emission parameters, generating a risk evidence vector containing the spatial grid number, time window number, risk-weighted concentration, data source credibility, and support for each risk level. Multiple risk evidence vectors within the same spatial grid and the same time window are aggregated into a pollution risk evidence unit.

[0185] In the multi-source data fusion stage, the system constructs a DSmT evidence framework based on pollution risk evidence units, using low-risk, medium-risk, and high-risk as basic risk propositions. Two overlapping boundary propositions, low-risk / medium-risk and medium-risk / high-risk, are set to accommodate uncertainties between adjacent risk levels. The system normalizes the risk level support and data source credibility of each data source after multiplying them, forming a basic trust allocation for the corresponding data source. When the risk-weighted concentration falls near the low-risk / medium-risk threshold, or near the medium-risk / high-risk threshold, the system transfers the corresponding boundary share to the corresponding overlapping boundary proposition. Subsequently, the system performs sequential conjunctive combination on multiple basic trust allocations within the same spatial grid and time window, and attributes the conflict volume resulting from an empty proposition intersection. Conflicts involving overlapping boundary propositions are included in the boundary conflict volume and are only proportionally redistributed between adjacent basic risk propositions; basic proposition conflicts caused by low-credibility data sources are included in the abnormal conflict volume and transferred to unknown risk propositions for isolation. Through this process, the system outputs the fusion support degree corresponding to low risk, medium risk, and high risk, and also outputs the fusion conflict intensity.

[0186] When the risk level fusion support of a spatial grid exceeds the event generation threshold, the system identifies the spatial grid and its corresponding time window as a pollution risk event. When the fusion conflict intensity exceeds the conflict threshold and increases within a continuous time window, the system identifies the spatial grid as a conflict-triggered candidate event. The system determines the event type based on the establishment of support triggering and conflict triggering, and writes the pollutant type, pollution source category, emission intensity level, meteorological diffusion conditions, topographic barrier coefficient, sensitive receptor distance level, risk dominance level fusion support, and fusion conflict intensity into the pollution risk event tag in a unified field order. Therefore, a pollution risk event not only includes information on whether it is abnormal, but also information on the type of pollutant, pollution source, diffusion conditions under which the abnormality occurred, proximity to sensitive receptors, and whether there are significant conflicts among multi-source evidence.

[0187] The system inputs pollution risk events and their labels into a labeled Hawkes process model, forming an event sequence according to spatial grid numbers. Each event is labeled with a vector based on its event type, trigger risk level, pollutant type, pollution source category, emission intensity level, meteorological diffusion conditions, topographic barrier coefficient, and sensitive receptor distance. The system maps fused conflict intensity to conflict levels and establishes support-triggered and conflict-triggered channels. Support-triggered events enter the support-triggered channel, conflict-triggered events enter the conflict-triggered channel, and composite-triggered events are replicated into two channels, each entering one channel. The system determines the trigger gain based on the trigger risk level, pollutant type, pollution source category, and channel type, and the trigger attenuation coefficient based on the conflict level. It then determines the model parameters through finite candidate set enumeration and log-likelihood optimization, recursively calculating the risk trigger intensity for each spatial grid within a future time window.

[0188] During the propagation analysis phase, the system reads the risk triggering intensity and fusion conflict intensity of each spatial grid and establishes an adjacency set for each spatial grid sharing its boundary. For the current grid and adjacent grids, the system calculates the difference between the risk triggering intensity of the adjacent grid and the risk triggering intensity of the current grid, and amplifies this difference using the fusion conflict intensity of the adjacent grids to obtain a coupling propagation discrimination value. The system selects the grid with the maximum value among the adjacent grids with positive coupling propagation discrimination values ​​as the main propagation direction grid and extends the propagation chain along this direction. When the risk triggering intensity of the current grid is lower than the propagation termination threshold or there is no positive coupling propagation discrimination value, the extension stops, and the set of grids covered by the propagation chain is determined as the pollution risk propagation range. The system then counts the longest continuous time period in which the risk triggering intensity of the starting grid and the covering grids of the propagation chain is not less than the duration threshold and the main propagation direction remains consistent within a continuous time window, and determines this as the risk duration.

