A water quality monitoring method based on multi-sensor data fusion

Abstract

This invention relates to the field of environmental monitoring technology and discloses a water quality monitoring method based on multi-sensor data fusion. The method includes: constructing and updating a dynamic causal graph reflecting the causal relationship between various water quality parameters based on the multi-source water quality monitoring data; performing fusion analysis on the real-time collected water quality parameter data based on the dynamic causal graph to generate a comprehensive water quality status index; performing pollution contribution inversion analysis on the real-time collected water quality parameter data based on the structure of the dynamic causal graph to identify the main suspected pollution sources; generating and executing a customized early warning strategy based on the comprehensive water quality status index and the main suspected pollution sources; and adaptively optimizing the early warning strategy generation mechanism based on the feedback of water quality improvement effect after the customized early warning strategy is executed. This invention can improve the accuracy of water quality monitoring based on multi-sensor data fusion.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to a water quality monitoring method based on multi-sensor data fusion. Background Technology

[0002] While multi-sensor data acquisition technology has been introduced into the field of water quality monitoring, enabling real-time monitoring of multiple parameters such as dissolved oxygen and chemical oxygen demand, existing monitoring methods can only complete the single acquisition and anomaly identification of water quality parameters. They can only answer "what anomaly has occurred" and "what impact might the anomaly have," but cannot accurately trace the source of pollution. Existing technologies lack analytical means to combine the causal relationship between water quality parameters with real-time anomaly data, and have not proposed concepts and inversion algorithms related to pollution contribution. It is difficult to quantify the contribution ratio of each parameter to water quality anomalies, and it is impossible to identify the main suspected pollution sources. As a result, the monitoring results can only serve as a signal of water quality anomalies, and cannot provide targeted decision-making basis for pollution control. Control actions often lack precision and are difficult to achieve efficient pollution control.

[0003] Existing water quality early warning systems mostly use preset, fixed templates for their warning strategies. These strategies can only be matched according to the degree of water quality anomaly, without considering the actual improvement effect after the strategy is implemented. There is no feedback mechanism for strategy optimization, and the effectiveness of the warning strategy is assumed to be constant. Such systems cannot quantitatively evaluate the rate of water quality improvement after strategy implementation, nor can they adjust the strategy generation mechanism based on actual treatment experience. This leads to a mismatch between the warning strategies and the actual water pollution scenarios, resulting in problems such as premature or excessive triggering of strategies or insufficient treatment efforts. The system's decision-making ability cannot be dynamically improved with the usage scenarios and treatment experience. Under long-term use, the effectiveness of early warning and treatment is difficult to guarantee. Therefore, how to improve the long-term stability and accuracy of water quality monitoring has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a water quality monitoring method based on multi-sensor data fusion to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a water quality monitoring method based on multi-sensor data fusion, comprising: S1, acquire multi-source water quality monitoring data, and construct and update a dynamic causal graph reflecting the causal relationship between various water quality parameters based on the multi-source water quality monitoring data; S2, Based on the dynamic cause-effect diagram, the real-time collected water quality parameter data are fused and analyzed to generate a comprehensive water quality status index; S3. Based on the structure of the dynamic cause-effect graph, the pollution contribution inversion analysis is performed on the real-time collected water quality parameter data to determine the main suspected pollution sources; S4. Based on the comprehensive water quality index and the main suspected pollution sources, generate and execute a customized early warning strategy; S5. Based on the feedback of water quality improvement effect after the implementation of the customized early warning strategy, the early warning strategy generation mechanism is adaptively optimized.

[0006] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention innovatively proposes the concept of "pollution contribution" and an inversion algorithm, enabling precise source tracing of water pollution and significantly improving the practicality and targeting of water quality monitoring. It breaks through the limitations of existing technologies that can only identify water quality anomalies and analyze their impact, shifting the research perspective from "what is the anomaly" and "what will happen" to "who caused the anomaly." It deeply integrates the topological structure, edge weights, and time delay coefficients of a dynamic causal graph with real-time water quality anomaly data. By calculating the initial anomaly responsibility value and iteratively backpropagating, the final responsibility vector is obtained, and then normalized and sorted to obtain the pollution contribution score. This accurately quantifies the contribution ratio of each water quality parameter to the overall anomaly, quickly identifying the main suspected pollution sources. It upgrades water quality monitoring results from simple anomaly signals to directional pollution control evidence, providing precise decision support for subsequent customized pollution treatment plans. This effectively solves the problems of insufficient source tracing capabilities and blind action in traditional monitoring.

