A water quality dynamic monitoring and pollution tracing early warning system and method

Abstract

This invention belongs to the field of water quality monitoring. It discloses a water quality dynamic monitoring and pollution source tracing and early warning system and method. The system comprises: constructing a multidimensional deep rock-water quality matrix based on long-term historical monitoring data; introducing the deep water-rock reaction thermodynamic saturation index and isotopic fractionation equilibrium constant constraints during principal component decomposition to construct a baseline dynamic evolution structure characterizing the evolution of the main control reaction factors; inputting multidimensional hydrochemical and isotopic data of the deep lithospheric fracture water to be measured into the baseline dynamic evolution structure to obtain its corresponding multivariate residual vector; generating a natural evolution deviation index and extracting the exogenous disturbance signal matrix; constructing a theoretical geochemical migration vector library of potential exogenous disturbance endmembers; performing exogenous disturbance discrimination based on the natural evolution deviation index; decomposing the disturbance source contribution based on restricted Bayesian endmember inversion; and outputting pollution source tracing results; thus achieving water quality dynamic monitoring and pollution source tracing and early warning.

Description

Technical Field

[0001] This invention relates to the field of water quality monitoring technology, and more specifically, to a water quality dynamic monitoring and pollution source tracing and early warning system and method. Background Technology

[0002] Deep lithospheric fissure water refers to groundwater bodies that exist within low-permeability bedrock fissure systems, are typically buried at depths of 1–3 km or even deeper, and have very weak dynamic exchange with surface water. These water bodies are relatively closed on a geological timescale, exhibiting long water ages, and are considered important media for studying deep life activities, rock-water interactions, and the safety of geological engineering environments.

[0003] Unlike shallow groundwater or surface water, deep fissure water typically exhibits extremely low dissolved organic carbon (TOC) content (often sub-milligrams per liter), low microbial abundance, and slow metabolic rates. The water body exists in a long-term thermodynamic quasi-equilibrium state, presenting a strongly reducing environment, with dissolved components dominated by rock-water reactions. Therefore, traditional water quality detection methods based on total indicators (such as COD and BOD) or culture counts have limited sensitivity and struggle to detect subtle abnormal changes.

[0004] With the development of deep-earth resources, underground energy storage projects, carbon dioxide geological sequestration, and radioactive waste geological disposal, deep fracture water may be affected by human activities such as drilling disturbances, underground engineering construction, or the injection of exogenous fluids. This can lead to trace amounts of exogenous substances entering the originally relatively closed water body, potentially introducing trace amounts of organic matter, oxygen, or young water components, thereby disrupting the original geochemical balance. However, due to the high mineralization, diverse elemental sources, and complex isotopic composition of deep fracture water, exogenous disturbance signals are often coupled with natural rock-water reaction processes, making it difficult to distinguish between natural geochemical evolution and anthropogenic pollution input. This poses a significant challenge to pollution source tracing and safety assessment.

[0005] Therefore, it is necessary to propose a dynamic water quality monitoring and pollution source tracing and early warning system and method. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a water quality dynamic monitoring and pollution source tracing and early warning system, comprising: The baseline modeling module is used to construct a multidimensional deep rock-water quality matrix based on long-term monitoring historical data. By introducing the deep water-rock reaction thermodynamic saturation index and isotopic fractionation equilibrium constant constraints during the principal component decomposition process, a baseline dynamic evolution structure is constructed to characterize the evolution of the main control reaction factors. The data deviation analysis module is used to input the multidimensional hydrochemical and isotopic data of the deep lithospheric fracture water to be measured into the baseline dynamic evolution structure and obtain its corresponding multivariate residual vector. The perturbation discrimination module is used to perform covariance normalization on the multivariate residual vector, generate the natural evolution deviation index, and extract the external perturbation signal matrix. The source tracing analysis module is used to construct a theoretical geochemical migration vector library of potential exogenous disturbance endmembers and to identify exogenous disturbances based on the natural evolution deviation index. When an exogenous disturbance event is determined to exist, the exogenous disturbance signal matrix is ​​used as input, and the disturbance source contribution is decomposed based on restricted Bayesian endmember inversion, outputting the pollution source tracing results.

[0007] Preferably, the baseline dynamic evolution structure includes: Each water sample in the long-term monitoring historical data is used as a row vector, and the indicators contained in each sample are used as column vectors to form a multidimensional deep rock water quality matrix. The selected indicators in each water sample data include major ions, trace elements, dissolved gas components and stable isotope ratios. Principal component decomposition is performed on the multidimensional deep rock-water quality matrix, and a regularization constraint term is introduced during the decomposition process. This regularization constraint term is used to not only maximize the variance when searching for the principal component direction, but also to force the load vector direction to match the evolution direction of the deep water-rock reaction thermodynamic saturation index and the trajectory of the isotope fractionation equilibrium constant, thereby obtaining the main control reaction factor. Establish nonlinear coupling equations among the main control reaction factors to form natural rock-water reaction path functions; Using the main control reaction factor as the coordinate axis, a multidimensional Euclidean space is formed; the natural rock-water reaction path function is embedded in this space as the core trajectory line; around the core trajectory line, the residual covariance at each node of the trajectory is used as the boundary envelope surface to define the range of natural fluctuations; and a mapping rule between the multidimensional deep rock-water quality matrix and this space is established. Finally, the core trajectory line, the boundary envelope surface, and the mapping rule are assembled together to form the baseline dynamic evolution structure. The mapping rule consists of a forward mapping matrix and a reverse mapping matrix. The forward mapping matrix is ​​the load vector matrix extracted by principal component decomposition, and the reverse mapping matrix is ​​the transpose of the load vector matrix.

[0008] Preferably, the index dimensions of the multidimensional hydrochemistry and isotope data are consistent with the index dimensions of the multidimensional water quality data matrix; Multidimensional hydrochemical and isotopic data are projected onto the space of the baseline dynamic evolution structure using a forward mapping matrix to obtain the actual observation points. The path prediction point with the shortest distance to the actual observation point is determined on the core trajectory line, and the vector pointing from the path prediction point to the actual observation point is denoted as the multivariate residual vector.

[0009] Preferably, the method for performing covariance normalization on the multivariate residual vector to generate the natural evolution deviation index includes: The local segment where the location path prediction point is located on the core trajectory line is extracted, and the set of residual vectors of water samples in this local segment is used as the local natural evolution residual covariance matrix. Eigenvalue decomposition is performed on the local natural evolution residual covariance matrix to obtain the eigenvector matrix and the corresponding eigenvalue diagonal matrix. A weighted transformation matrix is ​​constructed using the eigenvalue diagonal matrix. The multivariate residual vector of the deep lithospheric fracture water to be measured is then multiplied by the transpose of the eigenvector matrix and the weighted transformation matrix in turn to obtain the normalized residual vector. The Euclidean magnitude of the normalized residual vector is defined as the natural evolution deviation index.

