Automatic water quality monitoring method and system

Abstract

The invention relates to the technical field of water quality monitoring, in particular to an automatic water quality monitoring method and system.The method comprises the steps that multiple pieces of collected water quality monitoring data are combined pairwise, dynamic coupling strength is calculated, a topological network atlas is generated, and automatic extraction and structural characterization of the dynamic coupling relation among complex water quality parameters are achieved; the limitation of dependence on manual feature recognition traditionally is overcome; secondly, matching the topological network atlas with a preset pollution mode feature library, dynamically determining a newly added abnormal mode, and outputting an abnormal feature code set, thereby solving the key defect that a static model cannot recognize an unknown pollution mode; and finally, a water quality monitoring and early warning signal is output by fusing the pollution diffusion prediction result and the abnormal feature code set, so that bidirectional verification of data driving and a mechanism model is realized, and the early warning accuracy of a water quality abnormal phenomenon is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality monitoring, and in particular to a method and system for automatic water quality monitoring. Background Art

[0002] In the fields of environmental monitoring and public health, water quality monitoring plays an irreplaceable role in providing early warning of pollution incidents and ensuring drinking water safety. Current mainstream approaches rely on multi-parameter time-series analysis systems, deploying sensor networks to collect key indicators such as pH, dissolved oxygen, turbidity, and conductivity in real time, and detecting anomalies based on threshold comparisons or statistical models.

[0003] However, this type of method has significant limitations: on the one hand, faced with nonlinear and highly coupled water quality parameter interactions in complex water environments, it is necessary to rely on expert experience to manually design feature recognition; on the other hand, static models cannot adaptively learn new pollutant patterns, resulting in delayed or even missed identification of unknown abnormal conditions.

[0004] Therefore, how to automatically extract the key features of complex water quality data to accurately warn of abnormal water quality conditions is an urgent problem that technical personnel in this field need to solve. Summary of the Invention

[0005] In order to automatically extract key features of complex water quality data and accurately warn of abnormal water quality conditions, the present application provides a method and system for automatic water quality monitoring.

[0006] The present application provides a method and system for automatic water quality monitoring using the following technical solutions: A method for automatic water quality monitoring, comprising: By calculating the dynamic coupling strength between multiple water quality monitoring data collected, the target water quality monitoring data with dynamic coupling strength exceeding the preset coupling strength is selected, and directional edges are generated on the network vertices mapped by the target water quality monitoring data, and a topological network map is output; Match the topological network map with the preset pollution pattern feature library, determine the newly added abnormal pattern, and output the abnormal feature coding set; Obtain pollution diffusion prediction results, integrate the pollution diffusion prediction results and abnormal feature coding sets, and output water quality monitoring early warning signals.

[0007] Furthermore, by calculating the dynamic coupling strength between the multiple water quality monitoring data collected, selecting target water quality monitoring data whose dynamic coupling strength exceeds a preset coupling strength, generating directional edges on the network vertices mapped by the target water quality monitoring data, and outputting the topological network map, the steps include: Conduct cross-dimensional correlation analysis on the changing trends of any two water quality monitoring data and calculate the dynamic coupling strength between the two water quality monitoring data; When the dynamic coupling strength exceeds the preset coupling strength, a directional edge is generated on the network vertices mapped by the corresponding two target water quality monitoring data until multiple water quality monitoring data are traversed; Based on the generated multiple directional edges and the network vertices corresponding to the edges, a topological network graph is output.

[0008] Furthermore, the steps of performing cross-dimensional correlation analysis on the changing trends of any two water quality monitoring data and calculating the dynamic coupling strength between the two water quality monitoring data include: Identify the changing trends of any two water quality monitoring data and obtain the first trend feature and the second trend feature; Obtaining the degree of consistency between the first trend feature and the second trend feature in the direction of data change, the degree of matching in the data change rate, and the degree of coordination in the data fluctuation pattern, and generating a comparison result; The output dynamic coupling strength is calculated based on the comparison results.

[0009] Furthermore, the steps of matching the topological network map with the preset pollution pattern feature library, determining the newly added abnormal pattern, and outputting the abnormal feature coding set include: Traverse the topological network graph and parse it to obtain multiple key substructures; Dynamically match each key substructure with the preset pollution pattern feature library to obtain a correlation matching threshold; Obtain a target key substructure whose correlation matching threshold is lower than a preset correlation matching threshold, and generate a new abnormal pattern based on the target key substructure; Aggregate the newly added abnormal patterns and the fixed abnormal patterns in the preset pollution pattern feature library, and output the abnormal feature coding set.

[0010] Furthermore, the steps of traversing the topological network graph and parsing it to obtain multiple key substructures include: Based on the edge weight gradient mutation traversal detection topology network map, the path whose gradient change exceeds the preset gradient change amount is extracted as the candidate path; Adjacent nodes of each candidate path are aggregated to generate multiple key substructures that characterize the pollution conduction characteristics.

[0011] Furthermore, the steps of traversing and parsing the topological network graph to obtain multiple key substructures include: Based on the temporal conduction characteristics, the node-to-node association paths of the topological network graph are traversed and analyzed, and the conduction paths with time delay characteristics are identified as candidate paths; By backtracing the adjacent nodes of the candidate path, contamination conduction clusters are generated and output as multiple key substructures.

[0012] Furthermore, the steps of integrating the pollution diffusion prediction results and the abnormal feature coding set to output a water quality monitoring early warning signal include: Correct the confidence weight of the abnormal feature coding set, spatially fuse the corrected abnormal feature coding set with the pollution diffusion prediction results to generate a risk distribution map; According to the risk gradient mutation boundary of the risk distribution map, water quality monitoring early warning signals are output.

[0013] Furthermore, the steps of spatially fusing the modified abnormal feature coding set with the pollution diffusion prediction results to generate a risk distribution map include: According to the spatial location correlation, the modified abnormal feature coding set is dynamically weighted and fused with the pollution diffusion prediction results to output the spatial risk value distribution; The spatialized risk value distribution is passed through the risk gradient generator to construct a continuous spatialized risk value distribution; According to the spatial variation characteristics of the continuous spatial risk value distribution, the risk level boundary is extracted to obtain the risk distribution map.

[0014] Furthermore, the step of outputting a water quality monitoring warning signal according to the risk gradient mutation boundary of the risk distribution map includes: Analyze the risk gradient mutation boundary of the risk distribution map to obtain boundary characteristic attributes and risk transmission attributes; Through behavioral semantic mapping, the boundary characteristic attributes and risk conduction attributes are mapped into pollution behavior semantics; Based on the pollution behavior semantic matching response strategy, after generating the warning instruction, the water quality monitoring warning signal is output according to the warning instruction.

[0015] This application also proposes a water quality automatic monitoring system, comprising: The data coupling module is used to calculate the dynamic coupling strength between multiple water quality monitoring data collected, select target water quality monitoring data whose dynamic coupling strength exceeds the preset coupling strength, generate directional edges on the network vertices mapped by the target water quality monitoring data, and output a topological network map; The data matching module is used to match the topological network map with the preset pollution pattern feature library, determine the newly added abnormal pattern, and output the abnormal feature code set; The data decision module is used to obtain pollution diffusion prediction results, integrate the pollution diffusion prediction results and abnormal feature coding sets, and output water quality monitoring early warning signals.