[0007] 2. This invention innovatively introduces a "strategy-feedback" learning closed loop into the water quality early warning system, realizing adaptive optimization of the early warning strategy generation mechanism and continuously improving the system's decision-making capabilities with actual application. It breaks away from the existing industry status quo of fixed early warning strategies and pre-set effectiveness, clarifying the new understanding that the effectiveness of early warning strategies needs to be dynamically adjusted based on actual scenarios. By establishing a comprehensive strategy effectiveness evaluation model, after the customized early warning strategy is implemented, water quality data is continuously collected to calculate the actual improvement rate, which is then compared with the expected improvement rate to obtain an effectiveness evaluation value. Based on the effectiveness evaluation value and historical records, the adjustable parameters of the early warning strategy template are optimized and adjusted, allowing the system to learn and iterate from each actual handling experience, continuously adapting to the actual characteristics of water pollution scenarios. This effectively solves the problems of traditional early warning strategies being easily out of touch with reality and having unreasonable triggering timing or handling intensity, enabling the decision-making accuracy and governance effectiveness of the early warning system to continuously improve with the usage cycle, realizing the intelligent and dynamic evolution of water quality early warning and governance. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a water quality monitoring method based on multi-sensor data fusion, provided as an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0009] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0010] This application provides a water quality monitoring method based on multi-sensor data fusion. The executing entity of this water quality monitoring method based on multi-sensor data fusion includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the water quality monitoring method based on multi-sensor data fusion can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0011] Reference Figure 1 The diagram shown is a flowchart illustrating a water quality monitoring method based on multi-sensor data fusion according to an embodiment of the present invention. In this embodiment, the water quality monitoring method based on multi-sensor data fusion includes: S1, acquire multi-source water quality monitoring data, and construct and update a dynamic causal graph reflecting the causal relationship between various water quality parameters based on the multi-source water quality monitoring data; In this embodiment of the invention, the multi-source water quality monitoring data includes: Multi-source water quality monitoring data provides core input for the subsequent construction of dynamic cause-effect diagrams. Specifically, it includes water quality physicochemical parameter data directly measured by various sensors deployed in the water body. The main parameters include: dissolved oxygen, chemical oxygen demand, ammonia nitrogen, total phosphorus, pH value, water temperature, turbidity, and conductivity.

[0012] It should be noted that the parameters in the multi-source water quality monitoring data form the basis for the nodes in the dynamic causal graph, and their continuous temporal changes are the direct basis for analyzing the causal relationships between parameters and calculating the pollution contribution. In addition, the data also includes monitoring equipment operating status information, such as sensor signal strength and data timestamps, used to identify and mark anomalies during the data acquisition process, ensuring the reliability of the input data.

[0013] In this embodiment of the invention, the step of constructing and updating a dynamic causal graph reflecting the causal relationship between various water quality parameters based on the multi-source water quality monitoring data includes: Based on multi-source water quality monitoring data, the time series data of each water quality parameter are standardized and missing value imputation is performed to obtain a well-organized multi-parameter water quality time series dataset. Causal relationship mining is performed on the regularized multi-parameter water quality time series dataset to obtain a weighted directed graph reflecting the initial causal relationship between water quality parameters; Based on the regularized multi-parameter water quality time series dataset and the structure of the weighted directed graph, the time delay coefficient of the causal relationship of each pair of directed connections is calculated to obtain a dynamic causal graph with time delay coefficients. Based on the newly acquired real-time multi-source water quality monitoring data in the next acquisition, the edge weights and time delay coefficients in the dynamic causal graph are updated by a rolling window to update the dynamic causal graph.

[0014] It should be noted that standardization and missing value imputation are performed on raw water quality parameter data from different sources and with different dimensions. Normalization calculations based on the mean and standard deviation are performed to eliminate dimensional differences and map the data to a unified standard interval, thereby obtaining standardized parameter values. For missing data points caused by sensor failure or transmission interruption, the average value of data from adjacent stations is used to fill in the missing data points, so as to ensure the continuity and integrity of time series data and provide clean and consistent input data for subsequent causal relationship analysis.