[0010] Preferably, the method for extracting the external disturbance signal matrix includes: The coordinate values ​​of the multivariate residual vector are extracted as virtual factor scores. The virtual factor scores are de-standardized by the inverse mapping matrix to obtain the deviation values ​​of each index with physical dimensions. The deviation values ​​of each index constitute the external disturbance signal matrix.

[0011] Preferably, the method for identifying exogenous perturbations based on the natural evolution deviation index includes: Based on the residual covariance distribution characteristics of the boundary envelope surface in the baseline dynamic evolution structure, the confidence interval of natural evolution is obtained. Determine whether the natural evolution deviation index exceeds the natural evolution confidence interval. If it does, mark it as an exogenous disturbance event; if it does not, determine it as a natural geochemical evolution process.

[0012] Preferably, the theoretical geochemical migration vector library for constructing potential exogenous perturbation endmembers includes: Collect raw samples of potential external disturbance end-units and obtain the initial physicochemical parameters of the raw samples under preset multidimensional indicators. Among them, potential external disturbance end-units include at least engineering fluid end-units or shallow groundwater end-units. Thermodynamic environmental parameters of the deep baseline water body in the local section where the deep lithospheric fracture water is located are extracted from the baseline dynamic evolution structure; the initial physicochemical parameters and thermodynamic environmental parameters are input into the geochemical reaction simulation to simulate the water-rock reaction path and obtain the theoretical end-member composition of the original sample after the reaction under the deep thermodynamic equilibrium state. The theoretical end-member components after the reaction are compared with the corresponding components of the deep baseline water body by index-by-index difference calculation to obtain the deviation value of each end-member under each index. The deviation value constitutes the theoretical geochemical offset vector of each potential exogenous disturbance end-member. The theoretical geochemical migration vectors of all potential exogenous perturbation endmembers are aggregated, and constraints are set for each vector to construct a theoretical geochemical migration vector library.

[0013] Preferably, the process of decomposing the disturbance source contribution based on restricted Bayesian endmember inversion includes: Using the external disturbance signal matrix as the observation target, the theoretical geochemical migration vectors of each endmember in the theoretical geochemical migration vector library are called to construct the endmember linear mixing equation. We introduce specific restricted prior constraints for deep water environments, including physical feasible domain boundary constraints, deep thermodynamic equilibrium constraints, and extremely low organic carbon background constraints. Under the condition of satisfying the restricted prior constraints, we use the Markov chain Monte Carlo algorithm to iteratively sample the endmember linear mixing equations and solve for the posterior probability of the contribution ratio of each endmember. We output the posterior probability distribution of the contribution ratio of each potential external disturbance endmember as the pollution source tracing result.

[0014] Preferably, the method for iteratively sampling the endmember linear mixture equation using the Markov chain Monte Carlo algorithm to solve the posterior probability of the contribution ratio of each endmember includes: Within the range that satisfies the physical feasible domain boundary constraints, initialize the endmember contribution ratio vector, and generate candidate contribution ratio vectors in each iteration by adding random perturbations to the current ratio vector. The candidate contribution ratio vectors are subject to restricted prior constraints, including physical feasible domain boundary review, deep thermodynamic state equilibrium review based on theoretical redox potential shift, and extremely low organic carbon background review based on theoretical organic carbon shift. For the candidate contribution ratio vector that passes the review, calculate the matching likelihood between the corresponding theoretical mixed signal and the external disturbance signal matrix, calculate the acceptance probability in combination with the restricted prior constraints, and determine whether to update the candidate contribution ratio vector to the current state based on the acceptance probability. The process is repeated until the Markov chain converges. Statistical extraction is performed on the effective iterative samples after convergence to obtain the posterior probability distribution of the contribution ratio of each potential external disturbance endmember.

[0015] A method for dynamic water quality monitoring and pollution source tracing and early warning includes: S1. Construct a multidimensional deep rock water quality matrix based on long-term historical monitoring data; S2. By introducing the thermodynamic saturation index and isotopic fractionation equilibrium constant of deep water-rock reaction as constraints during the decomposition of principal components, a baseline dynamic evolution structure for characterizing the evolution of the main control reaction factors is constructed. S3. Input the multidimensional hydrochemical and isotopic data of the deep lithospheric fracture water to be measured into the baseline dynamic evolution structure to obtain its corresponding multivariate residual vector; S4. Perform covariance normalization on the multivariate residual vector to generate the natural evolution deviation index. S5. Extract the external disturbance signal matrix; S6. Construct a theoretical geochemical migration vector library for potential exogenous perturbation endmembers. S7. Based on the natural evolution deviation index, external disturbances are identified. When an external disturbance event is determined to exist, the external disturbance signal matrix is ​​used as input. The disturbance source contribution is decomposed based on restricted Bayesian endmember inversion, and the pollution source tracing results are output.

[0016] The technical effects and advantages of the water quality dynamic monitoring and pollution source tracing early warning system of this invention are as follows: To address the challenges of extremely low TOC content, slow microbial metabolism, and strong background noise in deep fissure water, a monitoring matrix was constructed using highly sensitive multidimensional hydrochemical and isotopic indicators. Through baseline modeling and residual analysis, minute chemical shifts introduced by exogenous substances can be extracted from the complex background of strong reducing and high mineralization, overcoming the sensitivity bottleneck of traditional methods that cannot detect or distinguish these shifts.

[0017] By using the thermodynamic saturation index and isotopic fractionation equilibrium constant of deep water-rock reaction as regularization constraints and embedding them into the principal component decomposition process, the constructed "baseline dynamic evolution structure" strictly follows the physicochemical path of water-rock reaction under natural conditions. By calculating the natural evolution deviation index and comparing it with the boundary envelope surface, natural fluctuations are distinguished from anthropogenic anomalies, avoiding the misjudgment of normal rock-water reactions as pollution.

