An ecological environment detection dynamic tracing method and system for multi-source heterogeneous data

Abstract

The present application belongs to the technical field of environmental pollution tracing, and discloses a kind of multi-source heterogeneous data ecological environment detection dynamic tracing method and system;Multi-source heterogeneous emission data of each discharge port in monitoring area is collected, source end fingerprint base library is constructed, and water quality observation data of confluence section is acquired to construct observation vector.Based on the flow data of each discharge port, the dynamic mixing ratio coefficient is calculated, and the isotope characteristic signal in the observation vector of the confluence section and the water quality fingerprint spectrum are executed reverse dilution compensation to generate the reconstructed feature vector.By matching the reconstructed feature vector with the source end fingerprint base library, a step-by-step residual stripping strategy is used to extract multiple contribution sources from complex mixed signals layer by layer until the residual energy is below the preset threshold.Based on the multi-level contribution source sequence and the dynamic mixing ratio coefficient, a dynamic tracing spectrum is generated and updated in real time.The present application can accurately identify and quantify the contribution of each pollution source in a very low signal-to-noise ratio environment, achieving real-time dynamic tracing in a true sense.

Description

Technical Field

[0001] This invention relates to the field of environmental pollution source tracing technology, and more specifically, to a dynamic source tracing method and system for ecological environment monitoring based on multi-source heterogeneous data. Background Technology

[0002] Existing technologies face a severe dilution effect challenge in tracing the sources of pollution in heterogeneous environments. This problem is particularly prominent in scenarios involving the confluence of wastewater in industrial parks or the confluence of multiple tributaries in rivers. Traditional fingerprint tracing technologies mainly rely on the unique isotope ratios and chemical fingerprint component proportions of pollution sources to establish "identification." However, when characteristic wastewaters from different enterprises mix at the confluence section, the unique fingerprint features of each source will experience severe mutual interference and masking. Further complicating matters, in river or pipe network systems, a large amount of clean background water will further dilute the already weak pollution source signals, causing the originally clearly distinguishable isotope features and chemical fingerprints to appear highly mixed and with a sharp decrease in signal intensity at the confluence point. This multiple dilution effect results in a mixed signal with an extremely low signal-to-noise ratio containing the superimposed contributions of multiple potential pollution sources. Existing clustering algorithms and pattern recognition technologies are inadequate when faced with such complex low signal-to-noise ratio environments. They are unable to effectively separate the independent features of each contributing source, nor can they accurately quantify the actual contribution ratio of each source. As a result, traditional source tracing methods lose their due discrimination ability and practical value in real-world environments with multiple mixed sources.

[0003] In view of this, the present invention proposes a dynamic source tracing method and system for ecological environment monitoring based on multi-source heterogeneous data to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a method for dynamic source tracing of ecological environment monitoring using multi-source heterogeneous data, comprising: Collect multi-source heterogeneous emission data from each discharge outlet within the monitoring area and water quality observation data from the confluence section. Construct a source fingerprint database based on the multi-source heterogeneous emission data and construct a confluence section observation vector based on the water quality observation data. The multi-source heterogeneous emission data includes flow data, isotope characteristic signals, and water quality fingerprint spectrum. The dynamic mixing ratio coefficient is calculated based on the flow data of each outlet, and the dynamic mixing ratio coefficient is used to perform reverse dilution compensation on the isotopic characteristic signal and water quality fingerprint in the confluence section observation vector to generate a reconstructed feature vector. The reconstructed feature vector is matched with the source fingerprint base library. Based on the matching results, a step-by-step residual stripping strategy is adopted to extract multi-level contribution sources. The step-by-step residual stripping strategy includes: identifying the main contribution source and deducting it from the reconstructed feature vector based on the theoretical contribution of the main contribution source to obtain the residual vector. The main contribution source identification and deduction are iteratively performed on the residual vector until the residual energy is lower than the preset threshold to obtain the multi-level contribution source sequence. Based on multi-level contributing source sequences and dynamic mixing ratio coefficients, a dynamic source tracing map is generated and output.

[0005] A dynamic source tracing system for ecological environment monitoring based on multi-source heterogeneous data, used to implement the aforementioned dynamic source tracing method for ecological environment monitoring based on multi-source heterogeneous data, includes: The data acquisition unit includes flow sensors, isotope sampling devices and online water quality monitoring instruments installed at each discharge outlet within the monitoring area, as well as a water quality observation device installed at the confluence section. The data acquisition unit is used to collect multi-source heterogeneous emission data from each discharge outlet and water quality observation data from the confluence section. The multi-source heterogeneous emission data includes flow data, isotope characteristic signals and water quality fingerprints. The data processing unit is communicatively connected to the data acquisition unit and includes: The base database construction subunit is used to construct a source fingerprint base database based on multi-source heterogeneous emission data and to construct a confluence section observation vector based on water quality observation data. The compensation calculation subunit is used to calculate the dynamic mixing ratio coefficient based on the flow data of each outlet, and to use the dynamic mixing ratio coefficient to perform reverse dilution compensation on the isotopic feature signal and water quality fingerprint in the observation vector of the confluence section, so as to generate a reconstructed feature vector. The source tracing and parsing subunit is used to match the reconstructed feature vector with the source fingerprint base library. Based on the matching results, a step-by-step residual stripping strategy is adopted to extract multi-level contribution sources. The step-by-step residual stripping strategy includes: identifying the main contribution source and deducting it from the reconstructed feature vector based on the theoretical contribution of the main contribution source to obtain the residual vector. The main contribution source identification and deduction are iteratively performed on the residual vector until the residual energy is lower than the preset threshold to obtain the multi-level contribution source sequence. The map generation subunit is used to generate dynamic source maps based on multi-level contributing source sequences and dynamic mixing ratio coefficients. The output unit is connected in communication with the data processing unit and is used to output dynamic traceability maps to the monitoring terminal.