[0189] During the risk level and early warning output phase, the system calculates a comprehensive risk value based on the risk trigger intensity and duration, and adjusts the risk level according to pollutant type, pollution source category, emission intensity level, meteorological diffusion conditions, topographic barrier coefficient, and sensitive receptor distance level. When the pollutant type belongs to the high-risk category or the sensitive receptor distance is within the closest range, the system raises the risk level by one level. When the risk trigger intensity exceeds the early warning threshold, and the fusion conflict intensity increases for at least two consecutive time windows, the system generates early warning information. The early warning information includes spatial grid number, current risk level, risk trigger intensity, risk duration, main propagation direction, and pollution risk event marker content, which is used by environmental management personnel to conduct pollution source verification, mobile monitoring retesting, enterprise emission control, and sensitive receptor protection.

[0190] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An environmental pollution risk assessment method based on multi-source data fusion, characterized in that, include: S1. Construct an environmental pollution observation dataset according to a unified spatial grid and a unified time window; S2. Extract pollutant concentration, spatial grid number, collection time, data source credibility and risk level support from the environmental pollution observation dataset to generate pollution risk evidence units; S3. Construct the DSmT evidence framework based on pollution risk evidence units to obtain the risk level fusion support and fusion conflict intensity of the corresponding spatial grid and time window. S4. Spatial grids whose risk level fusion support exceeds the event generation threshold are identified as pollution risk events. S5. Input the pollution risk event and pollution risk event label into the label Hawkes process model, and calculate the risk triggering intensity within the future time window based on the fusion conflict intensity correction event triggering gain and triggering attenuation coefficient. S6. Determine the direction and range of pollution risk propagation based on the gradient change of risk trigger intensity in adjacent spatial grids, and determine the duration of risk based on the duration of risk trigger intensity exceeding the duration threshold within a continuous time window. S7. Determine the environmental pollution risk level based on the risk trigger intensity, risk duration, and pollution risk event markers, and output early warning information when the risk trigger intensity exceeds the early warning threshold and the intensity of fusion conflict increases within at least two consecutive time windows.

2. The environmental pollution risk assessment method based on multi-source data fusion according to claim 1, characterized in that, S1 includes: S11. Acquire remote sensing data, ground monitoring data, meteorological data, pollution source emission data and manual inspection data of the target area within the monitoring period, and extract spatial coordinates, acquisition time and observation parameters respectively; S12. Establish an initial spatial grid based on the boundary of the target area, and calculate the risk sensitivity weight of each initial spatial grid according to the location of the pollution source, the emission intensity of the pollution source, the location of the sensitive receptor, and the type of the sensitive receptor. S13. Adjust the scale of the initial spatial grid according to the risk sensitivity weight, and divide the initial spatial grid with a higher risk sensitivity weight into smaller spatial grid units to obtain a unified spatial grid. S14. Determine the uniform time window length based on the frequency of pollutant concentration changes, the frequency of pollutant source emission changes, and the frequency of meteorological diffusion changes, and divide the monitoring period into continuous time windows according to the uniform time window length. S15. Map the remote sensing data to the corresponding spatial grid and time window according to the pixel center coordinates, and map the ground monitoring data to the corresponding spatial grid and time window according to the monitoring station coordinates and collection time. S16. Map meteorological data to corresponding spatial grids and time windows according to meteorological monitoring location and collection time; map pollution source emission data to corresponding spatial grids and time windows according to emission source location and emission time; and map manual inspection data to corresponding spatial grids and time windows according to inspection location and inspection time. S17. Using spatial grids and time windows as spatiotemporal indexes, remote sensing data, ground monitoring data, meteorological data, pollution source emission data, and manual inspection data mapped to the same spatiotemporal index are associated, and spatiotemporal integrity identifiers are generated based on the missing status of various types of data under the current spatiotemporal index. S18. Combine the spatiotemporal index, remote sensing observation parameters, ground monitoring parameters, meteorological diffusion parameters, pollution source emission parameters, manual inspection parameters, and spatiotemporal integrity identifier into environmental pollution observation records, and form an environmental pollution observation dataset from all environmental pollution observation records.