[0018] By constructing a theoretical geochemical migration vector library of potential exogenous disturbance endmembers, the process of potential pollution sources reaching a new equilibrium with high temperature, high pressure, and strong reducing conditions after entering the deep environment was simulated, and the theoretical migration of each endmember to water quality indicators was calculated. Subsequently, restricted Bayesian endmember inversion was used to decompose the observed exogenous disturbance signal matrix into the contribution ratio of each theoretical endmember, providing a reliable quantitative source tracing basis. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the water quality dynamic monitoring and pollution source tracing and early warning system of the present invention; Figure 2 This is a schematic diagram of the water quality dynamic monitoring and pollution source tracing and early warning method in this invention. Detailed Implementation

[0020] 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.

[0021] Example 1, please refer to Figure 1 and Figure 2 In this embodiment of the invention, a water quality dynamic monitoring and pollution source tracing and early warning system includes: The baseline modeling module is used to construct a multidimensional deep rock-water quality matrix based on long-term monitoring historical data. By introducing the deep water-rock reaction thermodynamic saturation index and isotopic fractionation equilibrium constant constraints during the principal component decomposition process, a baseline dynamic evolution structure is constructed to characterize the evolution of the main control reaction factors. The data deviation analysis module is used to input the multidimensional hydrochemical and isotopic data of the deep lithospheric fracture water to be measured into the baseline dynamic evolution structure and obtain its corresponding multivariate residual vector. The perturbation discrimination module is used to perform covariance normalization on the multivariate residual vector, generate the natural evolution deviation index, and extract the external perturbation signal matrix. The source tracing analysis module is used to construct a theoretical geochemical migration vector library of potential exogenous disturbance endmembers and to identify exogenous disturbances based on the natural evolution deviation index. When an exogenous disturbance event is determined to exist, the exogenous disturbance signal matrix is ​​used as input, and the disturbance source contribution is decomposed based on restricted Bayesian endmember inversion, outputting the pollution source tracing results.

[0022] The baseline dynamic evolution structure includes: Each water sample from the long-term monitoring historical data is treated as a row vector, and the indicators contained in each sample are treated as column vectors to form a multidimensional deep rock water quality matrix. It should be noted that before construction, the data values ​​need to be normalized and dimensionless to ensure that indicators of different magnitudes and units are comparable.

[0023] Long-term monitoring historical data refers to multiple water sample data obtained at different burial depths (such as 1km to 3km gradient) along the water flow direction in the same deep hydrogeological area. It represents clean data that has been accumulated over a long period of time and is considered to be in a "natural evolutionary state". The selected indicators in each water sample data should be able to fully characterize the three core dimensions of the deep water environment, namely water-rock reaction, redox, and evolutionary history. Specifically, it includes major ions, trace elements, dissolved gas components, and stable isotope ratios. It should be noted that major ions refer to the macroscopic products of water-rock reactions (such as silicate hydrolysis, salt dissolution, and cation exchange), which are the basis for obtaining the mineral saturation index and are used to eliminate background noise caused by natural fluctuations in mineralization. Trace elements (e.g., lithium (Li) originates from the alteration of deep clay minerals) have extremely low background values. Even trace amounts of shallow water mixing or drilling fluid leakage can cause drastic changes in the relative proportions of trace elements. Therefore, trace elements are key to identifying weak signals. Deep fissure water exists in a strongly reducing environment and is usually rich in CH4 and H2. Once external disturbances introduce trace amounts of oxygen or surface water, the balance of gas components is instantly disrupted (e.g., aerobic bacteria consume CH4 to produce CO2). The response rate of dissolved gases is much faster than that of ions, making it an effective indicator. Isotopes are not affected by simple concentration dilution. Due to their long water age, deep water's isotopic characteristics usually tend towards enrichment of heavy isotopes or specific radiogenic characteristics (e.g., high...). 87 Sr / 86 Sr), which has a completely different isotopic characteristic from shallow water or modern atmospheric precipitation, is therefore an indispensable data support for confirming the source.

[0024] In existing hydrogeochemical studies, principal component analysis (PCA) is a routine procedure. Running PCA on data yields several mathematical principal components. The biggest challenge with deep water data is multicollinearity (high concentrations of all ions changing in the same direction). However, conventional PCA only captures the direction of maximum data variance, and the extracted principal components (PCs) are often merely mathematical abstractions. They may even extract measurement noise or non-dominant processes as principal components, failing to correspond to actual geological processes and lacking mechanistic identification capabilities. Therefore, this design is as follows:

[0025] Principal component decomposition is performed on the multidimensional deep rock-water quality matrix, and a regularization constraint term is introduced during the decomposition process. This regularization constraint term is used to not only maximize the variance when searching for the principal component direction, but also to force the load vector direction (w) to match the evolution direction of the deep water-rock reaction thermodynamic saturation index and the trajectory of the isotope fractionation equilibrium constant, thereby obtaining the main control reaction factor. After this processing, if a certain mathematical principal component deviates from the evolution direction or the isotope fractionation law, its penalty term will increase sharply and be discarded by the algorithm, thus forcing the algorithm to split the independent factors of different reaction paths in the collinear data.

[0026] In simple terms, the mathematical formula for the PCA objective function after introducing a regularization penalty term can be expressed as follows: max(variance)−λ⋅f(w,SI,α), where f(w,SI,α) represents the distance function that measures the consistency between the evolution directions of the load vector w and SI and α, and λ is the penalty coefficient.

[0027] The deep water-rock reaction thermodynamic saturation index (SI) is a quantitative indicator that measures whether a specific mineral in groundwater is in a dissolved, precipitated, or equilibrium state. It is calculated by inputting the major ion concentrations, water temperature, and pH values ​​from a multidimensional deep rock-water quality matrix into geochemical simulation software (existing tools such as PHREEQC or Geochemist's Workbench). A saturation index < 0 indicates mineral dissolution, > 0 indicates precipitation, and ≈ 0 indicates equilibrium.

[0028] The isotopic fractionation equilibrium constant (α) is used to describe the fractionation of light or heavy isotopes (e.g., in specific water-rock reactions such as water-carbonate rock reactions and water-silicate rock alteration) during a particular water-rock reaction (e.g., water-carbonate rock reaction, water-silicate rock alteration). 18 O / 16 O) is a constant representing the distribution ratio between the solid (rock) and liquid (water) phases. It can be obtained directly from classical thermodynamic fractionation equations or empirical formulas in experimental petrology literature.