[0006] The technical effects and advantages of the present invention, which describes a dynamic source tracing method and system for ecological environment monitoring using multi-source heterogeneous data, are as follows: This invention effectively penetrates complex signals subject to multiple dilution interferences, extracting the true characteristic intensity and contribution ratio of each potential contributing source from seemingly chaotic convergence cross-sectional observation signals. This invention improves the accuracy and reliability of pollution source identification, establishing a dynamic source tracing system adaptable to real-world complex environmental conditions. This enables environmental regulatory departments to maintain high-precision source tracing capabilities even under complex operating conditions such as multi-source convergence, background dilution, and emission fluctuations. This invention changes the current passive situation in environmental regulation where "pollution is easy to detect, but the source is difficult to determine," providing strong technical support for accurately combating illegal discharge activities, scientifically formulating pollution control strategies, and effectively allocating environmental responsibility. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of a dynamic source tracing method for ecological environment monitoring using multi-source heterogeneous data according to the present invention; Figure 2 This is a schematic diagram of a dynamic source tracing system for ecological environment monitoring based on multi-source heterogeneous data, as described in this invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] This application provides a method and system for dynamic source tracing of multi-source heterogeneous data in ecological and environmental monitoring. The method is applicable to pollution source tracing in water environment monitoring areas such as rivers and lakes, enabling dynamic source tracing analysis of the confluence of multiple discharge outlets. The system execution entities include, but are not limited to: ecological and environmental monitoring platforms, water quality source tracing analysis systems, watershed pollution prevention and control systems, and intelligent environmental monitoring units, which can be considered as general processing nodes in this application.

[0010] Please see Figure 1 In this embodiment of the invention, the specific implementation process of a dynamic source tracing method for ecological environment monitoring based on multi-source heterogeneous data includes: Multi-source heterogeneous emission data from various discharge outlets within the monitoring area and water quality observation data from the confluence section are collected. A source fingerprint database is constructed based on the multi-source heterogeneous emission data, and a confluence section observation vector is constructed based on the water quality observation data. The multi-source heterogeneous emission data includes flow data, isotope characteristic signals, and water quality fingerprint spectrum.

[0011] The data collection process involves deploying flow meters, isotope sampling devices, and online water quality monitoring instruments at each discharge point within the monitoring area to acquire flow data, isotope characteristic signals, and water quality fingerprints in real time. Flow data reflects the discharge intensity of each discharge point, isotope characteristic signals provide stable source tracer markers, and water quality fingerprints contain combination patterns of various characteristic pollutant concentrations. Simultaneously, mixed water quality observation data is collected at the confluence section to record the comprehensive characteristics after the confluence. The construction of the source fingerprint database standardizes and stores the characteristic data of each discharge point, forming a matching and comparable feature library; the confluence section observation vector organizes the mixed observation data into a structured vector form, facilitating subsequent reverse analysis. These data provide two key elements for source tracing analysis: a "source feature library" and "mixed result samples," ensuring that the source tracing process has clear comparison benchmarks and reliable observational basis.

[0012] The dynamic mixing ratio coefficient is calculated based on the flow data of each outlet, and the dynamic mixing ratio coefficient is used to perform reverse dilution compensation on the isotopic characteristic signal and water quality fingerprint in the confluence section observation vector to generate a reconstructed feature vector.

[0013] The calculation process is based on the principle of water balance, using the ratio of real-time flow at each discharge point to the total flow at the confluence section as a dynamic mixing ratio coefficient, reflecting the volume proportion of each discharge point during the mixing process. Reverse dilution compensation is the core technology for eliminating the dilution effect of background water. For feature components in the confluence section observation vector affected by dilution from clean background water, the dynamic mixing ratio coefficient is applied for reverse amplification and baseline subtraction to restore the original feature intensity of each discharge point before mixing. The generation of the reconstructed feature vector involves recombining the compensated feature components with stable feature components unaffected by dilution, making them approximate the theoretical superposition state of the original features of each discharge point. This process is essentially a "demixing" operation, using mathematical inverse operations to deduce the source state from the mixing result, providing a more accurate comparison benchmark for subsequent feature matching and significantly improving the accuracy of source tracing.

[0014] The reconstructed feature vector is matched with the source fingerprint base library. Based on the matching results, a stepwise residual stripping strategy is adopted to extract multi-level contribution sources. The stepwise residual stripping strategy includes: identifying the main contribution source and deducting it from the reconstructed feature vector based on the theoretical contribution of the main contribution source to obtain the residual vector. The main contribution source identification and deduction are iteratively performed on the residual vector until the residual energy is lower than the preset threshold to obtain the multi-level contribution source sequence.

[0015] The matching process first calculates the multidimensional distance between the reconstructed feature vector and each candidate source in the source fingerprint database, selecting the candidate source with the smallest distance as the main contributing source at the current level. Then, based on the feature baseline and dynamic mixing ratio coefficient of this main contributing source, its theoretical contribution to the reconstructed feature vector is calculated and subtracted from the reconstructed feature vector to obtain the residual vector. The residual vector reflects the uninterpreted portion remaining after subtracting the main contributing source, and the residual energy of the residual vector is used as a measure of the remaining signal strength. The iterative process treats the residual vector as a new analysis object, repeatedly performing the main contributing source identification and subtraction operation, peeling away contributing sources level by level until the residual energy decreases below a preset threshold, indicating that the remaining signal is close to noise level. The multi-level contributing source sequence records the main contributing sources identified at each level and their contribution amounts, constituting a complete source tracing analysis result. This layer-by-layer peeling strategy effectively solves the source identification problem in multi-source mixed situations, decomposing the complex multi-source problem into a series of single-source identification problems, significantly improving the resolution and accuracy of source tracing.

[0016] Based on multi-level contributing source sequences and dynamic mixing ratio coefficients, a dynamic source tracing map is generated and output.

