A pollution source tracing method based on conductivity double feature matching

Abstract

The application discloses a pollution source tracing method based on conductivity double-feature matching, relates to the technical field of environmental monitoring and water pollution tracing, and is characterized in that: a plurality of emission ratio pollution source emission characteristic curves are simulated and constructed in a laboratory, a fingerprint library is established, dynamic weight weighted Euclidean distance and Softmax confidence are combined to realize accurate tracing, and the method has the advantages that: by simultaneously extracting the peak value feature in the concentration domain and the 10%-90% rise time core feature in the time domain from a pollution response waveform, a two-dimensional core feature vector with decoupled dimensions is formed, a three-dimensional matching identification system is jointly constructed by superimposing an emission ratio working condition parameter, the pollution source identification space is expanded from a single concentration dimension to a three-dimensional space combined with concentration, hydrodynamic force and emission intensity, in a homogenized pollution scene with similar wastewater concentrations, the 10%-90% rise time hydrodynamic force attribute can be relied on to effectively distinguish different pollution sources, and the problem of insufficient identification capability of the prior art in the same-concentration pollution source tracing scene is solved.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and water pollution source tracing technology, specifically a pollution source tracing method based on conductivity dual feature matching. Background Technology

[0002] In drainage pipe network systems, the illegal discharge of industrial wastewater is characterized by its concealment, intermittency, and suddenness. Traditional manual inspection and fixed-point sampling methods are difficult to achieve effective monitoring around the clock. As a comprehensive indicator reflecting the total ion strength of water bodies, conductivity has advantages such as high sensitivity, low monitoring cost, and continuous online data collection. In recent years, it has been widely used in the field of water quality anomaly monitoring and pollution source tracing in drainage pipe networks.

[0003] Patent document CN110297018A discloses a method and device for tracing the source of pollutant emissions from drainage pipe networks. The technical approach is as follows: multi-parameter monitoring modules containing conductivity meters are deployed at key nodes of the pipe network. Early warning thresholds are calculated through cloud-based multivariate regression analysis. When the conductivity exceeds the threshold, reverse source tracing alarm is triggered in combination with the spatial location of the node. This method simplifies the conductivity to a single instantaneous scalar value, which is only used as a basis for judging whether the standard is exceeded. It discards the waveform information of conductivity changing over time. The source tracing capability depends on the density of monitoring nodes. It can only locate the pipe section where the node is located and cannot distinguish the specific source among multiple upstream sewage outlets at the same node.

[0004] Patent document CN121856334A discloses a low-cost and rapid pollution source tracing method based on pipeline network topology analysis. It constructs a directed graph topology model of the pipeline network and uses an improved correlation algorithm to match the conductivity change patterns of upstream and downstream nodes, making preliminary use of the time series characteristics of conductivity. However, the core features extracted by this method are only statistical quantities such as the relative amplitude of conductivity and global numerical similarity. It does not separate morphological features directly related to the transport process from the waveform. When the wastewater concentrations of different pollution sources are similar, the amplitude difference of the conductivity signal disappears. Relying solely on concentration-related statistics cannot effectively distinguish the source. The source tracing capability is significantly reduced in the scenario of homogeneous discharge with similar wastewater concentrations.

[0005] Furthermore, the source tracing results of the aforementioned existing technologies all adopt a deterministic judgment mode to directly output a single conclusion, lacking quantitative assessment of the reliability of the source tracing results and interpretability analysis of the judgment basis. This makes it difficult to meet the requirements for the reliability of source tracing evidence in environmental law enforcement. At the same time, the cloud-based multivariate regression analysis of the former and the topological association matching algorithm of the latter both rely on strong computing resources, which is not conducive to achieving localized rapid source tracing response at monitoring terminals. In summary, the existing pipeline pollution source tracing technology based on conductivity has three main core defects: First, there is a fundamental bottleneck in the identification of pollution sources with the same concentration; second, the source tracing results adopt a deterministic judgment mode, lacking confidence quantification and interpretability, making it difficult to meet the evidence requirements of environmental law enforcement; and third, the algorithm relies on cloud computing resources, which is not conducive to localized deployment at edge terminals. In addition, the existing methods also have several shortcomings in engineering adaptability: they mostly use the peak value of absolute conductivity as a feature, which is easily affected by background water quality fluctuations; fingerprint databases are mostly single-point concentration calibrations, which are difficult to adapt to the actual working conditions of fluctuations in illegal discharge; the false alarm rate of single conductivity warning is high, and there is a lack of qualitative verification of chemical indicators.

[0006] There is an urgent need for an improved source tracing method that can accurately distinguish wastewater with similar concentrations, output quantitative confidence assessment results, and be deployed at edge terminals. To this end, we propose a pollution source tracing method based on conductivity dual feature matching. Summary of the Invention