[0029] Nonlinear coupling equations among the controlling reaction factors are established to form a natural rock-water reaction path function. Specifically, deep water-rock reactions do not occur in isolation but are mutually coupled. For example, mineral dissolution (controlling reaction factor 1) leads to changes in the isotopic composition of pore water (controlling reaction factor 3) and simultaneously consumes redox agents (controlling reaction factor 2). The establishment method includes: first, plotting two-dimensional or three-dimensional scatter plots between the extracted controlling reaction factors as variables; second, identifying their nonlinear behavior (such as exponential decay, logarithmic growth, and Langmuir isotherm adsorption curve characteristics), for example, as the carbonate dissolution factor increases, the isotope exchange factor exhibits asymptotically saturated nonlinear characteristics; then, using a nonlinear regression algorithm to fit the mathematical expressions between the factors; finally, combining the coupling equations among all the controlling reaction factors to form a set of parameterized trajectory equations describing the evolution of water from its initial state to a deep quasi-equilibrium state, denoted as the natural rock-water reaction path function.

[0030] Using the controlling reaction factor as the coordinate axis, a multidimensional Euclidean space is formed (e.g., the horizontal axis F1 is the carbonate-silicate co-dissolution factor, the vertical axis F2 is the deep reduction evolution factor, and the vertical axis F3 is the water-rock isotope exchange factor). The natural rock-water reaction path function is embedded in this space as the core trajectory line representing the baseline direction of natural evolution. Around this core trajectory line, the residual covariance at each node of the trajectory is used as the boundary envelope surface defining the range of natural fluctuations. A mapping rule between the multidimensional deep rock-water quality matrix and this space is established. Finally, the core trajectory line, the boundary envelope surface, and the mapping rule are assembled together to form the baseline dynamic evolution structure.

[0031] It should be noted that the method for setting the boundary envelope includes calculating the residual covariance around the trajectory line based on the multidimensional deep rock-water quality matrix obtained from long-term monitoring. The distance from the pipe wall to the center line represents the normal fluctuation range (confidence interval) of natural evolution. The pipe wall is not of uniform width; it is narrow in sections with dense data and stable evolution, and wide in sections with sparse data or transitional areas.

[0032] The mapping rules are automatically generated during principal component decomposition (PCD) and consist of two matrices: a forward mapping matrix and an inverse mapping matrix. The forward mapping matrix is ​​the loading vector matrix extracted from PCD. Multiplying the original index data (standardized multidimensional hydrochemical and isotopic data) of the water sample by this loading vector matrix yields the coordinates of the water sample in multidimensional Euclidean space (i.e., the scores of each controlling reactive factor). The inverse mapping matrix is ​​the transpose of the loading vector matrix. Multiplying the coordinate values ​​in space by this matrix and then performing inverse standardization (adding the historical mean and multiplying by the historical standard deviation) restores the predicted concentrations or ratios of the corresponding hydrochemical indicators.

[0033] The index dimensions of the multidimensional hydrochemistry and isotope data are consistent with the index dimensions of the multidimensional water quality data matrix. Multidimensional hydrochemical and isotopic data are projected onto the space of the baseline dynamic evolution structure using a forward mapping matrix to obtain the actual observation points. The path prediction point with the shortest distance to the actual observation point is determined on the core trajectory line, and the vector pointing from the path prediction point to the actual observation point is denoted as the multivariate residual vector.

[0034] It should be noted that the purpose of this step is to obtain the discrepancy between the actual situation and the situation that should exist in nature. The detailed steps include using a forward mapping matrix to transform the multidimensional hydrochemical and isotopic data of the "water sample to be tested" into a point in multidimensional Euclidean space, denoted as P_obs (actual observed value). Since the water sample to be tested must be in some evolutionary stage, the point on the trajectory line that best matches the degree of evolution of the water sample is found (the point with the shortest distance from P_obs to the trajectory line represents "the position that this drop of water should be in at the current evolutionary stage if it completely follows the natural water-rock reaction law"), denoted as P_pred (path function predicted value). In space, a line segment with an arrow pointing from P_pred to P_obs is drawn; this line segment is the multivariate residual vector. This vector contains the deviations in all directions of the main controlling reaction factors. If the deviation in a certain direction is particularly large, it indicates that the external force breaking the natural law in that direction is the strongest (for example, a large deviation in the redox factor indicates that oxygen-containing shallow water may have been mixed in).

[0035] Addressing the unique challenges of deep lithospheric fracture water—characterized by high mineralization, multicollinearity (simultaneous rise and fall of ions), and the ease with which weak exogenous signals are masked by natural evolutionary noise—conventional standardization methods (such as direct division by standard deviation) or ordinary Mahalanobis distance calculations fail because they cannot distinguish between directions that naturally exhibit dramatic fluctuations and directions that are extremely stable in their natural state but inevitably become abnormal upon change. Therefore, the standardization process in this design is as follows: Methods for generating the natural evolution deviation index by covariance normalization of multivariate residual vectors include: The local segment where the location path prediction point is located on the core trajectory line is used to extract the set of residual vectors of water samples within that local segment, which serves as the local natural evolution residual covariance matrix. This is because the amplitude and direction of the natural fluctuations of deep water are different at different evolutionary stages (such as the initial dissolution period and the later equilibrium period). Using local covariance ensures that the scale for measuring the deviation of the water sample being tested is a dynamic scale that is completely matched with its current evolutionary stage, rather than a static scale that is applied uniformly.

[0036] Eigenvalue decomposition is performed on the local natural evolution residual covariance matrix to obtain the eigenvector matrix and the corresponding eigenvalue diagonal matrix. The eigenvectors represent the orthogonal directions of the natural fluctuations in the deep water-rock reaction at the current stage (i.e., the independent evolution axes after decoupling), and the eigenvalues ​​represent the variance of the natural fluctuations along these independent axes. For example, the strongly coupled variables of carbonate dissolution and strontium release, which always rise and fall together, are transformed into independent orthogonal fluctuation axes to prevent small fluctuations in a certain natural evolution direction from being misjudged as a comprehensive anomaly because they involve multiple related ions.

[0037] Using the eigenvalue diagonal matrix Λ, a weighted transformation matrix W = Λ−1 / 2 is constructed. The multivariate residual vector of the deep lithospheric fracture water to be measured is then multiplied successively by the transpose of the eigenvector matrix V and the weighted transformation matrix R. → The normalized residual vector R is obtained. → _norm=W⋅VT⋅R → This is because in deep water-rock reactions, certain directions (such as major ion concentrations) naturally exhibit extremely large variances and eigenvalues. After Λ−1 / 2 weighting, these variances are significantly compressed (reduced) to prevent them from masking external signals. Conversely, certain directions (such as specific redox-sensitive indicators and trace isotope ratios) are extremely stable under natural conditions, with minimal variances and eigenvalues. After weighting, their minute deviations are significantly amplified (upgraded). This step essentially extracts and amplifies the thermodynamically sensitive weak anomalies that were originally submerged by high mineralization noise.

