Surface water pollution zoning prevention dynamic management system and method

Abstract

The invention relates to the crossing field of environmental information technology and water pollution control, and particularly discloses a surface water pollution zoning prevention and control dynamic management system and a surface water pollution zoning prevention and control dynamic management method. The system comprises a multi-source water quality sensing network, a hydrological topology modeling unit, a fusion prediction and traceability engine, a pollution evidence chain storage platform and a partition prevention and treatment decision terminal. The method comprises the following steps: constructing a hydrological topology directed graph with space-time attributes, fusing computational fluid mechanics physical constraints and a space-time diagram neural network, realizing efficient prediction and source inversion of pollution diffusion, and generating a non-tampering pollution evidence chain based on a block chain; and the partition prevention and treatment decision terminal divides high, medium and low risk areas according to the risk levels, and matches differentiation prevention and treatment instructions. According to the system and the method, precise tracing of second-level response, scientific and explainable dynamic simulation and law enforcement support with legal efficacy can be realized, and the intelligence and refinement level of water pollution prevention and control is improved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of environmental information technology and water pollution control, and specifically relates to a dynamic management system and method for the prevention and control of surface water pollution by zone. Background Technology

[0002] With increasingly stringent requirements for ecological and environmental protection, the zoned prevention and dynamic management of surface water pollution has become a core task for maintaining water ecological security and achieving precise governance. Modern water environment monitoring systems rely on sensor networks and numerical simulation technology to achieve real-time monitoring and risk warning of pollution status in rivers and lakes. When dealing with sudden pollution events under complex hydraulic conditions, the management system needs to possess efficient diffusion prediction capabilities and accurate pollution source tracing mechanisms, which places high demands on the effectiveness of model calculations, prediction accuracy, and the rigor of management logic.

[0003] Simulating water quality evolution and accurately locating pollution sources are crucial for achieving dynamic management. Traditional physical water quality models rely on complex numerical solutions to partial differential equations, resulting in excessively long computation times for large-scale water area simulations, making it difficult to meet the real-time decision-making needs of sudden pollution responses. Conventional deep learning models, while improving computational speed, often neglect the physical constraints of fluid motion, lack feature extraction methods for river topology in non-Euclidean spaces, and suffer from poor interpretability and overfitting issues in prediction results. Existing management systems mostly focus on passive monitoring and alarms, lacking tamper-proof evidence preservation and spatiotemporal logical tracing of the entire pollution diffusion process, leading to difficulties in evidence collection and insufficient evidence for source identification during environmental law enforcement. Therefore, there is a need to develop a dynamic management system and method for the zoned prevention and control of surface water pollution to address these challenges. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic management system for the prevention and control of surface water pollution by zone, which can solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A dynamic management system for zoning prevention and control of surface water pollution includes a multi-source water quality sensing network, a hydrological topology modeling unit, a fusion prediction and source tracing engine, a pollution evidence chain storage platform, and a zoning prevention and control decision-making terminal, wherein: The multi-source water quality sensing network is configured to collect water quality parameters, hydrological flow velocity and flow direction information at multiple monitoring points in surface water bodies in real time, and synchronously transmit the collected data to the hydrological topology modeling unit. The hydrological topology modeling unit constructs a directed graph structure with spatiotemporal attributes based on the spatial connectivity and hydrodynamic characteristics of the river system. Each monitoring node is used as a graph vertex and the water flow path is used as an edge, forming a hydrological topology model that can represent non-Euclidean spatial characteristics. The fusion prediction and source tracing engine integrates computational fluid dynamics solution logic and spatiotemporal graph neural network architecture. It uses the hydrological topology model as input and combines real-time data provided by the multi-source water quality sensing network to perform dynamic prediction of pollution diffusion and inversion calculation of pollution sources. The pollution evidence chain storage platform receives the spatiotemporal trajectory of pollution events, suspected emission locations, and associated water quality anomaly data output by the fusion prediction and tracing engine, generates tamper-proof pollution evidence records, and performs distributed storage. The zoned prevention and control decision terminal generates differentiated prevention and control instructions for different functional zones based on the prediction results of the fusion prediction and source tracing engine and the credible evidence provided by the pollution evidence chain storage platform, and pushes them to relevant law enforcement or operation and maintenance units.

[0006] Preferably, the fusion prediction and tracing engine has a built-in physical constraint module. This built-in physical constraint module applies mass and momentum conservation constraints to the water flow velocity field and pollutant transport process based on the discretization solution logic of the Navier-Stokes equations, ensuring that the prediction results conform to the basic laws of fluid mechanics.

[0007] Furthermore, the spatiotemporal graph neural network in the fusion prediction and tracing engine adopts a gated recursive mechanism and a temporal attention structure, which can dynamically extract the historical water quality sequence of each node in the hydrological topology graph and retain the temporal dependence of pollutant migration along the water flow direction during graph propagation.

[0008] Furthermore, the pollution evidence chain storage platform is based on a blockchain underlying architecture. It packages the source tracing results, original monitoring data hash values, timestamps, and geographical location information of each pollution event into blocks and writes them into a distributed ledger through a consensus mechanism to ensure the integrity, traceability, and non-repudiation of the evidence chain.