[0007] The purpose of this invention is to provide a pollution source tracing method based on dual-feature matching of conductivity.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a pollution source tracing method based on conductivity dual-feature matching, wherein the dual features refer to two core identification features: the relative rise peak in the concentration domain and the rise time of 10%-90% in the time domain, and the emission ratio is an auxiliary parameter for operating condition matching. The two features are combined to form a three-dimensional matching vector to complete the three-dimensional feature matching calculation. The method includes the following steps: Step 1: Deploy integrated conductivity and water temperature monitoring terminals with integrated edge computing units at the monitoring nodes of the drainage pipe network, collect raw conductivity time series data, and obtain standard conductivity time series after temperature correction and filtering noise reduction. Step 2: Construct a pollution event identification mechanism that combines end-of-pipe chemical indicators and process conductivity. The chemical monitoring indicators at the end-of-pipe emission outlet exceeding the preset threshold are used as the main early warning trigger condition. Simultaneously, the time series data of the conductivity monitoring nodes in the upstream pipeline network are retrieved. The starting and ending points of the waveform are accurately located using the combined judgment rule of the baseline 3σ noise threshold and the continuously rising slope threshold, and the complete pollution response waveform is captured. Step 3: Extract the peak features in the concentration domain and the 10%-90% rise time features in the time domain from the pollution response waveform to form a two-dimensional core feature vector with decoupled dimensions. Combine this with the emission ratio operating condition parameters to form a three-dimensional matching vector. The peak feature is the difference between the maximum conductivity value in the response waveform and the baseline background conductivity mean value before the pollution event. This is used to eliminate the interference of background conductivity fluctuations on feature stability under different monitoring scenarios. The 10%-90% rise time can independently support the differentiation of pollution sources under the same concentration scenario. Step 4: For each pollution source upstream of the monitoring node, construct corresponding emission characteristic curves through laboratory simulation, and build a pollution source characteristic fingerprint database after associating the pollution source information. Step 5: Perform dimensionless normalization on the real-time feature parameters and the fingerprint database feature parameters, dynamically adjust the weight coefficients of each feature parameter according to the degree of peak difference in the fingerprint database, match the real-time features with the emission feature curves of each pollution source, and complete the multi-dimensional feature matching by weighted Euclidean distance calculation to obtain the pollution source matching degree ranking result. Step 6: Use the Softmax function to convert the matching distance of each pollution source into a source tracing confidence level in the 0-1 interval, and combine the difference between the best confidence level and the second best confidence level to complete the confidence level classification of the source tracing results. Step 7: Calculate the contribution of emission ratio, peak value, and 10%-90% rise time to the source tracing results, and output quantitative source tracing results including confidence ranking, credibility level, and feature contribution analysis.

[0009] As a further aspect of the present invention: In step one, an integrated conductivity and water temperature monitoring terminal with an integrated edge computing unit is deployed at the target monitoring node of the drainage pipe network. Real-time conductivity values ​​and corresponding water temperature data of the water body are continuously collected at a fixed sampling frequency to form a raw conductivity time series with timestamps. All subsequent data preprocessing, feature extraction, matching calculation, and result output processes are completed locally on the terminal, without relying on cloud computing resources. Temperature correction is applied to the raw data, normalizing the measured conductivity at different water temperatures to a 25℃ reference value. The correction formula is: ; in, for The measured conductivity value at time t. To correct to the standard conductivity value at 25℃, This is the temperature correction factor for conductivity. for The water temperature at each time point is determined; the corrected conductivity sequence is then denoised using a moving average filter to obtain a stable standard conductivity time series.

[0010] As a further aspect of the present invention: In step two, the chemical monitoring indicators at the end emission outlet are used as the core basis for judging pollution events. When the concentration of the end chemical indicator exceeds the corresponding emission standard threshold, it is determined to be a pollution event warning trigger. Simultaneously, data analysis of the upstream conductivity node is initiated. Based on the conductivity data during the normal operation of the pipeline network, the baseline mean and baseline standard deviation are calculated. Based on the joint rule of the baseline 3σ noise threshold and the continuous rising slope threshold, the start and end boundaries of the pollution response waveform are located. When the conductivity values ​​of at least 3 consecutive sampling points exceed the sum of the baseline mean and 3 times the baseline standard deviation, and the conductivity rising slope per unit time is continuously greater than the preset slope threshold, the first sampling point that meets the conditions is marked as the waveform starting rising point. The conductivity change is continuously monitored. When the conductivity falls back to below the sum of the baseline mean and 3 times the baseline standard deviation and remains stable, it is marked as the waveform end point. The complete conductivity curve from the waveform starting rising point to the waveform end point is extracted as the response waveform of a single pollution event, excluding false triggers caused by background random fluctuations and short-term interference.

[0011] As a further aspect of the present invention: in step three, feature analysis is performed on the extracted single pollution response waveform, and two independent core feature parameters in the concentration domain and the time domain are extracted simultaneously. The first is the peak characteristic, which is the difference between the maximum conductivity value in the response waveform and the baseline background conductivity mean before the pollution event. This corresponds to the increase in conductivity caused by the pollutant, reflecting the intrinsic concentration and total ion strength of the wastewater, and can eliminate feature drift caused by differences in background water quality. The second is the 10%-90% rise time characteristic, which is the time span from the position of the baseline background conductivity mean superimposed with the 10% relative rise peak to the position of the baseline background conductivity mean superimposed with the 90% relative rise peak, based on the relative rise peak after background subtraction. This corresponds to the convection-diffusion process of the pollution front moving from the discharge outlet to the monitoring node, and can avoid the interference of baseline noise and peak plateau fluctuations on the extraction of time features. The two feature parameters together constitute a two-dimensional decoupled core feature vector, which, together with the emission ratio operating condition parameter, forms a three-dimensional matching vector. The 10%-90% rise time is not affected by the absolute concentration of pollutants.

[0012] As a further solution of the present invention: in step four, taking the conductivity monitoring node fixedly arranged in the pipe network as a reference, each pollution discharge subject upstream of the node is the corresponding pollution source; for each pollution source, a laboratory simulation method is adopted to construct a discharge characteristic curve: different discharge ratios, that is, the ratio of the pollution discharge amount to the background water amount, are adjusted under the simulated hydraulic conditions of the pipe network, the conductivity time-series data is collected by simulation under each discharge ratio, and temperature correction, waveform interception and feature extraction processing are sequentially completed to obtain a standardized feature vector after deducting the background baseline value under the corresponding working condition, which is taken as the feature point under this discharge ratio; after collecting multiple sets of standardized feature vectors corresponding to different discharge ratios, data fitting is performed to generate the discharge characteristic curve of the pollution source, which reflects the mapping relationship between the discharge ratio and the two characteristic parameters; the discharge characteristic curves of all pollution sources are stored in association with the corresponding pollution source identity information and discharge outlet position parameters, so as to construct a callable pollution source feature fingerprint library as a matching benchmark for subsequent real-time traceability.

