A water pollution multi-source tracing method and system fusing time sequence behavior fingerprints and chemical ratio fingerprints

Abstract

The application discloses a water pollution multi-source tracing method and system fusing time sequence behavior fingerprints and chemical ratio fingerprints, and belongs to the technical field of water pollution monitoring. The method comprises the following steps: obtaining water quality time sequence data of each pollution source discharge port and a monitoring point of a receiving water body, and obtaining pretreated water quality time sequence data through abnormality processing and flow normalization; generating chemical ratio fingerprints and time sequence behavior fingerprints based on the pretreated water quality time sequence data under a normal discharge state; performing time shift alignment on pollution time sequence segments and discharge time sequence segments according to a hydraulic transmission lag time, and generating event matching characteristics; inputting the chemical ratio fingerprints, the time sequence behavior fingerprints and the event matching characteristics into a tracing classification model to obtain a responsibility probability of each pollution source, and setting a contribution coefficient constraint when triggering multi-source demixing to obtain a pollution contribution proportion of each pollution source. The application improves the stability of pollution source identification and contribution proportion determination in a multi-source mixed discharge scene.

Description

Technical Field

[0001] This application belongs to the field of water pollution monitoring technology, specifically a method and system for tracing the source of water pollution from multiple sources by integrating temporal behavioral fingerprints and chemical ratio fingerprints. Background Technology

[0002] With the increasing number of polluting enterprises in industrial parks, it is common for multiple enterprises to share the same river, ditch, or other receiving water body. When a pollution incident occurs in the receiving water body, the pollutants in the water body are affected by factors such as discharge time, discharge flow rate, hydraulic transmission distance, dilution and diffusion process, and simultaneous discharge by multiple enterprises. It is difficult to reliably determine the source of pollution by simply comparing the results of a single sampling of the receiving water body with the test results of each enterprise's discharge outlet.

[0003] Among existing methods for tracing water pollution sources, some schemes establish a chemical fingerprint database of water samples discharged by enterprises and compare the similarity between polluted water samples and the fingerprint database after a pollution incident to determine the source of pollution. These methods typically establish static fingerprints based on the concentrations of multiple pollutants, pollutant concentration ratios, spectral characteristics, or other water quality characteristics, providing a basis for source tracing even when the pollution source is singular and the composition of enterprise emissions varies significantly.

[0004] However, in the actual operation of industrial parks, different enterprises may have similar production processes or similar pollutant compositions. Static chemical fingerprints based solely on pollutant concentrations or concentration ratios are easily affected by sampling time, discharge flow rate, dilution degree, and background water quality fluctuations. When multiple enterprises discharge intermittently or jointly in adjacent time periods, the pollutant concentrations detected at the receiving water monitoring points often appear as mixed data from multiple discharge sources. Static fingerprint comparison can only provide a similarity ranking and is insufficient to further determine the contribution ratio of each pollution source.

[0005] Furthermore, existing methods typically fail to incorporate the temporal patterns of corporate emissions into source tracing. Factors such as production shifts, cleaning cycles, order batches, and equipment start-up / shutdown status can cause emissions data to exhibit periodicity, peak periods, intermittency, and long-term trend variations. If these temporal emission patterns are not utilized during source tracing, misjudgment of responsible companies can easily occur in scenarios with similar chemical compositions or mixed emissions. Simultaneously, pollutants require a certain hydraulic transmission time to travel from the company's discharge outlet to the receiving water body monitoring point. If the emission timeline and the receiving water body pollution timeline are not aligned based on the hydraulic transmission time, it is difficult to determine whether there is a temporal correspondence between corporate emission events and receiving water body pollution events.

[0006] Therefore, existing water pollution source tracing technologies still suffer from problems such as a single static fingerprint dimension, insufficient utilization of emission time sequence patterns, difficulty in verifying the temporal correspondence between emission events and pollution events, and unstable determination of the contribution ratio of multi-source mixed emissions. Summary of the Invention

[0007] To address the above issues, this application provides a method and system for tracing multiple sources of water pollution by integrating temporal behavioral fingerprints and chemical ratio fingerprints. By performing anomaly processing, flow normalization, chemical ratio fingerprint generation, temporal behavioral fingerprint generation, hydraulic transmission lag alignment, responsibility probability identification, and constrained multi-source unmixing on the water quality time-series data of pollution source discharge outlets and receiving water bodies, a continuous tracing chain is formed from the historical emission characteristics of pollution sources to the contribution ratio of pollution events. This improves the stability of pollution source identification and contribution ratio determination in scenarios with similar chemical compositions and multi-source mixed emissions.

[0008] To achieve the above objectives, the technical solution adopted in this application is as follows: In a first aspect, this application provides a method for tracing the multiple sources of water pollution by fusing temporal behavioral fingerprints and chemical ratio fingerprints, including: Obtain water quality time series data from various pollution source discharge outlets and receiving water body monitoring points, perform anomaly processing and flow normalization on them, and obtain pretreated water quality time series data; Based on the time series data of pretreated water quality under normal discharge conditions of various pollution sources, the concentration ratios of multiple pollutants are calculated to generate chemical ratio fingerprints; Based on the time series data of pretreated water quality from various pollution sources, information on discharge cycle, peak period, intermittency index, concentration variation coefficient and discharge trend is extracted to generate time series behavioral fingerprints; Determine the hydraulic transmission lag time from each pollution source discharge outlet to the monitoring point of the receiving water body, and after a pollution event is triggered, perform time-shift alignment of the pollution time series segments of the receiving water body and the discharge time series segments of each pollution source based on the hydraulic transmission lag time to generate event matching features; By inputting chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features into the source tracing classification model, the probability of responsibility for each pollution source is obtained. The multi-source unmixing trigger indication is determined based on the comparison between the responsibility probability of each pollution source and the preset responsibility probability threshold. When the multi-source unmixing trigger indication is set to enter multi-source unmixing, the contribution coefficient of each pollution source is constrained based on the emission status of each pollution source at the emission time corresponding to the pollution event and the event matching characteristics. The concentration of pollutants in the receiving water body is unmixed and calculated to obtain the pollution contribution ratio of each pollution source.

[0009] Secondly, this application provides a multi-source water pollution tracing system that integrates temporal behavioral fingerprints and chemical ratio fingerprints, including: The data preprocessing unit is used to acquire water quality time series data from various pollution source discharge outlets and receiving water body monitoring points, perform anomaly processing and flow normalization on the data, and obtain preprocessed water quality time series data. The chemical fingerprint generation unit is used to calculate the concentration ratio of multiple pollutants and generate chemical ratio fingerprints based on the time series data of pretreated water quality under normal discharge conditions of each pollution source. The temporal fingerprint generation unit is used to extract discharge cycle, peak period, intermittency index, concentration variation coefficient and discharge trend information based on the pretreated water quality temporal data of each pollution source, and generate temporal behavioral fingerprints; The event matching unit is used to determine the hydraulic transmission lag time from each pollution source discharge outlet to the receiving water body monitoring point, and after a pollution event is triggered, to time-shift and align the pollution time sequence of the receiving water body and the discharge time sequence of each pollution source according to the hydraulic transmission lag time to generate event matching features; The responsibility identification unit is used to input chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features into the source tracing classification model to obtain the responsibility probability of each pollution source. The multi-source demixing unit is used to determine the multi-source demixing trigger indication based on the comparison result between the responsibility probability of each pollution source and the preset responsibility probability threshold. When the multi-source demixing trigger indication is to enter multi-source demixing, the unit sets constraints on the contribution coefficient of each pollution source based on the emission status and event matching characteristics of each pollution source at the emission time corresponding to the pollution event, and performs demixing calculation on the pollutant concentration of the receiving water body to obtain the pollution contribution ratio of each pollution source.

[0010] Compared with related technologies, this application has the following advantages: First, by performing anomaly processing and flow normalization on the water quality time series data of pollution source discharge outlets and receiving water body monitoring points, water quality data formed under different discharge flow rates, different sampling times, and different monitoring sections are converted into pre-processed water quality time series data with a unified caliber, reducing the interference of discharge flow fluctuations and abnormal sampling values ​​on subsequent chemical ratio fingerprints, time series behavioral fingerprints, and event matching features.

[0011] Second, by generating chemical ratio fingerprints based on pollutant concentration ratios and time-series behavioral fingerprints based on emission cycles, peak periods, intermittent indices, concentration variation coefficients, and emission trend information, pollution source identification no longer relies solely on static pollutant concentration similarity, but simultaneously utilizes pollutant composition relationships and emission behavior patterns to improve the ability to distinguish between enterprises with similar chemical compositions.

[0012] Third, by determining the hydraulic transmission lag time from each pollution source discharge outlet to the receiving water body monitoring point, and accordingly time-shifting and aligning the pollution time series segments of the receiving water body with the pollution discharge time series segments of each pollution source, a verifiable correspondence can be formed between pollution source discharge events and receiving water body pollution events in the time dimension, reducing the probability of pollution sources corresponding to times that did not occur being misjudged as responsible pollution sources.

[0013] Fourth, by inputting chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features into the source tracing classification model, the responsibility probability of each pollution source is obtained. Multi-source unmixing is triggered based on the responsibility probability. When there are multiple candidate pollution sources, the contribution coefficient constraint is further set based on the emission status and event matching features, so that the unmixing calculation is constrained by temporal evidence and emission evidence, thereby improving the stability of determining the pollution contribution ratio in multi-source mixed emission scenarios. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments are briefly introduced below.

[0015] Figure 1 This is a schematic diagram of a multi-source water pollution tracing application scenario provided in an embodiment of this application; Figure 2 This is a flowchart of a multi-source tracing method for water pollution provided in an embodiment of this application; Figure 3 This is a flowchart of a chemical ratio fingerprint and a temporal behavioral fingerprint generation method provided in an embodiment of this application; Figure 4 This is a schematic diagram of event time-shift alignment based on hydraulic transmission lag time provided in an embodiment of this application; Figure 5 This is a flowchart of a multi-source demixing calculation provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a multi-source water pollution tracing system provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application; Figure 8 This is a schematic diagram of the probability distribution of corporate responsibilities in a random forest model provided in an embodiment of this application; Figure 9 This is a schematic diagram illustrating the ranking of feature importance in a random forest model provided in an embodiment of this application; Figure 10 This is a schematic diagram illustrating the comprehensive pollution contribution ratio of various enterprises, provided in an embodiment of this application. Figure 11 This is a schematic diagram illustrating the decomposition of the contribution of each pollutant to each enterprise, provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and technical effects of the embodiments of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings. The following embodiments are used to illustrate this application and are not intended to limit the scope of protection of this application.

[0017] In this embodiment, the pollution source refers to the enterprise discharge outlet, park drainage branch pipe, production unit drainage node, or other discharge node that can be independently monitored, which directly or indirectly discharges wastewater into the receiving water body; the receiving water body monitoring point refers to the water quality monitoring location set in the river, ditch, drainage canal, or other receiving water body; the water quality time series data refers to the data of pollutant concentration, discharge flow rate, sampling time, monitoring location, and pollution source identification recorded in the order of sampling time; the pre-treated water quality time series data refers to the data formed after time normalization, anomaly processing, and flow normalization, which can be used for chemical ratio fingerprint generation, time series behavior fingerprint generation, and event matching feature generation.

[0018] The chemical ratio fingerprint in this application refers to the pollution source chemical composition characterization data formed based on the ratio relationship between the normalized concentration values ​​of multiple pollutants under normal emission conditions of the pollution source; the chemical ratio fingerprint includes at least the pollution source identifier, pollutant ratio combination, ratio center value, allowable deviation range of ratio, and fingerprint update time; the time-series behavioral fingerprint refers to the emission behavior characterization data formed in the time dimension based on the time-series data of pretreated water quality of the pollution source; the time-series behavioral fingerprint includes at least the pollution source identifier, emission cycle, peak period, intermittency index, concentration variation coefficient, emission trend information, and statistical interval identifier; the event matching feature refers to the matching characterization data formed after the pollution event is triggered, based on the hydraulic transmission lag time, by time-shifting and aligning the pollution time sequence of the receiving water body and the pollution source emission time sequence; the event matching feature includes at least the peak time deviation, time period overlap ratio, consistency of change direction, waveform correlation coefficient, and chemical ratio deviation.

[0019] like Figure 1 As shown, in one application scenario, an industrial park includes pollution sources A, B, and C, and a receiving water body monitoring point M. Pollution sources A, B, and C discharge wastewater into the same receiving water body through their respective discharge outlets. The receiving water body monitoring point M is located downstream of each discharge outlet. The water pollution multi-source tracing system acquires the time-series water quality data of the discharge outlets of pollution sources A, B, and C, and also acquires the time-series water quality data of the receiving water body monitoring point M. When a pollution event is triggered at the receiving water body monitoring point M, the water pollution multi-source tracing system outputs the responsibility probability and pollution contribution ratio of each pollution source based on the historical normal discharge data of each pollution source and the discharge data during the pollution event period. Figure 1It is used to reflect the data acquisition and source tracing relationship between pollution source discharge outlets, receiving water bodies, receiving water body monitoring points, and water pollution multi-source tracing systems.

[0020] like Figure 2 As shown in the figure, this embodiment provides a method for tracing the source of water pollution from multiple sources by fusing temporal behavioral fingerprints and chemical ratio fingerprints. The method includes steps S101 to S106. The method can be executed by a server, an edge computing device, a park water environment monitoring platform, or other computer equipment.

[0021] S101: Obtain time-series water quality data from various pollution source discharge outlets and receiving water body monitoring points, perform anomaly processing and flow normalization on the data, and obtain pre-treated water quality time-series data. The pre-treated water quality time-series data is subsequently used in S102 to generate chemical ratio fingerprints, in S103 to generate time-series behavioral fingerprints, and in S104 to generate event matching features. The method for obtaining pre-treated water quality time-series data includes the following steps: S1011, Obtain raw water quality sampling data from the discharge outlets of each pollution source and the monitoring points of the receiving water body, and generate raw water quality time series data.

[0022] Specifically, the water pollution multi-source tracing system acquires raw water quality sampling data from the discharge outlets of each pollution source and the monitoring points of the receiving water body according to the preset sampling intervals. The raw water quality sampling data includes the monitoring object identification, sampling time, pollutant type, pollutant concentration, discharge flow or cross-sectional flow, monitoring location and equipment status. The pollutant type includes at least one of chemical oxygen demand, ammonia nitrogen, total phosphorus, total nitrogen, conductivity, pH value, heavy metal ions, volatile phenols, petroleum or enterprise characteristic pollutants.

[0023] Pollutant concentrations are obtained from online water quality monitoring instruments, automatic sampling and testing results, or laboratory testing results; discharge flow rates are obtained from discharge outlet flow meters, pump station operation records, valve opening duration, pipeline design flow rates, or manually verified flow rates; monitoring locations are obtained from discharge outlet codes, monitoring point codes, geographic information system coordinates, or park drainage network archives; equipment status is obtained from online monitoring equipment operation logs, data acquisition terminal status records, or monitoring platform equipment status records.

[0024] After acquisition, the water pollution multi-source tracing system sorts the original water quality sampling data according to the monitoring object identification and sampling time to generate original water quality time series data; the original water quality time series data is used as input data for S1012 to perform sampling time correction.

