A system and method for estimating the degree of soil pollution of a contaminated site

Abstract

The application discloses a kind of pollution site soil pollution degree estimation system and method, it is related to soil environmental monitoring technical field;Its technical points are: by analyzing the pollution deposition information recorded in soil profile, the continuous soil pollution state is disassembled into multiple independent historical emission events, the pollution diffusion path and influence range of each emission event can be clearly traced back;Thus, the complete history of site pollution from occurrence to evolution is restored, so that the repair scheme can formulate differentiated control measures for pollution from different sources;Secondly, the non-target pollutants accompanying emission are used as the basis for time calibration, the chemical substances originally regarded as interference factors are converted into accurate time scale, and the multi-dimensional information contained in the soil samples is fully tapped;By establishing the corresponding relationship between soil depth and pollution time, the historical pollution event is accurately dated, which lays a solid foundation for subsequent pollution impact quantification analysis.

Description

Technical Field

[0001] This invention relates to the field of soil environmental monitoring technology, specifically to a system and method for predicting the degree of soil pollution in contaminated sites. Background Technology

[0002] Currently, the industry generally adopts a technical approach consisting of grid sampling, laboratory testing, and spatial interpolation to predict the degree of soil pollution. Staff first set up sampling points of a certain density according to the site area and pollution type, collect soil samples at different depths at each sampling point, send them to the laboratory to test the concentration of the target pollutant, and then use spatial interpolation algorithms to expand the discrete sampling point data into a continuous site pollution distribution map.

[0003] However, during the implementation of the above technical solution, at least the following technical problems were discovered: First, this technical approach only provides a snapshot of the pollution distribution at the time of sampling, which cannot distinguish the pollution contributions of different historical emission events within the site, nor can it reconstruct the complete process of pollutants migrating and spreading in the soil over time. For industrial sites with a long history of use and multiple emission events, this static prediction result is not only difficult to support accurate pollution source tracing and liability determination, but may also lead to deviations in remediation plans due to the inability to identify deep-seated historical pollution. Secondly, during laboratory testing, a large number of non-target pollutants emitted during industrial production are directly excluded as interfering impurities. These chemical substances often have specific emission time windows, which also contain extremely valuable time information. However, existing technologies ignore the value of this information and cannot establish a correspondence between soil depth and the time of pollution occurrence. As a result, the role of soil profile as a natural historical archive is not fully utilized. Therefore, we provide a soil pollution degree prediction system and method for contaminated sites. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: A soil contamination level prediction system for contaminated sites, the system comprising: The basic information collection unit collects data on the geographical location, historical evolution, topography, surrounding environment, and historical emission events of the contaminated site. The soil sample acquisition unit acquires concentration data of target pollutants and non-target labeled pollutants in soil at different sampling points and at different depths. The data standardization unit performs format standardization, outlier removal, and missing value completion on the collected data to generate a standardized dataset. The time series fingerprint construction unit, based on a standardized dataset, generates a multi-pollutant time series fingerprint map that reflects the variation of pollutant concentration with depth at each sampling point; The event decomposition and source tracing unit divides the continuous multi-pollutant time series fingerprint spectrum into independent pollution event fingerprint segments, establishes a correspondence with historical emission event archives, and verifies the decomposition. The non-target labeled pollutant calibration unit establishes an absolute time scale using the known emission time of the non-target labeled pollutants to calibrate the occurrence time of the pollution events obtained by the decomposition. The three-dimensional pollution impact matrix construction unit calculates the pollution contribution of a single event in the entire space and depth of the site based on the calibrated pollution event parameters and soil environmental parameters, and constructs an event-level three-dimensional pollution impact matrix. The time-series overlay unit overlays the pollution contribution of all pollution events in order of occurrence from earliest to latest, corrects for the mutual influence between different events, obtains a comprehensive pollution index at any location and depth of the site, and generates a pollution distribution map and a forecast report including single event impact analysis.

[0005] Furthermore, the historical emission event archive records the start and end times, emission location coordinates, emission method, target pollutant type and emission amount, and accompanying non-target labeled pollutant type and emission amount for each independent emission event; The historical emission event archive also includes a feature vector for each emission event, which contains the concentration ratio, peak value, and rate of change of all target pollutants emitted during that event. The non-targeted pollutants are chemical substances that were emitted only during historical periods at the site, are chemically stable in the soil environment without degradation reactions, do not react chemically with soil components, and are not listed in the soil pollution control catalog.

[0006] Furthermore, the soil sample acquisition unit divides the soil into multiple continuous sampling layers according to the principle of shallow layer densification and deep layer loosening, and the thickness of the sampling layer increases in a gradient with the depth. At each sampling point, original soil samples from each layer are collected simultaneously from top to bottom, and the sample collection thickness does not exceed two-thirds of the thickness of the corresponding sampling layer. The test detects heavy metals, organic pollutants, and inorganic pollutants in the target pollutants. Heavy metals include lead, cadmium, mercury, chromium, arsenic, copper, zinc, and nickel; organic pollutants include polycyclic aromatic hydrocarbons, polychlorinated biphenyls, and petroleum hydrocarbons; and inorganic pollutants include cyanides and sulfides.

[0007] Furthermore, the time series fingerprint construction unit arranges the standardized pollutant concentration data of each layer from shallow to deep to form an original concentration sequence. By calculating the concentration change rate of all pollutants between adjacent depth layers, a concentration change rate sequence is generated. The original concentration sequence and the concentration change rate sequence are horizontally spliced ​​together to form a two-dimensional multi-pollutant time series fingerprint map, where the map rows correspond to pollutant types and the columns correspond to soil depth layers.

[0008] Furthermore, the event splitting and tracing unit uses a mutation boundary identification algorithm to locate the depth boundary between different pollution events in the concentration change rate sequence of the multi-pollutant time series fingerprint spectrum, and divides the multi-pollutant time series fingerprint spectrum into multiple non-overlapping pollution event fingerprint segments. Specifically, for each pollution event fingerprint segment, the pollutant concentration ratio, peak value, and rate of change features are extracted to form a feature vector, and the correspondence is determined by calculating the cosine similarity with the feature vectors of historical emission events.

[0009] Furthermore, the calibration process of the non-targeted contaminant calibration unit includes: First, identify the depth locations where the concentration of all non-target marker pollutants exceeds the detection limit in the multi-pollutant time series fingerprint spectrum, and use their corresponding known emission times as time anchors; The average soil deposition rate for the corresponding time period is calculated by the depth difference and time difference between two adjacent time anchor points, and then the precise occurrence time of each pollution event fingerprint segment is estimated. If the difference between the estimated time and the historical time exceeds a preset threshold, the feature vector of the fingerprint fragment of the pollution event will be re-extracted and similarity matching will be performed until the time difference does not exceed the preset threshold.