[0016] Beneficial effects achieved: The present application proposes a method for automatic water quality monitoring, comprising: calculating the dynamic coupling strength between multiple collected water quality monitoring data, screening target water quality monitoring data whose dynamic coupling strength exceeds a preset coupling strength, generating directional edges on the network vertices mapped by the target water quality monitoring data, and outputting a topological network map; matching the topological network map with a preset pollution pattern feature library, determining newly added abnormal patterns, and outputting an abnormal feature coding set; obtaining pollution diffusion prediction results, fusing the pollution diffusion prediction results and the abnormal feature coding set, and outputting a water quality monitoring early warning signal.

[0017] That is, in this application, by combining multiple collected water quality monitoring data in pairs, calculating the dynamic coupling strength, and generating a topological network map, the dynamic coupling relationship between complex water quality parameters is automatically extracted and structured represented, overcoming the limitations of traditional reliance on manual feature recognition; then, by matching the topological network map with the preset pollution pattern feature library, and dynamically determining the newly added abnormal pattern, and outputting the abnormal feature coding set, the key defect of the static model that cannot identify unknown pollution patterns is solved; finally, by fusing the pollution diffusion prediction results and the abnormal feature coding set, the water quality monitoring early warning signal is output, realizing the two-way verification of data-driven and mechanism models, and significantly improving the accuracy of early warning of water quality anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of an automatic water quality monitoring method of the present application; Figure 2 This is a schematic diagram of the process of generating a topological network map based on multiple water quality monitoring data in this application; Figure 3 This is a schematic diagram of the topological network map output by this application; Figure 4 This is a flow chart of the application's output of an abnormal feature coding set based on a topological network graph and a preset pollution pattern feature library; Figure 5 This is a flow chart of the application for outputting water quality monitoring warning signals based on the abnormal feature coding set; Figure 6 It is a structural diagram of an automatic water quality monitoring system of the present application.

[0019] Description of Figure Numbers: 10. Data coupling module; 20. Data matching module; 30. Data decision module. DETAILED DESCRIPTION

[0020] The following is combined with Figure 1-6 This application is described in further detail.

[0021] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0022] In the description of this application, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on the specific circumstances.

[0023] The embodiment of the present application discloses a method for automatic water quality monitoring.

[0024] Please refer to Figure 1 In one embodiment of the present application, a method for automatic water quality monitoring includes steps S10 to S30: Step S10, by calculating the dynamic coupling strength between the multiple water quality monitoring data collected, screening the target water quality monitoring data whose dynamic coupling strength exceeds the preset coupling strength, generating directional edges on the network vertices mapped by the target water quality monitoring data, and outputting a topological network map.

[0025] Current water quality monitoring schemes mainly rely on single-parameter threshold alarms or multi-parameter statistical correlation analysis, which have the defect of being difficult to capture the nonlinear coupling effects between multiple parameters in complex water quality systems, resulting in insufficient ability to identify complex pollution events.

[0026] In response to the above-mentioned problems, this embodiment combines multiple collected water quality monitoring data, such as dissolved oxygen, pH value, conductivity, and other data that can reflect the characteristics of the water environment, calculates the dynamic coupling strength between the two water quality monitoring data, and screens the target water quality monitoring data whose dynamic coupling strength exceeds the preset coupling strength. At the network vertices mapped by the target water quality monitoring data, a directional edge is generated. Based on the edge and the corresponding network vertex, a topological network map that characterizes the dynamic association between parameters is constructed. The core of this method is to convert multiple discrete water quality monitoring data into a network structure. Because pollution events in real water environments often manifest as coordinated changes in data clusters, such as a sudden drop in dissolved oxygen accompanied by a surge in turbidity, or inverse fluctuations in pH and conductivity, traditional linear statistics cannot quantify such complex interactions.

[0027] This embodiment can automatically extract the implicit conduction paths and strength relationships between various water quality monitoring data, where network vertices map water quality monitoring data entities, and edge weights dynamically encode the spatiotemporal correlation of water quality monitoring data changes. This can effectively identify key features such as the ammonia nitrogen-dissolved oxygen correlation break caused by industrial wastewater and the chlorophyll a-temperature coupled oscillation in the early stages of algal blooms, providing high-dimensional feature input for pollution pattern matching and significantly improving the early warning capabilities for hidden pollution and new pollutants.

[0028] Step S20: Match the topological network map with the preset pollution pattern feature library, determine the newly added abnormal pattern, and output the abnormal feature coding set.

[0029] By matching the topological network map with the preset pollution pattern feature library, it aims to solve the key defect that conventional water pollution identification models cannot identify unknown pollution patterns.

[0030] By dynamically comparing abnormal substructures in the topological network map with known pollution patterns in the preset pollution pattern feature library, a new abnormal pattern generation mechanism is triggered when the matching degree is lower than the set threshold, and an abnormal feature coding set containing the coupling characteristics and topological attributes of water quality monitoring data is output. In this way, the adaptive expansion and incremental learning capabilities of the preset pollution pattern feature library are achieved. When an unknown pollution event occurs, the system can automatically generate the corresponding feature template and encode and store it, significantly improving the real-time capture capability and warning accuracy of new pollution patterns, and avoiding the risk of underreporting caused by model solidification in traditional solutions.

[0031] Step S30: Obtain the pollution diffusion prediction result, fuse the pollution diffusion prediction result and the abnormal feature coding set, and output a water quality monitoring early warning signal.

[0032] By obtaining pollution diffusion prediction results and fusing them with abnormal feature coding sets, the high false alarm rate and delayed response caused by the disconnect between water quality monitoring data and physical models in traditional water quality early warning systems can be overcome. This is because in actual pollution events, abnormal feature coding sets may be affected by sensor noise or local interference, generating spurious signals. Pollution diffusion results, based on hydrodynamic mechanisms, provide physical constraints. This allows for a dynamic weighted fusion of the real-time abnormal patterns represented by the abnormal feature coding sets and the pollutant migration trends quantified by the pollution diffusion prediction results. Physical rules are then used to verify the credibility of abnormal features and to modify the pre-set pollution pattern feature library.

[0033] The intelligent decision-making output under the mechanism constraint can drive the water quality monitoring prediction signal of the equipment action, significantly improve the accuracy of identifying complex pollution events, and avoid the low warning accuracy caused by the limitation of single water quality monitoring data.

[0034] It should be noted that the hydrodynamic mechanism can provide physical rules to constrain the migration and diffusion laws of pollutants based on the principles of fluid mechanics, and provide mandatory boundary conditions for the rationality verification and correction of water quality monitoring and early warning signals.

[0035] In this embodiment, by combining multiple collected water quality monitoring data in pairs, calculating the dynamic coupling strength, and generating a topological network map, the dynamic coupling relationship between complex water quality parameters is automatically extracted and structured represented, overcoming the limitations of traditional reliance on manual feature recognition; then, by matching the topological network map with a preset pollution pattern feature library, and dynamically determining new abnormal patterns, and outputting an abnormal feature coding set, the key defect of the static model that it cannot identify unknown pollution patterns is solved; finally, by fusing the pollution diffusion prediction results and the abnormal feature coding set, a water quality monitoring early warning signal is output, realizing two-way verification of data-driven and mechanism models, and significantly improving the accuracy of early warning of water quality anomalies.

[0036] In one possible implementation, refer to Figure 2 As shown, step S10 includes steps A10 to A30: Step A10: Perform cross-dimensional correlation analysis on the changing trends of any two water quality monitoring data to calculate the dynamic coupling strength between the two water quality monitoring data.

[0037] Firstly, sliding window processing combined with linear regression is used to extract the change trends of the first water quality monitoring data and the second water quality monitoring data, respectively, to generate the first trend feature and the second trend feature. Each trend feature contains the data change direction component, the data change rate scalar value and the data fluctuation pattern sequence.