[0025] S1012, the sampling time of the original water quality time series data is normalized to generate time-normalized water quality time series data; specifically, the water pollution multi-source tracing system establishes a unified time axis; the unified time axis is determined by the preset sampling start time, preset sampling end time and preset sampling interval; the preset sampling interval can be set according to the regulatory sampling requirements, the sampling cycle of online monitoring equipment or the time resolution required for pollution event identification.

[0026] For raw water quality time series data whose sampling time falls within a unified time axis, the water pollution multi-source tracing system retains the corresponding sampling values. For raw water quality time series data whose sampling time does not fall within a unified time axis but whose time difference with the adjacent unified sampling time is not greater than a preset time adjustment threshold, the water pollution multi-source tracing system adjusts the corresponding sampling values ​​to the nearest unified sampling time. The preset time adjustment threshold can be set based on a certain proportion of the preset sampling interval, the upper limit of the time synchronization error of the monitoring equipment, or the time deviation allowed by the regulatory platform.

[0027] For cases where multiple sampling values ​​exist for the same monitored object at the same unified sampling time, the sampling values ​​with normal equipment status are first selected, and then the sampling values ​​with the smallest difference between the sampling time and the unified sampling time are selected. If multiple sampling values ​​still exist, the sampling values ​​to be retained are determined according to the data source reliability level. The data source reliability level is preset in the order of online monitoring data, automatic sampling detection data, and manually supplemented detection data.

[0028] After the sampling time is normalized, the water pollution multi-source tracing system generates time-normalized water quality time-series data; the time-normalized water quality time-series data is used as input data for anomaly processing in S1013.

[0029] In one optional embodiment, when the sampling intervals of the pollution source discharge outlet and the receiving water body monitoring point are inconsistent, the water pollution multi-source tracing system adopts a time window aggregation method to form time-normalized water quality time series data. Specifically, the water pollution multi-source tracing system establishes a preset statistical time window centered on each sampling moment on a unified time axis, and performs a weighted average of multiple sampling values ​​falling within the preset statistical time window according to the sampling time distance to generate the pollutant concentration value corresponding to the unified sampling moment. The weighted averaged pollutant concentration value continues to enter S1013 for anomaly processing.

[0030] S1013, perform anomaly processing on time-normalized water quality time series data to generate anomaly-processed water quality time series data; specifically, the water pollution multi-source tracing system performs range anomaly judgment, missing measurement judgment, adjacent difference judgment, and historical quantile range judgment on time-normalized water quality time series data.

[0031] The judgment of abnormal measurement range involves comparing the pollutant concentration with the lower limit and upper limit of the corresponding monitoring instrument. The lower limit and upper limit of the measurement range are determined by the calibration parameters of the monitoring instrument or the applicable range of the detection method. When the sampled value is lower than the lower limit or higher than the upper limit of the measurement range, the water pollution multi-source tracing system will determine the corresponding sampled value as an abnormal measurement range sampled value and remove the abnormal measurement range sampled value.

[0032] Missing data determination is based on the duration of consecutive missing sample values. The preset missing data duration threshold is determined by the sampling interval and the allowed number of consecutive interpolation points. For example, if the allowed number of consecutive interpolation points is 2 and the sampling interval is 10 minutes, the preset missing data duration threshold is 20 minutes. When the consecutive missing data duration is not greater than the preset missing data duration threshold, the water pollution multi-source tracing system uses adjacent valid sample values ​​before and after the missing data duration for linear interpolation. When the consecutive missing data duration is greater than the preset missing data duration threshold, the water pollution multi-source tracing system does not perform interpolation and determines the corresponding time period as unusable for fingerprint generation.

[0033] The adjacent difference judgment compares the pollutant concentration difference between adjacent sampling times with a preset adjacent difference threshold. The preset adjacent difference threshold is determined by the historical quantile value of the adjacent sampling difference of the same pollution source and the same pollutant under normal emission conditions. When the adjacent difference amplitude is greater than the preset adjacent difference threshold and the equipment status is abnormal, the water pollution multi-source tracing system will identify the corresponding sampling value as an abnormal equipment sampling value and remove it. When the adjacent difference amplitude is greater than the preset adjacent difference threshold but the equipment status is normal, the water pollution multi-source tracing system will retain the corresponding sampling value and generate an event candidate identifier.

[0034] Historical quantile range determination involves comparing the current pollutant concentration with the historical quantile range formed under normal emission conditions for the same monitored object and the same pollutant. The historical quantile range can be formed from the 5th to the 95th historical quantile. If the current pollutant concentration exceeds the historical quantile range and the equipment is in normal condition, the corresponding sampled value is retained and used as a candidate input for pollution event identification.

[0035] After the anomaly handling is completed, time-series water quality data of the anomaly handling is generated; the time-series water quality data of the anomaly handling is used as input data for flow normalization in S1014.

[0036] S1014, normalize the flow rate of the abnormal water quality time series data to generate pretreated water quality time series data; specifically, the water pollution multi-source tracing system determines the reference discharge flow rate corresponding to each pollution source discharge outlet; the reference discharge flow rate is determined by the median discharge flow rate of the pollution source under normal discharge conditions, the permitted discharge flow rate recorded in the discharge permit, or the design discharge flow rate of the drainage facility; when the above data exist simultaneously, the median discharge flow rate under normal discharge conditions is used first; when the number of normal discharge samples is insufficient, the permitted discharge flow rate or the design discharge flow rate is used.

[0037] Subsequently, the pollutant concentrations at each sampling time are converted to the reference discharge flow rate according to the corresponding discharge flow rate to obtain normalized concentration values. Specifically, the water pollution multi-source tracing system first reads the measured pollutant concentration and corresponding discharge flow rate of a certain pollution source at the current sampling time, and then reads the reference discharge flow rate corresponding to that pollution source. When the discharge flow rate at the current sampling time is greater than the reference discharge flow rate, it indicates that the discharge load corresponding to the current sampling concentration is in a relatively increasing state. The water pollution multi-source tracing system increases the concentration characterization value at that sampling time according to the ratio between the current discharge flow rate and the reference discharge flow rate. When the discharge flow rate at the current sampling time is less than the reference discharge flow rate, it indicates that the discharge load corresponding to the current sampling concentration is in a relatively decreasing state. The water pollution multi-source tracing system decreases the concentration characterization value at that sampling time according to the ratio between the current discharge flow rate and the reference discharge flow rate. Through this conversion, the pollutant concentrations formed at different sampling times are all converted to normalized concentration values ​​under the same reference discharge flow rate.

[0038] In one example, for pollution source i, pollutant k, and sampling time t, the normalized concentration value is calculated as follows:

[0039] in, This represents the normalized concentration value of pollutant k corresponding to pollution source i at sampling time t; This represents the measured concentration of pollutant k corresponding to pollution source i at sampling time t; This represents the emission flow rate of pollution source i at sampling time t; This represents the reference emission flow rate corresponding to pollution source i; This represents a preset correction amount used to avoid the denominator being zero. It is a positive number and is less than the minimum effective value of the reference emission flow rate.

[0040] For the receiving water body monitoring points, the water pollution multi-source tracing system normalizes the flow rate of the monitoring section or the converted flow rate of the preset section to obtain the normalized concentration value corresponding to the receiving water body monitoring point; the flow rate of the monitoring section is calculated by the flow meter, the flow record of the hydrological station or the hydrodynamic model; the converted flow rate of the preset section is calculated by the section width, water depth and section velocity.

[0041] When the receiving water body monitoring point is not equipped with a cross-sectional flow meter, the multi-source water pollution tracing system can calculate the preset cross-sectional equivalent flow rate based on the cross-sectional width, average water depth, and average cross-sectional flow velocity of the monitoring cross-section. The cross-sectional width is obtained from river mapping data or on-site measurement data; the average water depth is obtained from water level gauge readings or manual depth sounding results; and the average cross-sectional flow velocity is obtained from flow meter, hydrological station data, or hydrodynamic models. The multi-source water pollution tracing system multiplies the cross-sectional width, average water depth, and average cross-sectional flow velocity to obtain the preset cross-sectional equivalent flow rate. When the monitoring cross-section is irregular, it can be divided into multiple sub-sections and the sub-sections can be summed to obtain the preset cross-sectional equivalent flow rate.

[0042] Pre-treated water quality time series data were used as input data for generating chemical ratio fingerprints in S102, time series behavioral fingerprints in S103, and event matching features in S104.

[0043] In some implementations, to illustrate the process of generating pretreated water quality time-series data, an example is given below; the measured chemical oxygen demand (COD) concentration of pollution source A at sampling time 10:00 is 120 mg / L, and the discharge flow rate is 80 m³ / h; the reference discharge flow rate for pollution source A is 100 m³ / h; the water pollution multi-source tracing system calculates the normalized COD concentration value of pollution source A at 10:00 according to the following method:

[0044] when When the reference discharge flow rate is negligible, the normalized chemical oxygen demand (COD) concentration at 10:00 for pollution source A is approximately 96 mg / L. If the normalized ammonia nitrogen concentration is 12 mg / L and the normalized total phosphorus concentration is 1.8 mg / L at the same sampling time, the water pollution multi-source tracing system will associate the sampling time 10:00, pollution source A, the normalized COD concentration of 96 mg / L, the normalized ammonia nitrogen concentration of 12 mg / L, the normalized total phosphorus concentration of 1.8 mg / L, and the corresponding discharge flow rate to form a pretreated water quality time series data corresponding to pollution source A at 10:00. This pretreated water quality time series data continues to serve as the input data for S102, S103, and S104.

[0045] In a further optional embodiment, the water pollution multi-source tracing system configures a data credibility level for pretreated water quality time series data. Specifically, the water pollution multi-source tracing system configures a first credibility level, a second credibility level, or a third credibility level for each piece of pretreated water quality time series data based on the sampling equipment status, flow data source, whether interpolation has been performed, whether alternative discharge flow has been used, and whether there are abnormal retention indicators. The first credibility level corresponds to data with normal equipment status, complete flow data, and no interpolation. The second credibility level corresponds to data with short-term interpolation and valid sampling values ​​before and after. The third credibility level corresponds to data with alternative discharge flow, manually verified flow, or abnormal retention indicators. The data credibility level is output in association with the pretreated water quality time series data and is used as an auxiliary input for the source tracing classification model in S105, or used to adjust the contribution coefficient constraints in S106.

[0046] S102, based on the time series data of pretreated water quality under normal discharge conditions of each pollution source, calculates the concentration ratio of multiple pollutants and generates chemical ratio fingerprints.

[0047] like Figure 3 As shown, the water pollution multi-source tracing system generates chemical ratio fingerprints and time-series behavioral fingerprints based on pretreated water quality time-series data. The chemical ratio fingerprint is used to characterize the stable chemical composition relationship of pollution sources, and is used to calculate the chemical ratio deviation in S104, and as input to the source tracing classification model in S105. Specifically, generating the chemical ratio fingerprint includes the following steps: S1021, Normal discharge status data is selected from the pretreated water quality time series data to generate a normal discharge sample set; specifically, the water pollution multi-source tracing system reads the monitoring object identifier, sampling time, normalized concentration value, equipment status, and production status records from the pretreated water quality time series data; the production status records come from the enterprise's production plan, equipment start-up and shutdown records, discharge permit execution ledger, online monitoring platform status identifiers, or manual verification records; when the equipment status corresponding to a certain sampling time is normal, and the production status record is routine production, routine cleaning, or stable discharge, the water pollution multi-source tracing system determines the corresponding sampling data as normal discharge status data.

[0048] When production status records are missing, the water pollution multi-source tracing system uses alternative judgment methods to determine normal discharge status data. These alternative judgment methods include: equipment status is normal; pollutant concentration is within the historical normal percentile range for the same pollution source and the same pollutant; and discharge flow rate is within the normal flow rate range. When all three conditions are met simultaneously, the water pollution multi-source tracing system will determine the corresponding sampled data as normal discharge status data.

[0049] After the screening is completed, the water pollution multi-source tracing system summarizes the normal emission status data according to the pollution source identification and generates a normal emission sample set corresponding to each pollution source; the normal emission sample set is used as the input data for S1022 to determine the pollutant ratio combination.

[0050] S1022, Based on the normal emission sample set, determine the pollutant ratio combination and generate an effective pollutant ratio combination set; specifically, the water pollution multi-source tracing system selects at least two pollutants from the pollutant set to form a pollutant ratio combination; each pollutant ratio combination includes a numerator pollutant and a denominator pollutant; for each pollutant ratio combination, the water pollution multi-source tracing system counts the number of times the denominator pollutant is effectively detected in the normal emission sample set.

[0051] The numerator and denominator pollutants can be determined based on pollutant detection stability, pollutant concentration magnitude, and pollution source discrimination capability. Specifically, the multi-source water pollution tracing system first removes pollutants with fewer than a preset effective detection frequency threshold from the pollutant set, and then selects pollutants with a high correlation to the pollution source's production process or historical pollution events as candidate pollutants from the remaining pollutants. For any two candidate pollutants, the multi-source water pollution tracing system prioritizes the pollutant with higher detection stability, lower low-concentration noise, and a higher proportion of effective concentration under normal discharge conditions as the denominator pollutant, and the other pollutant as the numerator pollutant. If both pollutants meet the denominator pollutant criteria, pollutant ratio combinations in two directions can be constructed respectively, and subsequent stability screening retains pollutant ratio combinations with lower dispersion and stronger pollution source discrimination capability. After determination, the multi-source water pollution tracing system generates a set of effective pollutant ratio combinations for S1023 to calculate pollutant concentration ratios.

[0052] The effective detection frequency threshold is set based on the minimum sample size for fingerprint statistics. The effective concentration threshold is determined based on the detection limit, quantitation limit, or minimum effective concentration set by the monitoring platform. The effective proportion threshold is determined based on the proportion of calculable ratios in the normal emission sample set. If the effective detection frequency of the denominator pollutant is not less than the preset effective detection frequency threshold, and the sampling proportion of the denominator pollutant with a normalized concentration value not less than the preset effective concentration threshold is not less than the preset effective proportion threshold, then the water pollution multi-source tracing system retains the corresponding pollutant ratio combination.

[0053] After screening, the water pollution multi-source tracing system generates a set of effective pollutant ratio combinations. This set of effective pollutant ratio combinations serves as the input data for S1023 to calculate pollutant concentration ratios.

[0054] S1023, calculate the pollutant concentration ratio based on the effective pollutant ratio combination set, and generate a pollutant concentration ratio sequence; specifically, for each pollution source, each effective pollutant ratio combination, and each sampling time, the water pollution multi-source tracing system reads the normalized concentration values ​​of the numerator pollutants and the normalized concentration values ​​of the denominator pollutants; when the normalized concentration value of the denominator pollutants is not less than the preset effective concentration threshold, the water pollution multi-source tracing system calculates the pollutant concentration ratio.

[0055] In one calculation example, the pollutant concentration ratio is calculated as follows:

[0056] in, This represents the ratio of pollutant concentrations between pollutant a and pollutant b at sampling time t for pollution source i. This represents the normalized concentration value of pollutant a corresponding to pollution source i at sampling time t; This represents the normalized concentration value of pollutant b corresponding to pollution source i at sampling time t; This indicates the preset correction amount, which is a positive number used to avoid the denominator being zero.