[0013] As a further solution of the present invention: in step five, when a real-time pollution event is captured and a real-time feature vector is extracted , which is the real-time relative uplift peak, which is the real-time 10%-90% rise time, the real-time discharge ratio under the corresponding working condition is obtained synchronously , the real-time discharge ratio can be measured by the flow monitoring device matched with the monitoring node, or estimated by combining the pipe network hydraulic model and based on the mass conservation relationship of the pollution response waveform. First, min-max dimensionless normalization processing is performed on the discharge ratio, the relative uplift peak and the 10%-90% rise time respectively to eliminate the dimensional difference and order of magnitude difference of the three parameters, and the normalization formula is: ; ; ; In the formula, is the discharge ratio parameter to be normalized, is the relative uplift peak parameter to be normalized, is the 10%-90% rise time parameter to be normalized, , are the normalized real-time discharge ratio, real-time relative uplift peak and 10%-90% rise time values, are the maximum value and the minimum value of the discharge ratio covered by all the pollution source discharge characteristic curves in the fingerprint library, , are the maximum value and the minimum value of the relative uplift peak covered by all the pollution source discharge characteristic curves in the fingerprint library, , The maximum and minimum values ​​of the rise time are the range of 10%-90% covered by the emission characteristic curves of all pollution sources in the fingerprint database. Before calculating the matching distance, the weighting coefficients are dynamically adjusted based on the peak differences of each contamination source in the fingerprint database: the maximum relative difference rate of peak values ​​in the fingerprint database is calculated, which is the ratio of the difference between the maximum and minimum peak values ​​to the maximum peak value; when the maximum relative difference rate is greater than a preset difference threshold, the peak weighting coefficient is increased. And reduce the rise time weighting coefficient by 10%-90%. Emissions ratio weighting factor Synchronous adjustment; when the maximum relative difference rate is less than or equal to the preset difference threshold, the peak weight coefficient is reduced. And increase the rise time weighting coefficient by 10%-90%. Emissions ratio weighting factor Synchronous and coordinated adjustment, emission ratio weighting coefficient Maintain a fixed baseline value, and only dynamically adjust the peak weight. Weighted by 10%-90% of the rise time The relative proportions of the three factors, with the sum of their weights always being 1, are used. The weighted Euclidean distance algorithm is employed to calculate the minimum matching distance between the normalized real-time feature parameters and the emission characteristic curves of each pollution source in the fingerprint database. The distance calculation formula is as follows: ; in, To normalize the peak value in real time, To achieve real-time normalized emission ratios, To normalize the rise time to 10%-90% in real time, , , For the first The normalized parameter of the matching point closest to the real-time feature vector on the emission characteristic curve of each pollution source. For real-time features and the first The matching distance of each pollution source is used to sort all pollution sources in ascending order of matching distance. The smaller the distance, the higher the matching degree, thus obtaining the pollution source matching degree ranking result.

[0014] As a further aspect of the present invention: In step six, the matching distances of each pollution source obtained in step five are input into the Softmax function for normalization transformation, mapping the abstract distance values ​​to confidence probabilities in the 0-1 interval. The smaller the distance, the higher the confidence level. The confidence level calculation formula is as follows: ; in, For the first The confidence level of tracing the source of pollution. This represents the total number of contamination sources in the fingerprint database. For the first The system matches the distance between pollution sources; it extracts the best confidence level (ranked first) and the second best confidence level (ranked second), calculates the difference between them, and classifies the credibility of the source tracing results according to a preset difference threshold: when the difference is greater than the first threshold, it is judged as high credibility; when the difference is between the first and second thresholds, it is judged as medium credibility; when the difference is less than the second threshold, it is judged as low credibility and indicates the risk of multi-source mixing, replacing the traditional deterministic hard judgment output mode.

[0015] As a further aspect of the present invention: In step seven, based on the weighting coefficients and single-parameter distance proportions in the weighted Euclidean distance, the contribution proportions of emission ratio, peak characteristics, and 10%-90% rise time characteristics to the source tracing matching results are calculated respectively; finally, a complete source tracing result is output, including a confidence ranking list of suspected pollution sources, the corresponding confidence value of each pollution source, the credibility level of this source tracing, and the contribution proportion analysis of characteristic parameters, forming a quantitative source tracing report that can be directly used for environmental law enforcement decisions.