[0038] Cross-dimensional correlation analysis compares the characteristics of parameters across different dimensions, such as ammonia nitrogen and dissolved oxygen. Specifically, cosine similarity is used to calculate the consistency of the first and second trend characteristics in the direction of data change. The degree of matching in data change rates is determined by subtracting the difference between the calculated rates and the normalized values. Dynamic time warping is then applied to quantify the degree of coordination of fluctuation patterns. The results of these three-dimensional comparisons are input into an integration function, which outputs the dynamic coupling strength, a comprehensive representation of the degree of mutual influence between the two water quality monitoring data sets.

[0039] This process directly supports the automated extraction of implicit conduction paths and intensity relationships between various water quality monitoring data, and updates all calculation results in real time through the stream processing framework.

[0040] Optionally, step A10 may include steps A11 to A13: Step A11: Identify the change trends of any two water quality monitoring data to obtain a first trend feature and a second trend feature.

[0041] First, through real-time time series analysis, such as sliding window processing combined with linear regression, the changing trends of any two water quality monitoring data are identified. The purpose is to extract the dynamic behavior characteristics of each water quality monitoring data. For example, the short-term change direction (upward or downward tendency), change rate (the amount of numerical change per unit time) and fluctuation pattern (sequence periodicity or amplitude characteristics) of each water quality monitoring data are calculated, thereby obtaining the first trend feature and the second trend feature.

[0042] Step A12: Obtain the consistency degree of the first trend feature and the second trend feature in the data change direction, the matching degree in the data change rate, and the coordination degree in the data fluctuation pattern, and generate a comparison result.

[0043] The degree of consistency between the first trend feature and the second trend feature in the direction of data change, the degree of matching in the data change rate, and the degree of coordination in the data fluctuation pattern are obtained, with the aim of generating a comprehensive comparison result to compare the similarity and correlation between the two trend features.

[0044] It should be noted that the degree of consistency between the first trend feature and the second trend feature in the direction of data change can be obtained by calculating the cosine similarity; the degree of matching between the first trend feature and the second trend feature in the data change rate can be obtained by calculating the difference calculation rate; the degree of coordination between the first trend feature and the second trend feature in the data fluctuation pattern can be obtained by applying the dynamic time normalization integer quantification fluctuation coordination.

[0045] Specifically, the consistency between the first trend feature and the second trend feature in the direction of data change can be calculated by cosine similarity as follows: The data change directions extracted by the first trend feature and the second trend feature are converted into direction vectors respectively, where the dimension of each direction vector corresponds to the sequence position of the real-time sliding window, and each real-time sliding window calculates its numerical change slope as the direction component; the dot product of the direction vector of the first trend feature and the direction vector of the second trend feature is calculated, and then divided by the product of the module lengths of the two vectors, and the cosine similarity is output.

[0046] The calculated cosine similarity represents the degree of consistency between the first and second trend features in the direction of data change. Specifically, it outputs 1 when the two vectors are in exactly the same direction, 0 when they are orthogonal, and -1 when they are in opposite directions. Each time new water quality monitoring data arrives, the direction vector within the sliding window is dynamically updated and the cosine similarity is recalculated, ensuring automated quantification of the degree of consistency in the direction of change trends between any two water quality monitoring data sets.

[0047] Specifically, the matching degree between the first trend feature and the second trend feature in the data change rate can be obtained by calculating the difference rate as follows: The data change rate scalar value included in the first trend feature and the data change rate scalar value included in the second trend feature are extracted, and the absolute difference between the two data change rate scalar values ​​is calculated. This absolute difference is compared with the maximum possible rate change range within the preset current monitoring period, and then normalized to a relative difference value between 0 and 1. A subtraction operation is performed on this relative difference, and the output value is converted into a matching metric.

[0048] The matching degree metric is defined as the degree of matching between the first trend feature and the second trend feature in terms of data change rate. Specifically, it outputs 1 when the two scalar values ​​are completely equal, and outputs a zero value when the rate difference reaches the maximum range.

[0049] By updating the preset rate range in real time, the calculation results within each monitoring cycle are guaranteed to dynamically adapt to the actual change range of water quality monitoring data, ensuring the timeliness and feasibility of the quantitative results of the rate matching degree, and supporting subsequent dynamic coupling intensity calculations.

[0050] Specifically, the coordination degree between the first trend feature and the second trend feature in the data fluctuation pattern can be obtained by applying dynamic time normalization to quantify the fluctuation coordination as follows: The data fluctuation pattern sequences of the first trend feature and the second trend feature are extracted through a sliding time window, wherein each data fluctuation pattern sequence is composed of amplitude features of multiple time points.

[0051] The Dynamic Time Warping algorithm is applied to calculate the minimum warped path cumulative distance between two data fluctuation pattern sequences. This minimum warped path cumulative distance reflects the difference in fluctuation patterns between the sequences. The minimum warped path cumulative distance is normalized by dividing it by the preset maximum possible fluctuation distance range. The normalized result is then subtracted from the result to convert it into a coordination value.

[0052] The coordination value is defined as the degree of coordination between the first trend feature and the second trend feature in the data fluctuation pattern, which is expressed as follows: when the two data fluctuation pattern sequences are exactly the same, the dynamic time warping cumulative distance is zero, and the normalized minus operation outputs a coordination value of 1, representing the highest degree of coordination; when the fluctuation pattern difference reaches the preset range boundary, the output coordination value is 0, representing the lowest degree of coordination.

[0053] The dynamic time-warped distance matrix is ​​used to adapt to the matching of non-equal-length sequences, and the maximum fluctuation distance range reference value is updated in real time to ensure that the quantification process of the fluctuation coordination degree can effectively capture the asynchronous fluctuation correlation between water quality monitoring data.

[0054] Step A13: Calculate the output dynamic coupling strength based on the comparison result.

[0055] Based on the comparison results of the generated first and second trend features, including the degree of consistency in the direction of data change, the degree of matching in the data change rate, and the degree of coordination in the data fluctuation pattern, a weighted average method is used to combine the consistency in the direction of data change, the degree of matching in the data change rate, and the degree of coordination in the data fluctuation pattern into a single value. This single value is the calculated output dynamic coupling strength, which aims to comprehensively quantify the degree of mutual influence between the two water quality monitoring data, serve the subsequent automated extraction of implicit transmission paths and strength relationships between the water quality monitoring data, and provide input for network vertex mapping and edge weight generation.

[0056] It should be noted that the above is achieved through a real-time numerical calculation engine, which ensures that the dynamic coupling strength is automatically recalculated after each new data update comparison result, meets the feasibility of streaming data processing, and supports pollution pattern matching and early warning capabilities.

[0057] Step A20: When the dynamic coupling strength exceeds the preset coupling strength, a directional edge is generated on the network vertices mapped to the corresponding two target water quality monitoring data until multiple water quality monitoring data are traversed.

[0058] It should be noted that the preset coupling strength is a pre-set coupling strength used to screen water quality monitoring data association pairs with practical conductive significance.

[0059] The dynamic coupling strength calculated in real time is used to quantify the changing trends of any two water quality monitoring data. Only when the dynamic coupling strength exceeds the preset coupling strength is it determined that there is a significant implicit conduction relationship between the two target water quality monitoring data. At this time, based on the dominant influence direction when generating the first trend feature and the second trend feature, for example, through the direction vector slope sign or the order of rate change, a directional edge is generated between the network vertices mapped by the target water quality monitoring data entities, pointing from the dominant change data vertex to the affected data vertex. In this way, a directed topological network map that can express the pollution conduction path is automatically constructed. In the specific implementation, the preset coupling strength can be configured through historical data analysis or expert experience, and the logic of generating directional edges is embedded in the real-time stream processing process to ensure technical feasibility, ultimately serving the pollution conduction pattern recognition and early warning.