[0057] After the calculation is completed, the water pollution multi-source tracing system organizes the pollutant concentration ratios according to the pollution source identification, pollutant ratio combination and sampling time, and generates a pollutant concentration ratio sequence; the pollutant concentration ratio sequence is used as the input data for S1024 to generate chemical ratio fingerprints.

[0058] S1024, Statistical processing is performed on the pollutant concentration ratio sequence to generate a chemical ratio fingerprint; Specifically, for each pollution source and each pollutant ratio combination, the water pollution multi-source tracing system calculates the ratio center value and the allowable deviation range of the pollutant concentration ratio sequence; The ratio center value can be the median or the mean; The allowable deviation range of the ratio is determined by the median ± a preset multiple interquartile range, the mean ± a preset multiple standard deviation, or a preset quantile interval; The preset multiple can be set based on the fluctuation range of the normal emission ratio of the same pollution source in historical review samples.

[0059] The multi-source water pollution tracing system can also screen the stability of pollutant ratio combinations. Specifically, for each pollutant ratio combination, the system reads the pollutant concentration ratio sequence formed in the normal emission sample set and calculates the dispersion based on this sequence. The dispersion can be represented by the coefficient of variation, the ratio of the interquartile range to the median, or the historical standard deviation of the pollutant concentration ratio sequence. After calculating the dispersion, the system compares it with a preset dispersion threshold. When the dispersion is not greater than the preset threshold, it indicates that the pollutant ratio combination has a stable ratio relationship under normal emission conditions, and the system retains the pollutant ratio combination and writes it into the chemical ratio fingerprint. When the dispersion is greater than the preset threshold, it indicates that the pollutant ratio combination fluctuates greatly under normal emission conditions, and the system does not include the pollutant ratio combination in the chemical ratio fingerprint. The preset dispersion threshold can be determined based on the dispersion distribution of stable pollutant ratio combinations in historical normal emission samples, or it can be set based on the experience value of the monitoring platform during the initial deployment phase of the park.

[0060] The water pollution multi-source tracing system combines pollution source identification, pollutant ratio combination, ratio center value, ratio allowable deviation range, number of samples participating in the statistics, and fingerprint update time to generate chemical ratio fingerprints. The chemical ratio fingerprints are used to calculate chemical ratio deviations in S104 and serve as input data for the source tracing classification model in S105. Figure 3 The fingerprinting method for normal emission sample set-pollutant ratio combination-chemical ratio corresponds to S1021 to S1024.

[0061] In some implementations, an example is given below to illustrate the process of generating chemical ratio fingerprints.

[0062] Pollution source A has three valid sampling times in the normal emission sample set, with corresponding normalized chemical oxygen demand (COD) concentrations of 96 mg / L, 104 mg / L, and 100 mg / L, and corresponding normalized ammonia nitrogen (AM) concentrations of 12 mg / L, 13 mg / L, and 10 mg / L, respectively. The water pollution multi-source tracing system calculates the COD to AM concentration ratios according to the pollutant concentration ratio formula, obtaining 8.0, 8.0, and 10.0. The water pollution multi-source tracing system then determines the median of the above concentration ratio sequence, 8.0. The ratio center value is set, and the allowable deviation range of the ratio is determined according to the quartile range or the preset deviation range to generate the chemical ratio fingerprint item corresponding to pollution source A. The chemical ratio fingerprint item includes pollution source A, chemical oxygen demand / ammonia nitrogen, ratio center value 8.0, allowable deviation range of the ratio, and number of samples participating in the statistics. The above chemical ratio fingerprint item is used in S104 to calculate the chemical ratio deviation between the pollution time series segment of the receiving water body and pollution source A, and is input into the source tracing classification model as part of the pollution source event feature vector in S105.

[0063] In one optional embodiment, when the difference in magnitude between the normalized concentration values ​​of the molecule and denominator pollutants in the same pollutant ratio combination within the normal emission sample set reaches a preset magnitude difference condition, the water pollution multi-source tracing system performs a logarithmic transformation on the pollutant concentration ratios to generate a chemical ratio fingerprint. The preset magnitude difference condition can be determined based on the multiple relationship between the median of the normalized concentrations of the molecule and denominator pollutants in the normal emission sample set; for example, when the ratio between the larger and smaller medians is not less than a preset magnitude multiple threshold, the preset magnitude difference condition is determined to have been met. Specifically, after calculating the pollutant concentration ratios, the water pollution multi-source tracing system takes the natural logarithm or the base-10 logarithm of the pollutant concentration ratios to obtain a logarithmic ratio, and calculates the ratio center value and the allowable deviation range based on the logarithmic ratio. The logarithmically transformed center value of the ratio and the allowable deviation range of the ratio are still used as components of the chemical ratio fingerprint, and are used to calculate the chemical ratio deviation in S104. In S105, they are used as components of the pollution source event feature vector and input into the source tracing classification model.

[0064] In a further optional embodiment, the water pollution multi-source tracing system generates chemical ratio sub-fingerprints according to production status; production status includes routine production status, equipment cleaning status, maintenance drainage status, and batch switching status. The water pollution multi-source tracing system first divides the normal discharge sample set into multiple status sample sets based on production status records, then calculates the center value and allowable deviation range of pollutant concentration ratios in each status sample set, generating chemical ratio sub-fingerprints for the corresponding production status. When the discharge time corresponding to a pollution event falls within a certain production status, the water pollution multi-source tracing system prioritizes using the chemical ratio sub-fingerprint corresponding to that production status for event matching and responsibility probability identification.

[0065] S103, based on the pretreated water quality time series data of each pollution source, extracts the discharge cycle, peak period, intermittency index, concentration variation coefficient and discharge trend information to generate time series behavioral fingerprints.

[0066] like Figure 3 As shown, step S103 is used to form a comprehensive pollution load sequence from the pretreated water quality time series data, and to extract emission cycle, peak period, intermittency index, concentration variation coefficient, and emission trend information based on the comprehensive pollution load sequence to generate a time series behavioral fingerprint. The time series behavioral fingerprint is used to characterize the emission behavior pattern of pollution sources in the time dimension and is used as input content for the source tracing classification model in step S105. Specifically, generating the time series behavioral fingerprint includes the following steps: S1031, Based on the pretreated water quality time series data, a comprehensive pollution load sequence is formed, and a comprehensive pollution load sequence of pollution sources is generated.

[0067] Specifically, the multi-source water pollution tracing system reads the pretreated water quality time-series data corresponding to each pollution source and determines the set of pollutants participating in the comprehensive pollution load calculation. The pollutant set consists of pollutants of regulatory concern, pollutants frequently detected in historical pollution events, and enterprise-specific pollutants. For each pollutant, the multi-source water pollution tracing system determines the corresponding pollutant weight and performs a weighted summation of the concentration values ​​of each pollutant to obtain the comprehensive pollution load value.

[0068] The pollutant weights are set based on the level of regulatory concern, the contribution of historical pollution events to identification, and the stability of pollutant detection. After the settings are completed, the water pollution multi-source tracing system normalizes the pollutant weights so that the weights of each pollutant involved in the calculation are non-negative and the sum of the weights is 1.

[0069] In one calculation example, the comprehensive pollution load value is calculated as follows:

[0070] in, This represents the comprehensive pollution load value of pollution source i at sampling time t. The set of pollutants involved in the calculation; This represents the preset pollutant weight corresponding to pollutant k. are non-negative and ; This represents the normalized concentration value of pollutant k corresponding to pollution source i at sampling time t.

[0071] After the calculation is completed, the water pollution multi-source tracing system arranges the comprehensive pollution load values ​​according to the sampling time sequence to generate a comprehensive pollution load sequence of pollution sources; the comprehensive pollution load sequence of pollution sources is used as the input data for extracting time-series behavioral features in S1032 to S1035.

[0072] S1032, perform periodic analysis on the comprehensive pollution load sequence of pollution sources to generate emission cycles; specifically, the water pollution multi-source tracing system establishes a set of candidate time lags; the set of candidate time lags is formed by sampling intervals, shift durations, production day cycles, cleaning cycles, or regulatory setting cycles; for each candidate time lag, the water pollution multi-source tracing system calculates the correlation between the comprehensive pollution load sequence and the lagged comprehensive pollution load sequence; the correlation calculation function can use the Pearson correlation coefficient.

[0073] The multi-source water pollution tracing system sorts the correlations of each candidate time lag in descending order of correlation, generating a correlation ranking sequence. It then reads the candidate time lag ranked first in the correlation ranking sequence and its corresponding maximum correlation. Subsequently, the system calculates the difference between the maximum correlation and the correlations of adjacent candidate time lags to obtain the period peak difference. Adjacent candidate time lags refer to one or two candidate time lags adjacent to the first-ranked candidate time lag in the candidate time lag set. When the maximum correlation is not less than a preset period correlation threshold and the period peak difference is not less than a preset period peak difference threshold, the system identifies the first-ranked candidate time lag as the emission cycle. When the maximum correlation is less than the preset period correlation threshold, or the period peak difference is less than the preset period peak difference threshold, the system outputs an identifier indicating that no stable emission cycle has been identified.

[0074] The preset periodic correlation threshold is used to determine whether the comprehensive pollution load sequence has an identifiable periodic correspondence. It can be determined by the correlation distribution of confirmed periodic samples under historical normal emission conditions, or it can be set based on empirical thresholds during the initial deployment phase of the park. The preset periodic peak difference threshold is used to determine whether the maximum correlation has peak distinguishability relative to adjacent candidate time lags. It can be determined by the correlation difference distribution between adjacent candidate time lags under historical normal emission conditions. When historical samples are lacking, the preset periodic correlation threshold and the preset periodic peak difference threshold can be initially set according to the empirical values ​​of the regulatory platform and updated after the accumulation of samples that have passed subsequent manual review.

[0075] In one calculation example, the emission cycle is calculated as follows:

[0076] in, This indicates the emission cycle corresponding to pollution source i; Represents the set of candidate time lags; This indicates the candidate time lag. This represents the correlation calculation function; This represents the comprehensive pollution load value corresponding to pollution source i at sampling time t. This indicates that pollution source i lags behind sampling time t. The comprehensive pollution load value after the calculation; argmax represents the candidate time lag selected from the set of candidate time lags that maximizes the correlation calculation result.

[0077] S1033, identify peak periods for the comprehensive pollution load sequence of pollution sources and generate peak periods; specifically, the water pollution multi-source tracing system determines a preset peak identification threshold based on the historical quantile of the comprehensive pollution load value under normal emission conditions of pollution sources; the preset peak identification threshold can be the 85th quantile, 90th quantile, or a regulatory threshold; subsequently, the water pollution multi-source tracing system determines the sampling time when the comprehensive pollution load value is not less than the preset peak identification threshold as the peak sampling time.

[0078] The multi-source water pollution tracing system further merges peak sampling times; when the time interval between adjacent peak sampling times is not greater than the preset peak merging interval, the multi-source water pollution tracing system merges adjacent peak sampling times into the same peak time period. The preset peak merging interval can be determined based on an integer multiple of the sampling interval.

[0079] After peak identification is completed, the water pollution multi-source tracing system generates peak time periods. The peak time periods are used as part of the time sequence behavior fingerprint generated in S1036, and are used in S104 to determine whether the back-dated time period corresponding to the pollution event corresponds to the historical high emission period of the pollution source.

[0080] S1034, The system performs statistical analysis on the emission status switching of the comprehensive pollution load sequence of pollution sources to generate an intermittency index. Specifically, the water pollution multi-source tracing system determines the emission status based on the comparison between the comprehensive pollution load value and a preset activation threshold. The preset activation threshold is determined by the low load quantile of normal emissions, the lower limit of permitted emissions, or the corresponding load value from equipment operation records. When the comprehensive pollution load value is not less than the preset activation threshold, the water pollution multi-source tracing system determines the corresponding sampling time as the activation state; when the comprehensive pollution load value is less than the preset activation threshold, the water pollution multi-source tracing system determines the corresponding sampling time as the deactivation state.

[0081] The water pollution multi-source tracing system compares the emission status of adjacent sampling points according to the sampling time sequence. When the emission status of two adjacent sampling points is inconsistent, the water pollution multi-source tracing system counts it as a state switch. Subsequently, the water pollution multi-source tracing system counts the number of switches between the open and closed states within the statistical interval and calculates the intermittency index. The intermittency index is used as a component of the time-series behavioral fingerprint generated by S1036.

[0082] In one calculation example, the intermittency index is calculated as follows:

[0083] in, Indicates the intermittency index of pollution source i; This indicates the number of times the state transitions between on and off within the statistical interval. This represents the number of sampling points within the statistical interval, and Greater than 1; when If the value is not greater than 1, the intermittency index is not calculated, and an indicator that the intermittency index cannot be calculated is output.

[0084] S1035, fluctuation and trend analysis are performed on the comprehensive pollution load sequence of pollution sources to generate concentration variation coefficient and emission trend information. Specifically, the water pollution multi-source tracing system reads the comprehensive pollution load sequence of pollution sources within a statistical interval, calculates the mean and standard deviation of each comprehensive pollution load value within the statistical interval, and calculates the concentration variation coefficient based on the mean and standard deviation. The concentration variation coefficient is used to characterize the degree of emission fluctuation of pollution sources within the statistical interval. Subsequently, the water pollution multi-source tracing system performs linear regression on the comprehensive pollution load sequence of pollution sources within the statistical interval, using the sampling time sequence as the independent variable and the comprehensive pollution load value as the dependent variable, and uses the linear regression slope as emission trend information.

[0085] In one calculation example, the concentration variation coefficient is calculated as follows:

[0086] in, The coefficient of variation of the concentration of pollution source i is represented by . This represents the overall pollution load sequence of pollution source i within the statistical interval; This represents the standard deviation of the comprehensive pollution load value within the statistical interval; This represents the mean value of the comprehensive pollution load within the statistical interval. This indicates a preset correction amount, used to avoid unstable calculation results when the mean is close to zero.

[0087] The water pollution multi-source tracing system further performs linear regression on the comprehensive pollution load sequence within the statistical interval. The linear regression uses the sampling time sequence as the independent variable and the comprehensive pollution load value as the dependent variable. The water pollution multi-source tracing system determines the linear regression slope as the emission trend information. The preset trend slope threshold is determined based on the quantile of the absolute value of the regression slope under historical normal emission conditions. When the regression slope is positive and the absolute value is greater than the preset trend slope threshold, the emission trend information is in the direction of increase. When the regression slope is negative and the absolute value is greater than the preset trend slope threshold, the emission trend information is in the direction of decrease. When the absolute value of the regression slope is not greater than the preset trend slope threshold, the emission trend information is in the direction of change that does not exceed the trend slope threshold.

[0088] After completing the fluctuation analysis and trend analysis, the water pollution multi-source tracing system generates concentration variation coefficient and emission trend information; the concentration variation coefficient and emission trend information are used as components of the time-series behavioral fingerprint generated by S1036.

[0089] In step S1036, the emission cycle, peak period, intermittency index, concentration variation coefficient, and emission trend information are combined to generate a time-series behavioral fingerprint. Specifically, the water pollution multi-source tracing system combines pollution source identification, statistical interval identification, emission cycle, peak period, intermittency index, concentration variation coefficient, emission trend information, and feature update time to generate a time-series behavioral fingerprint. After generation, the water pollution multi-source tracing system outputs the time-series behavioral fingerprint. The time-series behavioral fingerprint serves as input data for the source tracing classification model in step S105 and is used in step S106 to determine whether candidate pollution sources exhibit consistent emission behavior during the pollution event period. Figure 3 The integrated pollution load sequence in the data—emission cycle, peak period, intermittency index, concentration variation coefficient, emission trend information—corresponds to the time-series behavioral fingerprint processing path, which is S1031 to S1036.