[0010] Furthermore, the three-dimensional pollution impact matrix construction unit divides the site horizontal space into square grids of equal size and the soil depth into vertical grids of equal thickness. For each calibrated pollution event, the pollution contribution of the event in each grid is calculated by combining its emission parameters, pollutant migration characteristics and soil interlayer barrier characteristics through a pollutant migration and diffusion simulation method. The pollution contributions of all single events together constitute an event-level three-dimensional pollution impact matrix, where the three dimensions of the matrix correspond to the horizontal X coordinate, the horizontal Y coordinate and the soil depth, respectively.

[0011] Furthermore, the time-series overlay unit sorts all pollution events from earliest to latest according to their occurrence time, and sequentially overlays the pollution contribution of the current pollution event into the total pollution contribution matrix; After each pollution event is superimposed, the emission intensity, pollutant type and soil permeability of the event are considered. If it is a non-polluting solvent emission, the dilution effect correction coefficient on all previously superimposed events is calculated. At the same time, the migration disturbance correction coefficient is calculated to dynamically correct the pollution contribution of previous events. The total pollution contribution of each grid in the total pollution contribution matrix is ​​multiplied by the toxicity weight coefficient of the corresponding pollutant to obtain the comprehensive pollution index of each grid. The comprehensive pollution index is divided into four levels: no pollution, light pollution, moderate pollution and heavy pollution, according to a preset value range.

[0012] A method for predicting the degree of soil contamination at a contaminated site, comprising the following steps: Collect basic information about the contaminated site and a complete archive of historical emission events. The archive of historical emission events contains the feature vector of each emission event. Soil sampling layers are divided according to the principle of shallow densification and deep desaturation to obtain the concentration data of target pollutants and non-target labeled pollutants in the soil at different sampling points and at different depths. The collected data is standardized to obtain a standardized dataset with uniform format and no outliers or missing values; Based on a standardized dataset, a multi-pollutant time series fingerprint map reflecting the variation of pollutant concentration with depth at each sampling point is constructed. Identify the deep boundaries between different pollution events in the time series fingerprint of multiple pollutants, split them into multiple independent pollution event fingerprint segments, and achieve pollution event source tracing through feature vector cosine similarity matching; Time anchors were established using known emission times of non-targeted labeled pollutants to calculate the average soil deposition rate and calibrate the occurrence times of all pollution events. The site space is divided into three-dimensional grids. Based on the calibrated pollution event parameters, the pollution contribution of each event in each grid is calculated through pollutant migration and diffusion simulation, and an event-level three-dimensional pollution impact matrix is ​​constructed. The pollution contribution of all pollution events is superimposed in order of occurrence from earliest to latest. Each superimposed event corrects its impact on all previous events, resulting in a comprehensive pollution index at any location and depth of the site. The index is then divided into four levels: unpolluted, lightly polluted, moderately polluted, and heavily polluted, according to a preset value range. Generate a three-dimensional soil pollution level distribution map and a detailed forecast report including the impact range, contribution ratio, and evolution process of a single event.

[0013] Furthermore, the method also includes the following steps: When new soil sample data is obtained or new historical emission event records are added, the process of splitting, tracing, and calibrating is only re-executed for pollution events affected by the new data or records. The pollution contribution of the corresponding event in the event-level three-dimensional pollution impact matrix is ​​updated, and the comprehensive pollution index of the affected area is recalculated to perform incremental dynamic iteration of the prediction results.

[0014] This invention provides a system and method for predicting the degree of soil pollution in contaminated sites, which has the following beneficial effects: By analyzing the contaminated deposition information recorded in soil profiles, the continuous state of soil pollution is decomposed into multiple independent historical emission events, enabling a clear tracing of the pollution diffusion path and impact range of each emission event. This reconstructs the complete history of site pollution from its occurrence to its evolution, providing direct scientific evidence for pollution source tracing and liability determination, and allowing remediation plans to be tailored to different pollution sources. Secondly, by using non-target pollutants emitted along with the pollution as a benchmark for time calibration, chemical substances that were originally considered confounding factors are transformed into precise time scales, fully exploring the multi-dimensional information contained in soil samples. By establishing the correspondence between soil depth and pollution time, precise dating of historical pollution events is achieved, laying a solid foundation for subsequent quantitative analysis of pollution impacts. Furthermore, in practical applications, accurate pollution distribution results can be obtained without relying on excessively high sampling densities, effectively reducing the workload of soil sample collection and testing, and shortening the site investigation cycle. At the same time, the incremental update mechanism can recalculate only the affected parts when new data or new records appear, ensuring the timeliness and accuracy of the prediction results. It is not only suitable for the assessment of conventional industrial contaminated sites, but can also be extended to multiple scenarios such as remediation effect verification and long-term environmental quality monitoring. Attached Figure Description

[0015] Figure 1 This is an overall flowchart of the present invention; Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Taking an abandoned organic chemical site as an example, this site mainly produces intermediates for nitrobenzene dyes and has experienced production leaks and multiple unrecorded wastewater discharges. The soil contains varying degrees of heavy metal and chlorinated organic pollution. Furthermore, because the traditional method combining grid sampling and Kriging interpolation can only generate a static snapshot of pollution at the time of sampling, it cannot distinguish the sources of pollution at different times. This resulted in the initial remediation plan being revised three times due to the omission of deep historical pollution, yet still failing to meet the acceptance criteria. Therefore, a targeted solution is proposed, the details of which are as follows: The site basic information survey mainly involves reviewing enterprise business registration information, production licenses, environmental impact assessment approval documents, daily production ledgers, raw material procurement and warehousing records, product sales and delivery records, quarterly environmental discharge declarations, annual environmental monitoring reports, accident handling and closure records, and factory layout maps from different periods. It also collects basic data on the site's geographical location, historical development, topography, hydrogeology, and surrounding environment. Based on this, a historical emission event archive, verified for completeness, is constructed and used as the sole matching benchmark for tracing pollution events. This addresses the problem of existing technologies using historical records only as supplementary references or even completely ignoring their role. The archive not only records the start and end times, emission location coordinates, emission method, target pollutant type and single emission volume, and accompanying non-targeted pollutant type and single emission volume for each independent emission event, but also pre-generates a corresponding standard feature vector for each emission event. This feature vector consists of the concentration ratio characteristics, peak concentration characteristics, and concentration change rate characteristics of all target pollutants emitted in that event. These features are unique identifiers that distinguish different emission events, just like a fingerprint. To ensure the accuracy of the records, a three-party cross-validation method is used for integrity verification. This involves cross-checking and verifying data from production records, environmental protection records, and environmental monitoring records to ensure consistency among the three sources. This guarantees the accuracy of historical emission event records from the source and solves the industry-wide problem of inaccurate historical records leading to biased subsequent assessments in existing technologies. The specific process is as follows: First, the raw material consumption in the production ledger is compared with the pollutant emissions in the environmental protection file. If the difference exceeds 20%, it is marked as a suspicious record. Then, the authenticity of the suspicious record is further verified by combining the changes in pollutant concentrations in the surrounding environmental monitoring data during the same period. Finally, the missing key information is supplemented by interviewing the original enterprise management personnel. For example, during the investigation, it was found that a tank leak was recorded as a leak of 2 tons in the production record, but the data from the surrounding groundwater monitoring wells showed that the nitrobenzene concentration at that time rose sharply from 0.1 mg / L to 3.2 mg / L, which did not match the simulated leak of 2 tons. After reviewing the equipment maintenance records and insurance company claims reports at that time, the leak was finally corrected to 5.8 tons.