[0060] Step A30: output a topological network graph based on the generated multiple directional edges and the network vertices corresponding to the edges.

[0061] By traversing all pairs of water quality monitoring data, performing change trend correlation analysis on each pair of water quality monitoring data and calculating the dynamic coupling strength, whenever the dynamic coupling strength of any pair of target water quality monitoring data exceeds the preset coupling strength, a directional edge is generated between the two network vertices mapped thereto. After the traversal is completed, all directed edges of vertex pairs that meet the association strength conditions are integrated into an edge set, and all monitored water quality monitoring data entities are automatically mapped into a vertex set.

[0062] At this time, the vertex set and edge set are combined to construct a complete topological network graph structure, where each vertex represents a specific water quality monitoring data entity, such as the ammonia nitrogen concentration at a monitoring point, and each directional edge represents a conduction relationship that exceeds the preset coupling strength, such as the unidirectional impact of industrial wastewater discharge leading to a decrease in dissolved oxygen downstream.

[0063] Please refer to Figure 3 As shown in the figure, by traversing all pairs of water quality monitoring data, such as "industrial wastewater discharge" and "dissolved oxygen at downstream point B", "ammonia nitrogen concentration at monitoring point A" and "guide change", etc., a change trend correlation analysis is performed on each pair of water quality monitoring data and the dynamic coupling strength is calculated. Whenever the dynamic coupling strength of any pair of target water quality monitoring data exceeds the preset coupling strength, a directional edge is generated between the two network vertices mapped thereto. For example, if the correlation strength between industrial wastewater discharge and dissolved oxygen at downstream point B exceeds the standard, a pointing arrow (i.e., a directional edge) is generated between industrial wastewater discharge and dissolved oxygen at downstream point B.

[0064] After the traversal is completed, all vertex pairs and directed edges that meet the association strength conditions are integrated into an edge set, including Figure 3All solid lines with arrows in the graph, such as "Guided Change" → "Ammonia Nitrogen Monitoring", "Dissolved Oxygen at Downstream Point B" → "pH Value", "Dissolved Oxygen at Downstream Point B" → "Monitoring Point B", "Dissolved Oxygen at Downstream Point B" → "Dissolved Oxygen", "Monitoring Point C" → "pH Value", and all monitored water quality monitoring data entities are automatically mapped into vertex sets, namely "Industrial Wastewater Discharge", "Dissolved Oxygen at Downstream Point B", "Ammonia Nitrogen Monitoring", "pH Value", "Guided Change", "Dissolved Oxygen", "Monitoring Point B", "Monitoring Point C", and "Ammonia Nitrogen Concentration at Monitoring Point A" water quality monitoring data entities.

[0065] At this time, the vertex set and edge set are combined to construct a complete topological network graph structure, where each vertex represents a specific water quality monitoring data entity, such as the vertex "ammonia nitrogen concentration at monitoring point A" represents the ammonia nitrogen concentration index at monitoring point A, and each directional edge represents a conductive relationship that exceeds the preset coupling strength, such as "dissolved oxygen at downstream point B" → "industrial wastewater discharge", which clearly indicates that industrial wastewater discharge has caused the dissolved oxygen index at downstream point B to decrease).

[0066] It should be noted that Figure 3 The arrow-less connecting lines in the figure, such as the straight line from "Ammonia nitrogen concentration at monitoring point A" to "Monitoring point B", are essentially static binding relationships between water quality monitoring data entities and their attribute labels, and only indicate the subordinate relationship between water quality monitoring data entities and their exclusive indicators.

[0067] The topological network map is stored and rendered in real time through a graph database or graph computing engine to form a visual high-dimensional network structure. Its edge weights are automatically updated by the dynamic coupling strength, ultimately achieving automated extraction of implicit conduction paths between water quality parameters and providing dynamic topological feature support for pollution diffusion pattern matching and early warning.

[0068] In one possible embodiment, referring to Figure 4 As shown, step S20 includes steps B10 to B40: Step B10, traverse the topological network graph and analyze it to obtain multiple key substructures.

[0069] By systematically scanning and traversing the constructed topological network map, key substructures with specific pollution conduction significance are identified and separated.

[0070] The key substructures mainly include three categories: the first category is the unidirectional conduction cluster, such as the vertex sequence pointing from the ammonia nitrogen concentration at the industrial wastewater monitoring point to the downstream dissolved oxygen, which reveals the directional diffusion path of pollutants; the second category is the high-coupling intensity subnetwork, such as the bidirectional strong coupling vertex group formed by the chlorophyll a and temperature monitoring points, which characterizes the cooperative oscillation mechanism between parameters; the third category is topological anomaly features, such as the isolated vertices or missing edges formed by the broken ammonia nitrogen-dissolved oxygen correlation edge, indicating an event of pollution conduction obstruction.

[0071] By driving the extraction of key substructures through dynamic coupling strength values, key patterns such as the chlorophyll a-temperature closed-loop oscillation network formed in the early stage of algal blooms and the downstream dissolved oxygen collapse chain caused by industrial pollution are automatically output, providing high-dimensional spatiotemporal characteristics for pollution pattern matching and significantly improving the early warning capability for hidden pollution.

[0072] In a first feasible implementation, step B10 is implemented through steps B11 and B12: Step B11, based on the edge weight gradient mutation traversal detection topology network map, extract the path whose gradient change exceeds the preset gradient change amount as the candidate path.

[0073] First, the time series data of the edge weight of each edge in the topological network map is recorded. Each time the network is updated with new water quality monitoring data, the topological network map is traversed to calculate the instantaneous change gradient of all edge weights. The specific algorithm is: the difference between two consecutive calculated dynamic coupling strength values ​​is calculated, and then divided by the previous dynamic coupling strength value to obtain the normalized gradient value.

[0074] When a sudden change in the gradient value of consecutive edge weights on any path is detected—that is, when the normalized gradient value of an edge and the normalized gradient values ​​of its adjacent edges simultaneously exceed the normal fluctuation range within a short period of time—the gradient change of all adjacent edge weights on the path is immediately calculated, defined as the absolute difference between the current normalized gradient value and the historical average normalized gradient value. If the gradient change of three or more consecutive edges in a path exceeds the preset gradient change, the path is marked as a candidate path, thereby automatically capturing chain weight surges, such as in the ammonia nitrogen-dissolved oxygen transmission path downstream of an industrial pollution source.

[0075] At the same time, dynamic coupling strength continuous exceeding threshold polling is performed in parallel: the dynamic coupling strength value of the edges in each subnet of the topological network graph is monitored. When it is detected that the dynamic coupling strength of more than 70% of the edges in a subnet is greater than the preset coupling strength for three consecutive calculation cycles, the subnet is directly marked as a high coupling strength candidate substructure.

[0076] At the same time, a special scan of topological anomaly features is implemented: a cliff drop detection is performed on the dynamic coupling strength of each edge in the topological network map. When the current dynamic coupling strength value is ≤0.2 and the gradient change is <-0.7, the associated edge break event is marked. At the same time, the vertex association degree is counted, and isolated vertices are captured when the vertex association degree returns to zero for two consecutive cycles.

[0077] Step B12: Aggregate adjacent nodes of each candidate path to generate multiple key substructures that characterize pollution conduction characteristics.