[0015] It should be noted that the causal relationship mining first uses the PC algorithm to analyze the conditional independence between water quality parameters, constructs an acyclic causal skeleton graph, and identifies possible causal relationship directions. Then, based on the skeleton graph, the GES algorithm is used to score different directed graph structures using the Bayesian information criterion, and the causal graph structure with the highest score is searched as the optimal model, thus obtaining the initial weighted directed graph. After obtaining a fixed initial weighted directed graph, for each water quality parameter as a result, its observed value at a certain time is modeled as a linear weighted sum of the observed values ​​of all causal parameters at the corresponding time lag time, plus a random error term. The resulting regression coefficients are then defined as the weight values ​​of the corresponding causal edges. The mathematical expression for this model is as follows:

[0016] In the formula, Indicates the time of the child node The standardized value, This indicates that in a defined causal graph, the nodes... The set of all parent nodes, That is, the problem to be solved, from node to child node. edge weights, Represents a node In time The standardized value, Let be the time delay coefficient from node to node. Represents the regression residuals, where the regression residuals are... It is a direct measurement of the part of the data that was not explained by the model after the model has been fitted, that is, the identified parent nodes and their time-delay relationships. and For node indexing; This weighted directed graph is a graph theory-based data model where nodes represent different water quality parameters, directed edges between nodes represent the direct influence of one parameter on another, and edge weights represent the strength of the influence of one parameter on another.

[0017] Furthermore, the mathematical expression for the Bayesian information criterion scoring function used to evaluate graph structures in causal relationship mining is as follows:

[0018] In the formula, Represents a directed graph Bayesian information criterion score, This represents the number of samples in a regularized multi-parameter water quality time-series dataset. Representation diagram The number of independent parameters in the middle, Represents a given directed graph When the structure is defined, the maximum likelihood estimate of the dataset is given, where the Bayesian information criterion score is used, with a lower score indicating a better model.

[0019] It should be noted that the time delay coefficient calculation operation is performed on each directed edge from the cause node to the result node in the weighted directed graph, calculating the typical time interval required for the change in the cause node to propagate to the result node and produce an observable effect. The mathematical expression for calculating the time delay coefficient is as follows:

[0020] In the formula, Indicates from parameter to parameters The time delay coefficient, This represents the standardized value of the parameter over time. Indicates parameters In time The standardized value, This represents the cross-correlation function, used to measure the correlation between two time series at different time shifts. The correlation strength under, This represents the time shift that maximizes the function value. , To move forward or backward by different number of time steps, and For node indexing; The resulting dynamic cause-effect graph is a composite data structure containing water quality parameter nodes, directed edges, edge weights, and edge attribute time delay coefficients.

[0021] It should be noted that the specific process of the rolling window update operation is as follows: a data window of fixed time length is set, which defaults to the data of the most recent 30 days. As new data continues to arrive, the window scrolls forward. Each time the window is updated, the system re-executes the causal relationship mining operation based on the new data in the window to update the weights of the directed edges; and re-executes the time lag coefficient calculation operation to update the time lag coefficients. The essence of this operation is to use the latest local data to fine-tune and correct the causal model, so that the dynamic causal graph can reflect the latest and most authentic causal relationships and temporal characteristics between water quality parameters, thereby overcoming the interference of environmental changes, seasonal changes or long-term pollution trends on the fixed causal model.

[0022] S2, Based on the dynamic cause-effect diagram, the real-time collected water quality parameter data are fused and analyzed to generate a comprehensive water quality status index; In this embodiment of the invention, the step of fusing and analyzing real-time collected water quality parameter data based on the dynamic causal graph to generate a comprehensive water quality status index includes: Based on the real-time collected raw water quality parameter data and historical baseline statistics, the raw water quality parameter data is standardized to obtain a standardized current water quality parameter vector. The preliminary fusion score of the weighted adjacency matrix of the dynamic causal graph and the standardized current water quality parameter vector is calculated using the causal enhancement fusion algorithm. The initial fusion score is indexed to generate a comprehensive water quality index that represents the overall current water quality status.