[0017] The generation process first calculates the actual dynamic contribution rate of the main contributing sources at each level based on the multi-level contributing source sequence, and quantifies the relative importance of each source by combining the energy proportion during the residual stripping process. Then, a tree-like topology is constructed with the confluence section as the root node and the main contributing sources at each level as child nodes, using the contribution rate and confidence level as the weight attributes of the edges. Finally, the data is output to the monitoring terminal and periodically updated according to a sliding time window mechanism. This dynamic update mechanism enables the source tracing map to reflect the latest pollution source status and contribution changes in real time, realizing a dynamic evolutionary expression of the source tracing process. This dynamic map not only provides a static answer to "who is the pollution source," but also demonstrates the dynamic trend of "how the contribution changes," providing scientific decision support for environmental supervision and pollution prevention and control.

[0018] In this embodiment of the invention, the detailed implementation steps of collecting multi-source heterogeneous emission data from various discharge outlets within the monitoring area and water quality observation data from the confluence section, constructing a source fingerprint database based on the multi-source heterogeneous emission data, and constructing a confluence section observation vector based on the water quality observation data include: For each discharge outlet within the monitoring area, multi-source heterogeneous emission data are acquired according to a preset sampling period. Heterogeneous fusion and time alignment processing are performed on the multi-source heterogeneous emission data to extract stable isotope ratios and characteristic fingerprint component proportions.

[0019] The processing procedure first involves synchronously collecting flow data, isotope samples, and water quality indicators from each discharge outlet according to a preset sampling period (usually 15-30 minutes); then, timestamp alignment is performed on heterogeneous data with different sampling frequencies and data formats, mapping all data to a unified time base; next, stable isotope ratios (such as δ¹) are extracted from the isotope characteristic signals. 8 The data includes pollutants such as O and δD, as well as the proportions of characteristic fingerprint components in the water quality fingerprint spectrum (e.g., the proportion of each pollutant concentration relative to the total pollution load). Stable isotope ratios are source-conservative and not easily affected by environmental changes, providing reliable long-term tracer markers; the proportions of characteristic fingerprint components reflect the compositional characteristics of emissions, providing rich identification dimensions. This heterogeneous fusion processing integrates multi-source data with different time dimensions, physical dimensions, and variation characteristics into a unified feature representation, providing standardized input for subsequent database construction.

[0020] The stable isotope ratios and characteristic fingerprint component ratios of the same outlet are divided into stable isotope subsets and variable fingerprint subsets. The corresponding first baseline weights and second baseline weights are assigned to them respectively, and the source fingerprint baselines of the outlet are generated by combining them. The source fingerprint base library is composed of the source fingerprint baselines of all outlets.

[0021] The classification process divides the feature data of the same discharge outlet into two subsets based on the temporal stability of the features: a stable isotope subset containing conservative features such as isotope ratios with small temporal variations, and a variable fingerprint subset containing dynamic features such as the proportion of pollutant components that are greatly affected by operating conditions. The weighting process assigns a higher first baseline weight (typically 0.6-0.8) to the stable isotope subset and a lower second baseline weight (typically 0.2-0.4) to the variable fingerprint subset, reflecting the differences in reliability of different features in source tracing matching. The generation of the source fingerprint baseline combines the two subsets and their weights to form a complete feature description of the discharge outlet, including a feature value vector and a corresponding weight vector. The source fingerprint base library consists of the source fingerprint baselines of all discharge outlets within the monitoring area, forming a complete "source feature library." This classification weighting strategy fully utilizes the complementary advantages of different features, ensuring both long-term stability of source tracing (relying on stable features) and sensitivity to short-term changes (relying on variable features), significantly improving the adaptability and accuracy of the source tracing method.

[0022] For the confluence section, water quality observation data over a continuous period of time is acquired. Smoothing filtering is applied to the water quality observation data to remove abnormal fluctuations. The isotopic mixed signal and fingerprint mixed signal after confluence are extracted to construct the confluence section observation vector.

[0023] The processing begins with continuous water quality monitoring at the confluence section, covering a sufficient time window (typically 1-2 hours) to capture the dynamic characteristics of the mixing process. Then, smoothing filtering algorithms such as moving average or median filtering are applied to eliminate abnormal fluctuations caused by water flow disturbances and measurement noise, preserving the true trend of the mixing signal. Next, isotopic mixing signals (isotope ratios after confluence) and fingerprint mixing signals (proportions of pollutant components after confluence) are extracted from the filtered data. The construction of the confluence section observation vector organizes the extracted mixing signals into a vector form according to the same dimensions and order as the source fingerprint baseline, ensuring dimensional consistency for subsequent comparisons. This systematic data processing flow guarantees the quality and comparability of the observation vectors, providing a reliable "mixing result sample" for reverse analysis, which is a crucial prerequisite for accurate source tracing.

[0024] In this embodiment of the invention, the detailed implementation steps for calculating a dynamic mixing ratio coefficient based on the flow data of each discharge outlet, and using the dynamic mixing ratio coefficient to perform reverse dilution compensation on the isotopic characteristic signal and water quality fingerprint in the confluence section observation vector to generate a reconstructed feature vector include: Based on the water balance model, the volume ratio of each outlet at the current moment is calculated as the dynamic mixing ratio coefficient according to the real-time collected flow data of each outlet and the total flow data of the confluence section.

[0025] The calculation process is based on the principle of water balance. It reads the flow meter data from each outlet and the total flow data at the confluence section in real time, and calculates the proportion of each outlet's flow to the total flow. The calculation formula is: ; in, For the first Each outlet at time The dynamic mixing ratio coefficient, This is the flow rate data for this outlet. The total number of discharge outlets. The background clean water flow rate is typically calculated by subtracting the sum of the flow rates from each outlet from the total flow rate. The dynamic mixing ratio directly reflects the volumetric contribution of each outlet during the mixing process, serving as the quantitative basis for subsequent reverse compensation and source contribution assessment. A sliding window smoothing process is applied to the flow data to reduce the impact of instantaneous fluctuations and ensure the stability of the ratio. This dynamic calculation method based on real-time flow can adapt to flow changes, accurately track the time-varying characteristics of the contribution weights of each outlet, and provide a reliable mixing model for dynamic source tracing.