[0018] Existing technologies generally treat non-target pollutants as interfering factors and exclude them or even deliberately mask their signals. We, however, take the opposite approach, using them as a core tool for precise time calibration. Therefore, based on a complete historical emission event archive, we screen chemical substances that meet four conditions as non-target labeled pollutants. The four screening conditions and their underlying principles are as follows: (1) Emissions are only made at the site during specific historical periods to ensure that the time of the substance’s appearance corresponds strictly to the time of the specific pollution event and is not confused with pollution from other periods; (2) It is chemically stable in the soil environment and does not undergo degradation reaction, that is, it needs a half-life of >50 years to ensure that the substance can exist stably for a long time and accurately record the time of occurrence of pollution events. (3) It does not react chemically with soil components, that is, the adsorption partition coefficient Kd < 0.1 L / kg, so as to avoid changes in the concentration of substances due to adsorption, complexation and other chemical reactions, which would affect the accuracy of time calibration; (4) Not included in the National Soil Pollution Control List, ensuring that the substance will not be routinely detected as a target pollutant, thereby avoiding confusion with the signals of target pollutants. Among them, for abandoned organic chemical sites, the assessment finally identified three non-target marker pollutants: 1,1,2-trichloroethane used as a solvent, carbon tetrachloride used as an extractant, and 1,2-dichloropropane used as a reaction intermediate. The half-life of these three substances in the soil is more than 50 years, and they do not react with organic matter or clay minerals in the soil. They can exist stably in the soil for decades, which are time markers.

[0019] Soil sampling layers were divided according to the principle of shallow layer densification and deep layer thinning, with the thickness of the sampling layer increasing gradually with depth. Specifically, ten consecutive sampling layers were defined: 0-0.2m, 0.2-0.5m, 0.5-1m, 1-1.5m, 1.5-2m, 2-3m, 3-4m, 4-6m, 6-10m, and 10-15m. This division method ensures high resolution for shallow (0-2m) contamination while effectively reducing the workload and cost of deep (below 6m) sampling, thus resolving the contradiction in existing technologies where uniform sampling layer thickness leads to insufficient shallow resolution or excessively high deep costs. This was then combined with site history... The functional areas were distributed as follows: sampling points were densely arranged in a 5m×5m grid in key polluted areas such as the original production workshop, raw material warehouse, wastewater treatment plant, and sewage outlet; and sampling points were arranged in a 20m×20m grid in non-key areas such as office area and green belt. A direct-push drilling rig was used to simultaneously collect undisturbed soil samples from each layer from top to bottom at each sampling point. The sample collection thickness was controlled between 1 / 2 and 2 / 3 of the thickness of the corresponding sampling layer. If the collection thickness is too large, it will cause soil from different depth layers to mix and destroy the integrity of the polluted deposition sequence; if the collection thickness is too small, the sample will not be representative enough and will not accurately reflect the pollution status of that depth layer.

[0020] During sampling, a field blank sample is inserted for every 10 samples collected, and a transport blank sample is inserted for every 20 samples collected, to monitor cross-contamination during sampling and transportation. Collected soil samples are immediately sealed in brown glass bottles, stored in a refrigerated box, and sent to a CMA-accredited laboratory for testing within 24 hours. Laboratory testing strictly adheres to the "Technical Specification for Soil Environmental Monitoring" (HJ / T). (166-2004), each batch of samples includes 10% parallel samples, 1 laboratory blank, and 1 standard substance. The parallel samples are controlled with a relative deviation of ≤20%. The detection indicators cover the following: 8 heavy metals (lead, cadmium, mercury, chromium, arsenic, copper, zinc, and nickel); 3 organic pollutants (polycyclic aromatic hydrocarbons, polychlorinated biphenyls, and petroleum hydrocarbons); 2 inorganic pollutants (cyanide and sulfide); and the above 3 non-target labeled pollutants. Heavy metals are detected by inductively coupled plasma mass spectrometry (ICP-MS), organic pollutants are detected by gas chromatography-mass spectrometry (GC-MS), inorganic pollutants are detected by spectrophotometry, and non-target labeled pollutants are detected by gas chromatography-electron capture detector (GC-ECD).