[0009] Preferably, the hydrological topology modeling unit supports a dynamic update mechanism. When new monitoring points are added or the river structure changes, the node and edge connection relationships of the hydrological topology map are automatically reconstructed, and the input graph structure of the fusion prediction and source tracing engine is adjusted simultaneously.

[0010] Furthermore, the zoned prevention and control decision terminal is equipped with a risk level assessment module. Based on the predicted pollution concentration distribution, the location of sensitive water bodies, and the risk of population exposure, the risk level assessment module divides the management area into high, medium, and low risk zones, and matches each zone with corresponding emergency response strategies, sampling frequency adjustment plans, and sewage outlet investigation priorities.

[0011] Furthermore, the multi-source water quality sensing network includes fixed water quality buoys, mobile unmanned vessel-borne sensors, and shore-based video monitoring equipment. The three work together to achieve three-dimensional and full-coverage monitoring of key river sections, and the data acquisition frequency can be dynamically adjusted to a predetermined time interval according to the pollution warning level.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. The surface water pollution zoning prevention and control dynamic management system provided by the present invention integrates the physical constraints of computational fluid dynamics with the data-driven capabilities of spatiotemporal graph neural networks. While ensuring the scientific nature of pollution diffusion simulation, it improves computational efficiency, achieves dynamic prediction and accurate source tracing with a response time of seconds, and overcomes the dual defects of long computation time of traditional physical models and lack of interpretability of pure artificial intelligence models.

[0013] 2. The hydrological topology model constructed by the system can accurately characterize the non-Euclidean spatial characteristics of the river network, enabling the graph neural network to have stronger generalization ability and physical consistency when dealing with complex water system structures.

[0014] 3. By introducing blockchain technology to establish an immutable chain of pollution evidence, it provides environmental law enforcement with legally valid spatiotemporal coordinates and data credentials, realizing an upgrade from a passive alarm to a proactive accountability governance model.

[0015] 4. The zoned prevention and control decision-making terminal implements differentiated management and control strategies based on risk levels, which improves the precision and intelligence of water resource protection and provides reliable technical support for the rapid response and long-term prevention and control of sudden water pollution incidents. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the present invention, which integrates computational fluid dynamics physical constraints and spatiotemporal graph neural networks; Figure 3 This is a flowchart illustrating the main stages of surface water pollution monitoring, hydrological topology modeling, source tracing simulation, and evidence preservation decision-making in this invention. Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the multi-source water quality sensing network, the hydrological topology modeling unit, and the pollution evidence chain storage platform in this invention. Figure 5 This is a logical flow framework diagram generated based on risk level assessment and differentiated prevention and control strategies in this invention. Detailed Implementation

[0017] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0018] A dynamic management system for the prevention and control of surface water pollution by zone includes a multi-source water quality sensing network, a hydrological topology modeling unit, a fusion prediction and source tracing engine, a pollution evidence chain storage platform, and a zoned prevention and control decision terminal. The multi-source water quality sensing network is used to acquire in real time various water quality parameters, including dissolved oxygen, ammonia nitrogen, total phosphorus, and chemical oxygen demand, as well as dynamic hydrological data, including flow velocity field distribution, water level height, and cross-sectional geometric characteristics, through three-dimensional monitoring equipment deployed in the target water area. The multi-source water quality sensing network integrates a data cleaning and preprocessing submodule, which is configured to perform analog-to-digital conversion, outlier removal, and spatiotemporal correlation-based missing value interpolation compensation on the raw electrical signals collected by the sensors, ensuring that the data input to subsequent units has a high signal-to-noise ratio and temporal continuity. At the physical deployment level, the multi-source water quality sensing network consists of fixed water quality monitoring stations, adaptive water quality buoys, unmanned surface vessels equipped with integrated sensors, and shore-based hyperspectral visual recognition equipment. Among them, fixed monitoring stations are responsible for long-term high-frequency monitoring of control sections and cross-boundary sections; adaptive water quality buoys are distributed near the confluence of river tributaries and key sewage outlets, and can automatically adjust the sampling depth according to water level fluctuations; unmanned vessels are configured to automatically plan routes to suspected pollution core areas for refined sampling after receiving pollution warning instructions.

[0019] The hydrological topology modeling unit receives real-time data transmitted from the multi-source water quality sensing network and river vector data provided by the geographic information system, and constructs a mathematical topology representing the complex spatial connectivity of the river system. The unit defines each physical monitoring point or river characteristic section as a vertex in a directed graph, and defines the water flow path connecting adjacent vertices as an edge. To accurately depict the physical reality of surface water flow, the unit assigns a weight attribute to each edge when constructing the graph structure. This weight attribute is jointly determined by the average flow velocity of the river segment, the Manning roughness coefficient, the longitudinal gradient of the riverbed, and the hydrodynamic drag coefficient. The unit employs non-Euclidean space modeling technology, enabling it to handle complex river system topologies with tree-like branching, looping, or irregular confluence characteristics. This allows subsequent computational models to overcome the limitations of Euclidean space in handling river bends and confluences.