[0026] The target feature components affected by background water dilution are identified from the observation vector of the confluence section. Based on the dynamic mixing ratio coefficient and the preset background value, the target feature components are subjected to inverse amplification and baseline subtraction operations to eliminate the dilution effect of clean background water.

[0027] The identification process first distinguishes between concentration-type features susceptible to dilution (such as pollutant concentration) and ratio-type features unaffected by dilution (such as isotope ratios) from the observation vectors at the confluence section, marking the concentration-type features as target feature components. The reverse amplification operation, based on the inverse process of mixing and dilution, amplifies the observed mixed concentration according to a dynamic mixing ratio coefficient to compensate for the concentration reduction caused by background water dilution; the baseline subtraction operation subtracts the contribution of the background value, eliminating the influence of the baseline concentration of clean background water. The reverse compensation calculation formula is: ; in, For the first Compensation concentration for each discharge outlet characteristic, To observe concentrations at the confluence section, This is the mixing ratio coefficient for the background water. The background value is preset. This reverse compensation operation is essentially a mathematical inverse transformation of the mixing process, restoring the "diluted mixing result" to the "source state before dilution", eliminating the interference of background water dilution effect on feature matching, and significantly improving the accuracy of source identification.

[0028] The target feature components after inverse compensation are recombined with the stable feature components that are not affected by dilution to generate a reconstructed feature vector, which makes the reconstructed feature vector approximate the original feature superposition state of each outlet before mixing.

[0029] The reconstruction process recombines the reverse-compensated target feature components (concentration features with dilution effects eliminated) with stable feature components unaffected by dilution (conservative features such as isotope ratios) according to the dimensional order of the source fingerprint baseline, forming a structurally complete reconstructed feature vector. The physical meaning of the reconstructed feature vector is the theoretical superposition state of the original features from each outlet, that is, the result of simple superposition of the features from each outlet assuming no dilution or interaction. This reconstruction ensures that subsequent matching with the source fingerprint base library is established under "equal conditions," avoiding feature distortion caused by mixing and dilution effects, thus creating conditions for accurate source identification. The generation of the reconstructed feature vector is a key achievement of the reverse analysis, successfully transforming the "observed mixing results" into "directly comparable source features," laying a solid foundation for multi-source analysis.

[0030] In this embodiment of the invention, the detailed implementation steps of matching the reconstructed feature vector with the source fingerprint base library and extracting multi-level contribution sources based on the matching results using a step-by-step residual stripping strategy include: Calculate the multidimensional distance between the reconstructed feature vector and the source fingerprint baseline of each candidate source in the source fingerprint base library, and select the candidate source with the smallest multidimensional distance as the main contributing source at the current level.

[0031] The calculation process iterates through all candidate sources in the source fingerprint database. For each candidate source, a multidimensional distance is calculated between the source fingerprint baseline and the reconstructed feature vector, taking into account the differences between stable isotope subsets and volatile fingerprint subsets. The multidimensional distance is calculated using weighted Euclidean distance, with the following formula: ; in, For the first Multidimensional distance of candidate sources To reconstruct the eigenvector of the th dimensional components, as candidate sources Source fingerprint baseline dimensional components, and These are the dimensional index sets for the stable isotope subset and the variable fingerprint subset, respectively. and These correspond to the first and second baseline weights. The main contributor source selection process chooses the source with the smallest multidimensional distance from all candidate sources, indicating that the source's features are most similar to the reconstructed feature vector and are most likely the main contributor at the current level. This weighted multidimensional distance method fully utilizes feature classification and weight information, automatically emphasizing reliable features and weakening volatile features during the matching process, thus improving the accuracy and robustness of the matching.

[0032] Based on the source fingerprint baseline of the main contributor and its corresponding dynamic mixing ratio coefficient, the theoretical contribution vector of the main contributor in the reconstructed feature vector is calculated.

[0033] The calculation process extracts the source-end fingerprint baseline of the identified main contributing source and, combined with the corresponding dynamic mixing ratio coefficient, calculates the theoretical contribution that source should generate in the reconstructed feature vector. The theoretical contribution vector, obtained by multiplying the values ​​of each dimension of the source-end fingerprint baseline by the dynamic mixing ratio coefficient, reflects the feature strength that the source should contribute according to its flow weight. This calculation assumes that the features of each source maintain a linear superposition relationship during the mixing process, which is a reasonable simplification. The physical meaning of the theoretical contribution vector is "the amount of features that the main contributing source should provide if the reconstructed feature vector is entirely generated by this main contributing source," providing a subtraction benchmark for subsequent residual calculations.

[0034] Subtract the theoretical contribution vector from the reconstructed feature vector to obtain the residual vector at the current level, and then calculate the residual energy of the residual vector.

[0035] The calculation process involves subtracting the reconstructed eigenvector from the theoretical contribution vector dimension-by-dimensional to obtain the residual vector, which reflects the unexplained portion remaining after deducting the main contribution sources. The residual energy calculation uses a vector norm to measure the overall strength of the residual vector, typically employing the 2-norm (Euclidean norm). The residual energy formula is: ; in, For residual energy, The residual vector is the first dimensional components, denoted as the total dimension of the feature vector. The residual energy directly reflects the total intensity of the remaining signal in the current residual vector and serves as the basis for determining whether further stripping is necessary. High residual energy indicates the existence of unidentified contribution sources, while low residual energy indicates that the remaining signal is approaching noise levels. This quantitative evaluation method based on energy measurement provides a clear criterion for terminating the iteration.

[0036] If the residual energy is greater than or equal to the preset threshold, the residual vector is used as the new reconstructed feature vector. The main contribution source identification and theoretical contribution vector subtraction operations are repeated until the residual energy is less than the preset threshold. The output is a multi-level contribution source sequence composed of the main contribution sources extracted from each level.