[0016] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows: 1. This invention extracts the relative rise peak features in the concentration domain and the 10%-90% rise time features in the time domain simultaneously from the pollution response waveform to form a two-dimensional core feature vector with decoupled dimensions. This vector is then superimposed with emission ratio parameters to construct a three-dimensional matching and identification system. This expands the pollution source identification space from a single concentration dimension to a three-dimensional space combining concentration, hydrodynamics, and emission intensity. In homogeneous pollution scenarios with similar wastewater concentrations, different pollution sources can be effectively distinguished based on the hydrodynamic properties of the 10%-90% rise time. This improves the problem of insufficient identification capability of existing technologies in the same concentration pollution source tracing scenario and broadens the applicability of conductivity tracing methods. 2. This invention normalizes the weighted distance matching results using the Softmax function and outputs the source tracing confidence probability corresponding to each pollution source. It combines the difference between the best and second-best confidence scores to complete the credibility classification of the source tracing results and can analyze the contribution ratio of three types of features to the source tracing results. Unlike the traditional deterministic judgment mode, this output method enables the source tracing results to have quantitative confidence and feature-level interpretability, which is more suitable for the reliability assessment needs of source tracing evidence in environmental law enforcement. 3. This invention only needs to extract waveform feature parameters and complete the matching calculation through low-complexity vector distance operation. The entire feature extraction and source tracing calculation process can be deployed locally on the conductivity monitoring terminal with integrated edge computing unit without relying on large-scale computing resources in the cloud. Compared with the existing solution that relies on complex algorithm calculation in the cloud, it can effectively reduce the system deployment cost and improve the response speed and real-time performance of pollution source tracing. 4. This invention constructs emission characteristic curves covering multiple emission ratios through laboratory simulation, replacing the fingerprint database construction method of single-point calibration on site. It can adapt to the actual scenario of fluctuating illegal discharge volume, avoid the source tracing error caused by the mismatch between the single calibration concentration and the actual illegal discharge concentration, and further improve the engineering practicality of the method. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the method steps in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the process of constructing the pollution source feature fingerprint database in an embodiment of the present invention; Figure 3 This is a schematic diagram of the multidimensional feature matching and confidence calculation process in an embodiment of the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0019] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0020] Please see the appendix Figure 1 -Appendix Figure 3 This invention discloses a pollution source tracing method based on conductivity dual-feature matching, the method comprising the following steps: Step 1: Deploy integrated conductivity and water temperature monitoring terminals with integrated edge computing units at the monitoring nodes of the drainage pipe network, collect raw conductivity time series data, and obtain standard conductivity time series after temperature correction and filtering noise reduction. Step 2: Construct a pollution event identification mechanism that combines end-of-pipe chemical indicators and process conductivity. The chemical monitoring indicators at the end-of-pipe emission outlet exceeding the preset threshold are used as the main early warning trigger condition. Simultaneously, the time series data of the conductivity monitoring nodes in the upstream pipeline network are retrieved. The starting and ending points of the waveform are accurately located using the combined judgment rule of the baseline 3σ noise threshold and the continuously rising slope threshold, and the complete pollution response waveform is captured. Step 3: Extract the peak features in the concentration domain and the 10%-90% rise time features in the time domain from the pollution response waveform to form a two-dimensional core feature vector with decoupled dimensions. Combine this with the emission ratio operating condition parameters to form a three-dimensional matching vector. The peak feature is the difference between the maximum conductivity value in the response waveform and the mean baseline background conductivity value before the pollution event. This is used to eliminate the interference of background conductivity fluctuations on feature stability under different monitoring scenarios. The 10%-90% rise time can independently support the differentiation of pollution sources under the same concentration scenario. Step 4: For each pollution source upstream of the monitoring node, construct corresponding emission characteristic curves through laboratory simulation, and build a pollution source characteristic fingerprint database after associating the pollution source information. Step 5: Perform dimensionless normalization on the real-time feature parameters and the fingerprint database feature parameters, dynamically adjust the weight coefficients of each feature parameter according to the degree of peak difference in the fingerprint database, match the real-time features with the emission feature curves of each pollution source, and complete the multi-dimensional feature matching by weighted Euclidean distance calculation to obtain the pollution source matching degree ranking result. Step 6: Use the Softmax function to convert the matching distance of each pollution source into a source tracing confidence level in the 0-1 interval, and combine the difference between the best confidence level and the second best confidence level to complete the confidence level classification of the source tracing results. Step 7: Calculate the contribution of emission ratio, peak value, and 10%-90% rise time to the source tracing results, and output quantitative source tracing results including confidence ranking, credibility level, and feature contribution analysis. Example 1

[0021] Source tracing and verification in conventional multi-concentration pollution scenarios This example simulates a typical scenario where four companies have normal production emissions and wastewater concentrations that differ significantly.

[0022] Step 1: Data Acquisition and Temperature Correction Preprocessing The terminal continuously collects raw data. Before the pollution incident, the water body was operating stably. Data was collected at a specific moment: measured water temperature. ℃, measured conductivity Temperature correction factor Take 0.022 / ℃ (a typical value for conventional water bodies), and substitute it into the temperature correction formula: ; A moving average filter with a window size of 5 is used to denoise the corrected sequence, resulting in a stable standard conductivity time series. All calculations are performed on the local edge unit of the terminal without uploading to the cloud.

[0023] Step 2: Joint Early Warning Trigger and Waveform Extraction When chemical monitoring indicators at the end-of-pipe emission outlet exceed the emission standard threshold, a major pollution event warning is triggered, and time-series data from the corresponding conductivity monitoring nodes in the upstream pipeline network are retrieved simultaneously for detailed analysis.

[0024] Calculate the baseline mean during normal operation. Baseline standard deviation A 3σ noise threshold is used, i.e., the trigger threshold is... The threshold for the upward slope is set to .

[0025] The corrected conductivity at three consecutive sampling points was 902. 915 931 All exceeded 896 Furthermore, if the continuous upward slope is greater than the threshold, a pollution event is determined to be triggered, and the first point exceeding the standard is marked as the starting point of the waveform's upward movement (corresponding to the time). ).

[0026] Continuous monitoring continues until the conductivity falls back to the baseline range and remains stable. The complete conductivity curve from the initial rise point of the waveform to the end point of the waveform is extracted as the response waveform of a single pollution event, excluding false triggers caused by background random fluctuations and short-term interference.