[0020] The fusion prediction and source tracing engine, serving as the system's computational core, is configured as a hybrid architecture that deeply integrates physical and mechanical constraints with data-driven features. This engine comprises a physical constraint module and a spatiotemporal graph neural network module. The physical constraint module, based on a discretized expression of the Navier-Stokes equations, establishes constraint discriminants for mass and momentum conservation during water movement. Both the physical constraint module and the spatiotemporal graph neural network module require that, within any computational unit, the volume derivative of pollutant concentration must equal the algebraic sum of the diffusion and source terms. During model training and inference, prediction branches that do not conform to physical laws are pruned. The spatiotemporal graph neural network module is configured as a multi-layer architecture, including graph convolutional layers and temporal processing layers. The graph convolutional layers capture the spatial migration characteristics of pollutants between nodes in the hydrological topology graph at each time step, while the temporal processing layer extracts the nonlinear evolution of pollutant concentration over time through a gated loop structure. The fusion prediction and source tracing engine achieves physical consistency simulation of pollution diffusion processes and reverse inference of pollution source locations by using physical constraints as part of the loss function or as a correction factor for the hidden layer state.

[0021] The pollution evidence chain storage platform provides credible temporal and spatial evidence for all key outputs of the system. Based on a distributed ledger architecture, this platform encapsulates the source analysis report, raw water quality monitoring logs, monitoring equipment status parameters, and predicted pollution plume trajectory maps output by the fusion prediction and source tracing engine. Internally, the platform operates a hash operation subunit that transforms the aforementioned multi-dimensional data into a unique digital digest, combines it with the current legal timestamp and geographic location signature, and writes it into an immutable block via a consensus protocol. During the evidence storage process, the platform is also configured to synchronously store feedback information from law enforcement terminals, ensuring the evidence chain covers the entire lifecycle from pollution discovery, simulation prediction, source tracing and liability determination to law enforcement action.

[0022] The zonal prevention and control decision-making terminal is used to automatically generate targeted governance strategies based on prediction results and evidence chains. This terminal integrates a dynamic risk level assessment module, configured to calculate real-time risk assessment values ​​based on pollutant toxicity coefficients, water body functional zone classifications, downstream sensitive point distances, and estimated disaster areas. When the risk assessment value exceeds a preset first threshold, the terminal automatically triggers a high-risk zonal prevention and control strategy, including closing downstream water intakes, dispatching emergency interception pumping stations, and sending on-site inspection instructions to law enforcement agencies for sewage outlets. When the risk assessment value is between the first and second thresholds, a medium-risk prevention and control strategy is triggered, manifested as increased sampling frequency and enhanced operation of purification facilities. The terminal also features an interactive interface capable of displaying real-time visualized heat maps of water pollution distribution and supporting closed-loop monitoring and effect evaluation of the implementation of prevention and control instructions.

[0023] In further implementation details, the data transmission of the multi-source water quality sensing network follows a preset multi-level communication protocol. In normal operation, each monitoring point in the sensing network sends heartbeat packets and core water quality data at a first frequency. Once the fusion prediction and source tracing engine detects that the rate of change of water quality parameters exceeds a preset mutation threshold, the sensing network will be remotely triggered into an emergency mode, at which point the sampling and transmission frequency automatically switches to a higher second frequency. For extreme hydrological environments, the sensing network also integrates a disaster recovery storage mechanism. When the external communication link is interrupted, the data will be temporarily stored in local non-volatile memory, and a breakpoint resumption algorithm will be used to complete data synchronization after the link is restored.

[0024] The hydrological topology modeling unit incorporates an adaptive structure adjustment mechanism during model construction. When river flow interruption due to seasonal low water levels or changes in riverbed geometry due to dredging are detected, this adaptive structure adjustment mechanism can automatically identify the failure of topological connectivity and recalculate the adjacency matrix between nodes. For tidal river sections with complex confluence relationships, the hydrological topology modeling unit also introduces a phase delay factor to correct the dynamic impact of the reversal of water flow direction during high and low tides on the topological edge weights.

[0025] The core logic of the physical constraint module in the fusion prediction and source tracing engine lies in transforming the partial differential equations of classical fluid mechanics into regularization terms of a neural network. This module calculates the vector sum of the momentum change rate, pressure gradient force, gravity term, and viscous dissipation term at each grid point in the flow field, and determines the degree of deviation of the vector sum from the zero vector. If the deviation exceeds the allowable error tolerance, the weight matrix within the neural network is adjusted. During the source tracing and inversion process, the fusion prediction and source tracing engine employs a conjugate equation method combined with a genetic search algorithm, using the minimization of the sum of squared differences between the measured and simulated concentrations at the observation points as the objective function. It backtracks to possible discharge times in the time dimension and searches for the most probable geographical coordinates of the discharge in the spatial dimension.

[0026] Furthermore, the spatiotemporal graph neural network of the fusion prediction and source tracing engine employs a multi-head attention mechanism. This mechanism is configured to compute the correlation between nodes in parallel across multiple feature subspaces. One attention head focuses on capturing rapid transport features at close range and high flow velocities, while another attention head focuses on capturing cumulative effect features at long range and low flow velocities. By weighted fusion of the outputs of multiple attention heads, the model achieves extremely high generalization accuracy when simulating water pollution under different flow velocity conditions.