[0037] The iterative process first determines whether the current residual energy is greater than or equal to a preset threshold (usually set to 10-20% of the initial reconstructed feature vector energy). If the condition is met, the residual vector is used as a new analysis object, and the complete process of multidimensional distance calculation, main contributor source selection, theoretical contribution vector calculation, residual vector update, and residual energy evaluation is repeated. Each iteration identifies a level of main contributor source and records its identity, contribution amount, and matching degree. When the residual energy drops below the preset threshold, it is determined that the remaining signal is insufficient to support reliable source identification, and the iteration process terminates. The multi-level contributor source sequence is composed of the main contributor sources identified at each level arranged in the identification order, completely recording the hierarchical structure from major to minor contributions. This step-by-step stripping strategy decomposes the complex multi-source mixing problem into a series of simple "identification-subtraction-re-identification" steps, achieving effective separation of multiple sources through iteration, significantly improving the source tracing resolution capability, and enabling the identification of multiple coexisting pollution sources and their relative importance.

[0038] In this embodiment of the invention, before outputting the multi-level contribution source sequence composed of the main contribution sources extracted at each level, the method further includes: After determining the main source of contribution at each level, the matching degree of stable isotope subset and the matching degree of variable fingerprint subset corresponding to the main source of contribution are extracted respectively. Based on the first baseline weight and the second baseline weight, a weighted fusion is performed on the matching degree of stable isotope subset and the matching degree of variable fingerprint subset to obtain the comprehensive source tracing confidence of the main source of contribution.

[0039] After identifying the main contributing source, the assessment process calculates the matching degree between the stable isotope subset and the variable fingerprint subset of that source. The matching degree of the stable isotope subset is calculated based on the similarity of conservative features such as isotope ratios, typically using a normalized similarity index (e.g., 1 minus the normalized distance). The matching degree of the variable fingerprint subset is calculated based on the similarity of dynamic features such as the proportion of pollutant components. The weighted fusion process applies the first baseline weight and the second baseline weight to perform a weighted average of the matching degrees of the two subsets to obtain the comprehensive source tracing confidence score, as shown in the formula: ; in, To comprehensively assess the confidence level of source tracing, and These are the stable isotope subset matching degree and the variable fingerprint subset matching degree, respectively. and The corresponding weights are used. The comprehensive source tracing confidence score comprehensively reflects the matching quality of the two types of features, providing a quantitative evaluation of the reliability of the source tracing results. High confidence indicates a good match and reliable source tracing, while low confidence suggests a bias in the match and the need for further investigation. This weighted fusion method of classification evaluation fully utilizes the differentiated information of features, providing a more comprehensive and accurate confidence assessment than a single indicator.

[0040] If the overall source traceability confidence level is lower than the preset confidence threshold, the main contributing source is marked as a suspected variant source, and dynamic calibration is triggered on the volatile fingerprint subset of the suspected variant source. The dynamic calibration updates the volatile fingerprint subset and its second baseline weight based on the latest multi-source heterogeneous emission data of the discharge outlet.

[0041] The inspection process compares the overall source tracing confidence level with a preset confidence threshold (usually set to 0.7-0.8). When the confidence level is lower than the threshold, it indicates that the current matching quality is poor, possibly due to changes in source-end characteristics. The main contributing source is marked as a suspected variant source, triggering a dynamic calibration process. The dynamic calibration process updates the volatile fingerprint subset (features more easily changed by operating conditions) by extracting the current proportion of characteristic fingerprint components from the latest multi-source heterogeneous emission data of the discharge outlet and replacing the volatile fingerprint subset of that source in the source-end fingerprint base library. Simultaneously, the weight of the second baseline is re-evaluated based on the time stability of the latest data; if the latest features show greater volatility, the weight is reduced, and vice versa. This dynamic calibration mechanism enables the system to adaptively track the time-varying nature of source-end features, maintain the timeliness and accuracy of the source-end fingerprint base library, avoid source tracing inaccuracies caused by changes in the source end, and significantly improve the method's adaptability to dynamic environments.

[0042] The theoretical contribution vector and residual vector of the main contribution source are recalculated based on the updated volatile fingerprint subset.

[0043] After the volatile fingerprint subset is updated, the recalculation process re-executes the theoretical contribution vector calculation using the updated source fingerprint baseline (containing the new volatile fingerprint subset and adjusted second baseline weights) to obtain a contribution estimate based on the latest source features. Then, the residual vector and residual energy are recalculated to provide accurate input for subsequent levels of stripping. This real-time recalculation mechanism ensures that the entire step-by-step stripping process uses consistent and up-to-date source information, avoiding error accumulation caused by outdated features and guaranteeing the overall accuracy of the multi-level contribution source sequences. The combination of dynamic calibration and recalculation enables the system to have a self-correcting capability of "identifying problems - updating the model - re-analyzing," which is a crucial guarantee for achieving high-precision dynamic source tracing.

[0044] In this embodiment of the invention, the detailed implementation steps for generating and outputting a dynamic source tracing map based on multi-level contributing source sequences and dynamic mixing ratio coefficients include: Based on the outlet identifiers of the main contributing sources at each level in the multi-level contributing source sequence, and the residual energy proportion of each main contributing source in the stepwise residual stripping process, the actual dynamic contribution rate of each main contributing source is calculated.

[0045] The calculation process first extracts the outlet identifiers of the main contributing sources at each level in the multi-level contributing source sequence to determine the specific pollution source identity. Then, it calculates the residual energy proportion of each level during the step-by-step residual stripping process, i.e., the proportion of the residual energy difference before and after stripping at that level to the energy of the initial reconstructed feature vector. The actual dynamic contribution rate comprehensively considers the dynamic mixing ratio coefficient (reflecting flow contribution) and the residual energy proportion (reflecting feature contribution), obtained through their product or weighted combination, reflecting the comprehensive contribution of each source in both flow and feature dimensions. The main contributing source at the first level usually has the highest actual dynamic contribution rate, indicating its dominance in the mixing result; the contribution rates of subsequent levels decrease sequentially, reflecting the relative importance of secondary contributing sources. This contribution rate calculation method based on multi-dimensional information more accurately reflects the actual contribution of each source to pollution than simple flow proportion, providing a scientific priority ranking for pollution prevention and control.

