Water ecological pollution diffusion prediction method and system based on big data

Abstract

The invention relates to the technical field of environment monitoring, and discloses a water ecological pollution diffusion prediction method and system based on big data, and the method comprises the steps: constructing a sensing network of a water environment, obtaining real-time water environment data, carrying out the preprocessing of the data through an edge calculation node, and synchronously transmitting the data to a central platform; based on historical data and real-time water body environment data, establishing a dynamic prediction model of pollutant boundary condition parameters; according to the dynamic boundary condition parameters, constructing a pollutant propagation atlas, analyzing the spatial and temporal distribution trend of pollutant concentration and identifying the position of a pollution source; and according to a pollution source position identification result, executing aging prediction, and triggering a pollution early warning response in combination with a set threshold value. The space coverage rate and timeliness of pollution monitoring are improved, the intelligence and credibility of pollution tracking are enhanced, and the intelligent and scientific level of water ecological environment pollution prevention and control is improved.

Description

Technical Field

[0001] The present invention relates to the field of environmental monitoring technology, and in particular to a method and system for predicting the spread of water ecological pollution based on big data. Background Art

[0002] The diffusion of pollutants in rivers, lakes and other water bodies has obvious temporal and spatial evolution characteristics, and is affected by multiple factors such as flow velocity, water level, rainfall, and emissions, showing complex characteristics of strong nonlinearity and dynamic changes. Therefore, how to accurately and timely predict the diffusion trend of pollutants, identify potential pollution sources and carry out effective early warning has become a key technical requirement in water ecological environment protection.

[0003] In the existing technology, pollution diffusion models based on fluid mechanics principles (such as two-dimensional hydrodynamic models, finite element method simulations, etc.) are often used to analyze the migration and diffusion process of pollutants in water bodies. Although such methods have high theoretical accuracy, they are sensitive to initial boundary conditions, have complex modeling, heavy computational burdens, and are difficult to respond to dynamically changing hydrological conditions in real time. At the same time, pollution source identification often relies on expert experience or fixed rules, and lacks the ability to dynamically invert with big data.

[0004] In addition, traditional water quality monitoring mainly relies on a small number of fixed points, with insufficient temporal and spatial coverage, making it difficult to fully perceive the overall picture of pollution spread. With the development of remote sensing technology, IoT sensors and edge computing, it provides a basis for large-scale, high-frequency, and multi-dimensional data collection and processing, promoting the transformation of pollution spread prediction from "qualitative monitoring" to "intelligent analysis" and "early warning response".

[0005] However, there is still a lack of an intelligent method and system that can integrate multi-source water environment data, dynamically construct boundary conditions, identify pollution sources, predict pollution diffusion paths, and automatically trigger warnings. Therefore, it is urgent to build a water ecological pollution diffusion prediction system driven by big data with real-time processing capabilities and pollution warning functions to improve pollution prevention and control efficiency and ecological safety response capabilities. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problems solved by the present invention are: the existing water pollution diffusion prediction has a strong reliance on boundary conditions, inaccurate pollution source identification, poor prediction real-time performance and lack of intelligent early warning mechanism.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: A method for predicting the spread of water ecological pollution based on big data, comprising: constructing a sensor network of water environment, acquiring real-time water environment data, pre-processing the data using edge computing nodes, and synchronously transmitting the data to a central platform;

[0009] Based on historical data and real-time water environment data, establish a dynamic prediction model for pollutant boundary condition parameters;

[0010] According to the dynamic boundary condition parameters, construct a pollutant propagation map, analyze the spatio-temporal distribution trend of pollutant concentration and identify the location of the pollution source;

[0011] According to the identification result of the pollution source location, perform time-effect prediction, and trigger a pollution warning response in combination with a set threshold.

[0012] As a preferred solution of the water ecological pollution diffusion prediction method based on big data according to the present invention, wherein: the real-time water environment data includes water quality monitoring data, hydrodynamic related data, meteorological background data, remote sensing and image data, biological and ecological index data, and meta-information data;

[0013] The preprocessing includes performing noise filtering and outlier detection, removing sensor false alarm data, filling missing data using logical rules and model-driven mechanisms, and completing data cleaning operations to achieve standardized and normalized processing of the data; performing format unification and time-space dimension alignment processing on data from different sensors.