[0090] In some implementations, to illustrate the generation process of the temporal behavioral fingerprint, an example is given below; the pollutants involved in the calculation for pollution source A within a certain statistical interval include chemical oxygen demand (COD), ammonia nitrogen, and total phosphorus, with corresponding pollutant weights of 0.5, 0.3, and 0.2, respectively; if the normalized concentration of COD for pollution source A at 10:00 is 96 mg / L, the normalized concentration of ammonia nitrogen is 12 mg / L, and the normalized concentration of total phosphorus is 1.8 mg / L, then the comprehensive pollution load value can be calculated as follows:

[0091] The calculated comprehensive pollution load value of pollution source A at 10:00 is 51.96. The water pollution multi-source tracing system calculates the comprehensive pollution load values ​​at multiple sampling times in the same way, forming a comprehensive pollution load sequence for pollution source A. If the comprehensive pollution load sequence shows a peak value concentrated between 10:00 and 11:00 daily, and the autocorrelation analysis shows that the correlation for 24 hours meets the preset period identification condition, then the water pollution multi-source tracing system generates a time-series behavioral fingerprint with a 24-hour emission cycle and a peak period between 10:00 and 11:00 daily. If the number of times the on / off state switches within the statistical interval is 6, and the number of sampling points is 25, then the intermittency index is calculated as follows:

[0092] The intermittency index of pollution source A was calculated to be 0.25. The water pollution multi-source tracing system combines the emission cycle, peak period, intermittency index, concentration variation coefficient and emission trend information into the temporal behavior fingerprint of pollution source A, and continues to use it as the input data of the source tracing classification model in S105.

[0093] In one optional embodiment, the water pollution multi-source tracing system forms single-pollutant load sequences for multiple key pollutants and extracts the corresponding emission cycles, peak periods, intermittency indices, concentration variation coefficients, and emission trend information respectively. Subsequently, the water pollution multi-source tracing system combines the time-series features of multiple single pollutants according to pollutant weights to generate a time-series behavioral fingerprint. This optional embodiment is applicable to scenarios where a pollution source has different emission rhythms for different pollutants.

[0094] In a further optional embodiment, the water pollution multi-source tracing system sets an expiration period for time-series behavioral fingerprints. Specifically, the system determines whether the time-series behavioral fingerprints need to be updated based on enterprise production process change records, discharge permit change records, changes in online monitoring data distribution, and manual review results. When the deviation between the emission cycle, peak period, or intermittent index formed in the current statistical interval and the historical time-series behavioral fingerprints exceeds a preset fingerprint update threshold, the system generates a fingerprint update identifier and regenerates the time-series behavioral fingerprints based on the pre-treated water quality time-series data within the most recent preset time period. The updated time-series behavioral fingerprints continue to be used for S105 responsibility probability identification and S106 contribution coefficient constraints.

[0095] S104. Determine the hydraulic transmission lag time from each pollution source discharge outlet to the receiving water body monitoring point, and after a pollution event is triggered, perform time-shift alignment of the pollution time series segments of the receiving water body and the discharge time series segments of each pollution source based on the hydraulic transmission lag time to generate event matching features.

[0096] like Figure 4 As shown, S104 is used to establish a time correspondence between the pollution source discharge time series segment and the receiving water body pollution time series segment based on the hydraulic transmission lag time, so that a computable time series matching relationship is formed between the pollution source discharge event and the receiving water body pollution event; the event matching features are used as input data for the source tracing classification model in S105 and as the basis for contribution coefficient constraints in S106; S1041, the hydraulic path data from the pollution source discharge outlet to the receiving water body monitoring point is obtained, and hydraulic path parameters are generated.

[0097] Specifically, the multi-source water pollution tracing system acquires the hydraulic path length, cross-sectional velocity, water flow direction, pipeline connectivity, and outlet location between each pollution source discharge outlet and the receiving water body monitoring point. The hydraulic path length is obtained from river mapping data, pipeline topology data, or geographic information system path calculations. The cross-sectional velocity is obtained from real-time monitoring values ​​of flow meters, historical hydrological monitoring values, or hydrodynamic model calculations. The water flow direction is determined by the natural flow direction of the river, the gate opening status, the pumping station operating status, or hydrological monitoring data. The pipeline connectivity is determined by the park's drainage network map, on-site verification records, or regulatory platform facility archives. The outlet location is determined by the pollutant discharge permit registration information, geographic information system coordinates, or on-site verification positioning information.

[0098] After acquisition, the water pollution multi-source tracing system combines the hydraulic path length, cross-sectional flow velocity, and water flow direction to generate hydraulic path parameters; the hydraulic path parameters serve as input data for S1042 to determine the hydraulic transmission lag time.

[0099] S1042, determine the hydraulic transmission lag time based on the hydraulic path parameters, and generate the hydraulic transmission lag time; specifically, use the ratio between the hydraulic path length and the cross-sectional flow velocity as the initial hydraulic transmission lag time; in one calculation example, the initial hydraulic transmission lag time is calculated as follows:

[0100] in, This represents the initial hydraulic transmission lag time corresponding to pollution source i. This indicates the hydraulic path length from the discharge outlet of pollution source i to the monitoring point of the receiving water body, and the unit can be meters; This indicates the cross-sectional velocity of the corresponding hydraulic path, and the unit can be meters per second; This indicates a preset correction amount, used to avoid incalculable results due to zero cross-sectional flow velocity; when using meters and meters per second as units, The unit is seconds.

[0101] When tracer test data or historical pollution event alignment data are available, the water pollution multi-source tracing system further corrects the initial hydraulic transmission lag time based on the time difference between the peak emission time of the pollution source and the peak response time of the receiving water body; the historical lag time is formed by tracer tests, historical pollution peak alignment, or manually verified events.

[0102] In one calculation example, the hydraulic transmission lag time is calculated as follows:

[0103] in, This represents the hydraulic transmission lag time corresponding to pollution source i; This indicates the initial hydraulic transmission lag time. Indicates the historical lag time obtained based on historical pollution events or tracer experiments; This represents the preset fusion coefficient, and When the deviation between the current cross-sectional flow velocity and the historical cross-sectional flow velocity is less than a preset flow velocity deviation threshold, the flow velocity is reduced. To increase the weight of historical lag time; if the deviation between the current cross-sectional velocity and the historical cross-sectional velocity is not less than a preset velocity deviation threshold, then increase the weight of historical lag time. To increase the weight of the initial hydraulic transmission lag time.

[0104] The preset velocity deviation threshold can be determined based on the historical cross-sectional velocity fluctuation range under normal water level conditions for the same hydraulic path, or based on the allowable deviation between the cross-sectional velocity during the tracer test and the current cross-sectional velocity. Specifically, the water pollution multi-source tracing system can statistically analyze the standard deviation or quantile difference of historical cross-sectional velocities and use the preset multiple or quantile difference of the standard deviation as the preset velocity deviation threshold. When historical velocity statistics are lacking, the preset velocity deviation threshold can be determined based on the upper limit of velocity measurement error set by the monitoring platform or hydrological model.

[0105] After the calculation is completed, the water pollution multi-source tracing system generates the hydraulic transmission lag time corresponding to each pollution source; the hydraulic transmission lag time is used as the input data for time-shift alignment in S1044.

[0106] S1043, Based on the pretreated water quality time series data of the receiving water body monitoring points, pollution events are identified, and pollution time series segments of the receiving water body are generated; Specifically, the water pollution multi-source tracing system reads the pretreated water quality time series data corresponding to the receiving water body monitoring points and compares the pollutant concentration or comprehensive pollution load with the preset pollution event trigger threshold; The preset pollution event trigger threshold is determined by the regulatory limit, the background concentration quantile of the receiving water body, or the historical pollution event trigger value; The preset comprehensive pollution load threshold is determined by the background load distribution of multiple pollutants or the regulatory comprehensive evaluation requirements.

[0107] When the concentration of a certain pollutant is not less than the corresponding preset pollution event trigger threshold, or when the comprehensive pollution load formed by multiple pollutants is not less than the preset comprehensive pollution load threshold, the water pollution multi-source tracing system determines the pollution event trigger time. After the pollution event is triggered, the water pollution multi-source tracing system searches forward along the time axis for the most recent sampling time when the pollutant concentration falls back to the preset background fluctuation range, and determines it as the pollution event start time. The preset background fluctuation range is determined by the mean background concentration and allowable deviation of the preset background time period before the pollution event is triggered. The allowable deviation can be determined by the standard deviation of the background concentration or the historical background fluctuation quantile.

[0108] The water pollution multi-source tracing system searches backward along the time axis for the sampling moment when the pollutant concentration falls back to the preset background fluctuation range and continues to reach the preset recovery time, and determines this as the pollution event termination time; the preset recovery time is determined by an integer multiple of the sampling interval; the water pollution multi-source tracing system combines the pollution event start time, pollution event termination time, pollutant peak time, pollutant peak concentration, and pollutant change direction to generate a pollution time series segment of the receiving water body; the pollution time series segment of the receiving water body is used as the input data for time-shift alignment in S1044.

[0109] S1044, the time series of pollution in the receiving water body is backtracked based on the hydraulic transmission lag time to generate the emission time series of each pollution source; specifically, for each pollution source, the water pollution multi-source tracing system backtracks the time series of pollution in the receiving water body according to the hydraulic transmission lag time corresponding to the pollution source to determine the emission period corresponding to the pollution source; in order to cover sampling interval errors and hydraulic transmission speed fluctuations, the water pollution multi-source tracing system adds a preset time extension at both ends of the backtracking period.

[0110] The preset time extension is determined by the sampling interval, the range of cross-sectional flow velocity fluctuations, or the uncertainty range of hydraulic transmission lag time. Specifically, the preset time extension can be determined by a unified sampling interval and the hydraulic transmission lag time error. The water pollution multi-source tracing system first determines the hydraulic transmission lag time error based on the historical range of cross-sectional flow velocity fluctuations or tracer test errors, and then adds at least one unified sampling interval to the hydraulic transmission lag time error to obtain the preset time extension. If tracer test errors or historical flow velocity fluctuation data are lacking, the water pollution multi-source tracing system can set the preset time extension to one to three times the unified sampling interval and apply the same time extension rule to each pollution source in the same pollution event.

[0111] In one computational example, if the time series segment of the receiving water body contamination is... Then the emission time series segment corresponding to pollution source i is determined as follows:

[0112] in, This represents the starting point of the time sequence segment of pollution in the receiving water body. This represents the end point of the time sequence segment of pollution in the receiving water body; As a source of pollution The corresponding hydraulic transmission lag time; Reverse the preset time extension amount at the starting point; A time extension is preset for the backtracking termination point; the emission time sequence segments of each pollution source are used as input data for S1045 to calculate the event matching features.

[0113] S1045, based on the time series segments of pollution in the receiving water body and the emission time series segments of each pollution source, calculate matching parameters and generate event matching features; specifically, for each pollution source, the water pollution multi-source tracing system calculates the peak time deviation, time period overlap ratio, consistency of change direction, waveform correlation coefficient and chemical ratio deviation between the emission time series segment and the pollution time series segment of the receiving water body.

[0114] Peak time deviation is the absolute time difference between the peak emission time of the pollution source after correction for hydraulic transmission lag time and the peak pollution time of the receiving water body; the time period overlap ratio is the intersection time between the time period when the pollution source emission state is on and the backtracked pollution event time period divided by the total duration of the backtracked pollution event time period; the consistency of change direction is determined by comparing the change direction of the pollution load from the pollution source emission with the change direction of the pollution load in the receiving water body; the value is 1 when the two directions are consistent, and 0 when the two directions are inconsistent.

[0115] The waveform correlation coefficient is calculated from the aligned integrated pollution load sequence of pollution sources and the integrated pollution load sequence of receiving water bodies; the chemical ratio deviation is formed by the normalized difference between the pollutant concentration ratio in the pollution time series segment of the receiving water body and the ratio center value in the chemical ratio fingerprint of the pollution source; specifically, the water pollution multi-source tracing system first aligns the pollution source emission time series segment with the receiving water body pollution time series segment according to the hydraulic transmission lag time, and extracts the integrated pollution load sequence of pollution sources and the integrated pollution load sequence of receiving water bodies under the same time scale. If the sampling time scales of the two sequences are inconsistent, the water pollution multi-source tracing system will adjust the two sequences to the same time axis according to the sampling time adjustment method in S1012. Subsequently, the water pollution multi-source tracing system will use the paired comprehensive pollution load values ​​on the same time axis as the calculation object to calculate the correlation coefficient between the comprehensive pollution load sequence of the pollution source and the comprehensive pollution load sequence of the receiving water body, and use this correlation coefficient as the waveform correlation coefficient. The waveform correlation coefficient can be calculated using the Pearson correlation coefficient. When the number of effective paired sampling points of the two sequences is less than the preset minimum number of points, the water pollution multi-source tracing system will not calculate the waveform correlation coefficient and will generate a waveform correlation coefficient that cannot be calculated.

[0116] The multi-source water pollution tracing system first calculates the pollutant concentration ratio corresponding to the chemical ratio fingerprint of the pollution source in the pollution time series segment of the receiving water body. The calculation method is consistent with the calculation method of the pollutant concentration ratio in S1023. Subsequently, the multi-source water pollution tracing system reads the ratio center value and the ratio allowable deviation range corresponding to the same pollutant ratio combination in the pollution source chemical ratio fingerprint, and calculates the absolute difference between the pollutant concentration ratio of the receiving water body and the ratio center value. The multi-source water pollution tracing system then normalizes the absolute difference with the ratio allowable deviation range or a preset chemical ratio deviation threshold to obtain the chemical ratio deviation. If a certain pollutant ratio combination cannot be calculated in the pollution time series segment of the receiving water body, the multi-source water pollution tracing system does not use this pollutant ratio combination to calculate the chemical ratio deviation, and generates the chemical ratio deviation corresponding to the pollution source based on the mean or weighted mean of the deviations of the other calculable pollutant ratio combinations.

[0117] After the calculation is completed, the water pollution multi-source tracing system combines the peak time deviation, time period overlap ratio, change direction consistency, waveform correlation coefficient and chemical ratio deviation to generate event matching features. The event matching features are used as input data for the source tracing classification model in S105 and are used to set contribution coefficient constraints in S106. Figure 4 The pollution source emission time series - hydraulic transmission lag time - receiving water body pollution time series - event matching feature processing method corresponds to S1041 to S1045.

[0118] In some implementations, an example is given below to illustrate the process of time-shift alignment based on hydraulic transmission lag time.