[0046] Using the confluence section as the root node and the main contribution sources at each level as child nodes, a tree-like topology is constructed. The actual dynamic contribution rate and the corresponding comprehensive source tracing confidence are combined as the weight attributes of the edges between nodes to generate a dynamic source tracing graph.

[0047] The construction process uses the confluence section as the root node of the tree, representing the convergence point of pollution mixing. The main contributing sources at each level of the multi-level contributing source sequence are treated as child nodes, arranged according to identification order or contribution rate. Directed edges are established from the root node to each child node, representing the contribution relationship of each source to the confluence section. The edge weight attribute is designed as a binary tuple, containing two dimensions: actual dynamic contribution rate and comprehensive source tracing confidence. The former reflects the "quantity" of the contribution (how large the contribution is), and the latter reflects the "quality" of the source tracing (how reliable the results are). The dynamic source tracing map is generated using graphical visualization technology, presenting the tree-like topology in an intuitive graphical way. Node size reflects the contribution rate, and edge color or thickness reflects the confidence level, forming a clear and concise display of source tracing results. This tree-like topology clearly expresses key information such as "where the pollution comes from, how much each source contributes, and whether the source tracing is reliable," providing decision-makers with an easy-to-understand and use visualization tool.

[0048] The dynamic source tracing map is output to the monitoring terminal, and the dynamic mixing ratio coefficient and the weight attributes of the edges in the dynamic source tracing map are periodically updated based on the latest acquired multi-source heterogeneous emission data and water quality observation data according to the sliding time window mechanism, so as to realize the dynamic evolution expression of the source tracing process.

[0049] The output process sends the generated dynamic source tracing map to the monitoring terminal (such as the large display screen of the environmental monitoring center, mobile monitoring APP, etc.) via network interface or data bus for remote visualization. The dynamic update mechanism adopts a sliding time window strategy, setting a fixed update cycle (such as every hour or every 30 minutes). At the end of each update cycle, the latest multi-source heterogeneous emission data and water quality observation data are automatically acquired, and the complete process of dynamic mixing ratio coefficient calculation, reverse dilution compensation, stepwise residual stripping, and contribution rate assessment is re-executed to generate the latest source tracing results. Then, the weight attributes of the edges in the dynamic source tracing map (actual dynamic contribution rate and comprehensive source tracing confidence) are updated, and the node structure is adjusted as necessary (such as adding or removing contributing sources). The updated map is immediately pushed to the monitoring terminal for real-time display refresh. This periodic update mechanism enables the dynamic source tracing map to continuously reflect the latest pollution source status and contribution changes, not only showing the current snapshot but also presenting historical evolution trends.

[0050] In this embodiment of the invention, before performing reverse dilution compensation on the isotopic characteristic signals and water quality fingerprints in the confluence section observation vector using a dynamic mixing ratio coefficient, the method further includes: When a sudden increase or decrease in the flow data of any discharge outlet is detected, the outlet is determined to be in a non-steady-state emission state, and the start time of the non-steady-state emission is recorded.

[0051] The detection process monitors the flow data of each discharge point in real time, calculates the deviation ratio between the current flow and the previous average flow, and determines that a sudden increase or decrease in flow has occurred at the discharge point when the deviation exceeds a preset abrupt change threshold (usually 30-50%). This discharge point is marked as a non-steady-state emission state, distinguishing it from normal steady-state emission, and the precise timestamp of the state transition is recorded as the start time of the non-steady-state emission. Non-steady-state emissions typically correspond to special circumstances such as process adjustments, accidental emissions, or intermittent emissions, and their spatiotemporal propagation characteristics differ significantly from steady-state conditions. Timely identification of non-steady-state states is the trigger condition for subsequent spatiotemporal correction, providing crucial state information for accurate source tracing. This non-steady-state detection method based on flow monitoring is simple and effective, capable of capturing significant changes in emission status in real time.

[0052] Based on the spatial distance and hydrodynamic parameters from each discharge outlet to the confluence section within the monitoring area, a transmission delay model is constructed to calculate the characteristic transmission delay of each discharge outlet's discharge characteristics reaching the confluence section.

[0053] The modeling process first measures or queries the spatial distance (actual distance along the flow path) from each discharge outlet to the confluence section. Then, it acquires hydrodynamic parameters, including the average flow velocity of the river segment, the riverbed roughness coefficient, and the riverbed slope—factors affecting water transport. Based on hydraulic theory (such as Manning's formula or measured velocity data), a transport velocity model is constructed to calculate the time required for pollutants to travel from each discharge outlet to the confluence section. The characteristic transport delay considers the location differences of different discharge outlets; farther outlets correspond to longer transport delays, while closer outlets have shorter delays. The transport delay model can further consider complex factors such as dispersion effects and friction loss to improve the accuracy of delay estimation. This physical model-based method for calculating transport delay provides a quantitative basis for spatiotemporal alignment and is a key prerequisite for eliminating spatiotemporal misalignment.

[0054] Based on the feature transmission delay, the source fingerprint baseline of each discharge outlet at a historical moment is spatiotemporally aligned with the current observation vector of the confluence section. Then, based on the aligned source fingerprint baseline and confluence section observation vector, subsequent reverse dilution compensation and stepwise residual stripping operations are performed to eliminate spatiotemporal misalignment bias under non-steady-state emission scenarios.