[0014] As a preferred solution of the water ecological pollution diffusion prediction method based on big data according to the present invention, wherein: establishing the dynamic prediction model of pollutant boundary conditions includes: setting a time step, using the preprocessed historical data and real-time data of the past n time steps, and organizing them in a sliding window manner to form a water environment data sequence of continuous n time steps as the input of the dynamic prediction model;

[0015] The dynamic prediction model adopts a fusion prediction framework, which is composed of a parallel linear regression model and a deep learning model;

[0016] The linear regression model is to capture the linear change trend of water environment parameters and output a prediction result based on the water environment data sequence of the past n time steps;

[0017] The deep learning model is to use the GRU network structure to capture the non-linear dynamic characteristics of water environment parameters, predict the water environment data sequence, and support online incremental training to adapt to real-time data changes;

[0018] Perform weighted fusion on the prediction outputs of the linear regression model and the deep learning model to generate the final output of pollutant boundary condition parameters;

[0019] The pollutant boundary condition parameters include: water flow rate, flow rate, pollutant concentration and water level height at the boundary.

[0020] As a preferred embodiment of the big data-based water ecological pollution diffusion prediction method of the present invention, the dynamic prediction model further includes that when new real-time water environment data arrives, it inputs through an updated sliding window, and updates the linear regression model coefficients by using the recursive least squares method and updates the deep learning model parameters based on real-time data, so as to continuously optimize the prediction accuracy of the model.

[0021] As a preferred embodiment of the big data-based water ecological pollution diffusion prediction method of the present invention, constructing a pollutant propagation map includes generating a hydrological connectivity map based on dynamic boundary condition parameters, combining water body geographical information and basin structure. The nodes in the hydrological connectivity map represent key water area monitoring points, tributary confluences and drainage outlets, and the edges represent the flow relationships between water bodies.

[0022] The dynamic boundary condition parameters are used to dynamically assign values to the edges in the hydrological connectivity map to determine the water body flow direction, flow velocity and flow rate attributes. The pollutant concentration attributes of the nodes are also jointly determined by the dynamic boundary condition parameters and real-time collected data.

[0023] At a predetermined time step, based on the updated dynamic boundary condition parameters, a pollutant propagation map is generated, and a dynamic map sequence is continuously formed.

[0024] Perform spatio-temporal aggregation analysis on the dynamic map sequence, extract the change trajectories of pollutant concentrations at each node, judge the abnormal growth and diffusion direction of pollutant concentrations in the local area, and form a pollution diffusion path.

[0025] As a preferred embodiment of the big data-based water ecological pollution diffusion prediction method of the present invention, identifying the pollution source location includes: using the graph tracing algorithm to backtrack the pollution diffusion path, identifying the upstream node combination with the highest path overlap frequency as the pollution source candidate area, and making a judgment in combination with the temporal sequence information of the concentration peaks at each node and the water flow direction.

[0026] For the candidate pollution source area, retrieve the corresponding remote sensing image data for auxiliary verification, and compare the known sewage discharge facility information in the area in combination with the pre-stored environmental knowledge graph. Calculate the credibility of the pollution source identification result based on path continuity, temporal sequence rationality and hydrodynamic consistency, and finally confirm the pollution source location.

[0027] As a preferred embodiment of the big data-based water ecological pollution diffusion prediction method of the present invention, performing time effect prediction includes: using the geographical location of the pollution source to obtain the pollutant emission intensity and historical diffusion behavior.

[0028] Predict the expected arrival time, peak pollutant concentration, and duration of pollutant diffusion based on the pollution diffusion path; compare the peak pollutant concentration predicted for each node in the hydrological connectivity map with the preset safety threshold in real time to determine whether there is a risk of exceeding the peak pollutant concentration in the next time step.

[0029] When the peak pollutant concentration exceeds the preset threshold in the preset future time step, automatically generate a warning message; the warning message includes the pollution diffusion path, the expected time of exceeding the standard, the predicted peak concentration, and the corresponding risk level.

[0030] A water ecological pollution diffusion prediction system based on big data, wherein:

[0031] A data module that constructs a sensing network for the water environment, obtains real-time water environment data, performs preprocessing using edge computing nodes, and synchronously transmits it to the central platform.