3. The environmental pollution risk assessment method based on multi-source data fusion according to claim 1, characterized in that, S2 includes: S21. Read all observation records of the corresponding spatial grid and time window in the environmental pollution observation dataset; S22. Calculate the weighted pollution value based on pollutant concentration and pollution source emission data, and determine the spread potential of pollutants within the current spatial grid by combining meteorological diffusion parameters, thus forming a risk-weighted concentration. S23. Calculate the credibility of each data source based on the spatiotemporal integrity identifier, data source type, and the deviation between the collection time and the current time window, and associate it with the risk weighted concentration. S24. Map the risk-weighted concentration to low-risk, medium-risk, and high-risk threshold ranges to generate preliminary risk level support. S25. Adjust the initial risk level support based on manual inspection records and pollution source emission parameters to form the final risk level support. S26. Combine the credibility of the data source with the support of the final risk level to form a risk evidence vector; S27. For multiple data sources within the same spatial grid and time window, summarize the risk evidence vectors to generate a pollution risk evidence unit.

4. The environmental pollution risk assessment method based on multi-source data fusion according to claim 1, characterized in that, S3 includes: S31. For the same spatial grid number and the same time window number, read all risk evidence vectors within the pollution risk evidence unit, obtain the low-risk level support, medium-risk level support, high-risk level support and data source credibility corresponding to each data source, and multiply each risk level support with the data source credibility to obtain the credibility constraint support. Then, sum the credibility constraint support according to the three risk levels and normalize it to obtain the three-level normalized support of the data source. S32. Establish a risk identification framework, assuming the set of basic risk propositions is... Low risk, medium risk, high risk The DSmT proposition space is defined by a superpower set that allows intersection and union operations; within this proposition space, two overlapping boundary propositions are fixedly generated, namely low-risk propositions. Medium risk and medium risk High risk, and low risk Medium risk is defined as only accepting low / medium boundary uncertainty. High-risk is limited to accepting only medium / high boundary uncertainties; S33. Perform boundary attribution determination for each data source: Read the interval position of the risk-weighted concentration corresponding to the data source near the preset low / medium threshold and medium / high threshold. If the risk-weighted concentration falls within the low / medium boundary buffer, then the boundary share of the medium-risk level normalized support of the data source is transferred to the low-risk level. Medium risk; If the risk-weighted concentration falls within the medium / high boundary buffer, then the boundary share of the medium-risk level normalized support of this data source is transferred to the medium-risk level. High risk; the determination of the boundary share adopts the deterministic distance ratio rule, that is, the boundary share is equal to the inverse normalized result of the distance from the risk-weighted concentration to the two adjacent thresholds, and the support of each proposition is renormalized after the transfer; S34. Write the basic trust allocation formed by each data source into a unified structured BBA table. The BBA table contains: spatial grid number, time window number, data source identifier, proposition identifier, proposition support, and data source credibility. Using the spatial grid number and time window number as keys, merge the BBA tables of multiple data sources under the same key into a set of evidence to be fused. S35. For the evidence set to be fused, sequential conjunction and combination calculation is used: taking the BBA from any data source as the initial synthesis result, it is combined with the BBA from the remaining data sources one by one. During conjunction and combination, only the support of items with non-empty propositional intersection is accumulated, and the support of items with empty propositional intersection is accumulated to the conflict quantity. In the middle; among which the amount of conflict Accumulate according to the following formula: ； in, and Two data sources are respectively in the proposition Proposition Support level; S36. In calculating the amount of conflict During the process, conflict attribution marking is performed on each product term that generates a conflict, and the conflict is decomposed into boundary conflict quantities according to the following rules. Conflict quantity with abnormality : If at least one of the conflicting product terms is of low risk Medium risk or medium risk If the risk is high, then the product term should include the boundary conflict amount. ; If both propositions in a conflicting product term are basic risk propositions, and the credibility of the data source corresponding to either party is lower than a preset credibility threshold, then the product term is included in the abnormal conflict quantity. ; The remaining conflict product terms are included in the boundary conflict quantity. ; S37, Regarding the amount of border conflict Perform boundary-oriented reallocation: Reassignment occurs only among fundamental risk propositions adjacent to the boundary-overlapping propositions, where when The source involves low risk When the risk level is medium, the risk level is redistributed only to low and medium risk levels. when The source involves medium risk In high-risk situations, redistribution is only made to medium- and high-risk areas; The redistribution ratio adopts the proportional rule of the support of the corresponding basic risk propositions in the synthesis result, and the support of the three basic risk propositions is normalized after redistribution; S38. Regarding abnormal conflict quantities Perform abnormal isolation and reallocation: The entire process is transferred to unknown risk propositions, without assigning them to any basic risk propositions, and the unknown risk propositions are used as retained terms in the fusion result for normalization constraints; S39. The support of the three basic risk propositions of low risk, medium risk and high risk after normalization is determined as the risk level fusion support of the spatial grid and time window, and the fusion conflict intensity is defined as the weighted combination result of boundary conflict quantity and abnormal conflict quantity, wherein the weight of boundary conflict quantity is fixed to be greater than the weight of abnormal conflict quantity. S310 outputs the risk level fusion support and fusion conflict intensity of the corresponding spatial grid and time window.