[0119] The hydraulic path length from pollution source A's discharge outlet to the receiving water body monitoring point M is 2400 meters, corresponding to a cross-sectional flow velocity of 0.6 meters per second. The water pollution multi-source tracing system calculates the initial hydraulic transmission lag time using the following method:

[0120] when When the cross-sectional flow velocity is negligible, the initial hydraulic transmission lag time is approximately 4000 seconds, or about 66.7 minutes. If the pollution event time series segment of the receiving water body monitoring point M is from 14:10 to 15:00, and the preset time extension amounts are 10 minutes and 10 minutes respectively, then the water pollution multi-source tracing system determines the emission time series segment corresponding to pollution source A as approximately from 12:53 to 14:03. Subsequently, the water pollution multi-source tracing system reads the pretreated water quality time series data of pollution source A from 12:53 to 14:03, and calculates the peak time deviation, time period overlap ratio, consistency of change direction, waveform correlation coefficient, and chemical ratio deviation with the pollution time series segment of the receiving water body monitoring point M from 14:10 to 15:00, generating the event matching features corresponding to pollution source A. The event matching features are then input into the tracing classification model in S105, and used in S106 to set the contribution coefficient constraint of pollution source A.

[0121] In one optional embodiment, when there are multiple pipeline paths, tributary confluences, or gate scheduling between the pollution source discharge outlet and the receiving water body monitoring point, the water pollution multi-source tracing system determines the hydraulic transmission lag time based on the pipeline network topology and hydrodynamic model. Specifically, the water pollution multi-source tracing system inputs the discharge outlet location, pipeline connectivity, gate opening, pump station operating status, and cross-sectional water level into the hydrodynamic model, and the hydrodynamic model outputs the transmission time from each pollution source to the receiving water body monitoring point. The transmission time output by the hydrodynamic model is used as the hydraulic transmission lag time and is further used for backtracking of discharge time sequence segments and generation of event matching features.

[0122] The hydrodynamic model training method includes: the multi-source water pollution tracing system uses historical hydrological monitoring data, discharge outlet flow data, gate opening data, pump station operation data, cross-sectional water level data, and tracer test data as training or calibration data; among them, historical hydrological monitoring data, cross-sectional water level data, and cross-sectional flow velocity data are collected by hydrological monitoring stations or on-site flow meters; discharge outlet flow data are collected by discharge outlet flow meters or pump station operation records; gate opening data and pump station operation data are collected by park drainage facility operation records; tracer test data are formed by tracer dosing time, dosing location, monitoring point detection time, and tracer concentration change data; the input data of the hydrodynamic model includes discharge outlet location, pipe... The system outputs data including channel connectivity, cross-sectional water level, cross-sectional flow velocity, gate opening, pump station operating status, and discharge flow rate. This data includes the transmission time or range from each pollution source to the receiving water body monitoring point. During training or calibration, the multi-source water pollution tracing system uses the deviation between the model's output transmission time and the actual transmission time obtained from tracer experiments or historical pollution events as the calibration error, and adjusts the flow velocity correction parameters, diffusion parameters, or hydraulic resistance parameters in the model. When the calibration error is not greater than the preset transmission time error threshold, or when the preset number of calibration rounds is reached, a calibrated hydrodynamic model is obtained. This calibrated hydrodynamic model is used to output the hydraulic transmission lag time corresponding to each pollution source.

[0123] In a further optional embodiment, the water pollution multi-source tracing system sets an uncertain time range for the hydraulic transmission lag time. Specifically, the water pollution multi-source tracing system determines the lower and upper limits of the hydraulic transmission lag time based on the cross-sectional flow velocity fluctuation range, sampling interval, tracer test error, and water flow direction changes. When performing event matching, the water pollution multi-source tracing system performs time-shift alignment between the lower and upper limits of the hydraulic transmission lag time, and selects the alignment result that meets the preset matching conditions as the basis for generating event matching features. Through this further optional embodiment, it is possible to avoid the inability to form a time correspondence between emission events and pollution events due to a single hydraulic transmission lag time value error.

[0124] Furthermore, the multi-source water pollution tracing system calculates the event matching degree based on event matching characteristics. The event matching degree is used to represent the comprehensive matching degree between a pollution source's emission time series segment and the receiving water body's pollution time series segment in terms of time, waveform, direction of change, and chemical ratio. For pollution event e and pollution source i, the multi-source water pollution tracing system generates peak time score, time series overlap score, direction of change score, waveform correlation score, and chemical ratio score based on peak time deviation, time period overlap ratio, direction of change consistency, waveform correlation coefficient, and chemical ratio deviation, respectively.

[0125] The peak time score is determined based on the comparison between the peak time deviation and a preset peak time deviation threshold. When the peak time deviation is not greater than the preset threshold, the peak time score increases as the peak time deviation decreases; when the peak time deviation is greater than the preset threshold, the peak time score is 0. The time period overlap score is determined by the time period overlap ratio; the larger the overlap ratio, the closer the overlap score is to 1. The change direction score is determined based on the consistency of the change direction; when the change direction of the pollution load emitted by the pollution source is consistent with the change direction of the pollution load in the receiving water body, the change direction score is 1; when the two change directions are inconsistent, the change direction score is 0. The waveform correlation score is determined based on the waveform correlation coefficient; the larger the correlation coefficient, the closer the waveform correlation score is to 1. The chemical ratio score is determined based on the comparison between the chemical ratio deviation and a preset threshold. When the chemical ratio deviation is not greater than the preset threshold, the chemical ratio score increases as the deviation decreases; when the deviation is greater than the threshold, the chemical ratio score is 0.

[0126] In one computational example, the degree of event matching is calculated as follows:

[0127] in, Indicates a pollution incident Lower pollution source The corresponding event matching degree, and 0≤ ≤1; Indicates the pollution source index; Indicates the index of pollution events; Indicates peak time score; Indicates overlapping time periods in the scoring; The score indicates the direction of change; Indicates waveform-related scores; Indicates the chemical ratio score; , , , and These represent the weights corresponding to the peak time score, time period overlap score, direction of change score, waveform correlation score, and chemical ratio score, respectively. All weights are non-negative numbers. The weights can be set based on the ability of each scoring item in historical pollution events to distinguish the responsible pollution sources, or they can be set in an equal weighting manner during the initial deployment phase of the park.

[0128] In one specific scoring method, the peak time score is calculated as follows:

[0129] in, Indicates a pollution incident Lower pollution source The corresponding peak time deviation, wherein the peak time deviation is the pollution source The absolute time difference between the peak emission time, after correction for hydraulic transmission lag time, and the peak pollution time of the receiving water body; This indicates the preset peak time deviation threshold; This indicates taking the maximum value among the values ​​within the parentheses; when Not greater than When the peak time deviation is smaller, The closer to 1; when Greater than hour, Take 0.

[0130] The time period overlap score is calculated as follows:

[0131] in, Indicates a pollution incident Lower pollution source The corresponding time period overlaps in the scoring; Indicates pollution source The time period during which it was turned on and the pollution event The corresponding overlap duration between the emission periods is calculated. Indicates the pollution source corresponding to pollution event e. The total duration of the retrospective emission period; This represents a preset correction amount used to avoid a denominator of zero, and It is a positive number. Because... Not greater than The value range of the time period overlap score is 0 to 1.

[0132] The score for the direction of change is calculated as follows:

[0133] in, Indicates a pollution incident Lower pollution source The corresponding direction of change is scored; Indicates pollution source The direction of pollution load change within the corresponding emission time segment; This indicates the direction of pollution load change within a time series segment of the receiving water body. The direction of pollution load change can be determined based on the difference between the first and last values ​​of the comprehensive pollution load sequence within the corresponding time series segment or the slope of a linear regression. When the direction of pollution load change from the pollution source is consistent with the direction of pollution load change in the receiving water body, the direction of change score is 1; when they are inconsistent, the direction of change score is 0.

[0134] Waveform correlation scores are calculated as follows:

[0135] in, Indicates a pollution incident Lower pollution source Corresponding waveform-related scores; Indicates pollution source The waveform correlation coefficient between the discharge time series segment and the pollution time series segment of the receiving water body after alignment with the hydraulic transport lag time is given. This waveform correlation coefficient can be calculated using the Pearson correlation coefficient, and its value range is […]. 1,1];when When the value is positive, the waveform correlation score equals ;when When the value is zero or negative, the waveform correlation score is 0.

[0136] The chemical ratio score is calculated as follows:

[0137] in, Indicates a pollution incident Lower pollution source The corresponding chemical ratio score; Indicates a pollution incident Lower pollution source The corresponding chemical ratio deviation, wherein the chemical ratio deviation is determined by the ratio of pollutant concentrations in the time series segment of pollution in the receiving water body to the pollution source. The normalized difference between the center values ​​of the ratios in the chemical ratio fingerprint is formed; This indicates the preset chemical ratio deviation threshold; This indicates taking the maximum value among the values ​​within the parentheses. When... Not greater than When the chemical ratio deviation is smaller, The closer to 1; when Greater than hour, Take 0.

[0138] The water pollution multi-source tracing system combines peak time score, time period overlap score, change direction score, waveform correlation score, and chemical ratio score according to the above weights to obtain the event matching degree. Since each individual score is between 0 and 1, and; , , , and All are non-negative numbers and the sum of their weights is 1, therefore the event matching degree is... The value range is from 0 to 1. The multi-source water pollution tracing system correlates the calculated event matching degree with the event matching features and outputs the result. The event matching degree is used as a component of the pollution source event feature vector in S105, and is used in S1063 to determine the upper limit of the contribution coefficient corresponding to the candidate pollution source.

[0139] S105. Input the chemical ratio fingerprint, temporal behavioral fingerprint and event matching features into the source tracing classification model to obtain the responsibility probability of each pollution source.

[0140] S105 is used to input the historical chemical composition relationship of pollution sources, the historical emission behavior pattern of pollution sources, and the time period matching relationship of pollution events into the source tracing and classification model, and output the responsibility probability corresponding to each pollution source. The responsibility probability is used as the input data for S106 to determine the multi-source unmixing trigger indication; obtaining the responsibility probability of each pollution source includes the following steps: S1051, a pollution source event feature vector is constructed based on chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features, generating a pollution source event feature vector. Specifically, for each pollution source under the same pollution event, the water pollution multi-source tracing system reads the corresponding chemical ratio fingerprint, temporal behavioral fingerprint, event matching features, and data reliability level. The chemical ratio fingerprint provides the central value of the chemical ratio, the allowable deviation range of the ratio, and the chemical ratio deviation. The temporal behavioral fingerprint provides information on the emission cycle, peak period, intermittency index, concentration variation coefficient, and emission trend. The event matching features provide peak time deviation, period overlap ratio, consistency of change direction, and waveform correlation coefficient. The data reliability level provides the degree of data reliability when the pollution source participates in the tracing.

[0141] The water pollution multi-source tracing system combines the above data according to a preset feature order and performs normalization processing on features with inconsistent numerical scales. Continuous numerical features are normalized using maximum-minimum or mean-standard deviation normalization; judgment class features are converted to 0 or 1; and data confidence levels are converted to confidence level values. After the combination is completed, the water pollution multi-source tracing system generates pollution source event feature vectors. The pollution source event feature vectors are used as input data to the source tracing classification model in S1052; the preset feature order is predetermined according to the input dimension of the source tracing classification model and remains consistent during the training and application phases of the source tracing classification model.

[0142] S1052, input the pollution source event feature vector into the source tracing classification model to generate unnormalized scores for each pollution source; specifically, the source tracing classification model can be a gradient boosting decision tree model, a random forest model, a support vector machine model, a multilayer perceptron model, or other models capable of outputting classification scores; input the pollution source event feature vectors corresponding to each pollution source under the same pollution event into the source tracing classification model; the source tracing classification model outputs unnormalized scores for each pollution source based on the chemical ratio deviation, temporal behavior matching degree, event matching degree, and data credibility level in the feature vectors; the unnormalized scores of each pollution source are used as input data for responsibility probability normalization in S1053.

[0143] S1053, normalize the unnormalized scores of each pollution source to generate the responsibility probability of each pollution source; specifically, the water pollution multi-source tracing system normalizes the unnormalized scores of all candidate pollution sources under the same pollution event, so that the responsibility probabilities of each pollution source are on the same numerical scale; the set of candidate pollution sources is a non-empty set; after normalization, the sum of the responsibility probabilities of each pollution source under the same pollution event is 1; the pollution source responsibility probability is used to represent the possibility of the corresponding pollution source being the source of responsibility for the pollution event, and can be compared with the preset responsibility probability threshold.

[0144] In one calculation example, the probability of source liability for pollution is calculated as follows:

[0145] in, Let i be the probability of responsibility for pollution source i under pollution event e; The unnormalized score output by the source tracing classification model for pollution source i; This represents the unnormalized score output by the source tracing classification model for pollution source j; Let e ​​be the set of candidate pollution sources corresponding to pollution event e; The index of the pollution source in the candidate pollution source set; It is an exponential function; the liability probability of each pollution source is used as the input data for determining the multi-source unmixing trigger indication in S106.

[0146] S1054, The source tracing classification model is trained based on historical pollution event samples to generate a trained source tracing classification model; specifically, during the model training phase, the water pollution multi-source tracing system acquires historical pollution event samples, manually verified samples, or hydraulic diffusion simulation samples. Each training sample includes a pollution source event feature vector and responsibility labeling information; the responsibility labeling information can be single-source responsibility labeling or multi-source responsibility ratio labeling.

[0147] When the responsibility labeling information is a single-source responsibility label, the water pollution multi-source tracing system uses cross-entropy loss to correct the model parameters of the tracing classification model; when the responsibility labeling information is a multi-source responsibility ratio label, the water pollution multi-source tracing system uses mean squared error loss or cross-entropy loss to correct the model parameters of the tracing classification model; the training convergence conditions include the validation set loss no longer decreasing within a preset number of rounds, the model loss value being less than a preset loss threshold, or the number of training rounds reaching the preset training round limit.

[0148] After training is completed, the water pollution multi-source tracing system generates a trained source classification model; the trained source classification model is used in S1052 to score the feature vector of pollution source events.

[0149] In some implementations, an example is given below to illustrate the process of generating the pollution source liability probability. The candidate pollution source set corresponding to pollution event e includes pollution source A, pollution source B, and pollution source C. The source tracing classification model, based on chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features, outputs unnormalized scores of 2.1, 1.8, and 0.4 for pollution sources A, B, and C, respectively. The water pollution multi-source tracing system performs normalization processing according to the following formula:

[0150] After normalization, the pollution source responsibility probabilities corresponding to pollution source A, pollution source B, and pollution source C are approximately 0.52, 0.39, and 0.09, respectively. These pollution source responsibility probabilities serve as input data for determining the multi-source unmixing trigger indication in S106. If the preset responsibility probability threshold is 0.30, then both pollution source A and pollution source B meet the probability conditions for entering the candidate responsibility source judgment. If the preset probability difference threshold is 0.20, the responsibility probability difference between pollution source A and pollution source B is 0.13, which is less than the preset probability difference threshold, then S106 continues to determine whether to enter the multi-source unmixing.