[0021] After data collection, all collected data underwent standardization. First, data from different sources and formats, such as Excel spreadsheets, PDF reports, and raw laboratory data files, were converted into a unified JSON structured data format. A unified pollutant coding system and unit system were established, unifying the unit for all pollutant concentrations to mg / kg and the date format to YYYY-MM-DD. Then, a combination of logarithmic transformation and Grubbs' test was used to identify and remove outliers. This method is suitable for the natural log-normal distribution of soil pollutant concentrations. The specific process is as follows: First, the natural logarithmic transformation of all pollutant concentration values ​​is performed to convert the skewed distribution into an approximately normal distribution. Then, the mean (μ) and standard deviation (σ) of the transformed data are calculated, and the critical value G at a 95% confidence level is obtained by consulting the Grubbs critical value table. 95 If the absolute value of the difference between a certain detected value and the average value is greater than G 95 If the value is ×σ, it is considered a suspected outlier. All suspected outliers must be manually reviewed and verified by combining historical site leakage records, sampling point locations, and original laboratory test records. Outliers confirmed to be caused by sampling or testing errors are removed. Outliers confirmed to be true high-contamination points are retained and highlighted in subsequent analyses, mainly targeting sample data that were significantly contaminated during the sampling process. For missing values, they are handled according to data type: First, for missing target pollutant concentrations, the missing sampling point data is supplemented using the inverse distance weighted average of data from the same depth layer at three adjacent sampling points; the missing depth layer data is supplemented using the linear interpolation result of the data from two adjacent depth layers at the same sampling point. Second, for missing non-target markers, linear interpolation is used only when the marker is detected at the same depth layer at three adjacent sampling points; if there are undetected cases at adjacent points, they are marked as undetected and do not participate in the time anchor calculation to avoid generating false time markers. After generating a standardized dataset, it is synchronously distributed to all subsequent computing units through a message queue mechanism based on RabbitMQ. This ensures that all units use the same basic data for computation, avoiding computational errors caused by data inconsistency. As a result, asynchronous data distribution and parallel computing are achieved, which greatly improves the system's operating efficiency.

[0022] Based on standardized datasets, a multi-pollutant time-series fingerprint spectrum is constructed for each soil sampling point. This concept originates from sedimentary sequence analysis in geology, and its significance lies in digitizing the pollutant deposition information in the soil profile to form a quantifiable and analyzable feature spectrum, avoiding the inability to distinguish between different historical pollution events. The specific process is as follows: First, arrange the standardized pollutant concentration data of each layer into a one-dimensional original concentration sequence in order from shallow to deep, with the sequence length equal to the number of sampling layers; then calculate the concentration change rate of all pollutants between two adjacent depth layers. Use a segmented calculation method to avoid division by zero error and interference from extremely low concentrations. Two cases are required: (1) Upper layer concentration < method detection limit (MDL); then if the lower layer concentration ≥ MDL, the change rate = 1.0 (marked as the new pollution initiation boundary); if the lower layer concentration < MDL, the change rate = 0.0; (2) Upper layer concentration ≥ method detection limit; then the change rate = min((lower layer concentration - upper layer concentration) / upper layer concentration, 2.0), where the method detection limit (MDL) is the lowest detectable concentration of the laboratory detection method; an upper limit of 2.0 is set for the rate of change to avoid spurious abrupt changes caused by extremely low upper layer concentrations; this approach retains the sensitivity of the relative rate of change to the boundary, solves the computational collapse problem of the uncontaminated layer, and suppresses extreme value interference; furthermore, the lower layer concentration refers to the standardized concentration of contaminants in the deeper soil layer at the current sampling location, in mg / kg; the upper layer concentration refers to the standardized concentration of contaminants in the current shallow soil layer, also in mg / kg. g, as the calculation baseline, represents the absolute concentration difference between two adjacent pollutant layers, distinguishing between increasing and decreasing trends. The overall fractional structure transforms the absolute difference into a relative rate of change, eliminating interference from background pollutant concentrations and ensuring horizontal comparability of the variation amplitudes of different pollutants and sampling points. This indicator amplifies the trend of pollutant concentration changes with depth, highlighting its significance in identifying the boundaries between pollution events. While the concentrations of adjacent pollution events may not differ significantly, the rate of change can exhibit a marked abrupt change, which cannot be achieved using only concentration data. Finally, the original concentration sequence and the rate of change sequence are horizontally concatenated to generate a two-dimensional multi-pollutant time series fingerprint map. Rows correspond to different pollutant types, and columns correspond to different soil depths. Soil is a natural sedimentary system; pollutants emitted at different times gradually migrate downwards through atmospheric deposition, surface runoff, and groundwater infiltration, leaving traces in the soil at different depths, forming pollution rings similar to tree rings. The multi-pollutant time series fingerprint map is a digital record of these pollution rings, comprehensively reflecting the variation of pollutant concentrations with depth.

[0023] After the fingerprint map was constructed, a sliding window extreme value detection algorithm was used to locate the depth boundaries between different pollution events in the concentration change rate sequence. The sliding window length was dynamically calculated based on the average deposition rate of the site to ensure that pollution events with an interval of more than one year could be distinguished. The calculation formula was: Window length = rounded up (1.5 / average deposition rate / average sampling layer thickness), where 1.5 is the time resolution coefficient (corresponding to 1.5 years), and the average sampling layer thickness is the arithmetic mean of the thicknesses of all sampling layers. The sliding step size was fixed at one depth layer to ensure the continuity of boundary identification. The correction formula was: Window length = max(3, rounded up (1.5 / (average sampling layer thickness / average deposition rate)))) to ensure that the window length was not less than 3 to avoid misjudgment. If the window was too short, the misjudgment rate would increase, and normal concentration fluctuations would be identified as event boundaries. If the window was too long, the resolution would decrease, and it would be impossible to distinguish between adjacent pollution events.

[0024] The average rate of change of all pollutant concentrations within each window is calculated sequentially. When the average value of a window exceeds a preset abrupt change threshold, the depth location corresponding to that window is determined as the boundary of a pollution event. The abrupt change threshold is dynamically adjusted based on the site deposition rate and the frequency of pollution events, with an initial threshold set to 0.5. If the number of events after segmentation deviates from the number of historical records by more than 20%, the threshold is adjusted in increments of 0.1 until the deviation is less than 20%. Using these boundaries as limits, the fingerprint spectrum of each sampling point is segmented into multiple continuous and non-overlapping pollution event fingerprint segments. After segmentation, the integrity of the segmentation is verified using a pollutant-specific threshold to determine the continuity of the boundaries, avoiding the irrationality of a uniform threshold. The verification standard is that all pollution event fingerprint segments must completely cover the entire... The soil depth range is 0-15m, with no blank intervals. At the boundary between two adjacent segments, the absolute value of the pollutant concentration change rate for more than 70% is less than its own continuity threshold. In addition, the continuity threshold of a single pollutant is calculated as follows: calculate the coefficient of variation (standard deviation / mean) of the pollutant concentration at all sampling points, and take 1 / 2 of the coefficient of variation as the continuity threshold. If there are uncovered depth intervals or concentration change rates at the boundaries exceeding 0.1, it indicates that the segmentation is incomplete and the mutation threshold needs to be adjusted and the segmentation re-dismantled. For cases where the segmentation results of multiple sampling points are inconsistent, a majority voting method is used for fusion: if more than 70% of the sampling points identify event boundaries at the same depth, then that depth is determined to be the global event boundary; otherwise, the local boundary is retained and marked as a region-specific event.