[0078] First, based on the edge weight gradient mutation detection, the candidate paths are extracted, all their nodes are traversed and the adjacent node clustering operation is performed: if any two nodes in the candidate path meet the preset topological continuity conditions, such as the spatial coordinate distance is lower than the threshold, the hydrological flow direction is connected or the time window is synchronized, they will be aggregated into the same substructure.

[0079] The specific clustering method uses a breadth-first search algorithm, starting with the starting node of the candidate path as the root node. Based on the direction of the gradient change in the dynamic coupling strength between nodes, the search expands to adjacent nodes until the gradient change reaches a boundary node where the gradient change is lower than the preset gradient change. Each local network unit generated by the aggregation constitutes a key substructure that characterizes the conduction characteristics of a specific pollution. For example, a unidirectional cluster of ammonia nitrogen conduction clusters pointing to a dissolved oxygen collapse zone indicates an industrial pollution event, and a closed loop of chlorophyll a-temperature synchronous oscillations indicates the early stages of an algal bloom.

[0080] At the same time, the high coupling intensity candidate substructures marked by continuous super-threshold polling of dynamic coupling intensity include the chlorophyll a-temperature synchronous oscillation closed loop to characterize the early characteristics of algal blooms. In addition, the key substructures generated by the topological anomaly features constituted by isolated vertices or missing edges captured by special scanning of topological anomaly features indicate pollution conduction obstruction events.

[0081] The key substructures finally generated provide spatiotemporal conduction information for pollution pattern matching, significantly enhancing the ability to identify hidden pollution paths in the early stage.

[0082] In a second feasible implementation, step B10 can also be implemented through steps B13 and B14: Step B13: traverse and analyze the inter-node association paths of the topological network graph based on the timing conduction characteristics, and identify conduction paths with time delay characteristics as candidate paths.

[0083] First, the historical monitoring data time series of the starting node and the target node connected by each edge in the topological network graph are extracted. The fluctuation patterns of the two sequences are aligned through the dynamic time warping algorithm and the optimal warping path is calculated. The time difference between the fluctuation peak of the starting node data and the response peak of the target node is identified in the warping path and determined as the dynamic delay value. At the same time, the covariance of the two sequences in the fluctuation pattern is analyzed to eliminate the interference of random fluctuations.

[0084] When it is detected that the dynamic delay values ​​of consecutive edges on a path are all greater than the preset minimum dynamic delay threshold, and the covariance values ​​of all adjacent node pairs in the path exceed the preset covariance threshold, the path is determined to be a conduction path with time delay characteristics, that is, a candidate path.

[0085] For example, after a surge in the ammonia nitrogen concentration at an industrial wastewater monitoring point, the dissolved oxygen monitoring point downstream shows a downward trend after a fixed delay time. The above algorithm is used to identify the path from the industrial wastewater monitoring point to the downstream dissolved oxygen monitoring point that meets the delay condition. When traversing the topological network graph, a breadth-first search combined with delay constraints is used to automatically aggregate node pairs with continuous delay conduction.

[0086] The candidate paths finally extracted must meet two core characteristics: all edge dynamic delay values ​​are positive, and the overall covariance value of the path reflects stable conduction directionality. For example, the covariance value of the unidirectional conduction cluster is continuously higher than the threshold, which enables automatic capture of abnormal delayed conduction caused by concealed pipe drainage or unknown delay patterns of new pollutant migration, providing hidden conduction pattern characteristics for early warning.

[0087] Among them, the preset minimum dynamic delay threshold is the time window boundary value set based on the physical law of pollutant migration; the preset covariance threshold is the statistical critical value of the stable correlation of the quantitative fluctuation pattern.

[0088] In step B14, contamination conduction clusters are generated by backtracing adjacent nodes of the candidate path and output as multiple key substructures.

[0089] Based on the candidate paths identified by the preset minimum dynamic delay threshold and the preset covariance threshold, a backtracking algorithm is executed from the path end point of the candidate path, as described below: First, detect the direct upstream node B of node C (i.e., the end point of the candidate path) (based on the directional edge pointing to node C in the topological network graph) to verify whether the conduction relationship between node B and node C meets the contaminated conduction characteristics, such as the dynamic delay value exceeds the threshold and the covariance is greater than 0.65.

[0090] If the conditions are met, the system continues to trace back to node A, the upstream node of node B, and verifies the effectiveness of the transmission from node A to node B. During the tracing back process, adjacent nodes are aggregated based on preset spatial adjacency constraints, such as hydrological flow topological paths or geographic coordinate proximity. If all nodes on the path meet the continuity of the transmission characteristics, the tracing back path (node ​​A→B→C) and spatially adjacent homogeneous nodes are merged into a pollution transmission cluster.

[0091] For example, when generating key substructures for industrial pollution transmission clusters, node A (ammonia nitrogen at the discharge outlet) is the cluster core, while nodes B, C, and D form the collapsed zone cluster members. The pollution impact intensity value is calculated by averaging the dynamic coupling strength between nodes within the cluster. Each pollution transmission cluster is a key substructure representing a unified pollution event. This process uses graph database path tracing APIs, such as Neo4j's apoc.path.expandConfig, to perform dynamic clustering in real time. The final output is multiple key substructures that carry spatial extent, pollution intensity, and transmission direction, providing a refined pollution pattern fingerprint.

[0092] Step B20: Dynamically match each key substructure with a preset pollution pattern feature library to obtain a correlation matching threshold.

[0093] First, predefined standardized pollution pattern feature templates are extracted from the preset pollution pattern feature library. For example, the algal bloom outbreak pattern is a bidirectional strongly coupled closed-loop structure composed of the chlorophyll a vertex and the temperature vertex, and the industrial pollution pattern is a chain structure with the ammonia nitrogen vertex pointing to the dissolved oxygen collapse cluster through a unidirectional edge. For each key substructure to be matched, GraphSAGE (Graph Sample and Aggregated, a graph neural network model for graph node embedding learning) is used to convert its topological structure into a vectorized representation, including topological structure feature encoding such as vertex type, edge weight mean, and conduction direction distribution. At the same time, temporal conduction characteristics such as the dynamic coupling intensity gradient change of the key substructure are injected.

[0094] The cosine similarity calculation is performed on this vectorized representation and the vectors of all templates in the preset pollution pattern feature library, and the dynamic correlation value of the current key substructure to each pollution pattern is output, with a value range of -1~1.

[0095] Assume that the dynamic correlation matching threshold is 0.8, which is generated based on the statistical calibration of the matching rate between historical pollution events and pattern library templates. When the dynamic correlation value between the key substructure and a pollution template exceeds this threshold, such as the chlorophyll a-temperature closed-loop similarity of 0.9 matching the algal bloom outbreak pattern, it is determined to be successfully associated with the corresponding pollution pattern and triggers an early warning.

[0096] Step B30: obtaining a target key substructure whose correlation matching threshold is lower than a preset correlation matching threshold, and generating a new abnormal pattern based on the target key substructure.

[0097] After completing the dynamic correlation matching between each key substructure and the preset pollution pattern feature library, all target key substructures whose correlation matching threshold is lower than the preset correlation matching threshold, such as a historical calibration value of 0.8, are screened.