[0023] It should be noted that the standardization process uses the same mean and standard deviation parameters as those used when constructing the dynamic cause-effect graph to perform a linear transformation on the raw water quality parameter values ​​collected in real time.

[0024] It should be noted that the indexation mapping operation transforms the initial fusion score into a standardized index that is intuitive, easy to understand, and suitable for decision-making. Its mathematical expression is as follows:

[0025] In the formula, It is a comprehensive index of water quality status. This is the initial fusion score at the current moment. It is a reference baseline value. It is the curvature parameter; The comprehensive water quality index is a single quantitative indicator that integrates multiple abnormal parameters and their causal importance. The higher the value, the worse the water quality. It provides a unified, sensitive and physically meaningful decision-making basis for subsequent situation assessment and early warning triggering. The reference baseline is the median of all preliminary fusion scores over the past year, with a curvature parameter of 1.5.

[0026] In this embodiment of the invention, the mathematical expression of the causal enhancement fusion algorithm is as follows: ; In the formula, For the node at time Real-time anomaly degree, For the node at time The standardized value, To preset the abnormal threshold, For nodes The fusion weight is the adjustment factor, and the node is the fusion weight. The normalized causal centrality score, For nodes The normalized causal centrality score, To initially integrate the scores, and For node indexing, This represents the total number of nodes in the dynamic causal graph. For the current monitoring time, For parameters The abnormal direction coefficient takes a value of +1 or -1.

[0027] It should be noted that the causal centrality score of each parameter in the causal enhancement fusion algorithm...

[0028] In the formula, Let be the causal centrality vector. for The identity matrix, As the attenuation factor, A matrix consisting of edge weights The transpose of the matrix, elements ; Solving and normalizing yields the causal centrality vector. .

[0029] Furthermore, the preset anomaly threshold is determined based on the 95th percentile of historical data, and the adjustment factor is a set of preset candidate values, specifically: β=[0.5,1,2,3,5,10], with a default value of 2, which can be modified by the operator among the candidate values; the attenuation factor is obtained by calculating the largest eigenvalue in the transpose of the matrix formed by dividing 0.85 by the edge weights.

[0030] It should be noted that the calculation of the real-time anomaly degree is a one-way anomaly judgment based on the physical characteristics of different water quality parameters. For parameters with higher concentrations and greater risk, a specific setting is used. At this point, the formula calculates the degree to which it exceeds the threshold; for parameters where lower concentrations carry greater risk, the setting is... At this point, the formula calculates the degree to which it falls below the threshold. If it is within the normal range, the real-time anomaly level is 0.

[0031] S3. Based on the structure of the dynamic cause-effect graph, the pollution contribution inversion analysis is performed on the real-time collected water quality parameter data to determine the main suspected pollution sources; In this embodiment of the invention, the step of performing pollution contribution inversion analysis on the real-time collected water quality parameter data based on the structure of the dynamic causal graph to determine the main suspected pollution sources includes: Based on the standardized current water quality parameter vector and the preset anomaly threshold, the initial anomaly responsibility value of each water quality parameter is calculated to obtain the initial responsibility vector. Based on the topology, edge weights, and time delay coefficients of the dynamic causal graph, the initial responsibility vector is subjected to graph-based backpropagation iterative calculation to obtain the final responsibility vector. The final responsibility vector is normalized and sorted to obtain the pollution contribution score of each parameter. The main suspected pollution sources are determined based on the pollution contribution score.

[0032] It should be noted that the initial abnormality responsibility value represents the starting point for tracing pollution liability. The specific calculation method is as follows: For each water quality parameter, its standardized value is extracted from the standardized current water quality parameter vector and compared with the preset abnormality threshold of the parameter to obtain the initial abnormality responsibility value, which is used to distinguish between normal fluctuations and significant abnormalities; the initial abnormality responsibility values ​​obtained through this operation are combined to form the initial responsibility vector.