[0055] The alignment process, based on the current confluence section observation vector, calculates the historical time that each discharge outlet should align with, according to the characteristic transmission delay of each outlet. This is achieved by subtracting the transmission delay of the outlet from the current time, yielding the actual departure time of the emissions from that outlet. Then, it retrieves the source-end fingerprint baselines of each outlet at the corresponding historical time from the historical database. These historical baselines reflect the true characteristics of the emissions "arriving at the confluence section" at the emission time. After spatiotemporal alignment, a correct causal correspondence is established between the mixed features observed at the current confluence section and the source-end fingerprint baselines of each outlet at historical times. Subsequent reverse dilution compensation and stepwise residual stripping operations are performed based on the aligned data, ensuring an accurate temporal match between "what is observed at the confluence section and what each outlet has previously emitted," eliminating spatiotemporal misalignment caused by transmission lag. This spatiotemporal alignment mechanism is particularly important in non-steady-state emission scenarios, as rapidly changing emission characteristics without time correction can lead to severe source tracing errors. By introducing a transmission delay model and spatiotemporal alignment processing, the adaptability to dynamically changing scenarios is significantly improved, achieving true dynamic tracing and ensuring the accuracy and reliability of tracing under various complex working conditions.

[0056] The foregoing has described a dynamic source tracing method for ecological environment monitoring using multi-source heterogeneous data in the embodiments of this application. The following describes a dynamic source tracing system for ecological environment monitoring using multi-source heterogeneous data in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the dynamic source tracing system for ecological environment monitoring based on multi-source heterogeneous data in this application includes: The data acquisition unit includes flow sensors, isotope sampling devices and online water quality monitoring instruments installed at each discharge outlet within the monitoring area, as well as a water quality observation device installed at the confluence section. The data acquisition unit is used to collect multi-source heterogeneous emission data from each discharge outlet and water quality observation data from the confluence section. The multi-source heterogeneous emission data includes flow data, isotope characteristic signals and water quality fingerprints. The data processing unit is communicatively connected to the data acquisition unit and includes: The base database construction subunit is used to construct a source fingerprint base database based on multi-source heterogeneous emission data and to construct a confluence section observation vector based on water quality observation data. The compensation calculation subunit is used to calculate the dynamic mixing ratio coefficient based on the flow data of each outlet, and to use the dynamic mixing ratio coefficient to perform reverse dilution compensation on the isotopic feature signal and water quality fingerprint in the observation vector of the confluence section, so as to generate a reconstructed feature vector. The source tracing and parsing subunit is used to match the reconstructed feature vector with the source fingerprint base library. Based on the matching results, a step-by-step residual stripping strategy is adopted to extract multi-level contribution sources. The step-by-step residual stripping strategy includes: identifying the main contribution source and deducting it from the reconstructed feature vector based on the theoretical contribution of the main contribution source to obtain the residual vector. The main contribution source identification and deduction are iteratively performed on the residual vector until the residual energy is lower than the preset threshold to obtain the multi-level contribution source sequence. The map generation subunit is used to generate dynamic source maps based on multi-level contributing source sequences and dynamic mixing ratio coefficients. The output unit is connected in communication with the data processing unit and is used to output dynamic traceability maps to the monitoring terminal.

[0057] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0058] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0059] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A dynamic source tracing method for ecological environment monitoring using multi-source heterogeneous data, characterized in that, include: Collect multi-source heterogeneous emission data from each discharge outlet within the monitoring area and water quality observation data from the confluence section. Construct a source fingerprint database based on the multi-source heterogeneous emission data and construct a confluence section observation vector based on the water quality observation data. The multi-source heterogeneous emission data includes flow data, isotope characteristic signals, and water quality fingerprint spectrum. Based on the flow data of each outlet, a dynamic mixing ratio coefficient is calculated, and the dynamic mixing ratio coefficient is used to perform reverse dilution compensation on the isotopic feature signal and water quality fingerprint in the confluence section observation vector to generate a reconstructed feature vector. The reconstructed feature vector is matched with the source fingerprint base library. Based on the matching result, a step-by-step residual stripping strategy is adopted to extract multi-level contribution sources. The step-by-step residual stripping strategy includes: identifying the main contribution source and deducting it from the reconstructed feature vector based on the theoretical contribution of the main contribution source to obtain the residual vector. The main contribution source identification and deduction are iteratively performed on the residual vector until the residual energy is lower than a preset threshold to obtain a multi-level contribution source sequence. Based on the multi-level contributing source sequences and the dynamic mixing ratio coefficient, a dynamic source tracing map is generated and output.

2. The method according to claim 1, characterized in that, The data collected includes multi-source heterogeneous emission data from various discharge outlets within the monitoring area and water quality observation data from the confluence section. A source-end fingerprint database is constructed based on the multi-source heterogeneous emission data, and a confluence section observation vector is constructed based on the water quality observation data, including: For each discharge outlet within the monitoring area, multi-source heterogeneous emission data are acquired according to a preset sampling period. Heterogeneous fusion and time alignment processing are performed on the multi-source heterogeneous emission data to extract stable isotope ratios and characteristic fingerprint component ratios. The stable isotope ratio and characteristic fingerprint component ratio of the same outlet are divided into a stable isotope subset and a variable fingerprint subset, respectively assigned a first baseline weight and a second baseline weight, and combined to generate the source fingerprint baseline of the outlet. The source fingerprint base library is composed of the source fingerprint baselines of all outlets. For the confluence section, water quality observation data over a continuous period of time is acquired. The water quality observation data is then subjected to smoothing filtering to remove abnormal fluctuations. The isotopic mixed signal and fingerprint mixed signal after confluence are extracted to construct the observation vector of the confluence section.

3. The method according to claim 1, characterized in that, The process involves calculating a dynamic mixing ratio coefficient based on the flow data from each discharge outlet, and then using this dynamic mixing ratio coefficient to perform inverse dilution compensation on the isotopic characteristic signals and water quality fingerprints in the confluence section observation vector to generate a reconstructed feature vector, including: Based on the water balance model, the volume ratio of each outlet at the current moment is calculated as the dynamic mixing ratio coefficient based on the real-time collected flow data of each outlet and the total flow data of the confluence section. The target feature components affected by background water dilution are identified from the observation vector of the confluence section. Based on the dynamic mixing ratio coefficient and the preset background value, the target feature components are subjected to inverse amplification and baseline subtraction operations to eliminate the dilution effect of clean background water. The target feature components after inverse compensation are recombined with the stable feature components that are not affected by dilution to generate the reconstructed feature vector, so that the reconstructed feature vector approximates the original feature superposition state of each outlet before mixing.