[0027] The pollution evidence chain storage platform employs a hybrid storage scheme. Large volumes of raw surveillance video and high-frequency sensor waveform data are stored in an encrypted distributed file system, while key metadata such as summaries, source tracing conclusions, and prevention and control decision instructions are stored on the blockchain ledger. This design ensures data traceability and non-repudiation while addressing the storage pressure and synchronization efficiency issues faced by blockchain when processing massive amounts of unstructured data.

[0028] The zoning prevention and control decision-making terminal is also configured as a collaborative management platform, capable of conducting compliance reviews of prevention and control recommendations based on different water pollution prevention and control functional zoning (such as drinking water source protection areas, fishery water use areas, and industrial water use areas). For example, for drinking water source protection areas, the instructions generated by the decision-making terminal will have the highest priority and will automatically be associated with emergency plans to block all unnecessary upstream sewage outlets. This zoning prevention and control decision-making terminal also integrates a reinforcement learning-based strategy optimization submodule. This submodule continuously adjusts the weight parameters in the risk assessment algorithm by analyzing the handling process and final water quality restoration effects of historical pollution events, achieving continuous evolution and optimization of prevention and control strategies.

[0029] In the fusion prediction and source tracing engine, a Lagrange particle tracking operator is integrated to address the computational characteristics of non-Euclidean space. This Lagrange particle tracking operator is configured to simulate the random walk process of pollutant clumps within a graph structure. Unlike traditional grid-based Eulerian methods, this Lagrange particle tracking operator treats pollutants as discrete computational particles, using stochastic differential equations to describe the convection and turbulent diffusion processes of these particles in the water flow. By extracting features from the spatial distribution probability density of these computational particles, the spatiotemporal graph neural network can more meticulously characterize the edge diffusion features of the pollution plume, especially in river bends and shallow areas. This method reduces prediction errors caused by numerical dispersion.

[0030] The regional prevention and control decision-making terminal also considers the coupled influence of meteorological conditions when generating instructions. This terminal obtains rainfall forecast data for the next 48 hours through an interface. If a heavy rainfall event is predicted, the system will automatically correct the surface runoff generation coefficient and adjust the prevention and control instructions based on the corrected coefficient. For example, if it is predicted that heavy rainfall may cause overflow pollution from combined sewer systems, the decision-making terminal will issue instructions in advance to empty the pretreatment pond and activate the stormwater and sewage storage pond, realizing a shift from static prevention and control to dynamic pre-control based on meteorological coupling.

[0031] In the pollution evidence chain storage platform, the consensus mechanism employs an improved proof-of-stake algorithm. The weight of the stake depends not only on the storage contribution of participating nodes but also on the data quality score and operational stability score of the monitoring equipment. This design incentivizes monitoring units to maintain the operational accuracy of their equipment, ensuring the credibility of the data source for the storage platform.

[0032] The hydrological topology modeling unit also possesses three-dimensional spatial expansion capabilities. For deep-water lakes and reservoirs or large rivers with stratified flow characteristics, the modeling unit can expand the two-dimensional planar structure into a multi-layered three-dimensional structure based on the distribution of the thermocline in the vertical direction. In this three-dimensional structure, the vertical edges are assigned buoyancy correction weights based on the Businesk hypothesis, enabling the fusion prediction and source tracing engine to simulate the diffusion and sedimentation processes of pollutants in the water depth direction.

[0033] Example 2: As a supplement and evolution to Example 1, this example provides a dynamic management system for zoning prevention and control of surface water pollution based on edge computing and a hierarchical aggregation architecture. Building upon Example 1, this example focuses on optimizing the system's real-time response and data processing capabilities in large-scale watersheds.

[0034] A dynamic management system for the prevention and control of surface water pollution by zone includes an edge sensing access node cluster, a central computing hub, a distributed topology database, a blockchain logic verification network, and multi-level linkage prevention and control terminals. The edge sensing access node cluster is deployed in various sub-catchments within the watershed, with each edge node configured as an industrial control unit with independent computing capabilities. Each edge node integrates a lightweight water quality anomaly detection model. This model employs a one-dimensional convolutional neural network structure, enabling direct extraction of temporal features from current and voltage fluctuations collected by sensors at the device level, thus completing preliminary pollution event identification before data upload. The edge sensing access node cluster utilizes a dynamic networking protocol, automatically connecting mobile monitoring devices (such as underwater robots and monitoring drones) to the current network topology according to monitoring density requirements, achieving intensive monitoring of local water areas.

[0035] The central computing hub, acting as the system's global brain, is configured as a private cloud platform comprised of high-performance computing servers. The central computing hub runs a high-performance version of the fusion prediction and source tracing engine described in Example 1. Because edge nodes handle the initial data cleaning and local anomaly detection tasks, the central computing hub can concentrate computing resources to perform large-scale, full-basin simulation predictions. The central computing hub employs an asynchronous task scheduling mechanism, enabling it to simultaneously initiate multiple prediction processes, each corresponding to different hydrodynamic assumptions and pollution emission scenarios. By weighted fusion of multiple prediction results, it outputs a probabilistic diffusion prediction map with a confidence interval range.