[0027] Step 3: Extraction of two-dimensional feature parameters The complete pollution response waveform obtained was analyzed point by point, and two independent core feature parameters in the concentration domain and time domain were extracted simultaneously. These parameters, combined with the emission ratio operating condition parameters, were used to construct a three-dimensional matching vector. One characteristic is the relative rise peak, which is the difference between the maximum conductivity value in the response waveform and the baseline average conductivity before the pollution event. This corresponds to the actual rise in conductivity caused by the pollutant, reflecting the intrinsic concentration and total ionic strength of the wastewater, and can eliminate characteristic drift caused by differences in background water quality. In this embodiment, the maximum conductivity of the waveform is 1835. Therefore, the peak value is relatively raised in real time. ; Secondly, there is the 10%-90% rise time characteristic, which is based on the relative rise peak after background subtraction. The time span corresponding to the rise of conductivity from the position of the baseline background conductivity mean superimposed with the 10% relative rise peak to the position of the baseline background conductivity mean superimposed with the 90% relative rise peak corresponds to the convection-diffusion process of the pollution front moving from the emission outlet to the monitoring node. This characteristic can avoid the interference of baseline noise and peak plateau fluctuations on the extraction of time features, and its stability is better than traditional time features such as peak arrival time and left half peak width.

[0028] This embodiment extracts the real-time feature vector: real-time peak value. Real-time rise time 10%-90% Simultaneously estimate real-time emission ratio .

[0029] Step 4: Fingerprint database emission characteristic curve baseline data The laboratory simulation calibration of four companies has been completed in the early stage: based on the hydraulic conditions of the upstream pipeline network of the monitoring node, multiple sets of different emission ratio conditions were set for each pollution source. Conductivity time series data were repeatedly collected under each set of conditions, and temperature correction, waveform truncation and feature extraction were completed in sequence to obtain standardized feature points under the corresponding emission ratio. Emission characteristic curves of each company were generated by data fitting to reflect the mapping relationship between emission ratio and relative rise peak and 10%-90% rise time.

[0030] The following are typical operating condition characteristic values ​​that are close to the real-time emission ratio: Company A: At an emission ratio of 0.08, the relative peak rise is 360 μS / cm, and the rise time from 10% to 90% is 6.8 min; Company B: At an emission ratio of 0.12, the relative peak rise is 990 μS / cm, and the rise time from 10% to 90% is 10.4 min; Company C: At an emission ratio of 0.18, the relative peak rise is 1700 μS / cm, and the rise time from 10% to 90% is 12.7 min; Company D: At an emission ratio of 0.05, the relative peak rise is 50 μS / cm, and the rise time from 10% to 90% is 5.1 min; The aforementioned emission characteristic curves are associated with and stored in conjunction with the corresponding enterprise's identity information and emission outlet location parameters to construct a pollution source characteristic fingerprint database, which serves as a benchmark for real-time matching.

[0031] Step 5: Dimensionless Normalization and Dynamic Weight Matching First, determine the extreme values ​​of the fingerprint database features: the maximum value of the relative rise peak. minimum value ; Maximum rise time of 10%-90% minimum value Maximum emission ratio minimum value .

[0032] Min-max normalization is performed on real-time features to eliminate differences in units and orders of magnitude. ; Similarly, calculate the normalized values ​​of the standardized characteristics of each enterprise: Company A: The normalized value of the emission ratio is approximately 0.333. Company B: The normalized value of the emission ratio is approximately 0.556. Company C: The normalized value of the emission ratio is approximately 0.889. Company D: The normalized value of the emission ratio is approximately 0.167. Determine the degree of peak difference: Calculate the maximum relative difference rate of peak values ​​within the fingerprint database. The preset difference threshold is 30%. If the percentage is greater than 30%, it is considered a scenario with significant concentration differences, and the weighting coefficient is dynamically adjusted: peak weight. 10%-90% of the time weighting for the rise , This allows for rapid sorting based on concentration features, prioritizing matching.

[0033] The weighted Euclidean distance algorithm is used to calculate the distance between the real-time feature vector and all feature points on the emission characteristic curve of each pollution source. The minimum distance is taken as the final matching distance for that pollution source. The formula is as follows: ; Sort by distance in ascending order: Company B < Company A < Company C < Company D. The smaller the distance, the higher the matching degree, thus obtaining the pollution source matching degree ranking results.

[0034] Step 6: Softmax confidence transformation and confidence leveling Substitute all matching distances into the Softmax function for normalization, and calculate the traceability confidence score for each enterprise: ; The sum of the denominators is approximately 2.89. The final confidence scores for each company are as follows: Company A: ; Company B: ≈35.1% (optimal confidence level); Company C: (Second-best confidence level); Company D: ; The optimal confidence level and the second-best confidence level are extracted, and the difference between them is calculated to be 12.6%. The first threshold for confidence level is 10%, and the second threshold is 3%. Since 12.6% > 10%, the source tracing result is determined to be of high confidence.

[0035] Step 7: Feature Contribution Analysis and Result Output For the optimal matching result, company B, the feature contribution is calculated based on the proportion of each item in the weighted Euclidean distance: the proportion of the weighted distance item of each parameter to the square of the total distance is the contribution proportion of the corresponding parameter.

[0036] The calculations show that the contribution of emissions in this source tracing is approximately 12%, the contribution of peak characteristics is approximately 38%, and the contribution of the 10%-90% rise time characteristics is approximately 50%.

[0037] The final terminal outputs a complete source tracing report locally, which includes a list of suspected pollution sources ranked by confidence level, confidence level values ​​of each enterprise, credibility level, and analysis of the contribution ratio of the three parameters. The calculation time for the entire process of a single event is less than 1 second. Example 2

[0038] Source tracing and verification of pollution scenarios with similar characteristics and concentrations This embodiment simulates a homogeneous scenario where the wastewater concentrations of two companies are very similar, verifying the core capability of this method to distinguish between products with the same concentration by relying on the 10%-90% rise time characteristic.