[0151] In one optional embodiment, the source tracing classification model uses a rule-based scoring model instead of a machine learning classification model. Specifically, the water pollution multi-source tracing system reads the chemical ratio deviation, peak time deviation, time period overlap ratio, waveform correlation coefficient, emission cycle matching, and peak time period matching corresponding to each pollution source, and converts the above features into individual scores ranging from 0 to 1. For the chemical ratio deviation, the water pollution multi-source tracing system compares the pollutant concentration ratio in the pollution time series segment of the receiving water body with the ratio center value in the corresponding pollution source chemical ratio fingerprint. A chemical ratio score is generated when the chemical ratio deviation is not greater than a preset chemical ratio deviation threshold, and the smaller the chemical ratio deviation, the higher the score. The closer the chemical ratio score is to 1, the better. For peak time deviation, the water pollution multi-source tracing system compares the peak time of pollution source discharge (corrected by hydraulic transmission lag time) with the peak time of pollution in the receiving water body. When the peak time deviation is not greater than the preset peak time deviation threshold, a peak time score is generated. The smaller the peak time deviation, the closer the peak time score is to 1. For time period overlap ratio, the water pollution multi-source tracing system divides the intersection time of the time period corresponding to the pollution source's open state and the time period of the backtracked pollution event by the total time period of the backtracked pollution event to obtain the time period overlap ratio score. For waveform correlation coefficients, the multi-source water pollution tracing system calculates the correlation coefficients based on the time-shifted aligned comprehensive pollution load sequence of pollution sources and the comprehensive pollution load sequence of receiving water bodies, and converts the correlation coefficients into waveform scores. The larger the waveform correlation coefficient, the closer the waveform score is to 1. For emission cycle matching, the multi-source water pollution tracing system determines whether the retrospective time period corresponding to the pollution event falls within the emission cycle corresponding time period in the pollution source temporal behavior fingerprint, and generates an emission cycle score. For peak time period matching, the multi-source water pollution tracing system determines whether the retrospective time period corresponding to the pollution event overlaps with the peak time period in the pollution source temporal behavior fingerprint, and generates a peak time period score.

[0152] Furthermore, the water pollution multi-source tracing system performs a weighted summation of chemical ratio scores, peak time scores, time period overlap ratio scores, waveform scores, emission cycle scores, and peak time period scores according to preset scoring weights to obtain a comprehensive tracing score for each pollution source. The preset scoring weights are non-negative, and the sum of all scoring weights is 1. The preset scoring weights can be set based on the ability of each individual score in historical review samples to distinguish responsible pollution sources, or they can be set during the initial deployment phase of the park according to the weights of chemical ratio scores, peak time scores, time period overlap ratio scores, waveform scores, emission cycle scores, and peak time period scores. The water pollution multi-source tracing system normalizes the comprehensive tracing scores of each pollution source under the same pollution event to obtain the responsibility probability of each pollution source. The normalization process can be performed by dividing the comprehensive tracing score of each pollution source by the sum of the comprehensive tracing scores of all candidate pollution sources, so that the sum of the responsibility probabilities of each pollution source under the same pollution event is 1. This alternative embodiment is suitable for the initial deployment stage of parks where the number of historical pollution event samples is insufficient and stable training conditions are not yet available. Furthermore, after accumulating manual verification samples, the comprehensive source tracing score output by the rule scoring model and the manual verification annotation can be used together as training base data to train a machine learning-based source tracing classification model.

[0153] In a further optional embodiment, the water pollution multi-source tracing system feeds back the results of manual verification to the tracing classification model. Specifically, after regulatory personnel confirm the source of responsibility for a pollution incident based on on-site inspections, enterprise operation records, or supplementary sampling results, the water pollution multi-source tracing system associates and saves the pollution source event feature vector corresponding to the pollution incident with the manually verified responsibility labeling information. After reaching a preset number of new samples or a preset update cycle, the water pollution multi-source tracing system adds the new samples to the training sample set to incrementally update the tracing classification model. The updated tracing classification model continues to be used to generate the responsibility probability in subsequent pollution incidents.

[0154] like Figure 8 As shown, in one optional embodiment, after the water pollution multi-source tracing system calls the random forest model to generate the responsibility probability of each pollution source in S105, it also generates the random forest model's distribution data of the responsibility probability of each enterprise. The random forest model's distribution data of the responsibility probability of each enterprise includes pollution source identifiers, pollution source names, pollution source responsibility probabilities, and corresponding graphic markers; wherein, the pollution source responsibility probability is output by the random forest model based on the pollution source event feature vector, and the pollution source event feature vector is formed by combining chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features. The random forest model's distribution data of the responsibility probability of each enterprise is used for reading in S106 and participates in the generation of multi-source unmixing trigger indications.

[0155] exist Figure 8In the example shown, candidate pollution sources include E1 (chemical enterprise), E2 (electroplating enterprise), E3 (printing and dyeing enterprise), E4 (food processing enterprise), and E5 (metal surface treatment enterprise). The responsibility probabilities output by the random forest model are 8%, 35%, 40%, 6%, and 11%, respectively. The water pollution multi-source tracing system compares the responsibility probability of each enterprise with preset multi-source responsibility probability thresholds and preset single-source responsibility probability thresholds. The preset multi-source responsibility probability threshold is used to determine whether multiple pollution sources are jointly involved in the pollution event, while the preset single-source responsibility probability threshold is used to determine whether a single pollution source with a concentrated responsibility probability exists. In this embodiment, the preset multi-source responsibility probability threshold is 20%, and the preset single-source responsibility probability threshold is 70%. Since the responsibility probabilities of E2 electroplating enterprise and E3 dyeing and printing enterprise are 35% and 40% respectively, both of which are not less than the preset multi-source responsibility probability threshold of 20%, and both E2 electroplating enterprise and E3 dyeing and printing enterprise are in the discharge state at the time of the pollution event, the water pollution multi-source tracing system generates a multi-source unmixing trigger indication to enter the multi-source unmixing process, and regards E2 electroplating enterprise and E3 dyeing and printing enterprise as priority candidates in the subsequent candidate pollution source set determination process.

[0156] like Figure 9 As shown, in one optional embodiment, the water pollution multi-source tracing system, while generating the responsibility probability of each pollution source by calling the random forest model in S105, also reads the feature importance score output by the random forest model and generates feature importance ranking data. The feature importance ranking data includes feature identifier, feature name, feature type, Gini importance score, and ranking number; wherein, the feature type includes real-time event features, chemical ratio features, and time-series behavioral features, and the Gini importance score is used to characterize the degree of contribution of the corresponding feature to the classification results when the random forest model identifies pollution source responsibility. The water pollution multi-source tracing system sorts each feature according to the Gini importance score from largest to smallest to obtain feature importance ranking data, and provides the feature importance ranking data for the contribution coefficient constraint setting process in S106.

[0157] exist Figure 9In the example shown, the features ranked first include cross-correlation coefficient, COD / NH3-N ratio, intraday peak period, lag fit, main emission cycle, TOC / TP ratio, emission intermittency, weekly peak days, Pb / Cd ion ratio, coefficient of variation, Cr6+ / TN ratio, and trend slope. The water pollution multi-source tracing system, based on feature type identification, categorizes cross-correlation coefficient and lag fit into real-time event features, COD / NH3-N ratio, TOC / TP ratio, Pb / Cd ion ratio, and Cr6+ / TN ratio into chemical ratio features, and intraday peak period, main emission cycle, emission intermittency, weekly peak days, coefficient of variation, and trend slope into time-series behavior features. For real-time event features that are ranked first, the water pollution multi-source tracing system prioritizes reading the event matching features of the corresponding pollution source in S106; for chemical ratio features that are ranked first, the water pollution multi-source tracing system prioritizes reading the corresponding pollutant ratio deviation in S106; for time-series behavior features that are ranked first, the water pollution multi-source tracing system prioritizes reading the emission status and emission behavior matching results of the corresponding pollution source at the time of backtracking emission of the pollution event in S106.

[0158] When the prior ranking features of a pollution source show that it does not match the current pollution event, and the probability of responsibility for that pollution source is lower than a preset low probability threshold, the water pollution multi-source tracing system sets a strong constraint on the contribution coefficient of that pollution source in S106 and records the source of the constraint in the unmixing result. When the prior ranking features of a pollution source show that it matches the current pollution event, and the probability of responsibility for that pollution source is not less than a preset multi-source responsibility probability threshold, the water pollution multi-source tracing system considers that pollution source as a priority candidate in the candidate pollution source set. Therefore, the feature importance ranking data is not displayed as a separate result, but rather serves as input data for setting contribution coefficient constraints and generating the tracing evidence chain in subsequent multi-source unmixing.

[0159] S106. Based on the comparison between the responsibility probability of each pollution source and the preset responsibility probability threshold, a multi-source unmixing trigger indication is determined. When the multi-source unmixing trigger indication is to enter multi-source unmixing, constraints are set on the contribution coefficient of each pollution source according to the emission status and event matching characteristics of each pollution source at the emission time corresponding to the pollution event, and the concentration of pollutants in the receiving water body is unmixed to obtain the pollution contribution ratio of each pollution source.

[0160] like Figure 5 As shown, S106 is used to determine whether to enter multi-source unmixing based on the responsibility probability of each pollution source. When entering multi-source unmixing, it determines the set of candidate pollution sources based on the emission status of pollution sources and event matching characteristics, sets contribution coefficient constraints on the candidate pollution sources, and performs constrained unmixing calculation to obtain the pollution contribution ratio. Specifically, it includes the following steps: S1061, based on the ranking and threshold comparison of the responsibility probabilities of each pollution source, a multi-source unmixing trigger indication is generated. Specifically, the water pollution multi-source tracing system reads the responsibility probabilities of each pollution source output in S105 and sorts them from largest to smallest to obtain a responsibility probability ranking sequence. Subsequently, the water pollution multi-source tracing system sorts each pollution source in descending order of responsibility probability to generate a responsibility probability ranking sequence. When two pollution sources have the same responsibility probability, the water pollution multi-source tracing system determines their positions in the responsibility probability ranking sequence according to the degree of event matching from largest to smallest. When both the responsibility probability and the degree of event matching are the same, the water pollution multi-source tracing system determines their positions in the responsibility probability ranking sequence according to the preset pollution source identification order. The water pollution multi-source tracing system reads the responsibility probability corresponding to the pollution source with the first position in the responsibility probability ranking sequence and calculates the difference between the responsibility probability corresponding to the pollution source with the first position and the responsibility probability corresponding to the pollution source with the second position to obtain the responsibility probability difference.

[0161] A preset liability probability threshold is used to determine whether a pollution source should be included in the candidate liability source set. This threshold is determined by the probability distribution of correct liability sources in historical review samples, regulatory conservative identification requirements, or a manually set minimum liability probability requirement. A preset probability difference threshold is used to determine whether there is a probability separation between the first-ranked pollution source and the second-ranked pollution source. This threshold is determined by the distribution of the difference between the first and second liability probabilities in historical single-source pollution events.

[0162] When the responsibility probability of the first-ranked pollution source is not less than a preset responsibility probability threshold, and the difference in responsibility probability between the first-ranked pollution source and the second-ranked pollution source is not less than a preset probability difference threshold, the water pollution multi-source tracing system generates a single-source tracing indication. When the responsibility probabilities of at least two pollution sources are not less than the preset responsibility probability threshold, or the difference in responsibility probability between the first-ranked pollution source and the second-ranked pollution source is less than the preset probability difference threshold, the water pollution multi-source tracing system generates a multi-source demixing trigger indication to enter multi-source demixing.

[0163] S1062, a candidate pollution source set is determined based on the multi-source unmixing trigger indication, emission status, and event matching characteristics, and a candidate pollution source set is generated; specifically, when the multi-source unmixing trigger indication is to enter multi-source unmixing, the water pollution multi-source tracing system reads the emission status and event matching characteristics of each pollution source at the emission time corresponding to the pollution event; when the pollution source is in a non-open state at the corresponding emission time, the water pollution multi-source tracing system sets the contribution coefficient of the corresponding pollution source to zero; when the pollution source is in an open state and the event matching characteristics meet the preset matching conditions, the water pollution multi-source tracing system includes the corresponding pollution source in the candidate pollution source set.

[0164] The preset matching conditions include at least one of the following: the overlap ratio of time periods is not less than the preset overlap ratio threshold, the peak time deviation is not greater than the preset peak time deviation threshold, the waveform correlation coefficient is not less than the preset waveform correlation threshold, and the chemical ratio deviation is not greater than the preset chemical ratio deviation threshold. The above thresholds are determined by the distribution of the event matching characteristics corresponding to the true source of responsibility in historical review events, or by the regulatory conservative judgment requirements.

[0165] S1063, Based on the emission flow rate and event matching characteristics of each pollution source in the candidate pollution source set, a contribution coefficient constraint is set to generate contribution coefficient constraint conditions; Specifically, for each pollution source in the candidate pollution source set, the water pollution multi-source tracing system reads the corresponding emission flow rate, time period overlap ratio, waveform correlation coefficient, peak time deviation, and chemical ratio deviation; The water pollution multi-source tracing system calculates the event matching degree based on the peak time deviation, time period overlap ratio, waveform correlation coefficient, and chemical ratio deviation; The event matching degree ranges from 0 to 1; The smaller the peak time deviation, the larger the time period overlap ratio, the larger the waveform correlation coefficient, and the smaller the chemical ratio deviation, the closer the event matching degree is to 1.

[0166] The multi-source water pollution tracing system further determines the upper limit of the contribution coefficient based on the proportion of discharge flow and the degree of event matching. The proportion of discharge flow is the ratio of the discharge flow of a candidate pollution source to the sum of the discharge flows of all pollution sources in the candidate pollution source set. The multi-source water pollution tracing system calls a preset upper limit weight combination rule to combine the proportion of discharge flow and the degree of event matching into the upper limit of the contribution coefficient corresponding to the candidate pollution source. The preset upper limit weight combination rule includes flow weight and matching weight, the sum of which is 1. The upper limit of the contribution coefficient is the sum of the product of the proportion of discharge flow and the flow weight, and the product of the degree of event matching and the matching weight. The flow weight and matching weight are preset based on the influence of the flow contribution and event matching on the actual tracing results in historical pollution events at the monitoring section. When historical pollution event calibration data is lacking, the flow weight and matching weight are set with the same weight. The multi-source water pollution tracing system applies boundary restrictions to the generated upper limit of the contribution coefficient: it takes the value 1 when the upper limit of the contribution coefficient is greater than 1, and 0 when it is less than 0, so that the upper limit of the contribution coefficient corresponding to each candidate pollution source is limited to the range of 0 to 1.

[0167] The lower limit of the contribution coefficient for each candidate pollution source is determined based on the valid evidence status of the candidate pollution source. The multi-source water pollution tracing system identifies candidate pollution sources with an event matching degree not less than a preset valid matching threshold and an emission flow ratio not less than a preset valid flow ratio threshold as valid evidence pollution sources. For valid evidence pollution sources, the multi-source water pollution tracing system determines the lower limit of the contribution coefficient from the corresponding upper limit of the contribution coefficient according to a preset lower limit ratio, ensuring that the lower limit of the contribution coefficient is not greater than the corresponding upper limit of the contribution coefficient. For candidate pollution sources that do not meet the preset valid matching threshold or the preset valid flow ratio threshold, the multi-source water pollution tracing system sets the lower limit of the contribution coefficient to 0. The preset lower limit ratio is determined based on the statistical results of the lowest contribution ratio of credible pollution sources in historical pollution events at the monitoring section. When historical pollution event calibration data is lacking, the preset lower limit ratio is a fixed ratio value less than 1.