[0098] For each target key substructure, the topology analysis engine automatically extracts its core conduction features, including vertex combination types (such as the coupling of abnormal biological toxicity monitoring points and heavy metal vertices), edge weight distribution features (such as intermittent pulse oscillations in dynamic coupling strength), and spatiotemporal conduction patterns (such as reverse conduction timing in non-hydrological flow directions). Based on the extracted conduction features, new abnormal patterns are dynamically generated outside the preset pollution pattern feature library. The specific method is as follows: The topological skeleton of the target key substructure (retaining its directional edge and vertex association relationship) is abstracted into a universal pattern template, and its conduction dynamic characteristics (such as pulse oscillation parameters and non-reverse conduction timing characteristics) are injected. At the same time, the spatiotemporal conditions when the pattern is first triggered (such as heavy rain in the chemical plant catchment area) are recorded. The newly generated abnormal patterns are verified by spatial neighborhood and analyzed for temporal stability, and finally stored incrementally in the preset pollution pattern feature library in the form of encoded vectors, realizing automated expansion of pollution identification capabilities.

[0099] Step B40: Aggregate the newly added abnormal patterns and the fixed abnormal patterns in the preset pollution pattern feature library, and output an abnormal feature coding set.

[0100] First, all fixed anomaly patterns in the pre-defined pollution pattern feature library, such as the defined industrial pollution unidirectional conduction cluster template and the algal bloom bidirectional closed loop template, are merged into a unified set with new anomaly patterns dynamically generated through the new anomaly pattern process, such as the detected dissolved oxygen-pH atypical conduction ring pulse oscillation template. Multi-layer feature abstraction encoding is performed on each pattern in this unified set: For example, the first layer extracts its topological skeleton features, including vertex type combinations that can represent the distribution of vertex categories such as ammonia nitrogen, dissolved oxygen or heavy metals, directional edge connection relationships that characterize the ratio of unidirectional / bidirectional conduction, and substructure scale values ​​that represent the number of vertices and edge density.

[0101] The second layer injects dynamic behavior characteristics, including the dynamic coupling strength gradient change, such as the weight mutation slope of the collapsed cluster, the timing conduction delay statistics representing the mean and variance of the path delay value, and the fluctuation pattern covariance value.

[0102] The third layer is based on GraphSAGE, which converts the aforementioned feature combinations into fixed-dimensional vectorized anomaly feature encodings.

[0103] This eliminates redundant dimensions and ultimately generates a lightweight abnormal feature coding set, each of which corresponds to a unique pollution conduction pattern fingerprint. This coding set is synchronously updated to the real-time detection engine, providing a scalable pattern matching benchmark library for streaming water quality monitoring data, ensuring the early warning system's ability to dynamically capture hidden pollution and new pollutant characteristics.

[0104] In one possible implementation, refer to Figure 5 As shown, step S30 includes steps C10 to C20: Step C10: Correct the confidence weight of the abnormal feature coding set, spatially fuse the corrected abnormal feature coding set with the pollution diffusion prediction result, and generate a risk distribution map.

[0105] First, based on the verification results of the spatial coverage and temporal reproducibility of the abnormal pattern, a linear weighted correction is performed on the confidence weight of each code in the abnormal feature coding set. For example, the weight of a new abnormal pattern is increased by 0.1 every time the number of times it recurs within three days increases, and the weight of a fixed abnormal pattern is increased by 0.05 every time the coverage area of ​​a fixed abnormal pattern expands by one square kilometer.

[0106] The revised anomaly feature coding set carries the confidence weight vector, time-series conduction dynamic characteristics and topological structure fingerprint, and is input into the spatial fusion engine together with the pollution diffusion prediction results calculated by the hydrological diffusion model. The spatial fusion engine uses the Kriging spatial interpolation algorithm to align the physical migration grid of the pollution diffusion prediction results with the spatial projection points of the anomaly feature coding set, and then superimposes the anomaly feature intensity value with the confidence weight as the adjustment coefficient. The fusion calculation formula is the weighted product of the pollution prediction concentration value and the anomaly feature intensity value of each geographic coordinate point, and finally generates a risk distribution map that integrates the physical diffusion law and statistical anomaly characteristics.

[0107] It should be noted that the pollution diffusion prediction results may include the spatial distribution of pollutant concentration, migration direction and probability of retention hotspots; the physical migration grid is the isosurface of ammonia nitrogen diffusion concentration; the spatial projection points of the anomaly feature code set can be the geographic coordinates of dissolved oxygen collapse clusters, etc.

[0108] The risk distribution map is rendered in real time using vector tile technology, dynamically marking areas with enhanced hidden pollution characteristics, and achieving refined visual early warning of pollution risks in the temporal and spatial dimensions.

[0109] Optionally, step C10 may include steps C11 to C13: In step C11, based on the spatial location correlation, the modified abnormal feature coding set is dynamically weighted fused with the pollution diffusion prediction result to output the spatial risk value distribution.

[0110] Firstly, a geospatial index network of water quality monitoring stations is established, and the spatial coordinates of the modified anomaly feature coding set are mapped with the preset coupling strength of the spatial grid of the pollution diffusion prediction results.

[0111] The correlation calculation uses the inverse distance weight algorithm to assign weights to each spatial grid node within the preset spatial influence radius.

[0112] In the dynamic weight fusion stage, based on the modified confidence weight of the abnormal feature coding set and the basic weight of the pollution diffusion prediction result, a weighted superposition operation is performed according to the spatial correlation intensity to output a vector-type spatial risk value distribution covering the entire domain.

[0113] Among them, the spatial coordinates can be the center of the dissolved oxygen collapse cluster; the spatial grid nodes of the pollution diffusion prediction results can be the concentration points every 100 meters output by the hydrological diffusion model; the spatial grid node weighting can be that the closer the distance is to the center of the collapse cluster within 500 meters, the higher the weight of the grid node.

[0114] In addition, the specific formula for performing weighted overlay operations based on spatial correlation intensity is: It should be noted that R is the spatial risk value, C is the predicted concentration of pollution diffusion, is the basic weight, B is the abnormal feature strength, W is the modified confidence weight, and a is the position correlation coefficient, where the position correlation coefficient is obtained by linearly attenuating the distance from the grid node to the abnormal feature core point. The fusion process is continuously and dynamically updated, and the abnormal feature strength and pollution diffusion prediction value are synchronously adjusted when each round of water quality monitoring data is refreshed.

[0115] Step C12: The spatialized risk value distribution is passed through a risk gradient generator to construct a continuous spatialized risk value distribution.

[0116] The discrete spatialized risk value distribution is input into the risk gradient generator, which first converts the discrete risk points into a continuous field based on the Kriging spatial interpolation algorithm. Specifically, the risk value of each unsampled point is calculated as the weighted average of the risk values ​​of the neighboring monitoring points. The weight is dynamically determined by the spatial covariance function and the distance between the monitoring points. Then, the gradient smoothing algorithm is applied to eliminate the gradient jump noise caused by the uneven density of monitoring points and enhance the diffusion continuity along the hydrological flow path.

[0117] The risk gradient generator simultaneously injects spatial constraints to ensure that the gradient field forms a continuous risk conduction gradient along the river flow direction, and finally outputs a continuous spatial risk value distribution. This distribution provides a dynamic spatial risk feature map with centimeter-level accuracy for the identification of migration paths of hidden pollutants and emergency decision-making.

[0118] It should be noted that the gradient smoothing algorithm in this embodiment is anisotropic diffusion filtering; the spatial constraints are geographic spatial boundary rules such as river network topology, terrain elevation obstacles, etc.; the continuous risk transmission gradient is the decreasing gradient downstream of the industrial pollution source and the circular gradient in the reservoir stagnation area.

[0119] Step C13: extract the risk level boundary based on the spatial variation characteristics of the continuous spatialized risk value distribution to obtain a risk distribution map.