[0033] It should be noted that graph-based backpropagation iterative calculation is the core step in pollution contribution inversion. Its essence is to simulate the process of tracing "responsibility" or "pollution contribution" upstream along the causal chain. The specific algorithm of this backpropagation iterative calculation is as follows: the responsibility value of a node in the first step is equal to its responsibility value in the previous step, plus the responsibility traced back from all its downstream child nodes. The amount of responsibility traced back from each child node is equal to the current initial abnormal responsibility value of that child node, multiplied by the edge weight connecting the two nodes, divided by 1, and then added to the time delay coefficient between the two nodes. The longer the time delay coefficient, the longer the time delay of the causal effect, and the weaker the direct impact at the current moment. Therefore, the traced responsibility is correspondingly reduced. Then, a convergence determination is performed, and the iteration continues until the change in the responsibility value of all nodes is less than a preset termination threshold. The preset termination threshold is set to a value, at which point the distribution of responsibility in the network is considered to have reached a stable state, and the final responsibility vector is obtained.

[0034] It should be noted that normalization and pollution source identification convert the calculated absolute responsibility value into a comparable relative contribution and identify key pollution parameters. The specific steps are as follows: First, the contribution score is calculated: the final responsibility vector is normalized using Softmax to obtain a pollution contribution score vector; this operation maps all responsibility values ​​to a probability distribution, such that the pollution contribution score of each element is... And the sum of all scores is 1, which intuitively represents the contribution ratio of the current parameter to the current overall water quality anomaly. Subsequently, the main suspected pollution sources were identified: the main suspected pollution sources were identified as the parameters with the highest contribution scores; ultimately, the pollution contribution scores and the main suspected pollution sources together constituted the accurate pollution source tracing diagnosis results. They upgraded the traditional monitoring of "what abnormality occurred" to "which parameter or parameters mainly caused the abnormality", providing a direct and operable basis for the subsequent generation of targeted early warning strategies.

[0035] In this embodiment of the invention, the calculation process of the initial anomaly responsibility value is as follows: Calculate the difference between the standardized value of the parameter and the preset anomaly threshold; If the difference is greater than 0, it indicates that the water quality parameters of the current node are abnormal, and the difference between the standardized parameter value and the preset abnormal threshold is used as the initial abnormal responsibility value. If the difference is less than or equal to 0, the water quality parameters of the current node are considered normal, and the initial abnormality responsibility value is set to 0.

[0036] S4. Based on the comprehensive water quality index and the main suspected pollution sources, generate and execute a customized early warning strategy; In this embodiment of the invention, the step of generating and executing a customized early warning strategy based on the comprehensive water quality index and the main suspected pollution sources includes: The current water quality status level is obtained by comprehensively analyzing the water quality status index and the main suspected pollution sources using the situation assessment rules. Based on the water quality status level, a target template is matched for the main suspected pollution sources, and the main suspected pollution sources and the influence path information of the main suspected pollution sources in the dynamic cause-effect diagram are filled into the target template to generate a customized early warning strategy. The customized early warning strategy is executed, and the instruction information therein is distributed to the preset early warning release terminal.

[0037] It should be noted that the situation assessment rule is a pre-defined decision-making logic that combines quantitative indices with qualitative source tracing results to determine the comprehensive risk level. First, based on the numerical range of the comprehensive water quality index, it is mapped to a preliminary baseline risk level, as follows: For good, Slight pollution. Moderate pollution. It is severely polluted; Then, by combining the attributes of the main suspected pollution sources, such as whether the pollution type is toxic or whether it is easy to spread rapidly, the preliminary level is revised. First, a baseline risk level is determined based on the preset range into which the comprehensive water quality index falls. Then, the risk attributes of the main suspected pollution sources are queried. For example, if the main suspected pollution source is "cyanide", the risk attribute is "extremely high". If the risk attribute of the pollution source is higher than the current baseline risk level, the water quality status level is upgraded by one level; otherwise, the baseline risk level is maintained. The water quality status level obtained through this operation is a comprehensive judgment result that integrates the severity of pollution and the dangerous nature of the pollution, which is more accurate and comprehensive than the classification based on a single index.