[0038] The Euclidean modulus of the normalized residual vector (i.e., the square root of the vector length) is defined as the natural evolution deviation index. The final output scalar value is no longer a simple accumulation of concentration deviations, but a comprehensive anomaly metric that integrates dynamic benchmarks of local evolution stages, water-rock collinearity decoupling, and inverse amplification of thermodynamic sensitivity. It exhibits strong immunity to fluctuations in the natural evolution of deep water, while also demonstrating extremely high sensitivity to minute external disturbances (such as isotopic variations caused by the intrusion of trace amounts of shallow water).

[0039] Methods for extracting external disturbance signal matrices include: The coordinate values ​​of the multivariate residual vector are extracted as virtual factor scores. The virtual factor scores are de-standardized by the inverse mapping matrix to obtain the deviation values ​​of each index with physical dimensions. The deviation values ​​of each index constitute the external disturbance signal matrix.

[0040] It's important to note that if the residual vector remains in an abstract multidimensional Euclidean space, it's merely a collection of factor score differences, failing to directly identify which chemical substance is exceeding the limit. Therefore, it's necessary to restore the abstract bias to a specific list of chemical anomalies. The steps include extracting the coordinate values ​​of the multivariate residual vector, treating it as a set of virtual factor scores purely caused by external disturbances, multiplying this set of virtual factor scores by an inverse mapping matrix to obtain a standardized deviation vector containing all indicators from the multidimensional water chemistry and isotope data. This vector represents the standard deviation multiple of each indicator from the natural baseline. The standardized deviation vector is then restored to real physical units (e.g., mg / L, ‰), yielding the restored deviation values ​​for each indicator. These values ​​are then rearranged into a row vector, denoted as the external disturbance signal matrix. This matrix is ​​the pure pollution fingerprint extracted after stripping away the millions of years of natural water-rock reaction background from deep water. For example, Cl... − It deviated from +5mg / L. 18 O deviated by -0.3‰, Na + The deviation is 0.

[0041] The method for identifying exogenous perturbations based on the natural evolution deviation index includes: Based on the residual covariance distribution characteristics of the boundary envelope surface in the baseline dynamic evolution structure, a confidence interval for natural evolution is obtained. It should be noted that because the boundary envelope surface is not a uniformly wide circle, but rather varies in width (narrower where data is denser, wider in transitional areas), the tolerance for natural fluctuations differs at different evolutionary stages. Therefore, the confidence interval for natural evolution here is not the simple "mean ± several standard deviations" found in traditional univariate statistics. This confidence interval is a dynamic threshold that depends on where the current water sample is located within the envelope surface. Specific steps include: reprocessing the long-term monitoring historical data to calculate the natural evolution deviation index (i.e., historical residual modulus) for each historical water sample; collecting these historical deviation indices and analyzing their probability distribution (usually a chi-square distribution, or directly using an empirical probability distribution); and setting a confidence probability based on the needs of the engineering safety assessment, for example, setting it to 95%. This means that 5% of extreme natural fluctuations are allowed to be misjudged as anomalies, and the corresponding quantiles are extracted from the distribution as the upper limit of the threshold. Since the width of the boundary envelope at different trajectory nodes is adaptive, this threshold essentially represents a dynamic tolerance boundary. Finally, this upper limit of the threshold is defined as the boundary of the natural evolution confidence interval, and the confidence interval is obtained. When the natural evolution deviation index of the water sample exceeds this boundary, it indicates that it has deviated from the natural fluctuation envelope allowed by the baseline dynamic evolution structure, thus providing a quantitative judgment benchmark for subsequent identification of external disturbance events.

[0042] The system determines whether the natural evolution deviation index exceeds the confidence interval of natural evolution. If it does, it indicates that the observation point of the water sample in space has breached the boundary envelope, deviating from the maximum allowable natural fluctuation range of the natural rock-water reaction path function. This level of deviation cannot be explained by natural evolution, implying intrusion by exogenous substances such as drilling fluid or shallow water. This is then marked as an exogenous disturbance event. If it does not exceed the confidence interval, it means that the observation point of the water sample still falls within the boundary envelope, and its changes still follow the natural evolution law under the quasi-equilibrium state of deep thermodynamics, requiring no pollution warning. This is then determined to be a natural geochemical evolution process. This dynamic topological boundary discrimination mechanism effectively avoids false alarms caused by the complex evolution of deep water itself, greatly improving the accuracy of identifying weak exogenous disturbances.

[0043] Conventional source tracing techniques (such as the common Bayesian isotope mixing model) typically use raw test data of the exogenous fluid (e.g., directly obtaining ion concentrations from surface water) to match the water sample being tested. However, this approach completely fails in deep lithospheric fracture water environments. This is because once exogenous fluids (such as drilling fluids or shallow water) enter the high-temperature, high-pressure, and strongly reducing environment of deep water, they undergo rapid and intense geochemical reactions with the deep water or surrounding rocks (e.g., oxygen is rapidly consumed, carbonates precipitate due to pressure changes, and organic matter is degraded by deep microorganisms). The anomalous signals captured in the water sample are not the original characteristics of the exogenous fluid, but rather residual features of the reaction after the exogenous fluid has undergone deep environmental modification. Therefore, this approach does not use raw end-data; instead, it constructs a theoretical geochemical migration vector library. The specific implementation steps and methods are as follows: The theoretical geochemical migration vector library for constructing potential exogenous perturbation endmembers includes: Raw samples were collected from potential exogenous disturbance end-units to obtain their initial physicochemical parameters (such as initial temperature, pH value, and redox potential) under preset multidimensional indicators (consistent with the indicator dimensions of the multidimensional water quality data matrix). Potential exogenous disturbance end-units included at least engineering fluid end-units or shallow groundwater end-units. It should be noted that, considering the characteristics of deep engineering activities, possible disturbance sources were identified. Engineering fluid end-units included raw samples of drilling fluid, fracturing fluid, and underground energy storage injection fluids (such as CO2, H2, and compressed air). Shallow groundwater end-units included shallow oxygenated groundwater and atmospheric precipitation samples from the upper part of the target deep hydrogeological unit. Thermodynamic environmental parameters of the deep baseline water body in the local segment where the deep lithospheric fracture water to be measured is located are extracted from the baseline dynamic evolution structure. This is because the reaction that occurs after the entry of the exogenous fluid depends on the thermodynamic environment. Specifically, from the baseline dynamic evolution topology, typical characteristics of the deep water in the local segment where the current water sample to be measured is located (i.e., the segment where the path prediction point is located) are extracted, such as average temperature, pressure, background mineralization, partial pressure of major reducing gases (CH4, H2), and mineral composition of the surrounding rocks (such as iron-magnesium minerals and carbonate rock content). These constitute the thermodynamic environmental parameters for reaction simulation.