5. The environmental pollution risk assessment method based on multi-source data fusion according to claim 1, characterized in that, S4 includes: S41. Read the fusion support and fusion conflict intensity of low-risk, medium-risk and high-risk levels corresponding to each spatial grid and time window, and determine the risk dominance level. The risk dominance level is the risk level with the highest fusion support. S42. Determine the basic event generation threshold based on the risk-dominant level. When the risk-dominant level fusion support meets the following conditions: At that time, the corresponding spatial grid and time window are identified as support-triggered candidate pollution risk events; S43. In addition to support-triggered candidate contamination risk events, perform conflict trigger determination: when the fusion conflict intensity meets the conflict threshold. Furthermore, when the fusion conflict intensity of the spatial grid in the previous time window and the fusion conflict intensity in the current time window satisfy a monotonically increasing relationship, the corresponding spatial grid and time window are identified as conflict-triggered candidate pollution risk events. S44. Perform event confirmation for support-triggered candidate pollution risk events and conflict-triggered candidate pollution risk events: When both types of candidate events are established simultaneously, set the event type identifier to composite triggering type; when only one type of candidate event is established, set the event type identifier to the corresponding triggering type; and write the risk dominance level as the triggering risk level into the event record. S45. Read the pollutant type information in the environmental pollution observation dataset that is consistent with the spatial grid and time window corresponding to the event, and generate pollutant type identifiers according to the pollutant category coding rules. S46. Read the pollution source emission data and generate pollution source category identifiers according to the emission source industry category and emission process type. At the same time, generate emission intensity level identifiers according to the emission intensity range division results. S47. Read meteorological diffusion parameters and generate meteorological diffusion condition identifiers according to preset diffusion classification rules; read target area topographic data and generate topographic barrier coefficients according to topographic barrier determination rules. S48. Calculate the distance from the center point of the spatial grid to the nearest sensitive receptor and generate a sensitive receptor distance level label according to the distance interval division rule; S49. Combine the spatial grid number, time window number, event type identifier, trigger risk level, risk dominance level fusion support, fusion conflict intensity, pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient, and sensitive receptor distance level identifier in a unified field order to form a pollution risk event marking structure, and bind it with the corresponding spatial grid and time window to generate pollution risk event records.