4. The method according to claim 1, characterized in that, The step of matching the reconstructed feature vector with the source fingerprint base library, and extracting multi-level contribution sources based on the matching results using a step-by-step residual stripping strategy, includes: Calculate the multidimensional distance between the reconstructed feature vector and the source fingerprint baseline of each candidate source in the source fingerprint base library, and select the candidate source with the smallest multidimensional distance as the main contributing source at the current level; Based on the source fingerprint baseline of the main contributing source and its corresponding dynamic mixing ratio coefficient, the theoretical contribution vector of the main contributing source in the reconstructed feature vector is calculated. Subtract the theoretical contribution vector from the reconstructed feature vector to obtain the residual vector at the current level, and calculate the residual energy of the residual vector; If the residual energy is greater than or equal to a preset threshold, the residual vector is used as a new reconstructed feature vector, and the main contribution source identification and theoretical contribution vector subtraction operations are repeated until the residual energy is less than the preset threshold. The multi-level contribution source sequence composed of the main contribution sources extracted at each level is then output.

5. The method according to claim 4, characterized in that, Before the output consisting of the multi-level contribution source sequence extracted from each level, the following is also included: After determining the main source of contribution at each level, the stable isotope subset matching degree and the variable fingerprint subset matching degree corresponding to the main source of contribution are extracted respectively. Based on the first baseline weight and the second baseline weight, the stable isotope subset matching degree and the variable fingerprint subset matching degree are weighted and fused to obtain the comprehensive source tracing confidence of the main source of contribution. If the overall source tracing confidence is lower than the preset confidence threshold, the main contributing source is marked as a suspected variant source, and dynamic calibration is triggered on the volatile fingerprint subset of the suspected variant source. The dynamic calibration updates the volatile fingerprint subset and its second baseline weight based on the latest multi-source heterogeneous emission data of the discharge outlet. The theoretical contribution vector and residual vector of the main contribution source are recalculated based on the updated volatile fingerprint subset.

6. The method according to claim 1, characterized in that, The process of generating and outputting a dynamic source map based on the multi-level contributing source sequences and the dynamic mixing ratio coefficient includes: Based on the outlet identifiers of the main contributing sources at each level in the multi-level contributing source sequence, and the residual energy proportion of each main contributing source in the stepwise residual stripping process, the actual dynamic contribution rate of each main contributing source is calculated. Using the confluence section as the root node and the main contribution sources at each level as child nodes, a tree-like topology is constructed. The actual dynamic contribution rate and the corresponding comprehensive source tracing confidence are combined as the weight attributes of the edges between nodes to generate the dynamic source tracing graph. The dynamic source tracing map is output to the monitoring terminal, and the dynamic mixing ratio coefficient and the weight attributes of the edges in the dynamic source tracing map are periodically updated based on the latest acquired multi-source heterogeneous emission data and water quality observation data according to the sliding time window mechanism.

7. The method according to claim 3, characterized in that, Before performing inverse dilution compensation on the isotopic characteristic signal and water quality fingerprint spectrum in the confluence section observation vector using the dynamic mixing ratio coefficient, the method further includes: When a sudden increase or decrease in the flow data of any discharge outlet is detected, the discharge outlet is determined to be in a non-steady-state emission state, and the start time of the non-steady-state emission is recorded. Based on the spatial distance and hydrodynamic parameters from each discharge outlet to the confluence section within the monitoring area, a transmission delay model is constructed to calculate the characteristic transmission delay of each discharge outlet's discharge characteristics reaching the confluence section. Based on the aforementioned feature transmission delay, the source fingerprint baselines of each outlet at historical times are spatiotemporally aligned with the current observation vector of the confluence section. Then, based on the aligned source fingerprint baselines and confluence section observation vectors, subsequent reverse dilution compensation and stepwise residual stripping operations are performed.

8. A dynamic source tracing system for ecological environment monitoring based on multi-source heterogeneous data, used to implement the method described in any one of claims 1 to 7, characterized in that, include: The data acquisition unit includes flow sensors, isotope sampling devices, and online water quality monitoring instruments installed at each discharge outlet within the monitoring area, as well as a water quality observation device installed at the confluence section. The data acquisition unit is used to collect multi-source heterogeneous emission data from each discharge outlet and water quality observation data from the confluence section. The multi-source heterogeneous emission data includes flow data, isotope characteristic signals, and water quality fingerprints. A data processing unit, communicatively connected to the data acquisition unit, includes: The base database construction subunit is used to construct a source fingerprint base database based on the multi-source heterogeneous emission data, and to construct a confluence section observation vector based on the water quality observation data; The compensation calculation subunit is used to calculate the dynamic mixing ratio coefficient based on the flow data of each outlet, and to use the dynamic mixing ratio coefficient to perform reverse dilution compensation on the isotopic feature signal and water quality fingerprint in the confluence section observation vector to generate a reconstructed feature vector. The source tracing and parsing subunit is used to match the reconstructed feature vector with the source fingerprint base library, and extract multi-level contribution sources based on the matching result using a step-by-step residual stripping strategy. The step-by-step residual stripping strategy includes: identifying the main contribution source and deducting it from the reconstructed feature vector based on the theoretical contribution of the main contribution source to obtain the residual vector, and iteratively performing main contribution source identification and deduction on the residual vector until the residual energy is lower than a preset threshold to obtain a multi-level contribution source sequence. The map generation subunit is used to generate a dynamic source map based on the multi-level contributing source sequence and the dynamic mixing ratio coefficient; The output unit is communicatively connected to the data processing unit and is used to output the dynamic traceability map to the monitoring terminal.

Images