[0032] A pollutant boundary condition parameter module that establishes a dynamic prediction model for pollutant boundary condition parameters based on historical data and real-time water environment data.

[0033] A pollution diffusion simulation module that constructs a pollutant propagation map based on the dynamic boundary condition parameters, analyzes the temporal and spatial distribution trends of pollutant concentrations, and identifies the location of the pollution source.

[0034] An early warning module that performs pollution diffusion path simulation and time effect prediction based on the pollution source identification result, and triggers a pollution warning response in combination with the set threshold.

[0035] A computer device includes: a memory and a processor; the memory stores a computer program, and is characterized in that: when the processor executes the computer program, it implements the steps of the method described in any one of the present inventions.

[0036] A computer-readable storage medium stores a computer program, and is characterized in that: when the computer program is executed by a processor, it implements the steps of the method described in any one of the present inventions.

[0037] Advantages of the present invention: The method for predicting the diffusion of water ecological pollution based on big data provided by the present invention comprehensively acquires multi-source data such as water quality, hydrodynamic, meteorological, and remote sensing data by constructing a sensing network for the water environment, improving the spatial coverage and timeliness of pollution monitoring; preprocessing and standardizing the data through edge computing nodes, reducing the central computing pressure and enhancing the real-time response ability of the system; establishing a dynamic boundary condition prediction model by fusing historical data and real-time data, solving the problem of unstable boundary conditions in pollution diffusion simulation, and improving the accuracy and robustness of prediction; constructing a pollutant propagation map and combining it with a graph tracing algorithm for pollution source identification, enhancing the intelligence and credibility of pollution tracking; through pollution diffusion path simulation and timeliness prediction, and combining with a threshold-triggered early warning mechanism, realizing active early warning and rapid response to pollution risks, and overall improving the intelligent and scientific level of water ecological environmental pollution prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0039] Figure 1 It is the overall flowchart of a method for predicting the diffusion of water ecological pollution based on big data provided by the first embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0041] Example 1, referring to Figure 1 , which is an embodiment of the present invention, provides a method for predicting the diffusion of water ecological pollution based on big data, including:

[0042] S1: Construct a sensing network for the water environment, obtain real-time water environment data, and after preprocessing the data using edge computing nodes, synchronously transmit it to the central platform.

[0043] Deploy fixed sensors in key areas (including but not limited to river inlets, industrial discharge points, and downstream outlets) and necessary areas of the water area to be monitored. The fixed sensors are used to continuously monitor water quality parameters (such as pollutant concentration, pH value, dissolved oxygen, temperature, etc.) and fluid parameters (such as flow velocity and flow rate).

[0044] In cooperation with static sensors, deploy mobile sensing nodes in monitoring blind spots or boundary change areas, including unmanned boats and / or drones, which are configured with corresponding water quality, environment, and image acquisition devices to supplement the real-time monitoring data of fixed sensors.

[0045] Through the Internet of Things communication network, the original data collected by the above-mentioned fixed and mobile sensors is transmitted to the proximal edge computing node in real time.

[0046] Real-time water environment data includes water quality monitoring data, hydrodynamic-related data, meteorological background data, remote sensing and image data, biological and ecological index data, and meta-information data.

[0047] Water quality monitoring data includes pollutant concentration values (including but not limited to chemical oxygen demand, ammonia nitrogen, total phosphorus, total nitrogen, heavy metals, organic pollutants, etc.), pH value, conductivity, dissolved oxygen, turbidity, water temperature, and oxidation-reduction potential.

[0048] Hydrodynamic-related data includes flow velocity, flow direction, water level, water depth, cross-sectional flow rate, rainfall runoff, and tidal data.

[0049] Meteorological background data includes wind speed, wind direction, air temperature, humidity, air pressure, precipitation, and sunshine intensity, etc.

[0050] Remote sensing and image data includes surface remote sensing images, water color change maps, surface pollution zone maps, thermal infrared images, suspended matter images, etc. collected by satellites or drones.

[0051] Biological and ecological index data includes the number of plankton, chlorophyll a concentration in water, algae distribution, and cyanobacteria density, etc.

[0052] Meta-information data includes information such as the time stamp of monitoring data, acquisition device identification, longitude and latitude coordinates, sampling depth, and cross-section number.