[0038] It should be noted that strategy template matching and content filling are key to achieving customized early warning information. First, template matching is performed. The pre-built early warning strategy template library stores various early warning strategy frameworks for different combinations of "situation level - pollution source type." Each template is a structured text or instruction framework containing replaceable variable placeholders. The system uses binary pairs... Using the key, a query is performed in the template library for matching, where, Indicates the water quality status level. Indicates pollution source The category it belongs to, such as "nutrients", "heavy metals", "organic toxins", is used to match the most relevant target template; Then, information is populated. Based on the dynamic causal graph, the impact path information of the main suspected pollution sources is extracted. Starting from the node in the causal graph, a breadth-first search is performed along all outward directed edges to identify all its direct and indirect downstream impact nodes. The set of these nodes constitutes the potential impact range. Then, the specific parameter names of the main suspected pollution sources, the list of water quality parameters involved in their impact paths, and the possible impact time estimated based on edge weights and time delays are populated into the corresponding variable placeholders in the target template. Finally, after the strategy is generated and filled in, the original framework template is transformed into a specific, customized early warning strategy that includes a clear source of pollution, scope of impact, and risk warnings. This strategy not only includes the warning level, but also diagnostic information on where and what causes what kind of risk. For example, the generated customized early warning strategy might be: Orange warning: The current ammonia nitrogen (NH3-N) concentration in the water is abnormally high, which is expected to lead to a significant decrease in dissolved oxygen content downstream in the next 2-4 hours. It is recommended to conduct key inspections of upstream sewage outlets and prepare for oxygenation.

[0039] In this embodiment of the invention, the water quality status level is a classification variable, which includes four levels: "blue", "yellow", "orange" and "red", corresponding to the four levels of the baseline risk level in the status assessment rules: good, slightly polluted, moderately polluted and heavily polluted.

[0040] S5. Based on the feedback of water quality improvement effect after the implementation of the customized early warning strategy, the early warning strategy generation mechanism is adaptively optimized.

[0041] In this embodiment of the invention, the adaptive optimization of the early warning strategy generation mechanism based on the water quality improvement feedback after the execution of the customized early warning strategy includes: During the preset observation period after the implementation of the customized early warning strategy, water quality data are continuously collected and a comprehensive water quality status index is calculated. The actual water quality improvement rate is calculated based on the time series of the comprehensive water quality status index. Based on the type and intensity of the customized early warning strategy and the attributes of the main suspected pollution sources, the expected water quality improvement rate is obtained by calculation through a preset expected effect prediction model. Based on the actual water quality improvement rate and the expected water quality improvement rate, the water quality improvement effectiveness is evaluated to obtain the effectiveness evaluation value of the customized early warning strategy. Based on the performance evaluation value and the corresponding historical performance records, the adjustable parameters of the strategy template from which the customized early warning strategy in the early warning strategy template library originates are optimized and adjusted to update the early warning strategy generation mechanism.

[0042] It should be noted that the calculation of the actual water quality improvement rate is a direct quantitative assessment of the effectiveness of the early warning strategy. First, a fixed observation period is set, which is 24 hours by default. Starting from the time the customized early warning strategy is implemented, water quality data is continuously collected and the comprehensive water quality index is calculated in real time to obtain the index sequence during the observation period. Subsequently, a linear fit was performed on the change of the index sequence over time, and the slope of the fitted line is the actual rate of water quality improvement; if the slope is negative, it indicates that the index is decreasing and the water quality is improving; if the slope is positive, it indicates that the index is increasing and the water quality is deteriorating; the absolute value of the slope reflects the speed of improvement or deterioration.

[0043] It should be noted that the mathematical expression for the expected outcome prediction model is as follows:

[0044] In the formula, To achieve the desired rate of water quality improvement, Based on improving the rate constant, For strategy type coefficients, For pollution source category coefficients, This represents the strategy strength coefficient. The basic improvement rate constant represents the typical rate of decline in the comprehensive water quality index after implementing standard intensity routine water quality improvement measures in the water area. The default value is -2.0 index points / day. A negative sign indicates a decrease in the index and water quality improvement. The strategy type coefficient is extracted based on the specific measure type of the early warning strategy: production restriction and emission reduction, ecological water replenishment, and emergency pesticide application. The corresponding values ​​are extracted for each measure type. When the measure type is ecological water replenishment When the measure type is emergency drug administration ; The pollution source category coefficient is set based on the categories of the main suspected pollution sources identified: nutrients, heavy metals, and organic matter. The value for nutrients is set accordingly. ; Corresponding values ​​for heavy metals ; values ​​corresponding to organic compounds ; The strategy intensity coefficient is set based on the action intensity level specified in the early warning strategy: good, mild, moderate, and severe. The value for "good" is [value missing]. The corresponding value for mild cases is... The corresponding value for moderate is... The value corresponding to severe cases is... .