[0120] Firstly, the risk gradient field of the continuous spatial risk value distribution is calculated through the Sobel convolution kernel, and the spatial location where the risk value suddenly changes is identified. Based on the gradient extreme points, the candidate boundary lines of the preset risk level threshold are delineated. At the same time, the geographical constraints are integrated to correct the boundary topology.

[0121] Based on the preset risk level threshold, the continuous field is segmented into isosurfaces to generate risk level boundaries that conform to the natural diffusion pattern. Finally, the risk level boundaries and risk gradient mutation lines are superimposed on the basic geographic base map. The vector-raster fusion technology is used to output a risk distribution map with centimeter-level accuracy, providing a dynamic spatial decision-making basis for hidden pollution interception and emergency resource allocation.

[0122] Among them, the spatial location may be the edge of a steep drop in risk value downstream of an industrial pollution source; the candidate boundary line may be the boundary of an area where the risk value corresponding to the high risk level threshold is greater than 0.8; the preset risk level threshold may be divided according to the degree of loss caused by historical pollution incidents, for example, low risk is 0~0.3, medium risk is 0.3~0.6, and high risk is 0.6~1.

[0123] Step C20: outputting a water quality monitoring warning signal according to the risk gradient mutation boundary of the risk distribution map.

[0124] In this embodiment, the continuous risk gradient field obtained by processing the risk distribution map with the gradient smoothing algorithm is analyzed in real time, and a spatial gradient operator is applied to detect the risk gradient mutation boundary position where the risk value changes sharply.

[0125] When the intensity of a detected risk gradient mutation boundary exceeds a preset mutation threshold, the system automatically extracts the geographic coordinates covering that boundary and associates them with the topological network vertices of the water quality monitoring entities mapped within that spatial range. The risk difference between the inside and outside of the risk gradient mutation boundary and the pollutant types identified by the associated anomaly feature coding set are used to generate a water quality monitoring warning signal that includes spatial positioning signals, pollution feature signals, and timeliness signals.

[0126] Among them, the spatial positioning signal expands to form a circular warning area with the geographical coordinates of the risk gradient mutation boundary as the center; the pollution characteristic signal marks the dominant pollution parameters and risk levels; and the timeliness signal predicts the arrival time of pollution diffusion based on the boundary gradient change rate.

[0127] Optionally, step C20 may include steps C21 to C23: Step C21, analyzing the risk gradient mutation boundary of the risk distribution map to obtain boundary characteristic attributes and risk conduction attributes.

[0128] A continuous risk gradient field is calculated using a spatial gradient operator to detect the location of the risk gradient mutation boundary. The spatial geometric characteristics of this boundary are extracted as boundary characteristic attributes, including boundary morphological type, gradient intensity distribution extreme points, total boundary length, and mean curvature. Risk conduction attributes are also calculated based on the angle between the boundary direction and the hydrological flow direction. Specifically, the conduction direction, conduction stability, and the dominant pollution type determined by the associated anomaly feature code set are extracted.

[0129] Boundary characteristic attributes quantify the characteristics of spatial anomaly patterns, and risk conduction attributes characterize the dynamic behavior patterns of pollution diffusion. Both provide dual decision-making inputs of spatial structure and conduction dynamics for the subsequent generation of water quality monitoring and early warning signals.

[0130] Among them, the boundary morphology type can be tongue-shaped protrusion or ring-shaped closure; the conduction direction can be downstream unidirectional or reverse diffusion; the conduction stability is the standard deviation of the boundary gradient intensity; the dominant pollution type can be set to heavy metal or dissolved oxygen collapse type.

[0131] Step C22: Mapping the boundary characteristic attributes and risk conduction attributes into pollution behavior semantics through behavior semantic mapping.

[0132] Specifically, by calling the preset pollution behavior semantic rule library, the boundary morphology type in the boundary feature attribute and the conduction direction in the risk conduction attribute are matched to the preset rule items. For example, tongue-shaped advance + downstream one-way mapping is "industrial pollution downstream advance behavior". At the same time, the conduction stability and the dominant pollution type are integrated to perform compound condition judgment. For example, high stability + dissolved oxygen collapse type is mapped to "biochemical oxygen-consuming pollution retention behavior".

[0133] This mapping process converts geometric features and dynamic conduction patterns into understandable natural language descriptions through a real-time semantic matching engine, outputting pollution behavior semantics such as "heavy metal reverse infiltration behavior," "algal bloom annular diffusion behavior," and "industrial pollution downstream sudden behavior." These pollution behavior semantics are directly linked to the early warning instruction generation module to achieve interpretable expression of the pollution situation.

[0134] Step C23: After generating an early warning instruction based on the pollution behavior semantic matching response strategy, a water quality monitoring early warning signal is output according to the early warning instruction.

[0135] The pollution behavior semantics output by behavioral semantic mapping are matched with the associated response rules in the preset response strategy library. For example, sudden industrial pollution behavior triggers a three-level interception response, generating an early warning instruction set including target space positioning instructions, pollution feature control instructions, and timeliness instructions.

[0136] The coordinate range of the circular warning area is determined based on the spatial positioning instructions of the early warning instructions. The list of dominant pollution parameters and risk level labels are loaded based on the pollution characteristic control instructions. The diffusion time prediction model is driven by the timeliness instructions to output the pollution arrival time.

[0137] Finally, the above elements are integrated to generate a standardized water quality monitoring early warning signal, which is pushed to the supervision terminal in real time through the Internet of Things protocol and simultaneously triggers the automatic sampling equipment to perform targeted monitoring, realizing a closed-loop transformation from pollution behavior semantics to executable early warning signals.

[0138] This application also provides a water quality automatic monitoring system, referring to Figure 6 As shown, the automatic water quality monitoring system includes: The data coupling module 10 is used to calculate the dynamic coupling strength between the multiple water quality monitoring data collected, select the target water quality monitoring data whose dynamic coupling strength exceeds the preset coupling strength, generate directional edges on the network vertices mapped by the target water quality monitoring data, and output a topological network map; The data matching module 20 is used to match the topological network map with the preset pollution pattern feature library, determine the newly added abnormal pattern, and output the abnormal feature code set; The data decision module 30 is used to obtain the pollution diffusion prediction result, fuse the pollution diffusion prediction result and the abnormal feature coding set, and output the water quality monitoring early warning signal.

[0139] Optionally, the data coupling module 10 is further configured to: Conduct cross-dimensional correlation analysis on the changing trends of any two water quality monitoring data and calculate the dynamic coupling strength between the two water quality monitoring data; When the dynamic coupling strength exceeds the preset coupling strength, a directional edge is generated on the network vertices mapped by the corresponding two target water quality monitoring data until multiple water quality monitoring data are traversed; Based on the generated multiple directional edges and the network vertices corresponding to the edges, a topological network graph is output.

[0140] Optionally, the data coupling module 10 is further configured to: Identify the changing trends of any two water quality monitoring data and obtain the first trend feature and the second trend feature; Obtaining the degree of consistency between the first trend feature and the second trend feature in the direction of data change, the degree of matching in the data change rate, and the degree of coordination in the data fluctuation pattern, and generating a comparison result; The output dynamic coupling strength is calculated based on the comparison results.

[0141] Optionally, the data matching module 20 is further configured to: Traverse the topological network graph and parse it to obtain multiple key substructures; Dynamically match each key substructure with the preset pollution pattern feature library to obtain a correlation matching threshold; Obtain a target key substructure whose correlation matching threshold is lower than a preset correlation matching threshold, and generate a new abnormal pattern based on the target key substructure; Aggregate the newly added abnormal patterns and the fixed abnormal patterns in the preset pollution pattern feature library, and output the abnormal feature coding set.