[0036] The distributed topology database is used to store and manage the complex and ever-changing hydrological topology relationships across the entire watershed. This database employs a storage architecture based on a non-relational graph database, supporting rapid querying and updating of terabyte-level graph node and edge data. By subscribing in real-time to flow monitoring data from the hydrological department and rainfall data from the meteorological department, the database dynamically maintains the weight matrix of each edge in the topology graph. When torrential rains and floods cause river overflows and create new flow paths, the distributed topology database can automatically establish temporary topological edges between affected nodes using logical operators, ensuring the accuracy of the fusion prediction and source tracing engine's simulations under extreme weather conditions.

[0037] The blockchain logic verification network described herein is an enhanced implementation of the evidence storage function. Unlike the single-chain evidence storage in Embodiment 1, this network adopts a master-slave chain parallel architecture. The master chain is used to record the conclusions of major pollution incidents, administrative penalty recommendations, and general instructions for prevention and control decisions; the slave chain is deployed at the edge sensing node level to record high-frequency equipment self-inspection information, hash fingerprints of original monitoring values, and early warning trigger records at the edge. The master and slave chains are periodically anchored in state through a cross-chain relay protocol. This design improves the system's throughput when handling massive high-frequency evidence storage requests, while ensuring strong consistency of core evidence across the entire network.

[0038] The multi-level coordinated prevention and control terminal is configured to support a multi-level collaborative office system at the provincial, municipal, county, and on-site operation levels. Internally, this terminal operates a resource scheduling algorithm based on game theory. In the event of a major water pollution incident spanning multiple administrative regions, this algorithm can automatically calculate the coordinated prevention and control responsibilities and optimal resource allocation plan for each administrative region based on their water resource reserves, emergency material reserves, treatment facility capabilities, and the degree of downstream damage. The terminal also integrates augmented reality-assisted enforcement functions. On-site personnel can scan the polluted water body using handheld devices, and the system can overlay an invisible pollutant cloud map generated by a fusion prediction engine onto the real-world image in real time, guiding enforcement personnel to accurately locate the discharge outlet.

[0039] In the fusion prediction and tracing engine of this embodiment, the physical constraint module is further refined into a momentum term constraint submodule, a scalar transport constraint submodule, and an energy balance constraint submodule. The momentum term constraint submodule is configured to calculate the combined influence of pressure gradient, Coriolis force, and bottom drag on the fluid velocity vector in the flow field, ensuring that the flow field generated by the neural network meets the curl and divergence requirements of fluid dynamics. The scalar transport constraint submodule is specifically designed for non-conservative substances such as chemical oxygen demand and heavy metals, converting the biochemical degradation rate, sedimentation rate, and sediment release rate of the substances into kinetic constraint terms. Through this multi-dimensional physical intervention, the simulation results generated by the system are not only statistically optimized but also rigorous and reliable in terms of physicochemical mechanisms.

[0040] For the hydrological topology modeling unit, this embodiment introduces a self-healing topology mapping technology. When a monitoring node goes offline due to hardware damage, power depletion, or communication failure, this self-healing topology mapping technology can reconstruct the pseudo-observations of the offline node in virtual space based on the spatiotemporal correlation of historical data and using the observations of neighboring nodes through Kriging interpolation or mean estimation methods, thereby maintaining the integrity of the topology map. This self-healing capability ensures that the management system can maintain continuous dynamic monitoring and analysis capabilities even in harsh environments.

[0041] The pollution evidence chain storage platform in this embodiment also has an automatic triggering function based on smart contracts. When the source tracing result output by the fusion prediction and source tracing engine points to a specific pollution discharge permit holder, and the confidence level of the pollution behavior exceeds a preset safety threshold, the storage platform can automatically activate a preset smart contract. This smart contract is configured to automatically lock the remote monitoring data of the pollution discharge outlet of the pollution discharge permit holder, send a pre-storage notification to the regulatory authorities, and automatically call the interface of the relevant environmental law enforcement departments when necessary, achieving seamless connection from technical discovery to legal response.

[0042] In the regional prevention and control decision-making terminal, a sensitivity analysis mechanism is incorporated into the generation process of prevention and control strategies. The system is configured to perform sensitivity calculations on each decision variable (such as the flood discharge of the upstream reservoir, the opening degree of the downstream intercepting sewer, and the dosage of emergency agents). The system automatically selects the most cost-effective decision combination for improving water quality and displays it to managers in the form of a decision tree on the terminal interface. Managers can quickly obtain optimized prevention and control strategy solutions by adjusting weight preferences (such as cost priority, time priority, or environmental protection priority).

[0043] This embodiment also designs a local fine-grained grid overlay technique for small-scale, highly complex water environments such as urban rivers. When the system detects water quality anomalies in certain sensitive small-scale water areas (such as landscape lakes or areas around urban drinking water intake points), the hydrological topology modeling unit automatically inserts encrypted micro-level sub-nodes between the macro-level topology nodes in that sensitive small-scale area, while the central computing hub dynamically allocates more neural computing resources for fine-grained simulation of the local area. This hybrid modeling approach of "macro-topology + local micro-level" achieves a balance between the breadth of basin-wide management and the depth of prevention and control in key areas.