[0044] Initial physicochemical and thermodynamic environmental parameters are input into a geochemical reaction simulation to simulate the water-rock reaction pathway and obtain the theoretical end-member composition of the original sample under deep thermodynamic equilibrium. Specifically, this step simulates the virtual reaction of the original end-member in the deep environment. Geochemical reaction simulation can be achieved using mature water-rock reaction simulation software (such as PHREEQC or Geochemist's Workbench) to perform a mixed titration simulation of the original end-member composition with deep local environmental parameters. This obtains the theoretically desirable geochemical component concentrations and isotope ratios (i.e., the theoretical end-member composition) that the end-member fluid should possess after fully adapting to the deep thermodynamic quasi-equilibrium state.

[0045] The theoretical end-member composition after reaction is subtracted from the corresponding composition of the deep baseline water body index by index, yielding the deviation value of each end-member under each index. These deviation values ​​constitute the theoretical geochemical migration vector for each potential exogenous disturbance end-member. It should be noted that the exogenous disturbance does not result in absolute concentration, but rather in a relative change to the baseline. For each exogenous end-member, the following calculation is performed: the theoretical end-member composition after reaction is subtracted from the composition of the deep baseline water body index by index. All index deviation values ​​are arranged in a uniform order to form a one-dimensional row vector. This vector is the theoretical geochemical migration vector for that end-member, representing the actual migration trajectory that a unit of this exogenous fluid, after mixing into deep water and undergoing complete transformation of the deep environment, will trace in the multidimensional index space.

[0046] The theoretical geochemical migration vectors of all potential exogenous disturbance endmembers are aggregated, and constraints are set for each vector to construct a theoretical geochemical migration vector library. It should be noted that constraints are set for each vector based on physical reality. For example: the contribution ratio must be ≥0 (meaning negative emissions are not allowed); the sum of the contribution ratios of all endmembers must be ≤1 (or a specific threshold is set according to the mixing law); specific indicator boundaries are set (such as oxygen consumption cannot exceed the total amount of reducing gases originally present in the deep environment). This scheme recreates the change process of exogenous fluids in the deep environment through reaction path simulation, showing its changed appearance (theoretical migration vectors). When the exogenous disturbance signal matrix (real anomaly fingerprint) is compared based on the migration vectors, the Bayesian model can accurately calculate which fluid and in what proportion has been mixed into the deep old water.

[0047] The process of decomposing the perturbation source contribution based on restricted Bayesian endmember inversion includes: Using the external disturbance signal matrix as the observation target, theoretical geochemical migration vectors of each endmember from the theoretical geochemical migration vector library are retrieved to construct an endmember linear mixing equation. Specifically, this involves establishing a mixing equation with the "external disturbance signal matrix" as the observation target. The external disturbance signal matrix is ​​set as the observation vector Y, representing the deviation values ​​of various indicators from the natural baseline in the water sample. The theoretical geochemical migration vectors of all potential external disturbance endmembers (such as drilling fluid, shallow oxygenated water, etc.) are retrieved from the theoretical geochemical migration vector library and denoted as follows: =X 1,X Let 2,...,Xn be the observed anomaly Y, which is a mixture of these n endmembers in a certain proportion. Right now

[0048] ,in, ε is the contribution ratio of the i-th endmember (the unknown to be solved), and ε is the model error (which follows a normal distribution).

[0049] We introduce specific constrained prior constraints for deep water environments, including physical feasible domain boundary constraints, deep thermodynamic equilibrium constraints, and extremely low organic carbon background constraints. It should be noted that conventional models typically only set these constraints. ≥0 and ∑ =1, which can lead to multiple solutions and distortions in complex deep water environments. Therefore, this scheme incorporates the following three-dimensional constrained priors, with the physical feasible region constraint directly introduced from the constraints set by the theoretical geochemical migration vector. The deep thermodynamic state prior is based on the characteristics of the strongly reducing environment of deep fracture water, setting prior probabilities including the contribution ratio of a certain endmember (such as shallow water). This can lead to a significantly higher redox potential in the mixed water than the measured deviation. The model will assign a very low prior probability to this group, suppressing it. The extremely low organic carbon background prior is based on the extremely low TOC characteristics of deep water. If the calculated contribution ratio of a certain end-member (such as drilling fluid containing organic matter) is... The reconstructed TOC deviation is much larger than the actual TOC deviation in the external disturbance signal matrix, and the model will automatically reduce the weight of the drilling fluid endmember. Under the condition of satisfying the constrained prior constraints, the Markov chain Monte Carlo algorithm is used to iteratively sample the endmember linear mixing equation to solve the posterior probability of the contribution ratio of each endmember; the posterior probability distribution of the contribution ratio of each potential external disturbance endmember is output (outputting the mean or 95% confidence interval of the average contribution ratio of each potential external disturbance endmember, such as: the mean contribution ratio of shallow oxygenated groundwater is 62% or confidence interval 55%~69%, the mean contribution ratio of drilling fluid is 18% or confidence interval 12%~24%, and the remaining 20% ​​is attributed to the model's unexplained residuals), as the pollution source tracing result. Specifically, the dominant pollution source is determined based on the overlap of the posterior probability distributions of each endmember. If the contribution ratio of shallow water is significantly higher than that of drilling fluid and rarely overlaps with it, the final conclusion is: "This external disturbance event was mainly caused by the mixing of shallow oxygen-containing groundwater, with the possibility of trace drilling fluid leakage." At this point, the precise source tracing is complete.

[0050] Due to the complex physicochemical asymmetric constraints introduced by the deep water environment, the endmember contribution ratio cannot be directly solved using conventional algebraic equations. Therefore, a Markov chain Monte Carlo algorithm must be used for numerical optimization, with the following steps: The method for iteratively sampling the endmember linear mixture equation using the Markov chain Monte Carlo algorithm to solve the posterior probability of the contribution ratio of each endmember includes: Within the constraints of the physical feasible domain boundary, an initial endmember contribution ratio vector is initialized, and in each iteration, a candidate contribution ratio vector is generated by adding a random perturbation to the current ratio vector. Specifically, a set of initial endmember contribution ratio vectors is randomly generated, denoted as f(0)=[f1(0),f2(0),...,fn(0), which represents the starting point of the Markov chain, i.e., an assumed initial suspect assignment. Secondly, based on the current state f(t) of the t-th iteration, a new set of candidate contribution ratio vectors, denoted as f∗, is generated by adding a small random perturbation (usually following a multivariate normal distribution or a uniform distribution). This step is equivalent to tentatively changing the suspect share of each endmember.