6. The environmental pollution risk assessment method based on multi-source data fusion according to claim 1, characterized in that, S5 includes: S51. Read the pollution risk event records, group them by spatial grid number, and arrange them in ascending order by time window number within each spatial grid to form an event sequence. ; S52, For each event Constructing tag vectors The marker vector is generated by concatenating a fixed number of fields in the following order: event type identifier, trigger risk level, pollutant type identifier, pollution source category identifier, emission intensity level identifier, meteorological diffusion condition identifier, terrain barrier coefficient level identifier, and sensitive receptor distance level identifier. S53. Mapping the intensity of fusion conflicts to conflict levels. The conflict level is divided into three segments based on a fixed threshold: when the fusion conflict intensity is less than... Time to take When the fusion conflict intensity is greater than or equal to and less than Time to take When the fusion conflict intensity is greater than or equal to Time to take and will Write event ; S54. Establish a dual-channel Hawkes structure, with a fixed channel set including support trigger channels. Conflict triggering channel Channel attribution follows a deterministic rule: only channels with an event type identifier indicating support-triggered behavior are included. When the event type is identified as conflict-triggered, only [the event] enters [the relevant area]. When the event type is identified as composite triggering, it is copied into two channel events and entered separately. and ; S55. Define the risk trigger intensity for each spatial grid. The total intensity is obtained by summing the contributions from both channels, and the intensity expression is fixed as follows: ； in, Background intensity; For the event The ownership of the passage; The trigger gain is determined jointly by the marker vector and the channel; The trigger attenuation coefficient is determined by the conflict level; S56, Constructing Trigger Gain The determination process adopts a deterministic rule of segmented table lookup + product binding: Determine the base value of the level based on the trigger risk level. The base value of the level is derived from the preset discrete set. Values; By tag vector The pollutant type identification and pollution source category identification determine the category coefficient. The category coefficients are read directly from the preset mapping table; By channel Determine the channel coefficient Channel coefficients are derived from a preset discrete set. Values; Determine the trigger gain as and to Apply upper bound Constraints, truncation occurs when the upper bound is exceeded. ; S57, Constructing the trigger decay coefficient The determination process adopts a deterministic rule of conflict level-attenuation level table: a preset attenuation level table. And satisfy ,when Pick ,when Pick ,when Pick For compound-triggered events, the attenuation level is fixed down by one level within the conflict triggering channel, i.e., when... Pick ,when Pick ,when Still take S58. Construct a parameter setting process, employing a deterministic solution using finite candidate set enumeration + log-likelihood selection: Preset candidate set... For Preset candidate set For Preset candidate set For Traverse within each spatial grid. The combination of events is used to calculate the log-likelihood value of the event sequence under the combination, and the combination with the largest log-likelihood value is selected as the final parameter value of the spatial grid. S59. Based on the final parameter values, recursively calculate for each spatial grid within the future time window according to the time window step size. And the corresponding future time window As the risk trigger intensity output, the output results include the spatial grid number, the future time window number, and the risk trigger intensity value.

7. The environmental pollution risk assessment method based on multi-source data fusion according to claim 1, characterized in that, S6 includes: S61. Read the risk triggering intensity value of each spatial grid within each time window. Let the spatial grid number be... The time window number is Spatial grid In the time window The risk trigger intensity within is denoted as Simultaneously read the spatial grid In the time window Internal fusion conflict intensity value ; S62, for each spatial grid Establish adjacency set Adjacency set To be compatible with spatial grids The set of all adjacent spatial grids sharing a boundary; S63, in the same time window Inside, for spatial grids Each adjacent spatial grid in its adjacent set Calculate the coupling propagation discrimination value: First, calculate the difference in risk triggering intensity between the adjacent spatial grid and the current spatial grid, i.e., using minus Obtain the difference; then use the fusion conflict intensity value of adjacent spatial grids. The difference is amplified by multiplying the difference by one and adding the fusion conflict strength value to obtain the coupled propagation discrimination value for propagation determination. S64. Among adjacent spatial grids with positive coupling propagation discriminant values, select the adjacent spatial grid with the largest coupling propagation discriminant value as the main propagation pointing grid, and then... The direction pointing to the grid of the main propagation is determined as the main propagation direction of pollution risk. S65, using spatial grids Starting with a grid, propagation chain expansion is performed along the main propagation direction: when the risk triggering strength of the current grid is less than the propagation termination threshold. The propagation terminates when there are no adjacent grids with positive coupling propagation criteria. The entire set of spatial grids covered by the propagation chain is determined as the range of pollution risk propagation, where the propagation termination threshold is... This is the preset minimum propagation strength threshold; S66, Set the duration threshold as The duration threshold To determine the risk trigger intensity threshold for the sustained risk status; within a continuous time window sequence, each window of the propagation chain starting grid and the propagation chain covering grid is checked to see if the risk trigger intensity is not less than the sustained threshold and the main propagation direction remains consistent in adjacent time windows. Time windows that simultaneously meet both conditions are marked as stable propagation time windows. S67. Calculate the length of the longest continuous time window segment of the stable propagation time window, and determine the time length corresponding to this continuous time window segment as the risk duration. S68. Output the main propagation direction of pollution risk, the propagation range of pollution risk, and the duration of risk corresponding to the output spatial grid number and time window number.

Images