[0044] Example 3: In this example, the system is further configured as an environmental management digital twin system with autonomous learning and evolution capabilities. Based on Examples 1 and 2, this environmental management digital twin system, by introducing a continuous learning mechanism and knowledge graph technology, achieves in-depth analysis and long-term prediction of the evolution patterns of surface water environment.

[0045] A dynamic management system for the prevention and control of surface water pollution by zone includes a digital twin mirror unit, a multi-dimensional knowledge graph library, an evolutionary fusion computing engine, a blockchain integrity verification framework, and a panoramic prevention and control simulation terminal. The digital twin mirror unit is configured to construct a virtual mapping in the cloud that is completely synchronized with the physical river system. This digital twin mirror unit not only includes the river's hydrodynamic parameters but also integrates land use types around the basin, pipeline distribution maps, production processes of key polluting enterprises, and historical meteorological and hydrological big data. Through real-time data streams acquired from the multi-source water quality sensing network, the digital twin mirror unit drives the virtual river model to evolve synchronously, enabling managers to intuitively observe the real-time dynamic migration process of pollutants at different water layers and in different river sections in virtual space.

[0046] The multidimensional knowledge graph database stores domain expert knowledge, historical governance cases, environmental laws and regulations, and the migration and transformation patterns of chemical substances related to water pollution prevention and control. This database employs semantic modeling technology, defining "pollutants," "exceedance limits," "meteorological conditions," "prevention measures," and "final effects" as entities, and establishing a complex knowledge network through semantic associations. When a new pollution event occurs, the system can automatically find the most similar historical cases to the current situation through the knowledge graph's inference engine and provide valuable prevention and control suggestions.

[0047] The evolutionary fusion computing engine, based on the original CFD and GNN fusion, adds an online evolution module based on meta-learning. This online evolution module is configured to automatically adjust the topological connection weights and penalty factors of physical constraint terms of the spatiotemporal graph neural network according to the residual feedback between actual observations and predicted values. This means that during use, the system will automatically evolve the optimal prediction parameters best suited to the current environment based on the water quality evolution characteristics of different watersheds and seasons, achieving adaptive optimization of the model for each river.

[0048] In this embodiment, the blockchain integrity verification framework is configured as a hierarchical data asset management platform. Besides achieving basic immutable evidence storage, this framework also incorporates zero-knowledge proof technology. This enables the system to prove to law enforcement agencies whether a company's pollution emissions exceed legal thresholds without disclosing sensitive privacy information such as its production processes. In this way, the privacy of data producers is protected while ensuring the public credibility of environmental enforcement evidence.

[0049] The panoramic prevention and control simulation terminal employs a 3D visualization engine to integrate complex predictive data, source tracing results, and prevention and control recommendations into a panoramic interactive space. This terminal supports two working modes: "spatiotemporal retrospection" and "scheme simulation." In retrospection mode, managers can view water quality evolution and pollution discharge evidence over any past time period, much like playing back a video. In simulation mode, managers can manually change prevention and control parameters in the model (such as simulating the closure of a sewage outlet or increasing the ecological flow of a section of river). The system will then display in real time the expected effects of these parameter changes on future water quality improvement, thus assisting decision-makers in selecting the optimal prevention and control scheme.

[0050] In this embodiment, the evolutionary fusion computing engine also integrates a source strength uncertainty analysis submodule based on Bayesian inference. Considering that emissions are often difficult to obtain precisely in sudden pollution events, this source strength uncertainty analysis submodule is configured to set the discharge rate as a random variable following a specific distribution. Through multiple iterative calculations, the engine can output the maximum posterior estimate of the pollution source strength and the corresponding confidence interval, providing more scientific and convincing evidence for subsequent pollution liability determination.

[0051] The digital twin mirror unit also includes a virtual simulation operator based on deep reinforcement learning. This operator is configured to automatically simulate various extreme pollution scenarios (such as secondary water pollution from a fire in an upstream chemical industrial park, or a large-scale oil spill at a cross-border section) in a normal, pollution-free environment, and autonomously find the optimal emergency response path. The experiential data generated from these virtual simulations is fed back to a multi-dimensional knowledge graph database, ensuring that the system has a wealth of practical governance solutions in place before a real crisis occurs.

[0052] In the aforementioned zoned prevention and control decision-making terminal, the issuance of prevention and control instructions adopts an encrypted transmission mechanism based on role-based access control. For maintenance personnel and law enforcement personnel with different responsibilities, the terminal automatically filters out irrelevant and redundant information, providing only the operating interface and instructions within their scope of responsibility. For example, grassroots river chiefs receive river patrol tasks and suspected site investigation instructions based on geographical location; while technical experts receive detailed physical constraint simulation reports and multi-source data comparison charts.

[0053] The multi-source water quality sensing network described in Example 3 has been intelligently upgraded. Each monitoring station is equipped with an edge computing-based visual analysis unit, which can identify abnormal water color, foam accumulation, and floating object characteristics around the sewage outlet in real time. These visual features, as unstructured information, are converted into text tags and uploaded along with the numerical information from the sensors, providing richer multimodal input for the fusion prediction and source tracing engine.