[0045] It should be noted that the water quality improvement effectiveness assessment is used to calculate the ratio of the actual effect of the customized early warning strategy to the expected effect, thereby obtaining a standardized effectiveness evaluation value. When the effectiveness evaluation value is greater than 1, it means that the actual effect of the customized early warning strategy is better than expected; when the effectiveness evaluation value is equal to 1, it means that the effect of the customized early warning strategy meets expectations; when the effectiveness evaluation value is less than 1, it means that the effect of the customized early warning strategy does not meet expectations.

[0046] It should be noted that the optimization and adjustment of strategy template parameters is the core step in realizing the self-evolution of the early warning strategy generation mechanism. First, the strategy template corresponding to the customized early warning strategy executed this time and the calculated performance evaluation value are associated with the scene features and stored in the historical learning log. Then, the log is analyzed. For any strategy template, the moving average of its performance evaluation value for several recent executions is calculated. The optimization algorithm fine-tunes the adjustable parameters of the template based on the deviation between the moving average and the preset performance target. A typical optimization rule is as follows: If the moving average of a template's recent multiple executions is lower than the performance target, it is considered that the preset anomaly threshold for triggering the strategy may be too sensitive, causing the strategy to be triggered too early or too frequently. Therefore, the preset anomaly threshold is appropriately increased so that the strategy is only triggered in more severe situations in the future. Conversely, if the average performance is higher than the target, it may indicate that the strategy is not triggered in a timely manner, and the preset anomaly threshold can be appropriately decreased. The adjustment amount is the difference between the moving average and the performance target multiplied by 0.01.

[0047] In the several embodiments provided by this invention, it should be understood that the disclosed method can be implemented in other ways.

[0048] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0049] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, and technology that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A water quality monitoring method based on multi-sensor data fusion, characterized in that, The method includes: S1, acquire multi-source water quality monitoring data, and construct and update a dynamic causal graph reflecting the causal relationship between various water quality parameters based on the multi-source water quality monitoring data; S2, Based on the dynamic cause-effect diagram, the real-time collected water quality parameter data are fused and analyzed to generate a comprehensive water quality status index; S3. Based on the structure of the dynamic cause-effect graph, the pollution contribution inversion analysis is performed on the real-time collected water quality parameter data to determine the main suspected pollution sources; S4. Based on the comprehensive water quality index and the main suspected pollution sources, generate and execute a customized early warning strategy; S5. Based on the feedback of water quality improvement effect after the implementation of the customized early warning strategy, the early warning strategy generation mechanism is adaptively optimized.

2. The water quality monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The multi-source water quality monitoring data includes: Multi-source water quality monitoring data provides core input for the subsequent construction of dynamic cause-effect diagrams. Specifically, it includes water quality physicochemical parameter data directly measured by various sensors deployed in the water body. The main parameters include: dissolved oxygen, chemical oxygen demand, ammonia nitrogen, total phosphorus, pH value, water temperature, turbidity, and conductivity.

3. The water quality monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The construction and updating of the dynamic causal graph reflecting the causal relationships among various water quality parameters based on the multi-source water quality monitoring data includes: Based on multi-source water quality monitoring data, the time series data of each water quality parameter are standardized and missing value imputation is performed to obtain a well-organized multi-parameter water quality time series dataset. Causal relationship mining is performed on the regularized multi-parameter water quality time series dataset to obtain a weighted directed graph reflecting the initial causal relationship between water quality parameters; Based on the regularized multi-parameter water quality time-series dataset and the structure of the weighted directed graph, the time delay coefficient of the causal relationship of each pair of directed connections is calculated to obtain a dynamic causal graph with time delay coefficients. Based on the newly acquired real-time multi-source water quality monitoring data in the next acquisition, the edge weights and time delay coefficients in the dynamic causal graph are updated by a rolling window to update the dynamic causal graph.