[0025] For each pollution event fingerprint segment, the concentration ratio, peak value, and rate of change of all target pollutants are extracted to form the feature vector of that segment. Then, the cosine similarity between this feature vector and the feature vector of each emission event in the historical emission event archive is calculated. The formula for calculating the cosine similarity is cosθ = (A·B) / (|A|×|B|), where θ is the spatial angle between the two feature vectors A and B, which is only a mathematical definition and is not assigned a separate value in engineering; A is the feature vector extracted from the current pollution event fingerprint segment, composed of multi-dimensional features such as pollutant concentration ratio, peak value, and rate of change; B is the pre-stored standard feature vector in the historical emission event archive, and the two have completely identical dimensions; A·B is the dot product (inner product) of the two vectors, calculated using the formula... The formula is the sum of the product of corresponding dimension values, used to characterize the overall overlap of vectors; |A| and |B| are the magnitudes (norms) of vectors A and B, respectively, which are the normalization parameters of the vectors themselves. Their function is to eliminate the influence of the overall numerical magnitude of the vectors on similarity, so that the similarity is determined only by the trend of feature distribution. Compared with other similarity indicators such as Euclidean distance, cosine similarity focuses more on the direction of the vectors rather than their magnitude, which can effectively eliminate the influence of differences in emissions from different pollution events, making it more suitable for tracing the source of pollution events; historical emission events with a cosine similarity greater than 0.9 are identified as the source tracing results corresponding to the segment. The threshold of 0.9 is determined based on statistical analysis: when the similarity is greater than 0.9, the feature difference between the two events is less than 5%, and the misclassification rate is less than 1%; In addition, supplementary processing logic is provided for two special cases. First, for multi-event overlapping matching: if the similarity of multiple historical events exceeds 0.9, the event with the highest similarity is taken as the primary matching event. The pollution contribution matrix of the second-highest similarity event is multiplied by a coefficient of (secondary similarity / primary similarity) and then added to the pollution contribution matrix of the primary event grid by grid to obtain the merged event pollution contribution. Second, for low-similarity events: if the similarity of a certain segment with all historical events is less than 0.9, it is marked as an unrecorded event. For unrecorded events, the occurrence time is estimated based on adjacent time anchors and deposition rate, the emission location is determined based on the grid with the highest pollution concentration in that event, and the emission amount is calculated back from the total pollution contribution. All parameters are stored in the historical emission event archive and treated as a separate event in subsequent pollution contribution calculations.

[0026] To accurately calibrate the timing of pollution incidents, an absolute timescale is established using the known emission times of non-targeted labeled pollutants. Traditional techniques cannot establish a precise correlation between soil depth and pollution occurrence time, only providing a rough estimate of the pollution's age. However, using non-targeted labeled pollutants achieves precise time calibration with an error of no more than 30 days. The specific process is as follows: In multi-pollutant time-series fingerprinting, the depth locations where the concentration of all non-target labeled pollutants exceeds the method detection limit are identified. Based on the known usage time range of each non-target labeled pollutant, the absolute time points corresponding to these depth locations are determined, forming multiple absolute time anchors. For example, 1,1,2-trichloroethane appears at a depth of 8.2-9.5 m, corresponding to the time periods of 1985-1995; carbon tetrachloride appears at a depth of 4.7-5.8 m, corresponding to the time periods of 1995-2005; and 1,2-dichloropropane appears at a depth of 2.1-2.9 m, corresponding to the time periods of 2005-2015. The concentration-weighted average depth is used as the anchor depth to accurately reflect the deposition depth corresponding to the main emission time of the pollutant. The calculation formula is: Anchor Depth =Σ(depth i × concentration i) / Σ(concentration i); where depth i is the midpoint depth of the sampling layer, and concentration i is the measured concentration of the marker in that layer; Calculations show that the anchor depth for 1,1,2-trichloroethane is 8.72 m, corresponding to the year 1990; the anchor depth for carbon tetrachloride is 5.18 m, corresponding to the year 2000; and the anchor depth for 1,2-dichloropropane is 2.46 m, corresponding to the year 2010. This method is more accurate than the midpoint method. After this correction, the anchor depth deviation is less than 0.15 m, and the time error is less than 6 months. When the estimated deposition rates of different non-targeted marker pollutants are inconsistent, a dynamic weighted average method is used for calibration. The weights are automatically calculated based on the detection accuracy and chemical stability of the markers. The formula is: weight i = (1 / MDL) i ×lg(T1 / 2 i )) / Σ(1 / MDL j ×lg(T1 / 2 j ), where MDL i The method detection limit is T1 / 2 for the i-th marker. Higher detection accuracy results in a smaller MDL and a larger weight. i Let represent the half-life of the i-th marker in the soil. The better the stability, the longer the half-life, and the greater the weight. For example, 1,1,2-trichloroethane MDL = 0.001 mg / kg, half-life = 60 years; carbon tetrachloride MDL = 0.001 mg / kg, half-life = 55 years; 1,2-dichloropropane MDL = 0.002 mg / kg, half-life = 50 years; the calculated weights are 0.36, 0.33, and 0.31, respectively.

[0027] The average soil deposition rate at a given soil sampling point within a given time period is calculated by dividing the depth difference between two adjacent absolute time anchor points by the corresponding time difference. The formula is: Deposition rate (m / year) = (Lower anchor point depth - Upper anchor point depth) / (Lower anchor point time - Upper anchor point time), where the lower anchor point depth represents the anchor point location with a longer time period and greater soil depth, in meters; the upper anchor point depth represents the anchor point location with a more recent time period and shallower soil depth, in meters, and the difference between the two is used as two markers. The vertical depth difference of the anchor points; the lower anchor point time and the upper anchor point time are the historical years corresponding to the two anchor points, in years. The difference between the two is the time interval between the two pollution events. Taking the anchor points in 1990 and 2000 as examples, the lower anchor point depth is 8.85m, the upper anchor point depth is 5.25m, and the time difference is 10 years. Therefore, the deposition rate is (8.85-5.25) / 10=0.36m / year. Similarly, the average soil deposition rate between 2000 and 2010 is calculated to be 0.27m / year. 5m / year; This multi-anchor calibration method can accurately reflect the changes in soil deposition rate at different times, solving the problem of large time calibration errors caused by the assumption of a constant deposition rate in existing technologies. Then, based on the calculated average soil deposition rate, the precise occurrence time corresponding to each pollution event fingerprint segment is estimated. If the difference between the estimated time and the time recorded in the historical emission event archive exceeds a preset threshold, the feature vector of the pollution event fingerprint segment is re-extracted and similarity matching is performed until the time difference does not exceed the preset threshold. The preset threshold is determined based on the completeness of the site's historical records, and the value ranges from 15 to 90 days. The more complete the historical records, the smaller the threshold can be set, and the higher the calibration accuracy. In this assessment, since the site's historical records are relatively complete, the preset threshold is set to 30 days. The final calibration results show that the maximum difference between the estimated time and the historical record time for all pollution events is 27 days, which meets the accuracy requirements. The preset threshold for the difference between the estimated time and the historical record time is 15 to 90 days, determined based on the completeness of the site's historical records.