[0039] The emission conditions of Company A and Company B were adjusted so that the relative peak rise of the wastewater conductivity of the two companies was basically the same under the same emission ratio. The characteristic point data of the corresponding emission characteristic curves are as follows: When the emission ratio is 0.09, the relative peak rise of Company A is 355 μS / cm, and the rise time from 10% to 90% is 6.7 min. When the emission ratio is 0.09, Company B has a relative peak rise of 350 μS / cm and a rise time of 10%-90% of the emission ratio of Company B, which is 10.3 min.

[0040] Company A was selected to conduct a simulated illegal discharge. After the end-of-pipe chemical indicators triggered a pollution warning, the monitoring terminal captured the pollution response waveform and extracted real-time characteristics: the relative rise peak was 352 μS / cm, the 10%-90% rise time was 6.6 min, and the corresponding emission ratio was 0.09.

[0041] The calculated peak relative difference rate is approximately 1.4%, which is less than the preset difference threshold of 30%. This is considered a scenario with the same concentration, and the weighting coefficient is dynamically adjusted: Peak Weight =0.2, 10%-90% rise time weight =0.6, emission ratio weight =0.2, which makes the matching process rely primarily on time-domain hydrodynamic features to achieve accurate differentiation.

[0042] After dimensionless normalization and weighted Euclidean distance calculation, and then substituting into the Softmax formula, the final confidence level of company A is approximately 90.2%, and the confidence level of company B is approximately 9.8%.

[0043] If the difference between the best and second-best confidence scores is greater than the first threshold, it is judged as high confidence. According to the contribution analysis, the contribution of the 10%-90% rise time characteristic accounts for about 78%, the contribution of the peak characteristic accounts for about 12%, and the contribution of the emission ratio accounts for about 10%.

[0044] The final source tracing result accurately points to Company A, proving that in the same concentration scenario with extremely low peak feature differentiation, this method can achieve effective source tracing through the hydrodynamic characteristics of 10%-90% rise time, breaking through the fundamental bottleneck of existing technologies. Example 3

[0045] Edge terminal local deployment verification This embodiment verifies the feasibility of deploying and the performance of the method on low-configuration edge computing terminals, and verifies the advantages of the algorithm's lightweight design.

[0046] The test terminal uses an industrial-grade low-power edge monitoring terminal, with a general-purpose microcontroller as the main controller, equipped with limited running memory and storage space, and no external computing acceleration unit.

[0047] The complete calculation process of this method was compiled and embedded into the terminal firmware, including the entire process of joint early warning response, extraction of relative rise peak and 10%-90% rise time, emission characteristic curve matching, and three-dimensional weighted distance calculation. A stable operation test was carried out for 72 consecutive hours, during which 20 random pollution events were simulated.

[0048] Test results show that the terminal can stably complete the entire process of data acquisition, temperature correction, event recognition, feature extraction, matching calculation and result output without any abnormalities such as program crashes or computational overflows; the calculation time for a single pollution event is consistently in the low millisecond range, far lower than the minute-level sampling interval; the memory usage is extremely low, and there is ample remaining storage space; there is no need to upload the original time series data to the cloud, only the final source tracing results need to be uploaded, and the amount of data transmission is significantly reduced compared to cloud computing solutions.

[0049] Tests have shown that this method has low algorithm complexity and low resource consumption, and can be directly implemented on existing low-cost conductivity monitoring terminals for localized real-time traceability without additional hardware costs, demonstrating high engineering applicability and promotional value.

[0050] Comparative Example Comparison of source tracing performance using existing correlation algorithms This comparative example uses the improved correlation tracing method published in CN121856334A as a control. Parallel tests were conducted under the same test scenario as Examples 1 and 2 to compare the differences in tracing performance between the two.

[0051] In typical multi-concentration scenarios, the comparative method uses the relative variation of conductivity and the correlation between upstream and downstream to trace the source, achieving a final source tracing accuracy of approximately 85%. However, misjudgments are prone to occur between enterprises with small concentration differences. This method requires calling the pipeline topology database for multi-node correlation calculations for a single source tracing, which significantly increases the computation time compared to the method of this invention. Furthermore, it cannot be completed locally on the edge terminal and requires data to be uploaded to the cloud for processing. In addition, its reliance on conductivity as a single indicator for early warning results in a false alarm rate far higher than the combined early warning mechanism of end-point chemical indicators and conductivity of this invention, and it also cannot be completed locally on the edge terminal, requiring data to be uploaded to the cloud for processing.

[0052] In scenarios with the same concentration, the comparative method relies solely on the amplitude and variation characteristics of conductivity. Since the concentrations of the two companies are very close, the global similarity of the time-series waveforms is extremely high, and the correlation and differentiation are extremely low. Ultimately, the accuracy of tracing the source is less than 50%, which basically loses the ability to effectively identify the source. This is significantly different from the accuracy of nearly 90% of the present invention.

[0053] Furthermore, the comparative method only outputs a single pollution source determination result, without confidence quantification assessment, and cannot provide feature contribution analysis as a basis for determination. The source tracing process has low transparency and weaker evidentiary value than the present invention.

[0054] Supplementary comparative experiment: Robustness verification under background water quality fluctuation conditions To further verify the anti-interference capability of this method, a background water quality fluctuation test group was set up: by adjusting the baseline conductivity of the background water, different background water quality conditions such as rainy season and dry season were simulated, and the source tracing accuracy of the present invention and the comparative method were tested respectively.

[0055] Test results show that under background conductivity fluctuations, the comparative method uses the absolute peak value as the core feature, and the feature value deviates significantly with the baseline drift, resulting in a significant decrease in matching accuracy. In contrast, the present invention uses the relative peak value, which eliminates the influence of the baseline background, significantly improving feature stability. The source tracing accuracy is basically unaffected by background water quality fluctuations, demonstrating stronger scenario adaptability and anti-interference robustness.