[0168] For pollution sources not included in the candidate pollution source set, the water pollution multi-source tracing system will fix their corresponding contribution coefficients to 0 and will not participate in subsequent unmixing calculations. After the settings are completed, the water pollution multi-source tracing system generates contribution coefficient constraints. These constraints include the candidate pollution source set, the lower limit of the contribution coefficient for each candidate pollution source, the upper limit of the contribution coefficient for each candidate pollution source, and a summation constraint on the contribution coefficients. Specifically, the contribution coefficient of each candidate pollution source must be no less than its corresponding lower limit and no greater than its corresponding upper limit, and the sum of the contribution coefficients of all candidate pollution sources in the candidate pollution source set must be 1. The contribution coefficient constraints serve as the constraint input for the unmixing calculation in S1064.

[0169] S1064, a constrained unmixing objective function is constructed based on the background deduction concentration of the receiving water body and the source term concentration of the candidate pollution sources, and the constrained unmixing objective function is generated; specifically, the water pollution multi-source tracing system first determines the background deduction concentration of each pollutant during the pollution event period of the receiving water body; the background deduction concentration is obtained by deducting the mean or median of the background concentration within the preset background time period before the pollution event is triggered; the preset background time period consists of multiple sampling intervals before the pollution event is triggered, and does not include the time period of the identified pollution event.

[0170] Subsequently, the water pollution multi-source tracing system reads the normalized pollutant concentration of the candidate pollution source at the corresponding emission time and uses it as the candidate pollution source term concentration; the candidate pollution source term concentration comes from the emission time sequence segment in S104 aligned according to the hydraulic transmission lag time.

[0171] The multi-source pollution tracing system for water pollution uses the background concentration of the receiving water body (subtracted from the concentration of pollutants) as the observed quantity, the concentration of candidate pollution source terms as the source term quantity, and the contribution coefficient of each pollution source as the variable to be determined, constructing a constrained unmixing objective function. In one calculation example, the constrained unmixing objective function is as follows:

[0172] The constraints are:

[0173] in, This represents the contribution coefficient corresponding to pollution source i; This represents the set of pollutants involved in the unmixing process; This represents the weight corresponding to pollutant k. It is a non-negative number and has been normalized. Indicates a pollution incident Background subtracted concentration of pollutant k at monitoring points in the receiving water bodies; Indicates a pollution incident The corresponding set of candidate pollution sources. Indicates pollution source The pollutant k-normalized concentration at the corresponding discharge time after alignment with hydraulic transmission lag time; This indicates the preset penalty coefficient. It is a non-negative number and can be adjusted based on historical unmixing residuals; This indicates the degree of event matching corresponding to pollution source i, and ; Indicates pollution source The corresponding upper limit of the contribution coefficient, and The constrained unmixing objective function is used as the input for S1065 to solve for the pollution contribution ratio.

[0174] S1065, Solve the constrained unmixing objective function to generate the pollution contribution ratio of each pollution source; specifically, the water pollution multi-source tracing system solves the constrained unmixing objective function under the constraints that the contribution coefficients are non-negative, the contribution coefficients are no greater than the upper limit of the contribution coefficients, and the sum of the contribution coefficients is one. The solution method can be constrained least squares solution, quadratic programming solution, or iterative projection solution; after the solution is completed, the water pollution multi-source tracing system determines the contribution coefficient corresponding to each candidate pollution source as the pollution contribution ratio.

[0175] After the solution is completed, the water pollution multi-source tracing system generates the pollution contribution ratio of each pollution source. The pollution contribution ratio and the pollution source responsibility probability output by S105 are used together as the pollution event tracing output data. The pollution source responsibility probability is used to represent the possibility of the pollution source being responsible, and the pollution contribution ratio is used to represent the quantitative contribution of the pollution source to the concentration of pollutants in the receiving water body.

[0176] like Figure 10As shown, in one optional embodiment, after the water pollution multi-source tracing system completes the constrained least squares objective function solution in S106, it also generates comprehensive pollution contribution ratio data for each enterprise. The comprehensive pollution contribution ratio data for each enterprise includes a pollution event identifier, a pollution source identifier, a pollution source name, a comprehensive pollution contribution ratio, a source of contribution coefficient constraints, and an unmixing algorithm identifier. The comprehensive pollution contribution ratio is calculated from the contribution coefficient of the corresponding pollution source. The source of contribution coefficient constraints records whether the pollution source is subject to emission status constraints, event matching feature constraints, or temporal consistency constraints. The unmixing algorithm identifier records that the current pollution contribution ratio is calculated using NNLS unmixing. The comprehensive pollution contribution ratio data for each enterprise is used in S1066 for unmixing residual verification and serves as part of the tracing output data.

[0177] exist Figure 10 In the example shown, the water pollution multi-source tracing system performs NNLS unmixing calculations on E2 electroplating enterprise, E3 dyeing and printing enterprise, and other candidate pollution sources based on the responsibility probability distribution output in S105 and the candidate pollution source set determined in S106. The unmixing results show that the comprehensive pollution contribution ratio of E2 electroplating enterprise is 45%, the comprehensive pollution contribution ratio of E3 dyeing and printing enterprise is 52%, and the total comprehensive pollution contribution ratio of other enterprises is 3%. The water pollution multi-source tracing system associates and saves the above comprehensive pollution contribution ratios with the corresponding pollution source identifiers, identifies E2 electroplating enterprise and E3 dyeing and printing enterprise as the main contributing pollution sources of this pollution event, and records the total contribution of other enterprises as the remaining contribution item.

[0178] In this embodiment, the total pollution contribution ratio of other enterprises includes the sum of the contribution ratios of candidate pollution sources that do not meet the main contribution display conditions; the main contribution display conditions include a pollution contribution ratio not less than a preset main contribution ratio threshold, or a pollution source responsibility probability not less than a preset multi-source responsibility probability threshold and a pollution contribution ratio ranking first. For pollution sources that are not in a discharge state, or pollution sources whose event matching characteristics do not meet the preset matching conditions, the water pollution multi-source tracing system restricts their contribution coefficients to zero or within the corresponding value range according to the contribution coefficient constraints set in S106, and records the source of the constraint in the pollution contribution ratio data of each enterprise. Thus, Figure 10 The overall pollution contribution ratio shown is not displayed as a separate result, but rather as input data for the S1066 residual verification and the final source tracing output.

[0179] like Figure 11As shown, in one optional embodiment, after generating the comprehensive pollution contribution ratio data of each enterprise, the water pollution multi-source tracing system also generates enterprise-specific contribution decomposition data for each pollutant. The enterprise-specific contribution decomposition data for each pollutant includes a pollution event identifier, pollutant identifier, pollutant name, pollution source identifier, pollution source name, pollutant component contribution ratio, discharge cessation constraint flag, and contribution reconstruction value. The pollutant component contribution ratio characterizes the contribution percentage of the corresponding pollution source to the background concentration of the corresponding pollutant after deduction. The discharge cessation constraint flag records whether the corresponding pollution source's contribution coefficient is restricted at the time of the pollution event's retrospective discharge. The contribution reconstruction value is obtained by multiplying the pollution contribution ratio of the corresponding pollution source by the concentration of the corresponding pollution source component. The enterprise-specific contribution decomposition data for each pollutant is read by S1066 when calculating the reconstructed value of the pollutant concentration in the receiving water body and the unmixing residual.

[0180] Specifically, for each pollutant, the multi-source water pollution tracing system reads the pollution contribution ratio of each pollution source and the corresponding source term concentration of the candidate pollution source obtained in S1065. It then sums the reconstructed contribution values ​​of each pollution source for the same pollutant to obtain the reconstructed pollutant concentration value of the receiving water body corresponding to that pollutant. Next, it divides the reconstructed contribution value of each pollution source to that pollutant by the reconstructed pollutant concentration value of the receiving water body corresponding to that pollutant to obtain the pollutant component contribution ratio of the corresponding pollution source. When the reconstructed pollutant concentration value of the receiving water body is zero, the multi-source water pollution tracing system marks the pollutant component contribution ratio corresponding to that pollutant as uncalcifiable and excludes that pollutant from the unmixing residual calculation in S1066, or assigns a corresponding missing item label according to the preset missing item processing rules.

[0181] exist Figure 11 In the example shown, the monitored pollutants include COD, NH3-N, TP, TN, TOC, pH, turbidity, DO, Pb2+, and Cd / Cr. The multi-source water pollution tracing system calculates the contribution ratios of E2 electroplating enterprise, E3 dyeing and printing enterprise, and other enterprises to each of the above-mentioned monitored pollutants, and displays the data in a stacked manner. For COD, NH3-N, TP, TN, TOC, pH, and turbidity, both E2 electroplating enterprise and E3 dyeing and printing enterprise have corresponding contributions. For Pb2+ and Cd / Cr, the contribution ratio of E2 electroplating enterprise is ranked first, and the multi-source water pollution tracing system writes the contribution ratios of pollutants corresponding to Pb2+ and Cd / Cr as evidence for tracing the source of metallic pollutants into the tracing output data. For enterprises that were in a state of suspension at the time of the pollution event's retrospective discharge, the multi-source water pollution tracing system limits their corresponding contribution coefficients to zero and records the suspension constraint mark in the enterprise contribution decomposition data for each pollutant. Therefore, Figure 11 The data shown is not presented as a standalone result, but rather as input data for pollutant concentration reconstruction, unmixing residual verification, and source tracing report sub-items in S1066.

[0182] S1066, perform residual verification on the pollution contribution ratio to generate source tracing output data; specifically, the water pollution multi-source tracing system multiplies the pollution contribution ratio of each pollution source with the concentration of the candidate pollution source term and then performs a weighted summation to obtain the reconstructed value of the pollutant concentration in the receiving water body; the water pollution multi-source tracing system compares the reconstructed value of the pollutant concentration in the receiving water body with the background deducted concentration in the receiving water body to obtain the unmixing residual; the unmixing residual can be formed by the weighted sum of squares between the observed value and the reconstructed value, or it can be formed by the relative error.

[0183] The preset unmixing residual threshold is determined by the unmixing residual distribution of historically verified events. Specifically, the water pollution multi-source tracing system acquires several historical pollution events that have undergone manual verification and whose verification conclusion is "passed," and reads the unmixing residual values ​​formed after unmixing calculations for each historical pollution event. The unmixing residual values ​​are sorted and statistically analyzed to obtain the unmixing residual distribution. The water pollution multi-source tracing system determines the preset unmixing residual threshold based on the preset quantile position in the unmixing residual distribution, or based on the sum of the mean of the unmixing residuals and a preset multiple of the standard deviation. This preset unmixing residual threshold is used to characterize the upper limit of residuals allowed in historically verified events. Historically verified events refer to historical pollution events whose pollution source tracing conclusions have been manually verified or confirmed by law enforcement. The unmixing residual value is used to characterize the deviation between the measured pollutant concentration at the monitoring section and the weighted sum of the concentrations contributed by each pollution source in the historical pollution event. By determining the preset unmixed residual threshold by statistically analyzing the unmixed residual distribution from historical review events, the residual judgment of subsequent unmixed results can be based on a comparable range formed by verified events.

[0184] When the unmixing residual is not greater than the preset unmixing residual threshold, the water pollution multi-source tracing system outputs the pollution event identifier, the candidate pollution source set, the responsibility probability of each pollution source, and the pollution contribution ratio of each pollution source; when the unmixing residual is greater than the preset unmixing residual threshold, the water pollution multi-source tracing system outputs a manual verification identifier and indicates the possibility of unmonitored pollution sources, background water quality changes, or hydraulic transmission lag time deviation.

[0185] After completing the residual verification, the water pollution multi-source tracing system generates tracing output data; Figure 5 The pollution source responsibility probability, multi-source unmixing trigger indication, candidate pollution source set, contribution coefficient constraint, constrained least squares unmixing, and pollution contribution ratio processing method correspond to S1061 to S1066.

[0186] In some implementations, the multi-source demixing process in this embodiment is described, and an example is given below; Pollution incident In step S105, the responsibility probabilities of pollution sources A, B, and C are output as 0.52, 0.39, and 0.09, respectively; the preset responsibility probability threshold is 0.30, and the preset probability difference threshold is 0.20. Since both pollution sources A and B are not less than the preset responsibility probability threshold, and the responsibility probability difference between pollution sources A and B is less than the preset probability difference threshold, the water pollution multi-source tracing system generates a multi-source unmixing trigger indication to enter multi-source unmixing. Subsequently, the water pollution multi-source tracing system reads the emission status of pollution sources A and B at the emission time corresponding to the pollution event. If both pollution sources A and B are in the open state, and the event matching characteristics meet the preset matching conditions, the water pollution multi-source tracing system includes pollution sources A and B in the candidate pollution source set. If pollution source C is not in the open state, the contribution coefficient of pollution source C is set to zero.

[0187] Furthermore, the multi-source water pollution tracing system uses the background subtracted concentrations of chemical oxygen demand, ammonia nitrogen, and total phosphorus during the pollution event period of the receiving water body as observations, and the normalized pollutant concentrations of pollution sources A and B at the aligned discharge time as source terms, to construct a constrained unmixing objective function:

[0188] The constraints are:

[0189] If the contribution coefficient of pollution source A is 0.64 and the contribution coefficient of pollution source B is 0.36, then the pollution contribution ratio of pollution source A is 64%, the pollution contribution ratio of pollution source B is 36%, and the pollution contribution ratio of pollution source C is 0, as output by the multi-source water pollution tracing system. The above pollution contribution ratios and the responsibility probabilities of each pollution source are used together as the tracing output data.

[0190] In an alternative embodiment, when the number of pollutants eligible for unmixing in a pollution event is less than the number of candidate pollution sources, resulting in insufficient constraints on the constrained least squares objective function, the water pollution multi-source tracing system uses a constraint allocation method based on event matching degree and emission load ratio to determine the pollution contribution ratio. Specifically, the water pollution multi-source tracing system reads the event matching degree corresponding to each candidate pollution source generated in S1045, and reads the comprehensive pollution load value of each candidate pollution source at the emission time corresponding to the pollution event or within the corresponding emission time segment. The comprehensive pollution load value is normalized to the concentration of each pollutant according to the method described in S1031. The value is weighted by the corresponding pollutant weights; when the corresponding emission time series includes multiple sampling times, the water pollution multi-source tracing system can take the mean, peak, or integral value of the comprehensive pollution load value within the emission time series as the comprehensive pollution load value corresponding to the candidate pollution source, and all candidate pollution sources in the same pollution event use the same value taking method; subsequently, the water pollution multi-source tracing system calculates the ratio of the comprehensive pollution load value corresponding to a certain candidate pollution source to the sum of the comprehensive pollution load values ​​corresponding to all candidate pollution sources in the candidate pollution source set, to obtain the comprehensive pollution load proportion corresponding to the candidate pollution source; the water pollution multi-source tracing system then... The event matching degree and the proportion of comprehensive pollution load are each assigned a preset combined weight. Both preset combined weights are non-negative and their sum is 1. The event matching weight is obtained by multiplying the event matching degree corresponding to the candidate pollution source by the preset combined weight corresponding to the event matching degree. The emission load weight is obtained by multiplying the proportion of comprehensive pollution load corresponding to the candidate pollution source by the preset combined weight corresponding to the proportion of comprehensive pollution load. The event matching weight and the emission load weight are added together to obtain the initial contribution weight corresponding to the candidate pollution source. After calculating the initial contribution weight for each candidate pollution source, the water pollution multi-source tracing system will assign a specific candidate pollution source to the appropriate source. The initial contribution weight corresponding to the selected pollution source is calculated as a ratio to the sum of the initial contribution weights of all candidate pollution sources in the candidate pollution source set, to obtain the pollution contribution ratio corresponding to the candidate pollution source. After normalization, the sum of the pollution contribution ratios of all candidate pollution sources under the same pollution event is 1, and the pollution contribution ratio of pollution sources not included in the candidate pollution source set is set to 0. The preset combination weight can be set according to the difference in the explanatory power of the event matching degree and emission load ratio on the pollution contribution ratio in the historical review pollution events, or the event matching degree and the comprehensive pollution load ratio can be set to equal weights in the initial deployment stage of the park.