[0051] The candidate contribution ratio vectors are subject to restricted prior constraints, including physical feasible domain boundary checks (e.g., checking for negative values ​​or sums greater than 1 in f*; if present, the candidate state is rejected, and the original state f(t) is retained for the next iteration), deep thermodynamic equilibrium checks based on theoretical redox potential shifts (e.g., calculating the theoretical redox potential shift under the mixing ratio based on the candidate ratio f* and the theoretical shift vector library; if the shift severely violates the measured state of the current deep water strong reducing environment, for example, theoretical calculations indicate that severe oxidation should occur, but the measured state does not have this characteristic, then the candidate state is given an extremely low prior probability, significantly reducing its acceptance probability), and extremely low organic carbon background checks based on theoretical organic carbon shifts (similarly, calculating the theoretical organic carbon TOC shift; if f* leads to the contribution of an organic endmember, such as drilling fluid, causing the TOC deviation to exceed the actual TOC anomaly in the external disturbance signal matrix, it is also given an extremely low prior probability). For a candidate contribution ratio vector that passes the review, the matching likelihood between the corresponding theoretical mixed signal and the external disturbance signal matrix is ​​calculated. The acceptance probability is calculated in combination with the constrained prior constraints. Based on the acceptance probability, it is determined whether to update the candidate contribution ratio vector to the current state. Specifically, if the candidate state f* passes the constraint review, the algorithm will calculate its likelihood, that is, calculate the degree of matching between the theoretically mixed anomalous signal and the actually observed external disturbance signal matrix under the candidate ratio f* (usually evaluated by the probability density of a multidimensional normal distribution, the smaller the difference, the higher the probability). Then, combining the likelihood and the prior probability, the acceptance probability of transitioning from f(t) to f* is calculated. A random number between 0 and 1 is generated. If it is less than the acceptance probability, the candidate state is accepted and f(t+1) = f* is set; otherwise, it is rejected and f(t+1) = f(t) is set.

[0052] The process is repeated until the Markov chain converges. Statistical extraction is then performed on the valid iterative samples after convergence to obtain the posterior probability distribution of the contribution ratio of each potential external disturbance endmember. Specifically, the Markov chain gradually forgets its initial state and samples back and forth around the optimal ratio combination that best matches the actual anomalous signal and does not violate deep physical laws. Statistical analysis is performed on the large number of valid samples retained, and a frequency distribution histogram of the contribution ratio of each endmember is plotted. This distribution is the desired posterior probability distribution, which not only gives the most likely mean contribution of each endmember but also its confidence interval (e.g., the 95% confidence interval), thus completing the quantitative decomposition from uncertain aliasing signals to high-confidence source tracing results.

[0053] Example 2, please refer to Figure 1 and Figure 2 In this embodiment of the invention, a method for dynamic water quality monitoring and pollution source tracing and early warning includes: S1. Construct a multidimensional deep rock water quality matrix based on long-term historical monitoring data; S2. By introducing the thermodynamic saturation index and isotopic fractionation equilibrium constant of deep water-rock reaction as constraints during the decomposition of principal components, a baseline dynamic evolution structure for characterizing the evolution of the main control reaction factors is constructed. S3. Input the multidimensional hydrochemical and isotopic data of the deep lithospheric fracture water to be measured into the baseline dynamic evolution structure to obtain its corresponding multivariate residual vector; S4. Perform covariance normalization on the multivariate residual vector to generate the natural evolution deviation index. S5. Extract the external disturbance signal matrix; S6. Construct a theoretical geochemical migration vector library for potential exogenous perturbation endmembers. S7. Based on the natural evolution deviation index, external disturbances are identified. When an external disturbance event is determined to exist, the external disturbance signal matrix is ​​used as input. The disturbance source contribution is decomposed based on restricted Bayesian endmember inversion, and the pollution source tracing results are output.

[0054] 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 water quality dynamic monitoring and pollution source tracing and early warning system, characterized in that, include: The baseline modeling module is used to construct a multidimensional deep rock-water quality matrix based on long-term monitoring historical data. By introducing the deep water-rock reaction thermodynamic saturation index and isotopic fractionation equilibrium constant constraints during the principal component decomposition process, a baseline dynamic evolution structure is constructed to characterize the evolution of the main control reaction factors. The data deviation analysis module is used to input the multidimensional hydrochemical and isotopic data of the deep lithospheric fracture water to be measured into the baseline dynamic evolution structure and obtain its corresponding multivariate residual vector. The perturbation discrimination module is used to perform covariance normalization on the multivariate residual vector, generate the natural evolution deviation index, and extract the external perturbation signal matrix. The source tracing analysis module is used to construct a theoretical geochemical migration vector library of potential exogenous disturbance endmembers and to identify exogenous disturbances based on the natural evolution deviation index. When an exogenous disturbance event is determined to exist, the exogenous disturbance signal matrix is ​​used as input, and the disturbance source contribution is decomposed based on restricted Bayesian endmember inversion, outputting the pollution source tracing results.

2. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 1, characterized in that, The baseline dynamic evolution structure includes: Each water sample in the long-term monitoring historical data is used as a row vector, and the indicators contained in each sample are used as column vectors to form a multidimensional deep rock water quality matrix. The selected indicators in each water sample data include major ions, trace elements, dissolved gas components and stable isotope ratios. Principal component decomposition is performed on the multidimensional deep rock-water quality matrix, and a regularization constraint term is introduced during the decomposition process. This regularization constraint term is used to not only maximize the variance when searching for the principal component direction, but also to force the load vector direction to match the evolution direction of the deep water-rock reaction thermodynamic saturation index and the trajectory of the isotope fractionation equilibrium constant, thereby obtaining the main control reaction factor. Establish nonlinear coupling equations among the main control reaction factors to form natural rock-water reaction path functions; Using the main control reaction factor as the coordinate axis, a multidimensional Euclidean space is formed; the natural rock-water reaction path function is embedded in this space as the core trajectory line; around the core trajectory line, the residual covariance at each node of the trajectory is used as the boundary envelope surface to define the range of natural fluctuations; and a mapping rule between the multidimensional deep rock-water quality matrix and this space is established. Finally, the core trajectory line, the boundary envelope surface, and the mapping rule are assembled together to form the baseline dynamic evolution structure. The mapping rule consists of a forward mapping matrix and a reverse mapping matrix. The forward mapping matrix is ​​the load vector matrix extracted by principal component decomposition, and the reverse mapping matrix is ​​the transpose of the load vector matrix.

3. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 2, characterized in that, The index dimensions of the multidimensional hydrochemistry and isotope data are consistent with the index dimensions of the multidimensional water quality data matrix. Multidimensional hydrochemical and isotopic data are projected onto the space of the baseline dynamic evolution structure using a forward mapping matrix to obtain the actual observation points. The path prediction point with the shortest distance to the actual observation point is determined on the core trajectory line, and the vector pointing from the path prediction point to the actual observation point is denoted as the multivariate residual vector.

4. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 3, characterized in that, The method for generating the natural evolution deviation index by performing covariance normalization on the multivariate residual vector includes: The local segment where the location path prediction point is located on the core trajectory line is extracted, and the set of residual vectors of water samples in this local segment is used as the local natural evolution residual covariance matrix. Eigenvalue decomposition is performed on the local natural evolution residual covariance matrix to obtain the eigenvector matrix and the corresponding eigenvalue diagonal matrix. A weighted transformation matrix is ​​constructed using the eigenvalue diagonal matrix. The multivariate residual vector of the deep lithospheric fracture water to be measured is then multiplied by the transpose of the eigenvector matrix and the weighted transformation matrix in turn to obtain the normalized residual vector. The Euclidean magnitude of the normalized residual vector is defined as the natural evolution deviation index.

5. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 4, characterized in that, The method for extracting the external disturbance signal matrix includes: The coordinate values ​​of the multivariate residual vector are extracted as virtual factor scores. The virtual factor scores are de-standardized by the inverse mapping matrix to obtain the deviation values ​​of each index with physical dimensions. The deviation values ​​of each index constitute the external disturbance signal matrix.

6. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 5, characterized in that, The method for identifying exogenous perturbations based on the natural evolution deviation index includes: Based on the residual covariance distribution characteristics of the boundary envelope surface in the baseline dynamic evolution structure, the confidence interval of natural evolution is obtained. Determine whether the natural evolution deviation index exceeds the natural evolution confidence interval. If it does, mark it as an exogenous disturbance event; if it does not, determine it as a natural geochemical evolution process.

7. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 6, characterized in that, The theoretical geochemical migration vector library for constructing potential exogenous perturbation endmembers includes: Collect raw samples of potential external disturbance end-units and obtain the initial physicochemical parameters of the raw samples under preset multidimensional indicators. Among them, potential external disturbance end-units include at least engineering fluid end-units or shallow groundwater end-units. Thermodynamic environmental parameters of the deep baseline water body in the local section where the deep lithospheric fracture water is located are extracted from the baseline dynamic evolution structure; the initial physicochemical parameters and thermodynamic environmental parameters are input into the geochemical reaction simulation to simulate the water-rock reaction path and obtain the theoretical end-member composition of the original sample after the reaction under the deep thermodynamic equilibrium state. The theoretical end-member components after the reaction are compared with the corresponding components of the deep baseline water body by index-by-index difference calculation to obtain the deviation value of each end-member under each index. The deviation value constitutes the theoretical geochemical offset vector of each potential exogenous disturbance end-member. The theoretical geochemical migration vectors of all potential exogenous perturbation endmembers are aggregated, and constraints are set for each vector to construct a theoretical geochemical migration vector library.

8. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 7, characterized in that, The process of decomposing the perturbation source contribution based on restricted Bayesian endmember inversion includes: Using the external disturbance signal matrix as the observation target, the theoretical geochemical migration vectors of each endmember in the theoretical geochemical migration vector library are called to construct the endmember linear mixing equation. We introduce specific restricted prior constraints for deep water environments, including physical feasible domain boundary constraints, deep thermodynamic equilibrium constraints, and extremely low organic carbon background constraints. Under the condition of satisfying the restricted prior constraints, we use the Markov chain Monte Carlo algorithm to iteratively sample the endmember linear mixing equations and solve for the posterior probability of the contribution ratio of each endmember. We output the posterior probability distribution of the contribution ratio of each potential external disturbance endmember as the pollution source tracing result.

9. The water quality dynamic monitoring and pollution source tracing and early warning system according to claim 7, characterized in that, The method for iteratively sampling the endmember linear mixture equation using the Markov chain Monte Carlo algorithm to solve the posterior probability of the contribution ratio of each endmember includes: Within the range that satisfies the physical feasible domain boundary constraints, initialize the endmember contribution ratio vector, and generate candidate contribution ratio vectors in each iteration by adding random perturbations to the current ratio vector. The candidate contribution ratio vectors are subject to restricted prior constraints, including physical feasible domain boundary review, deep thermodynamic state equilibrium review based on theoretical redox potential shift, and extremely low organic carbon background review based on theoretical organic carbon shift. For the candidate contribution ratio vector that has passed the review, calculate the matching likelihood between the corresponding theoretical mixed signal and the external disturbance signal matrix, calculate the acceptance probability in combination with the restricted prior constraints, and determine whether to update the candidate contribution ratio vector to the current state based on the acceptance probability. The process is repeated until the Markov chain converges. Statistical extraction is performed on the effective iterative samples after convergence to obtain the posterior probability distribution of the contribution ratio of each potential external disturbance endmember.

10. A method for dynamic water quality monitoring and pollution source tracing and early warning, used to implement the dynamic water quality monitoring and pollution source tracing and early warning system described in claims 1-9, characterized in that, include: S1. Construct a multidimensional deep rock water quality matrix based on long-term historical monitoring data; S2. By introducing the thermodynamic saturation index and isotopic fractionation equilibrium constant of deep water-rock reaction as constraints during the decomposition of principal components, a baseline dynamic evolution structure for characterizing the evolution of the main control reaction factors is constructed. S3. Input the multidimensional hydrochemical and isotopic data of the deep lithospheric fracture water to be measured into the baseline dynamic evolution structure to obtain its corresponding multivariate residual vector; S4. Perform covariance normalization on the multivariate residual vector to generate the natural evolution deviation index. S5. Extract the external disturbance signal matrix; S6. Construct a theoretical geochemical migration vector library for potential exogenous perturbation endmembers. S7. Based on the natural evolution deviation index, external disturbances are identified. When an external disturbance event is determined to exist, the external disturbance signal matrix is ​​used as input. The disturbance source contribution is decomposed based on restricted Bayesian endmember inversion, and the pollution source tracing results are output.

Images