[0056] Results Summary This invention constructs a pollution source fingerprint by extracting core waveform features from both concentration and time dimensions, and superimposes emission intensity parameters to form a three-dimensional matching system. This fundamentally overcomes the identification bottleneck of existing technologies in pollution scenarios with the same concentration, significantly expanding the applicable scenarios of conductivity-based source tracing methods. The accompanying probabilistic confidence output and feature contribution analysis mechanism change the traditional hard-determination output mode, making the source tracing results more transparent and of greater evidentiary value, thus meeting the actual needs of environmental law enforcement. Furthermore, the pollution early warning mechanism combining end-of-pipe chemical indicators and conductivity effectively reduces the false alarm rate of single conductivity-based early warnings. The use of relative peak rise features eliminates feature drift caused by background water quality fluctuations, and the use of 10%-90% rise time features avoids interference from baseline noise and peak plateaus. The emission characteristic curve constructed by combining emission ratio parameters can adapt to actual scenarios of fluctuating illegal discharge volumes, further improving the method's engineering practicality and adaptability to complex operating conditions.

[0057] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention.

[0058] Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A pollution source tracing method based on conductivity dual-feature matching, characterized in that, The method includes the following steps: Step 1: Deploy integrated conductivity and water temperature monitoring terminals with integrated edge computing units at the monitoring nodes of the drainage pipe network, collect raw conductivity time series data, and obtain standard conductivity time series after temperature correction and filtering noise reduction. Step 2: Construct a pollution event identification mechanism that combines end-of-pipe chemical indicators and process conductivity. The chemical monitoring indicators at the end-of-pipe emission outlet exceeding the preset threshold are used as the main early warning trigger condition. Simultaneously, the time series data of the conductivity monitoring nodes in the upstream pipeline network are retrieved. The starting and ending points of the waveform are accurately located using the combined judgment rule of the baseline 3σ noise threshold and the continuously rising slope threshold, and the complete pollution response waveform is captured. Step 3: Extract the peak features in the concentration domain and the 10%-90% rise time features in the time domain from the pollution response waveform to form a two-dimensional feature vector with decoupled dimensions. Combine this with the emission ratio operating condition parameters to form a three-dimensional matching vector. The peak feature is the difference between the maximum conductivity value in the response waveform and the baseline background conductivity mean value before the pollution event. The 10%-90% rise time can independently support the differentiation of pollution sources under the same concentration scenario. In step three, feature analysis is performed on the extracted single pollution response waveform, and two independent core feature parameters in the concentration domain and time domain are extracted simultaneously. The first characteristic is the peak value, which is the difference between the maximum conductivity value in the response waveform and the baseline background conductivity before the pollution event, corresponding to the magnitude of conductivity increase caused by pollutants. The second characteristic is the 10%-90% rise time, which is the time span from the position of the baseline background conductivity plus a 10% relative rise peak to the position of the baseline background conductivity plus a 90% relative rise peak, based on the relative rise peak after background subtraction. This corresponds to the convection-diffusion process of the pollution front moving from the emission outlet to the monitoring node. The two characteristic parameters together constitute a dimensionally decoupled two-dimensional feature vector, which, together with the emission ratio operating condition parameter, forms a three-dimensional matching vector. The 10%-90% rise time is not affected by the absolute concentration of pollutants. Step 4: For each pollution source upstream of the monitoring node, construct corresponding emission characteristic curves through laboratory simulation, and build a pollution source characteristic fingerprint database after associating the pollution source information. Step 5: Perform dimensionless normalization on the real-time feature parameters and the fingerprint database feature parameters, dynamically adjust the weight coefficients of each feature parameter according to the degree of peak difference in the fingerprint database, match the real-time features with the emission feature curves of each pollution source, and complete the multi-dimensional feature matching by weighted Euclidean distance calculation to obtain the pollution source matching degree ranking result. In step five, when real-time contamination events are captured and real-time feature vectors are extracted... back, To increase the peak value in real time, To obtain the real-time emission ratio under the corresponding operating conditions for the 10%-90% rise time. First, the emission ratio, relative rise peak, and 10%-90% rise time are subjected to min-max dimensionless normalization to eliminate the dimensional and order-of-magnitude differences among the three parameters. The normalization formula is as follows: ； ； ； In the formula, The emission ratio parameter to be normalized, The relative rise peak parameter to be normalized, The rise time parameter is between 10% and 90% and needs to be normalized. , The values ​​represent the normalized real-time emission ratio, real-time relative peak rise, and 10%-90% rise time. These represent the maximum and minimum emission ratios covered by the emission characteristic curves of all pollution sources in the fingerprint database. , This represents the maximum and minimum values ​​of the relative rise peaks covered by the emission characteristic curves of all pollution sources in the fingerprint database. , The maximum and minimum values ​​of the rise time are the range of 10%-90% covered by the emission characteristic curves of all pollution sources in the fingerprint database. Before calculating the matching distance, the weighting coefficients are dynamically adjusted based on the peak differences of each contamination source in the fingerprint database: the maximum relative difference rate of peak values ​​in the fingerprint database is calculated, which is the ratio of the difference between the maximum and minimum peak values ​​to the maximum peak value; when the maximum relative difference rate is greater than a preset difference threshold, the peak weighting coefficient is increased. And reduce the rise time weighting coefficient by 10%-90%. Emissions ratio weighting factor Synchronous adjustment; when the maximum relative difference rate is less than or equal to the preset difference threshold, the peak weight coefficient is reduced. And increase the rise time weighting coefficient by 10%-90%. Emissions ratio weighting factor Synchronous and coordinated adjustments are made; the weighted Euclidean distance algorithm is used to calculate the minimum matching distance between the normalized real-time feature parameters and the emission characteristic curves of each pollution source in the fingerprint database. The distance calculation formula is as follows: ; in, To normalize the peak value in real time, To achieve real-time normalized emission ratios, To normalize the rise time to 10%-90% in real time, , , For the first The normalized parameter of the matching point closest to the real-time feature vector on the emission characteristic curve of each pollution source. For real-time features and the first The matching distances of each pollution source are used to sort all pollution sources in ascending order of matching distance; the smaller the distance, the higher the matching degree. This yields the pollution source matching degree ranking results. Step 6: Use the Softmax function to convert the matching distance of each pollution source into a source tracing confidence level in the 0-1 interval, and combine the difference between the best confidence level and the second best confidence level to complete the confidence level classification of the source tracing results. Step 7: Calculate the contribution of emission ratio, peak value, and 10%-90% rise time to the source tracing results, and output quantitative source tracing results including confidence ranking, credibility level, and feature contribution analysis.