[0053] The preprocessing includes performing noise filtering and outlier detection, eliminating sensor false alarm data, filling in missing data using logical rules and model-driven mechanisms, and completing data cleaning operations to achieve the standardized and normalized processing of data. Perform format unification and time-space dimension alignment processing on data from different sensors.

[0054] Furthermore, through the collaborative deployment of fixed and mobile sensing nodes, the dynamic perception ability for key areas and boundary change areas is enhanced. With the help of edge computing nodes, the raw data is preprocessed, including noise filtering, anomaly detection, missing value filling, and format standardization, to ensure the real-time, integrity, and consistency of the data, providing a high-quality and structured data foundation for subsequent pollution diffusion prediction and pollution source identification, thereby improving the real-time response ability and analysis accuracy of the entire system.

[0055] S2: Based on historical data and real-time water environment data, establish a dynamic prediction model for pollutant boundary condition parameters.

[0056] When the system starts, it initializes a fixed-length data buffer to store the monitoring data of the most recent n time steps. The data buffer can be implemented using a first-in-first-out queue structure, and its length is equal to the time step n. The data at each time point includes multiple preprocessed feature variables, such as environmental parameters like water flow rate, water flow, pollutant concentration, water level height, water temperature, pH value, dissolved oxygen, etc.

[0057] The system extracts the standardized monitoring data of the most recent n time points from the historical database and fills them into the buffer in chronological order to form a complete time series window. This window serves as the first set of inputs for the model to generate an initial pollutant boundary prediction result.

[0058] As the system continues to run, whenever new real-time monitoring data is received, the system automatically appends this data to the end of the buffer and removes the earliest piece of data at the head of the buffer, thereby implementing the update operation of the sliding window. Each time the window slides, it always retains the n latest time step data to ensure that the model input maintains the latest time series state.

[0059] The update of the sliding window can be triggered based on a time period (such as every 5 minutes or every 10 minutes) or event-driven (such as detecting a sudden change in pollutant concentration). Whenever the window update is completed, it is pushed as a new input sequence to the dynamic prediction model, including a parallel linear regression sub-model and a deep learning sub-model.

[0060] The input sequence is organized in a two-dimensional data structure, where the first dimension is the time step and the second dimension is multiple feature variables at each time point. This sequence is directly fed into the model for prediction without being split into single-point data, effectively retaining the context characteristics of the time series and enhancing the model's learning ability for the spatio-temporal change trend of pollutants.

[0061] The sliding window is continuously updated according to the data acquisition rhythm, and the model prediction results will be fed back to the pollution diffusion simulation module. Meanwhile, the system supports error tracking and model incremental learning mechanisms. When there is a deviation between the sliding window input and the actual observation, the system can dynamically adjust the model parameters or trigger an online retraining operation to further enhance the system's response ability to sudden pollution events.

[0062] Set the time step, and use the preprocessed historical data and real-time data of the past n time steps. Through the sliding window method, organize and form a water environment data sequence of continuous n time steps as the input of the dynamic prediction model.

[0063] The dynamic prediction model adopts a fusion prediction framework, which consists of a parallel linear regression model and a deep learning model.

[0064] The linear regression model is to capture the linear change trend of water environment parameters and output prediction results based on the water environment data sequence of the past n time steps.

[0065] The deep learning model is to use the GRU network structure to capture the non-linear dynamic characteristics of water environment parameters, predict the water environment data sequence, and support online incremental training to adapt to real-time data changes.

[0066] Weightedly fuse the prediction outputs of the linear regression model and the deep learning model to generate the final output of the pollutant boundary condition parameters.

[0067] The output of the linear model is The output of the GRU model is The fused output result is:

[0068]

[0069] Among them, represents the predicted boundary water flow rate, represents the predicted boundary flow rate, represents the predicted boundary pollutant concentration, represents the predicted boundary water level height.

[0070] Among them, the fusion weight α t can adaptively change and is defined as follows:

[0071]

[0072] Among them, Δ t represents the absolute value of the difference between the prediction results of the two models at time t, reflecting the deviation degree of the model output. β represents the sensitivity control factor of the fusion weight.

[0073] The pollutant boundary condition parameters include: the water flow rate, flow rate, pollutant concentration, and water level height at the boundary.