4. The water quality monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The process of fusing and analyzing real-time collected water quality parameter data based on the dynamic causal graph to generate a comprehensive water quality status index includes: Based on the real-time collected raw water quality parameter data and historical baseline statistics, the raw water quality parameter data is standardized to obtain a standardized current water quality parameter vector. The preliminary fusion score of the weighted adjacency matrix of the dynamic causal graph and the standardized current water quality parameter vector is calculated using the causal enhancement fusion algorithm. The initial fusion score is indexed to generate a comprehensive water quality index that represents the overall current water quality status.

5. A water quality monitoring method based on multi-sensor data fusion as described in claim 4, characterized in that, The mathematical expression of the causal enhancement fusion algorithm is as follows: ; In the formula, For the node at time Real-time anomaly degree, For the node at time The standardized value, To preset an abnormal threshold, For nodes The fusion weight is the adjustment factor, and the node is the fusion weight. The normalized causal centrality score, For nodes The normalized causal centrality score, To initially integrate the scores, and For node indexing, This represents the total number of nodes in the dynamic causal graph. For the current monitoring time, For parameters The abnormal direction coefficient takes a value of +1 or -1.

6. The water quality monitoring method based on multi-sensor data fusion as described in claim 4, characterized in that, Based on the structure of the dynamic causal graph, the pollution contribution inversion analysis is performed on the real-time collected water quality parameter data to identify the main suspected pollution sources, including: Based on the standardized current water quality parameter vector and the preset anomaly threshold, the initial anomaly responsibility value of each water quality parameter is calculated to obtain the initial responsibility vector. Based on the topology, edge weights, and time delay coefficients of the dynamic causal graph, the initial responsibility vector is subjected to graph-based backpropagation iterative calculation to obtain the final responsibility vector. The final responsibility vector is normalized and sorted to obtain the pollution contribution score of each parameter. The main suspected pollution sources are determined based on the pollution contribution score.

7. A water quality monitoring method based on multi-sensor data fusion as described in claim 6, characterized in that, The calculation process for the initial anomaly responsibility value is as follows: Calculate the difference between the standardized value of the parameter and the preset anomaly threshold; If the difference is greater than 0, it indicates that the water quality parameters of the current node are abnormal, and the difference between the standardized parameter value and the preset abnormal threshold is used as the initial abnormal responsibility value. If the difference is less than or equal to 0, the water quality parameters of the current node are considered normal, and the initial abnormality responsibility value is set to 0.

8. A water quality monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The process of generating and executing a customized early warning strategy based on the comprehensive water quality index and the main suspected pollution sources includes: The current water quality status level is obtained by comprehensively analyzing the water quality status index and the main suspected pollution sources using the situation assessment rules. Based on the water quality status level, a target template is matched for the main suspected pollution sources, and the main suspected pollution sources and the influence path information of the main suspected pollution sources in the dynamic cause-effect diagram are filled into the target template to generate a customized early warning strategy. The customized early warning strategy is executed, and the instruction information therein is distributed to the preset early warning release terminal.

9. A water quality monitoring method based on multi-sensor data fusion as described in claim 8, characterized in that, The water quality status level is a classification variable that includes four levels: "blue," "yellow," "orange," and "red," which correspond to the four levels of the baseline risk level in the status assessment rules: good, slightly polluted, moderately polluted, and heavily polluted.

10. A water quality monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The adaptive optimization of the early warning strategy generation mechanism based on the water quality improvement feedback after the implementation of the customized early warning strategy includes: During the preset observation period after the implementation of the customized early warning strategy, water quality data are continuously collected and a comprehensive water quality status index is calculated. The actual water quality improvement rate is calculated based on the time series of the comprehensive water quality status index. Based on the type and intensity of the customized early warning strategy and the attributes of the main suspected pollution sources, the expected water quality improvement rate is obtained by calculation through a preset expected effect prediction model. Based on the actual water quality improvement rate and the expected water quality improvement rate, the water quality improvement effectiveness is evaluated to obtain the effectiveness evaluation value of the customized early warning strategy. Based on the performance evaluation value and the corresponding historical performance records, the adjustable parameters of the strategy template from which the customized early warning strategy in the early warning strategy template library originates are optimized and adjusted to update the early warning strategy generation mechanism.

Images