[0054] To address the blockchain integrity verification framework, this embodiment also introduces a verifiable random function to randomly select audit nodes from across the network to participate in consensus. This mechanism prevents malicious nodes from colluding to tamper with specific pollution discharge records. The system also establishes a token-based monitoring incentive mechanism. For site owners who consistently maintain high-precision data uploads, the system will automatically allocate corresponding governance credit points, which can serve as a positive reference in subsequent environmental approvals or pollution rights trading.

[0055] In summary, through the system architecture design of multiple embodiments, this invention realizes a closed-loop dynamic management of surface water pollution, from accurate perception, physical consistency simulation, scientific source tracing, tamper-proof evidence storage to intelligent decision-making.

[0056] Those skilled in the art should understand that, although the modules and units in the above embodiments have been specifically divided and described, in actual systems engineering practice, these functional modules can be flexibly recombinated, merged, or further refined according to the configuration of hardware resources, the requirements of computing performance, and the specificity of the application scenario. For example, the hydrological topology modeling unit can be integrated with the fusion prediction and source tracing engine in the same high-performance computing node, or the functions of the pollution evidence chain storage platform can be deployed in edge gateway hardware with a trusted execution environment.

[0057] In the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0058] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0059] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A dynamic management system for zoning prevention and control of surface water pollution, characterized in that, include: A multi-source water quality sensing network is configured with monitoring equipment deployed in three dimensions to collect water quality parameters, hydrological flow velocity, water level and flow direction information of monitoring points in surface water bodies in real time, and transmit the collected data synchronously to the hydrological topology modeling unit. The hydrological topology modeling unit is connected to the multi-source water quality sensing network. Based on the spatial connection relationship and hydrodynamic characteristics of the river system, it constructs a directed graph structure with spatiotemporal attributes, taking each monitoring node as the graph vertex and the water flow path as the edge, forming a hydrological topology model that represents non-Euclidean spatial characteristics. The integrated prediction and source tracing engine communicates with the hydrological topology modeling unit and is configured to use the hydrological topology model as input, combined with real-time collected data, to perform dynamic prediction of pollution diffusion and pollution source inversion calculation by executing physical constraint logic and extracting temporal features. The pollution evidence chain storage platform is connected to the fusion prediction and tracing engine. It is used to receive the spatiotemporal trajectory of pollution events, suspected emission locations and related water quality anomaly data, and generate tamper-proof pollution evidence records through hash operations and consensus mechanisms and perform distributed storage. The zoned prevention and control decision-making terminal is connected to the fusion prediction and tracing engine and the pollution evidence chain storage platform, respectively, and is used to generate differentiated prevention and control instructions for different functional zones based on the prediction results and the pollution evidence records.

2. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, The multi-source water quality sensing network includes a data cleaning and preprocessing submodule, a fixed water quality monitoring station, an adaptive water quality buoy, an unmanned shipborne sensor, and a shore-based hyperspectral visual recognition device. The data cleaning and preprocessing submodule is configured to perform analog-to-digital conversion, outlier removal, and interpolation compensation for missing values ​​based on spatiotemporal correlation on the raw electrical signals collected by the sensors. The fixed water quality monitoring stations are deployed at control sections and cross-boundary sections for high-frequency monitoring; The adaptive water quality buoy is deployed at the confluence of river tributaries and at sewage outlets, and is configured to automatically adjust the sampling depth according to water level fluctuations. The unmanned vessel-borne sensor is configured to move to the suspected core pollution area for mobile monitoring according to an automatically planned route after receiving a pollution warning command; The shore-based hyperspectral visual recognition device is configured to identify abnormal water color, foam accumulation, and floating object characteristics around the sewage outlet in real time, and convert the visual features into text labels as unstructured input information.

3. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, When constructing the directed graph structure, the hydrological topology modeling unit is configured to assign a weight attribute to each edge. The weight attribute is determined by the average flow velocity, Manning roughness coefficient, longitudinal slope of the riverbed, and hydrodynamic drag coefficient of the corresponding river segment. The hydrological topology modeling unit also includes an adaptive structural adjustment module and a three-dimensional space expansion module; The adaptive structure adjustment module is configured to automatically identify failed nodes and recalculate the adjacency matrix between nodes when the river channel is interrupted or the riverbed geometry changes, causing the topological connection relationship to fail. The three-dimensional spatial expansion module is configured to expand the two-dimensional planar structure into a multi-layer three-dimensional structure based on the distribution of thermoclines in the vertical direction for water bodies with stratified flow characteristics, and to assign buoyancy correction weights based on the Businsk assumption on the vertical edges to simulate the diffusion and sedimentation of pollutants in the water depth direction.

4. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, The fusion prediction and tracing engine includes a physical constraint module, which has a built-in discriminant based on the discretization solution logic of the Navier-Stokes equations, configured to establish constraints on the mass conservation and momentum conservation during the water movement process. The physical constraint module requires that, within any computational unit, the volume derivative of the pollutant concentration must be equal to the algebraic sum of the diffusion term and the source term. It also calculates the vector sum of the momentum change rate, pressure gradient force, gravity term, and viscous dissipation term at each grid point in the flow field, and determines whether the deviation of the vector sum from the zero vector is within a preset error tolerance. In this way, prediction branches that do not conform to physical laws are pruned. During the source tracing and inversion process, the physical constraint module adopts the conjugate equation method combined with the genetic search algorithm, with the objective function being to minimize the sum of squares of the difference between the measured concentration and the simulated concentration at the observation point. It traces back the time of pollution discharge in the time dimension and searches for the geographical coordinates of the discharge in the spatial dimension.

5. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, The fusion prediction and tracing engine includes a spatiotemporal graph neural network module, which is configured as a multi-layer architecture, including graph convolutional layers and temporal processing layers. The graph convolutional layer is used to capture the spatial migration characteristics of pollutants between nodes of the hydrological topology graph at each time step. The time-series processing layer extracts the nonlinear evolution of pollutant concentration over time through a gated loop structure and a multi-head attention mechanism. The multi-head attention mechanism is configured to compute the correlation between nodes in parallel within multiple feature subspaces, wherein the first attention head focuses on capturing fast transport features at close range and high flow rates, and the second attention head focuses on capturing cumulative effect features at long range and low flow rates. The fusion prediction and tracing engine also integrates a Lagrange particle tracking operator, which is configured to treat pollutants as discrete computational particles, use stochastic differential equations to describe the convection and turbulent diffusion process of particles in water flow, and extract features from the spatial distribution probability density of computational particles to characterize the edge diffusion features of the pollution plume.

6. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, The pollution evidence chain storage platform is based on a distributed ledger architecture, including a hash operation subunit and a hybrid storage submodule; The hash operation subunit is used to convert the source analysis report, original water quality monitoring log, monitoring equipment status parameters and pollution plume trajectory map into a unique digital digest, and generate a block by combining the current legal timestamp and geographical location signature; The hybrid storage submodule is configured to store large-capacity raw surveillance video and high-frequency sensor raw waveform data in an encrypted distributed file system, and store the summary information, tracing conclusions and prevention and control decision instructions of the raw waveform data on the blockchain ledger. The pollution evidence chain storage platform adopts a master-slave chain parallel architecture. The master chain is used to record the judgment conclusions of major pollution events and the general instructions for prevention and control decisions. The slave chain is deployed at the edge sensing node level to record equipment self-inspection information and edge early warning trigger records. The master and slave chains are periodically anchored in state through a cross-chain relay protocol.

7. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, The zoned prevention and control decision-making terminal is equipped with a risk level assessment module, a resource scheduling sub-module, and a strategy optimization sub-module. The risk level assessment module is configured to calculate the risk assessment value based on the pollutant toxicity coefficient, water body functional zone classification, downstream sensitive point distance, and estimated disaster area. When the risk assessment value exceeds the preset first threshold, a high-risk zone prevention and control strategy is automatically triggered, including closing the downstream water intake, dispatching emergency interception pumping stations, and sending on-site inspection instructions to law enforcement agencies. When the risk assessment value is between the first threshold and the preset second threshold, a medium-risk prevention and control strategy is triggered, including increasing the sampling frequency and activating the enhanced operation mode of the purification facilities. The resource scheduling submodule adopts a game theory-based resource scheduling algorithm, which is configured to calculate the collaborative prevention and control responsibility and resource allocation plan based on the water resource reserves, emergency material reserves and treatment facility capabilities of each administrative region when a cross-administrative region pollution incident occurs. The strategy optimization submodule uses reinforcement learning algorithms to analyze the effects of historical actions and dynamically adjusts the weight parameters in the risk assessment algorithm.

8. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, The data transmission of the multi-source water quality sensing network follows a multi-level communication protocol. Under normal operating conditions, each monitoring point sends core water quality data at the first frequency. When the fusion prediction and tracing engine detects that the rate of change of water quality parameters exceeds the preset mutation threshold, the multi-source water quality sensing network is controlled to enter emergency mode and switches the sampling and transmission frequency to a second frequency higher than the first frequency. The multi-source water quality sensing network also integrates a disaster recovery storage mechanism, which is configured to temporarily store data in local non-volatile memory when the external communication link is interrupted, and use the breakpoint resume algorithm to complete data synchronization after the link is restored. Each monitoring station in the multi-source water quality sensing network is equipped with an edge computing unit. The edge computing unit has a built-in lightweight water quality anomaly detection model, which is configured to complete the preliminary identification of pollution events before data is uploaded, and supports dynamic networking protocols to access mobile monitoring devices.

9. The dynamic management system for zoning prevention and control of surface water pollution according to claim 1, characterized in that, The system also includes a digital twin mirror unit and a multidimensional knowledge graph library; The digital twin mirror unit is configured to build a virtual mapping synchronized with the physical river system in the cloud, integrating river hydrodynamic parameters, land use types around the basin, pipeline distribution map and production process flow of key polluting enterprises, and using real-time data from the multi-source water quality sensing network to drive the synchronous evolution of the virtual river model. The multidimensional knowledge graph database uses semantic modeling technology to define pollutants, exceedance multiples, meteorological conditions, and prevention and control measures as entities and establish semantic associations. When a pollution incident occurs, the system uses the inference engine of the knowledge graph to find historical cases similar to the current situation and provide prevention and control suggestions.

10. A dynamic management method for zoned prevention and control of surface water pollution, characterized in that, The dynamic management system for zoning prevention and control of surface water pollution, as described in any one of claims 1-9, is used to achieve dynamic management of zoning prevention and control of pollution.

Images