2. The pollution source tracing method based on conductivity dual-feature matching according to claim 1, characterized in that, In step one, an integrated conductivity and water temperature monitoring terminal with an integrated edge computing unit is deployed at the target monitoring node of the drainage pipe network. It continuously collects real-time conductivity values ​​and corresponding water temperature data of the water body at a fixed sampling frequency, forming a raw conductivity time series with timestamps. All subsequent data preprocessing, feature extraction, matching calculation, and result output processes are completed locally on the terminal. Temperature correction is applied to the raw data, normalizing the measured conductivity at different water temperatures to a 25℃ reference value. The correction formula is: ； in, for The measured conductivity value at time t. To correct to the standard conductivity value at 25℃, This is the temperature correction factor for conductivity. for The water temperature at each time point is determined; the corrected conductivity sequence is then denoised using a moving average filter to obtain a stable standard conductivity time series.

3. The pollution source tracing method based on conductivity dual-feature matching according to claim 1, characterized in that, In step two, the chemical monitoring indicators at the end emission outlet are used as the core basis for judging pollution events. When the concentration of the end chemical indicator exceeds the corresponding emission standard threshold, it is determined that a pollution event warning is triggered. Simultaneously, the data analysis of the upstream conductivity node is started. Based on the conductivity data during the normal operation of the pipeline network, the baseline mean and baseline standard deviation are calculated. Based on the joint rule of the baseline 3σ noise threshold and the continuously rising slope threshold, the start and end boundaries of the pollution response waveform are located. When the conductivity values ​​of at least three consecutive sampling points exceed the sum of the baseline mean and three times the baseline standard deviation, and the slope of conductivity increase per unit time is continuously greater than the preset slope threshold, the first sampling point that meets the conditions is marked as the waveform start point; the conductivity change is continuously monitored, and when the conductivity falls back to below the sum of the baseline mean and three times the baseline standard deviation and remains stable, it is marked as the waveform end point; the complete conductivity curve from the waveform start point to the waveform end point is extracted as the response waveform of a single pollution event.

4. The pollution source tracing method based on conductivity dual-feature matching according to claim 1, characterized in that, In step four, the conductivity monitoring node fixed in the pipeline network is used as the benchmark, and each sewage discharge entity upstream of the node is the corresponding pollution source. For each pollution source, a discharge characteristic curve is constructed using laboratory simulation: under simulated pipeline hydraulic conditions, different discharge ratios are adjusted, i.e., the ratio of pollutant discharge to background water volume. Conductivity time-series data is collected under each discharge ratio, and temperature correction, waveform truncation, and feature extraction are performed sequentially to obtain a standardized feature vector after deducting the background baseline value under the corresponding operating condition, which is used as the feature point under that discharge ratio. After collecting multiple sets of standardized feature vectors corresponding to different discharge ratios, the discharge characteristic curve of the pollution source is generated through data fitting to reflect the mapping relationship between the discharge ratio and the dual feature parameters. The discharge characteristic curves of all pollution sources are associated and stored with the corresponding pollution source identity information and discharge outlet location parameters to construct a callable pollution source feature fingerprint database.

5. The pollution source tracing method based on conductivity dual-feature matching according to claim 1, characterized in that, In step six, the matching distances of each pollution source obtained in step five are input into the Softmax function for normalization transformation, mapping the abstract distance values ​​to confidence probabilities in the 0-1 interval. The smaller the distance, the higher the confidence level. The confidence level calculation formula is as follows: ； in, For the first The confidence level of tracing the source of pollution. This represents the total number of contamination sources in the fingerprint database. For the first The distance between pollution sources is matched; the best confidence level ranked first and the second best confidence level ranked second are extracted, the difference between the two is calculated, and the credibility of the source tracing results is classified according to the preset difference threshold: when the difference is greater than the first threshold, it is judged as high credibility; when the difference is between the first threshold and the second threshold, it is judged as medium credibility; when the difference is less than the second threshold, it is judged as low credibility and a risk of multi-source mixing is indicated.

6. The pollution source tracing method based on conductivity dual-feature matching according to claim 1, characterized in that, In step seven, based on the weighting coefficients in the weighted Euclidean distance and the proportion of single-parameter distances, the contribution proportions of emission ratio, peak characteristics, and 10%-90% rise time characteristics to the source tracing matching results are calculated respectively. Finally, a complete source tracing result is output, including a confidence ranking list of suspected pollution sources, the corresponding confidence value of each pollution source, the credibility level of this source tracing, and the contribution proportion analysis of characteristic parameters, forming a quantitative source tracing report that can be directly used for environmental law enforcement decisions.

Images