[0028] Because existing technologies cannot distinguish the contributions of different pollution events, only the total pollution concentration can be obtained; therefore, after completing time calibration, an event-level three-dimensional pollution impact matrix is ​​constructed. This event-level matrix allows for a quantitative characterization of the impact range of each historical emission event. First, the site's horizontal space is divided into 1m square grids, totaling 120×100=12000 horizontal grids. The soil depth is divided into 0.2m thick vertical grids, totaling 75 vertical grids, ultimately forming a 120×100×75 three-dimensional grid system. The grid size can be dynamically adjusted according to the site area, pollution level, and required accuracy: when the computational load exceeds the server's performance threshold, the grid size increases in steps of 0.5m; when the pollution concentration gradient in a local area is greater than 10mg / (m³), the grid size is adjusted accordingly. When the weight is kg, the grid size in this area is reduced in steps of 0.2m. In this assessment, the 1m×1m×0.2m grid size can keep the computational load within a reasonable range while ensuring computational accuracy. For each calibrated pollution event, the convection-dispersion equation is solved using the finite difference method, taking into account its emission location, emission intensity, vertical migration rate of pollutants, horizontal diffusion coefficient, and soil interlayer resistance coefficient, to calculate the pollution contribution of the event in each three-dimensional grid. The convection-dispersion equation is the basic equation describing the migration and diffusion of pollutants in porous media, and its form is: (θc) / t= / x(θD_L× c / x)+ / y(θD_T× c / y)+ / z(θD_V× c / z)- / z(θv_zc)-ρ_b s / t, where, (θc) / t represents the change in the total mass of pollutants in the soil medium per unit time; where θ is the soil porosity (dimensionless), representing the proportion of pore volume to total volume in the soil; c is the concentration of pollutants in the liquid phase of the soil, in mg / kg; t is the time, in years; partial differential symbol Represents minute changes and is used to describe dynamic migration processes; / x(θD_L× c / x) represents the longitudinal dispersion term, describing the molecular diffusion and mechanical dispersion of pollutants in the horizontal X direction; D_L is the longitudinal dispersion coefficient, in meters. 2 / Year; / y(θD_T× c / y) represents the lateral dispersion term, describing the dispersion of pollutants in the horizontal Y direction; D_T is the lateral dispersion coefficient, in meters. 2 / Year; / z(θD_V× c / z) represents the vertical dispersion term, describing the dispersion of pollutants along the Z-direction of soil depth; D_V is the vertical dispersion coefficient, in meters.2 / Year;- / z(θv_zc) is the vertical convection term; the negative sign indicates that pollutants migrate downwards along the depth. v_z is the vertical seepage velocity of soil pore water, in m / year. This term simulates the process of groundwater flow driving pollutants downwards; -ρ_b s / t represents the adsorption sink term, and ρ_b represents the soil dry bulk density, in kg / m³. 3 ; s represents the concentration of pollutants adsorbed in the soil solid phase, in mg / kg. This term describes the process of pollutants being adsorbed by soil particles and released from the liquid phase. The initial condition is set to t=0, where the pollutant concentration in the emission location grid is the initial concentration, and the concentration in the other grids is 0. The boundary conditions are set to zero flux boundary at the site boundary and free outflow boundary at the bottom boundary. Since the pollutants of interest are all recalcitrant substances, the influence of degradation is ignored during the assessment. The pollution contribution of each pollution event in all three-dimensional grids is stored as a three-dimensional matrix. All single-event three-dimensional matrices together constitute the event-level three-dimensional pollution impact matrix. This matrix fully records the pollution impact of each historical emission event in the entire space and depth of the site, and is the basis for subsequent time-series overlay calculations.

[0029] Next, we perform temporal overlay and dynamic correction of pollution events. This is because existing technologies simply add up all pollution contributions, ignoring the interactions between different events, leading to significant prediction deviations. Our dynamic correction mechanism addresses this issue. All pollution events are sorted in ascending order of occurrence. The pollution contribution of each current event is then sequentially overlaid into the total pollution contribution matrix. After overlaying each event, based on its emission intensity, pollutant type, and soil permeability, and if it involves non-polluting solvent emissions, a dilution correction coefficient is calculated for all previously overlaid events. Simultaneously, migration disturbances are also calculated. The dynamic correction factor dynamically adjusts the pollution contribution of previous events. The impact of subsequent events on previous pollution is categorized into two completely different types, each employing a different correction logic. If the subsequent event releases the same pollutants as the previous event, the concentration is directly accumulated without dilution, and the correction factor is 1.0. Only when the subsequent event involves the release of non-polluting solvents such as fire water leaks or cooling water leaks will it dilute the previous pollutants. The correction factor is calculated as follows: Dilution Correction Factor = Total Pollution Amount in the Previous Stage / (Total Pollution Amount in the Previous Stage + Water Intake from the Subsequent Solvent × Solubility), where the total pollution amount in the previous stage is in kg, and the water intake from the subsequent solvent is in m³. 3 The unit of solubility is kg / m³ 3The migration disturbance correction factor is derived based on the convection-dispersion equation and reflects the degree to which additional water flow caused by later events drives earlier pollutants downward. The calculation formula is: Migration disturbance correction factor = 1 + (Δv z × time interval) / average sampling layer thickness, where Δv z The additional vertical flow velocity increase caused by the later events is expressed in m / year; the time interval is the interval between two pollution events, expressed in years; the average sampling layer thickness is 1.5 m; for example, all later events were contaminant leaks, with no solvent leaks, therefore the dilution correction factor is 1.0 for all; the 1995 leak event caused a local flow velocity increase of 0.018 m / year, and the time interval between it and the 1985 event was 10 years, so the migration disturbance correction factor was calculated as 1 + (0.018 × 10) / 1.5 = 1.12, therefore the overall correction factor for the 1985 event is 1. 0 × 1.12 = 1.12; This dynamic correction mechanism can accurately reflect the impact of later pollution events on earlier pollution, significantly improving the prediction accuracy; Multiply the total pollution contribution of each grid in the total pollution contribution matrix by the toxicity weight coefficient of the corresponding pollutant, and normalize using standard values ​​to obtain the comprehensive pollution index of that grid. The calculation formula is: Comprehensive pollution index = Σ(pollutant concentration i × toxicity weight i) / Σ(national standard screening value i × toxicity weight i), where: the national standard screening value is taken from the "Soil Environmental Quality Construction Land Soil Pollution Risk Control Standard (Trial)" (GB (GB 36600-2018); the toxicity weight coefficients are determined according to the "Technical Guidelines for Risk Assessment of Soil Pollution in Construction Land" (HJ 25.3-2019), with cadmium = 10, lead = 5, and nitrobenzene = 3. This normalization method allows the comprehensive index to directly reflect the ratio of the pollution level to the national standard: a comprehensive index ≤ 1 indicates that the screening value specified in GB 36600-2018 has not been exceeded, and > 1 indicates that the screening value has been exceeded and a risk assessment is required. The pollution level is divided into four levels according to the comprehensive index: unpolluted: comprehensive index ≤ 0.25; slightly polluted: 0.25 < comprehensive index ≤ 0.5; moderately polluted: 0.5 < comprehensive index ≤ 1.0; heavily polluted: comprehensive index > 1.0. For example, a typical grid comprehensive index of 0.62 indicates moderate pollution, not exceeding the screening value; 1.21 indicates heavy pollution, exceeding the screening value; and 0.21 indicates unpolluted, which directly corresponds to the national standard and can be directly used for remediation plan formulation.