[0191] In a further optional embodiment, the water pollution multi-source tracing system performs reliable output control on the pollution contribution ratio. Specifically, after obtaining the pollution contribution ratio of each pollution source, the water pollution multi-source tracing system first calculates the unmixing residual based on the difference between the background concentration of the receiving water body and the concentration of each candidate pollution source term after weighted reconstruction according to the pollution contribution ratio. Then, it statistically analyzes the number of pollution sources in the candidate pollution source set, the event matching degree corresponding to each candidate pollution source, and the proportion of missing key pollutants participating in the unmixing. Among them, the proportion of missing key pollutants is the number of key pollutants in the pollution event for which no effective concentration value was obtained. The ratio between the quantity and the preset total quantity of key pollutants, where key pollutants are determined by pollutants of regulatory concern, enterprise-specific pollutants, and pollutants used for chemical ratio fingerprint calculation; when the unmixing residual is not greater than the preset unmixing residual threshold, the proportion of missing key pollutants is not greater than the preset missing proportion threshold, and the matching degree of the pollution source corresponding to each output pollution contribution proportion in the candidate pollution source set is not less than the preset matching degree threshold, the water pollution multi-source tracing system determines the pollution contribution proportion as a reliable output result and outputs the pollution contribution proportion and reliable output identifier; when the unmixing residual is greater than the preset unmixing residual threshold, When the proportion of missing key pollutants exceeds a preset missing proportion threshold, or when the matching degree between the pollution source and the corresponding event for the output pollution contribution proportion is less than a preset matching degree threshold, the water pollution multi-source tracing system will determine the pollution contribution proportion as the output result to be reviewed, and output the pollution contribution proportion, a low confidence indicator, and a manual review requirement indicator. The preset unmixing residual threshold is determined by the distribution of unmixing residuals from historically reviewed pollution events, the preset missing proportion threshold is determined by the minimum integrity requirement of key pollutants required for unmixing, and the preset matching degree threshold is determined by the event matching degree of the responsible pollution source in historically reviewed pollution events. The distribution is determined; when a low-confidence flag is output, the water pollution multi-source tracing system generates a verification reason based on the item that triggered the low-confidence flag. If the unmixing residual is greater than the preset unmixing residual threshold, the verification reason includes no monitoring of pollution sources, changes in background water quality, or deviations in pollutant attenuation parameters. If the proportion of missing key pollutants is greater than the preset missing proportion threshold, the verification reason includes missing key pollutants. If the event matching degree is less than the preset matching degree threshold, the verification reason includes deviations in hydraulic transmission lag time or insufficient matching of discharge time sequence. This avoids outputting the pollution contribution ratio as a definitive conclusion when the unmixing conditions are insufficient.

[0192] like Figure 6 As shown in the embodiments of this application, a multi-source water pollution tracing system that integrates temporal behavioral fingerprints and chemical ratio fingerprints is also provided. The system includes a data preprocessing unit, a chemical fingerprint generation unit, a temporal fingerprint generation unit, an event matching unit, a responsibility identification unit, and a multi-source demixing unit.

[0193] The data preprocessing unit acquires water quality time-series data from various pollution source discharge outlets and receiving water body monitoring points, performs time normalization, anomaly processing, and flow normalization on the water quality time-series data, and obtains preprocessed water quality time-series data. The output of the data preprocessing unit is input to the chemical fingerprint generation unit, the time-series fingerprint generation unit, and the event matching unit, respectively.

[0194] The chemical fingerprint generation unit calculates the concentration ratios of various pollutants based on pretreated water quality time-series data under normal discharge conditions from each pollution source, generating a chemical ratio fingerprint. The chemical ratio fingerprint includes at least the pollution source identifier, pollutant ratio combination, ratio center value, and allowable deviation range. The chemical fingerprint generation unit's input and outputs include an event matching unit and a responsibility identification unit.

[0195] The time-series fingerprint generation unit extracts discharge cycle, peak period, intermittency index, concentration variation coefficient, and discharge trend information from pretreated water quality time-series data of each pollution source to generate a time-series behavioral fingerprint. The time-series fingerprint generation unit outputs to and inputs to a responsibility identification unit and a multi-source demixing unit.

[0196] The event matching unit is used to determine the hydraulic transmission lag time from each pollution source discharge outlet to the receiving water body monitoring point. After a pollution event is triggered, it performs time-shift alignment of the pollution time series segments of the receiving water body and the discharge time series segments of each pollution source based on the hydraulic transmission lag time to generate event matching features. The event matching unit's input and output are the responsibility identification unit and the multi-source demixing unit.

[0197] The responsibility identification unit inputs chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features into the source tracing classification model to obtain the responsibility probability of each pollution source. The output of the responsibility identification unit is input into the multi-source demixing unit.

[0198] The multi-source unmixing unit is used to determine the multi-source unmixing trigger indication based on the comparison between the responsibility probability of each pollution source and the preset responsibility probability threshold. When the multi-source unmixing trigger indication is to enter multi-source unmixing, it sets constraints on the contribution coefficient of each pollution source based on the emission status of each pollution source at the emission time corresponding to the pollution event and the event matching characteristics, and performs unmixing calculation on the pollutant concentration of the receiving water body to obtain the pollution contribution ratio of each pollution source.

[0199] The water pollution multi-source tracing method described in S101 to S106 above is executed in a computer device, such as... Figure 7As shown, the computer equipment includes a processor, a memory, a communication interface, a water quality time-series data input terminal, and a source tracing result output terminal. The communication interface is used to receive water quality time-series data uploaded by monitoring points at each pollution source discharge outlet and receiving water body. The memory is used to store at least one computer program, preprocessed water quality time-series data, chemical ratio fingerprints, time-series behavioral fingerprints, event matching features, a source tracing classification model, and source tracing output data. The processor is used to load and execute at least one computer program in the memory to implement the water pollution multi-source tracing method described in S101 to S106 above. The source tracing result output terminal is used to output pollution event identifiers, the responsibility probability of each pollution source, the pollution contribution ratio of each pollution source, and manual verification identifiers.

[0200] In one implementation, the computer equipment is a park water environment monitoring platform server. The park water environment monitoring platform server receives data uploaded by online monitoring equipment at discharge outlets, online monitoring equipment at receiving water body monitoring points, and flow monitoring equipment through a communication interface. The processor executes a multi-source water pollution tracing method and finally outputs the probability of responsibility and the proportion of pollution contribution to the monitoring terminal through the tracing result output terminal.

[0201] In one alternative embodiment, the computer device is an edge computing device installed on-site in the industrial park; the edge computing device directly receives real-time monitoring data from the discharge outlet and the receiving water body monitoring point, performs anomaly processing, flow normalization, event matching and preliminary responsibility probability identification locally, and uploads the source tracing output data to the regulatory platform.

[0202] In a further optional embodiment, the computer device is also connected to a manual review terminal; the manual review terminal is used to receive the source tracing output data and return the manual review status. After receiving the manual review pass status, the computer device associates and saves the pretreated water quality time series data, chemical ratio fingerprint, time series behavioral fingerprint, event matching features, responsibility probability, pollution contribution ratio, and review status of the corresponding pollution event, for subsequent updates to the source tracing classification model, chemical ratio fingerprint, or time series behavioral fingerprint.

[0203] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for tracing the multi-source origins of water pollution by fusing temporal behavioral fingerprints and chemical ratio fingerprints, characterized in that, include: Obtain water quality time series data from various pollution source discharge outlets and receiving water body monitoring points, perform anomaly processing and flow normalization on them, and obtain pretreated water quality time series data; Based on the time series data of pretreated water quality under normal discharge conditions of various pollution sources, the concentration ratios of multiple pollutants are calculated to generate chemical ratio fingerprints; Based on the time series data of pretreated water quality from various pollution sources, information on discharge cycle, peak period, intermittency index, concentration variation coefficient and discharge trend is extracted to generate time series behavioral fingerprints; Determine the hydraulic transmission lag time from each pollution source discharge outlet to the monitoring point of the receiving water body, and after a pollution event is triggered, perform time-shift alignment of the pollution time series segments of the receiving water body and the discharge time series segments of each pollution source based on the hydraulic transmission lag time to generate event matching features; By inputting chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features into the source tracing classification model, the probability of responsibility for each pollution source is obtained. The multi-source unmixing trigger indication is determined based on the comparison between the responsibility probability of each pollution source and the preset responsibility probability threshold. When the multi-source unmixing trigger indication is set to enter multi-source unmixing, the contribution coefficient of each pollution source is constrained based on the emission status of each pollution source at the emission time corresponding to the pollution event and the event matching characteristics. The concentration of pollutants in the receiving water body is unmixed and calculated to obtain the pollution contribution ratio of each pollution source.

2. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, Flow normalization involves using the reference discharge flow rate corresponding to each pollution source as a benchmark, multiplying the pollutant concentration at each sampling time by the corresponding discharge flow rate to obtain the pollution load value, and then converting the pollution load value into a normalized concentration value under the reference discharge flow rate. The water quality time series data corresponding to the receiving water body monitoring point are normalized to the same caliber according to the monitoring section flow rate or the pre-set section converted flow rate, so that the pollution source discharge outlet and the receiving water body monitoring point form comparable pretreated water quality time series data.

3. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, Generating a chemical ratio fingerprint involves selecting at least two pollutants from pretreated water quality time series data to form a pollutant ratio combination, and calculating the corresponding pollutant concentration ratio when the denominator pollutant concentration is not less than a preset effective concentration threshold; performing median and discrete range statistics on multiple pollutant concentration ratios formed by the same pollution source under normal discharge conditions to obtain a chemical ratio fingerprint that includes pollutant ratio terms, ratio center values, and allowable deviation ranges of ratios.

4. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, Generating time-series behavioral fingerprints involves dividing the pretreated water quality time-series data corresponding to the pollution source into multiple statistical intervals according to date, shift, or production batch; determining the emission cycle based on the autocorrelation peak position of the pollution load sequence within each statistical interval; determining the peak time period based on the set of sampling times where the pollution load is not less than the preset peak identification threshold; determining the intermittency index based on the number of times the emission status switches between on and off states; and generating concentration variation coefficients and emission trend information based on the pollution load mean, standard deviation, and time regression slope.

5. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, Determining the hydraulic transmission lag time involves acquiring information on the hydraulic path length, cross-sectional velocity, and water flow direction from each pollution source discharge outlet to the receiving water body monitoring point, and forming an initial hydraulic transmission lag time based on the ratio between the hydraulic path length and the cross-sectional velocity. When tracer test data or historical pollution event alignment data are available, the initial hydraulic transmission lag time is corrected based on the time difference between the peak discharge time of the pollution source and the peak response time of the receiving water body to obtain the hydraulic transmission lag time corresponding to each pollution source.

6. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, The generation of event matching features includes determining the start time, end time, peak pollutant time, and pollutant change direction of the pollution time series segment of the receiving water body after the pollution event is triggered; for each pollution source, the corresponding discharge time series segment is back-calculated according to the hydraulic transmission lag time, and the peak time deviation, time overlap ratio, change direction consistency, waveform correlation coefficient, and chemical ratio deviation between the discharge time series segment and the pollution time series segment of the receiving water body are calculated to generate event matching features.

7. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, Determining the multi-source unmixing trigger indication includes sorting each pollution source according to its responsibility probability, determining the responsibility probability of the first-sorted pollution source and the difference in responsibility probabilities between adjacent sorted pollution sources; when the responsibility probability of the first-sorted pollution source is not less than a preset responsibility probability threshold and the difference in responsibility probabilities is not less than a preset probability difference threshold, a single-source tracing indication is determined; when there are at least two pollution sources whose responsibility probabilities are not less than a preset responsibility probability threshold, or whose difference in responsibility probabilities is less than a preset probability difference threshold, a multi-source unmixing trigger indication is determined to enter multi-source unmixing.

8. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, The constraints on the contribution coefficients of each pollution source include determining the set of candidate pollution sources based on the emission status of the pollution sources at the emission time corresponding to the pollution event; setting the contribution coefficient to zero for pollution sources that are not in the emission state; setting the range of contribution coefficient values ​​for pollution sources that are in the emission state and whose event matching characteristics meet the preset matching conditions; the range of contribution coefficient values ​​is jointly determined by the emission flow rate corresponding to the pollution source, the time period overlap ratio in the event matching characteristics, and the chemical ratio deviation.

9. The method for tracing the multiple sources of water pollution according to claim 1, characterized in that, The unmixing calculation of pollutant concentration in the receiving water body includes using the background subtracted concentration of each pollutant during the pollution event period as the observation, the normalized pollutant concentration of the candidate pollution sources at the corresponding discharge time as the source term, and the contribution coefficient of each pollution source as the variable to be solved, to construct a constrained least squares objective function; and solving the objective function within the constraints of non-negative contribution coefficients, a sum of contribution coefficients of one, and candidate pollution source constraints to obtain the pollution contribution ratio of each pollution source.

10. A multi-source tracing system for water pollution that integrates temporal behavioral fingerprints and chemical ratio fingerprints, characterized in that, include: The data preprocessing unit is used to acquire water quality time series data from various pollution source discharge outlets and receiving water body monitoring points, perform anomaly processing and flow normalization on the data, and obtain preprocessed water quality time series data. The chemical fingerprint generation unit is used to calculate the concentration ratio of multiple pollutants and generate chemical ratio fingerprints based on the time series data of pretreated water quality under normal discharge conditions of each pollution source. The temporal fingerprint generation unit is used to extract discharge cycle, peak period, intermittency index, concentration variation coefficient and discharge trend information based on the pretreated water quality temporal data of each pollution source, and generate temporal behavioral fingerprints; The event matching unit is used to determine the hydraulic transmission lag time from each pollution source discharge outlet to the receiving water body monitoring point, and after a pollution event is triggered, to time-shift and align the pollution time sequence of the receiving water body and the discharge time sequence of each pollution source according to the hydraulic transmission lag time to generate event matching features; The responsibility identification unit is used to input chemical ratio fingerprints, temporal behavioral fingerprints, and event matching features into the source tracing classification model to obtain the responsibility probability of each pollution source. The multi-source demixing unit is used to determine the multi-source demixing trigger indication based on the comparison result between the responsibility probability of each pollution source and the preset responsibility probability threshold. When the multi-source demixing trigger indication is to enter multi-source demixing, the unit sets constraints on the contribution coefficient of each pollution source based on the emission status and event matching characteristics of each pollution source at the emission time corresponding to the pollution event, and performs demixing calculation on the pollutant concentration of the receiving water body to obtain the pollution contribution ratio of each pollution source.

Images