[0142] Optionally, the data matching module 20 is further configured to: Based on the edge weight gradient mutation traversal detection topology network map, the path whose gradient change exceeds the preset gradient change amount is extracted as the candidate path; Adjacent nodes of each candidate path are aggregated to generate multiple key substructures that characterize the pollution conduction characteristics.

[0143] Optionally, the data matching module 20 is further configured to: Based on the temporal conduction characteristics, the node-to-node association paths of the topological network graph are traversed and analyzed, and the conduction paths with time delay characteristics are identified as candidate paths; By backtracing the adjacent nodes of the candidate path, contamination conduction clusters are generated and output as multiple key substructures.

[0144] Optionally, the data decision module 30 is configured to: Correct the confidence weight of the abnormal feature coding set, spatially fuse the corrected abnormal feature coding set with the pollution diffusion prediction results to generate a risk distribution map; According to the risk gradient mutation boundary of the risk distribution map, water quality monitoring early warning signals are output.

[0145] Optionally, the data decision module 30 is further configured to: According to the spatial location correlation, the modified abnormal feature coding set is dynamically weighted and fused with the pollution diffusion prediction results to output the spatial risk value distribution; The spatialized risk value distribution is passed through the risk gradient generator to construct a continuous spatialized risk value distribution; According to the spatial variation characteristics of the continuous spatial risk value distribution, the risk level boundary is extracted to obtain the risk distribution map.

[0146] Optionally, the data decision module 30 is further configured to: Analyze the risk gradient mutation boundary of the risk distribution map to obtain boundary characteristic attributes and risk transmission attributes; Through behavioral semantic mapping, the boundary characteristic attributes and risk conduction attributes are mapped into pollution behavior semantics; Based on the pollution behavior semantic matching response strategy, after generating the warning instruction, the water quality monitoring warning signal is output according to the warning instruction.

[0147] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for automatic water quality monitoring, characterized in that: include: By calculating the dynamic coupling strength between the two pairs of the collected multiple water quality monitoring data, the target water quality monitoring data whose dynamic coupling strength exceeds the preset coupling strength is selected, and directional edges are generated on the network vertices mapped by the target water quality monitoring data, and a topological network map is output; Matching the topological network map with a preset pollution pattern feature library, determining a new abnormal pattern, and outputting an abnormal feature coding set; Obtain a pollution diffusion prediction result, fuse the pollution diffusion prediction result with the abnormal feature code set, and output a water quality monitoring early warning signal.

2. The automatic water quality monitoring method according to claim 1, characterized in that: The steps of calculating the dynamic coupling strength between each of the plurality of water quality monitoring data collected, screening target water quality monitoring data whose dynamic coupling strength exceeds a preset coupling strength, generating directional edges on the network vertices mapped by the target water quality monitoring data, and outputting a topological network map include: Conduct cross-dimensional correlation analysis on the changing trends of any two water quality monitoring data and calculate the dynamic coupling strength between the two water quality monitoring data; When the dynamic coupling strength exceeds the preset coupling strength, generating directional edges on the network vertices mapped by the corresponding two target water quality monitoring data respectively until the plurality of water quality monitoring data are traversed; The topological network graph is output based on the generated multiple directional edges and the network vertices corresponding to the edges.

3. The automatic water quality monitoring method according to claim 2, characterized in that: The step of performing cross-dimensional correlation analysis on the change trends of any two water quality monitoring data and calculating the dynamic coupling strength between the two water quality monitoring data includes: Identifying the change trends of the arbitrary two water quality monitoring data to obtain a first trend feature and a second trend feature; Obtaining the degree of consistency between the first trend feature and the second trend feature in the direction of data change, the degree of matching in the data change rate, and the degree of coordination in the data fluctuation pattern, and generating a comparison result; The dynamic coupling strength is calculated and outputted according to the comparison result.

4. The automatic water quality monitoring method according to claim 1, characterized in that: The steps of matching the topological network map with a preset pollution pattern feature library, determining a newly added abnormal pattern, and outputting an abnormal feature code set include: Traversing and parsing the topological network graph to obtain multiple key substructures; Performing dynamic correlation matching on each of the key substructures with the preset pollution pattern feature library to obtain a correlation matching threshold; Acquire the target key substructure whose correlation matching threshold is lower than the preset correlation matching threshold, and generate the newly added abnormal pattern based on the target key substructure; Aggregate the newly added abnormal pattern and the fixed abnormal pattern in the preset pollution pattern feature library, and output the abnormal feature code set.

5. The automatic water quality monitoring method according to claim 4, characterized in that: The step of traversing the topological network graph and parsing it to obtain multiple key substructures includes: The topological network graph is detected based on edge weight gradient mutation traversal, and paths whose gradient changes exceed a preset gradient change amount are extracted as candidate paths; Adjacent nodes of each candidate path are aggregated to generate a plurality of key substructures representing pollution conduction characteristics.

6. The automatic water quality monitoring method according to claim 4, characterized in that: The step of traversing the topological network graph and parsing it to obtain multiple key substructures includes: Traversing and analyzing the associated paths between nodes of the topological network graph based on the time sequence conduction characteristics, and identifying conduction paths with time delay characteristics as candidate paths; By backtracking the adjacent nodes of the candidate path, a contaminated conduction cluster is generated and output as a plurality of the key substructures.

7. The automatic water quality monitoring method according to claim 1, characterized in that: The step of fusing the pollution diffusion prediction result and the abnormal feature code set to output a water quality monitoring early warning signal includes: Correcting the confidence weight of the abnormal feature coding set, spatially fusing the corrected abnormal feature coding set with the pollution diffusion prediction result to generate a risk distribution map; The water quality monitoring early warning signal is output according to the risk gradient mutation boundary of the risk distribution map.

8. The automatic water quality monitoring method according to claim 7, characterized in that: The step of spatially fusing the modified abnormal feature code set with the pollution diffusion prediction result to generate a risk distribution map includes: According to the spatial location correlation, the modified abnormal feature coding set is dynamically weighted fused with the pollution diffusion prediction result to output a spatial risk value distribution; Passing the spatialized risk value distribution through a risk gradient generator to construct a continuous spatialized risk value distribution; According to the spatial variation characteristics of the continuous spatialized risk value distribution, the risk level boundary is extracted to obtain the risk distribution map.

9. The automatic water quality monitoring method according to claim 7, characterized in that: The step of outputting the water quality monitoring early warning signal according to the risk gradient mutation boundary of the risk distribution map includes: Analyzing the risk gradient mutation boundary of the risk distribution map to obtain boundary characteristic attributes and risk conduction attributes; Mapping the boundary characteristic attributes and the risk conduction attributes into pollution behavior semantics through behavior semantic mapping; Based on the pollution behavior semantic matching response strategy, after generating an early warning instruction, a water quality monitoring early warning signal is output according to the early warning instruction.

10. A water quality automatic monitoring system, characterized in that: include: A data coupling module is used to calculate the dynamic coupling strength between the multiple water quality monitoring data collected, select the target water quality monitoring data whose dynamic coupling strength exceeds the preset coupling strength, generate directional edges on the network vertices mapped by the target water quality monitoring data, and output a topological network map; A data matching module is used to match the topological network map with a preset pollution pattern feature library, determine the newly added abnormal pattern, and output an abnormal feature code set; The data decision module is used to obtain the pollution diffusion prediction result, fuse the pollution diffusion prediction result and the abnormal feature coding set, and output a water quality monitoring early warning signal.

Images