[0030] After calculating the degree of contamination, a 3D visualization technology is used to generate a 3D soil contamination distribution map of the contaminated site. Different colors represent different contamination levels, clearly marking the boundaries of areas with different contamination levels, the specific locations of each sampling point, and the comprehensive contamination index. Users can interactively rotate and zoom the view to view the contamination distribution at any depth and profile. A detailed preliminary report is also generated, including a basic introduction to the site, data collection and processing procedures and quality control information, historical contamination event tracing results and time calibration information, the impact range of each emission event and its contribution ratio in each area, and the dynamic evolution of the site's contamination process. The system will provide forecasts of pollution diffusion trends over the next 30 years, along with targeted pollution prevention and remediation recommendations. Furthermore, it will establish a unified management system for the entire process, with a three-tiered user access control system: administrators, operators, and visitors. Administrators will have the highest privileges, including user management and system configuration. Operators will perform data entry and analysis. Visitors will only be able to view the forecast results. All data generated during system operation will be uniformly stored and managed, with daily automatic backups and off-site backups. The system will monitor CPU usage, memory usage, and disk space usage in real time, and promptly send email and SMS alerts when system anomalies occur.

[0031] When new soil sample data is acquired or new historical emission event records are added, the system initiates an incremental dynamic update process. Existing technologies require reprocessing all data after acquiring new data, which is time-consuming and labor-intensive. Our approach, however, only recalculates the affected portion. The triggering conditions are: the amount of newly acquired soil sample data exceeds 10% of the original data, or the added historical emission event involves a major leak, such as a leak exceeding 1 ton. Before executing the update, a global parameter change assessment is performed. If the new data only affects local pollution concentrations and does not change global parameters such as soil deposition rate or non-target marker anchor locations, then only the affected time interval and spatial range are locally recalculated. If the new data causes a change in the global deposition rate exceeding 10%, or adds valid non-target marker anchors, then all global parameters are updated first, and the occurrence time of all pollution events is recalibrated. The pollution contribution of the affected area is recalculated to generate an updated 3D pollution distribution map and forecast report. For example, in 2024, during a supplementary site investigation, an unrecorded solvent leak from 1988 was discovered, and eight deep soil samples were collected. The affected time period was determined to be 1985–1995, and the affected spatial area was approximately 2000 square meters in the northwest of the site, with an impact radius of 50 meters. Only this part was recalculated, and the entire update process took only about 2 hours, without the need to reprocess all data, greatly improving work efficiency. When there are no non-targeted labeled pollutants that meet the conditions at the site, a label-free alternative calibration method can be used. Soil samples are collected at different depths, one sample every 2 meters, with each sample weighing no less than 200g. The soil organic matter content and carbon isotope composition are determined using radiocarbon dating. 14 C) Determine the soil deposition time of key depth layers to establish a preliminary time-depth relationship. Then, combine the site's historical emission records to cross-validate and correct the preliminary time-depth relationship. Finally, estimate the precise occurrence time of each pollution event fingerprint segment. The time calibration accuracy of this method is about 6 months, which can meet the prediction requirements of most contaminated sites.

[0032] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0033] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0034] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A system for predicting the degree of soil contamination at contaminated sites, characterized in that, The forecasting system includes: The basic information collection unit collects data on the geographical location, historical evolution, topography, surrounding environment, and historical emission events of the contaminated site. The soil sample acquisition unit acquires concentration data of target pollutants and non-target labeled pollutants in soil at different sampling points and at different depths. The data standardization unit performs format standardization, outlier removal, and missing value completion on the collected data to generate a standardized dataset. The time series fingerprint construction unit, based on a standardized dataset, generates a multi-pollutant time series fingerprint map reflecting the variation of pollutant concentration with depth at each sampling point. The time series fingerprint construction unit arranges the standardized pollutant concentration data of each layer from shallow to deep to form an original concentration sequence. By calculating the concentration change rate of all pollutants between adjacent depth layers, a concentration change rate sequence is generated. The original concentration sequence and the concentration change rate sequence are horizontally stitched together to form a two-dimensional multi-pollutant time series fingerprint map, where the map rows correspond to pollutant types and the columns correspond to soil depth layers. The event segmentation and tracing unit divides the continuous multi-pollutant time series fingerprint spectrum into independent pollution event fingerprint segments, establishes a correspondence with historical emission event archives, and verifies the segmentation. The event segmentation and tracing unit uses a mutation boundary identification algorithm to locate the depth boundary between different pollution events in the concentration change rate sequence of the multi-pollutant time series fingerprint spectrum, dividing the multi-pollutant time series fingerprint spectrum into multiple non-overlapping pollution event fingerprint segments. Specifically, for each pollution event fingerprint segment, the pollutant concentration ratio, peak value, and change rate features are extracted to form a feature vector, and the correspondence is determined by calculating the cosine similarity with the feature vectors of historical emission events. The non-targeted labeled pollutant calibration unit establishes an absolute time scale using the known emission times of the non-targeted labeled pollutants and calibrates the occurrence times of the pollution events obtained from the decomposition. The calibration process of the non-targeted labeled pollutant calibration unit includes: First, identify the depth locations where the concentration of all non-target marker pollutants exceeds the detection limit in the multi-pollutant time series fingerprint spectrum, and use their corresponding known emission times as time anchors; The average soil deposition rate for the corresponding time period is calculated by the depth difference and time difference between two adjacent time anchor points, and then the precise occurrence time of each pollution event fingerprint segment is estimated. If the difference between the estimated time and the historical time exceeds a preset threshold, the feature vector of the fingerprint fragment of the pollution event will be re-extracted and similarity matching will be performed until the time difference does not exceed the preset threshold. The three-dimensional pollution impact matrix construction unit calculates the pollution contribution of a single event in the entire space and depth of the site based on the calibrated pollution event parameters and soil environmental parameters, and constructs an event-level three-dimensional pollution impact matrix. The time-series overlay unit overlays the pollution contribution of all pollution events in order of occurrence from earliest to latest, corrects for the mutual influence between different events, obtains a comprehensive pollution index at any location and depth of the site, and generates a pollution distribution map and a forecast report including single event impact analysis.

2. The soil pollution degree prediction system for contaminated sites according to claim 1, characterized in that, The historical emission event archive records the start and end times, emission location coordinates, emission method, target pollutant type and emission amount, and accompanying non-target labeled pollutant type and emission amount for each independent emission event; The historical emission event archive also includes a feature vector for each emission event, which contains the concentration ratio, peak value, and rate of change of all target pollutants emitted during that event. Non-targeted pollutants are chemical substances that were emitted only during historical periods at the site, are chemically stable in the soil environment without degradation reactions, do not react chemically with soil components, and are not listed in the soil pollution control catalog.

3. The soil pollution level prediction system for contaminated sites according to claim 1, characterized in that, The soil sample acquisition unit divides the soil into multiple continuous sampling layers according to the principle of shallow densification and deep descaling, and the thickness of the sampling layer increases in a gradient with the depth. At each sampling point, original soil samples from each layer are collected simultaneously from top to bottom, and the sample collection thickness does not exceed two-thirds of the thickness of the corresponding sampling layer. The test detects heavy metals, organic pollutants, and inorganic pollutants in the target pollutants. Heavy metals include lead, cadmium, mercury, chromium, arsenic, copper, zinc, and nickel; organic pollutants include polycyclic aromatic hydrocarbons, polychlorinated biphenyls, and petroleum hydrocarbons; and inorganic pollutants include cyanides and sulfides.

4. The soil pollution level prediction system for contaminated sites according to claim 1, characterized in that, The three-dimensional pollution impact matrix construction unit divides the site horizontal space into square grids of equal size and the soil depth into vertical grids of equal thickness. For each calibrated pollution event, the pollution contribution of the event in each grid is calculated by combining its emission parameters, pollutant migration characteristics and soil interlayer barrier characteristics through a pollutant migration and diffusion simulation method. The pollution contributions of all single events together constitute an event-level three-dimensional pollution impact matrix, where the three dimensions of the matrix correspond to the horizontal X coordinate, the horizontal Y coordinate and the soil depth, respectively.

5. The soil pollution degree prediction system for contaminated sites according to claim 4, characterized in that, The time-series overlay unit sorts all pollution events from earliest to latest according to their occurrence time, and then overlays the pollution contribution of the current pollution event into the total pollution contribution matrix in turn; After each pollution event is superimposed, the emission intensity, pollutant type and soil permeability of the event are considered. If it is a non-polluting solvent emission, the dilution effect correction coefficient on all previously superimposed events is calculated. At the same time, the migration disturbance correction coefficient is calculated to dynamically correct the pollution contribution of previous events. The total pollution contribution of each grid in the total pollution contribution matrix is ​​multiplied by the toxicity weight coefficient of the corresponding pollutant to obtain the comprehensive pollution index of each grid. The comprehensive pollution index is divided into four levels: no pollution, light pollution, moderate pollution and heavy pollution, according to a preset value range.

6. A method for predicting the degree of soil pollution at a contaminated site, applied to the soil pollution degree prediction system for a contaminated site as described in claim 1, characterized in that, The steps of this estimation method are as follows: Collect basic information about the contaminated site and a complete archive of historical emission events. The archive of historical emission events contains the feature vector of each emission event. Soil sampling layers are divided according to the principle of shallow densification and deep desaturation to obtain the concentration data of target pollutants and non-target labeled pollutants in the soil at different sampling points and at different depths. The collected data is standardized to obtain a standardized dataset with uniform format and no outliers or missing values; Based on a standardized dataset, a multi-pollutant time series fingerprint map reflecting the variation of pollutant concentration with depth at each sampling point is constructed. Identify the deep boundaries between different pollution events in the time series fingerprint of multiple pollutants, split them into multiple independent pollution event fingerprint segments, and achieve pollution event source tracing through feature vector cosine similarity matching; Time anchors were established using known emission times of non-targeted labeled pollutants to calculate the average soil deposition rate and calibrate the occurrence times of all pollution events. The site space is divided into three-dimensional grids. Based on the calibrated pollution event parameters, the pollution contribution of each event in each grid is calculated through pollutant migration and diffusion simulation, and an event-level three-dimensional pollution impact matrix is ​​constructed. The pollution contribution of all pollution events is superimposed in order of occurrence from earliest to latest. Each superimposed event corrects its impact on all previous events, resulting in a comprehensive pollution index at any location and depth of the site. The index is then divided into four levels: unpolluted, lightly polluted, moderately polluted, and heavily polluted, according to a preset value range. Generate a three-dimensional soil pollution level distribution map and a detailed forecast report including the impact range, contribution ratio, and evolution process of a single event.

7. The method for predicting the degree of soil pollution at contaminated sites according to claim 6, characterized in that, The estimation method also includes the following steps: When new soil sample data is obtained or new historical emission event records are added, the process of splitting, tracing, and calibrating is only re-executed for pollution events affected by the new data or records. The pollution contribution of the corresponding event in the event-level three-dimensional pollution impact matrix is ​​updated, and the comprehensive pollution index of the affected area is recalculated to perform incremental dynamic iteration of the prediction results.

Images