[0074] When new real-time water environment data arrives, by updating the sliding window input and using the recursive least squares method to update the linear regression model coefficients and incrementally train and update the deep learning model parameters based on the real-time data, the continuous optimization of the model prediction accuracy is achieved.

[0075] Furthermore, by constructing a dynamic prediction model that fuses historical and real-time water environment data, high-precision and real-time updates of pollutant boundary conditions (such as water flow rate, flow rate, concentration, water level, etc.) are achieved. The sliding window mechanism is used to ensure that the model always runs based on the latest time-series data. Combining the linear regression model to capture the stable trends of environmental parameters and the deep learning model (GRU) to identify complex non-linear change characteristics, and the prediction results of the two are fused through adaptive weights. This design significantly enhances the system's perception and prediction ability of pollution diffusion input conditions, provides stable and accurate data support for subsequent pollution propagation simulation and early warning, and through the error feedback and incremental update mechanism, realizes the continuous self-adaptation and intelligent evolution of the model under sudden environmental changes.

[0076] S3: Based on the dynamic boundary condition parameters, construct a pollutant propagation map, analyze the spatio-temporal distribution trend of pollutant concentration, and identify the location of the pollution source.

[0077] Constructing the pollutant propagation map includes: using the dynamic boundary condition parameters as the driving core, combining the water body geographic information and the basin structure to generate a hydrological connectivity map. The nodes in the hydrological connectivity map represent the monitoring points of each key water area, the confluence of tributaries, and the drainage outlets, and the edges represent the flow relationships between water bodies. The directionality, flow rate, and flux of the edges are all assigned and updated in real time by the dynamic boundary condition parameters.

[0078] The pollutant boundary condition parameters output in stage S2 Among them, the predicted pollutant concentration is used as the dynamic attribute of node v i in the map; the predicted water flow rate and flow rate are mapped to the directionality and transmission ability of edge e ij ; the predicted water level height is used to determine the effective flux state of the edge, dynamically control the activity of the edge, and ensure that the map reflects the current real water flow environment.

[0079] The system reconstructs and updates the propagation map every time step based on the output of the latest dynamic boundary condition parameters to form a continuous dynamic map sequence. The real-time collected pollutant concentration data is used as the calibration basis for node attributes to further improve the accuracy of the map reflecting the actual migration state of pollutants.

[0080] The system performs spatio-temporal aggregation analysis on the dynamic graph sequence to extract the migration trajectories of pollutants between nodes and the concentration evolution trend. If the pollutant concentration at a certain node rises significantly in consecutive time steps and the direction of the connected edge flows from the upstream to this node, the system marks it as a potential pollution impact node.

[0081] In the pollution source identification stage, based on the current pollution path, the system calls the graph tracing algorithm, and based on the diffusion path driven by the dynamic boundary condition parameters in the dynamic graph, traces the pollution migration trajectory backward to identify the upstream nodes with a relatively high frequency of path overlap. Combining the concentration peak time, propagation time required, and hydrodynamic direction of the nodes, the pollution propagation path set of candidate node s is screened.

[0082] For the candidate pollution source area, the corresponding remote sensing image data is retrieved for auxiliary verification, and combined with the pre-stored environmental knowledge graph to compare the information of known pollution discharge facilities in the area. The credibility of the pollution source identification result is calculated based on path continuity, temporal sequence rationality, and hydrodynamic consistency, and finally the location of the pollution source is confirmed.

[0083] After the preliminary path tracing is completed, the credibility scoring formula is defined as follows:

[0084]

[0085] Among them, represents the set of pollution propagation paths passing through candidate node s. γ p represents the maximum pollutant concentration on path p. T p represents the time required for the pollutant to propagate from the start point to the end point of the path. δ(p, s) ∈ {0, 1} represents whether the path passes through candidate node s. If so, it is 1; otherwise, it is 0.

[0086] The credibility R s greater than 0.8 indicates that this node is a real pollution source. The system combines this score with the frequency of path overlap in the graph tracing path, the concentration temporal sequence characteristics, and the knowledge graph matching result as an important reference basis for calculating the credibility of the pollution source.

[0087] Furthermore, by constructing a pollutant propagation graph and combining the graph tracing algorithm, the quantitative modeling and visual tracking of the spatio-temporal propagation path of pollutants are realized. On this basis, a credibility scoring mechanism based on pollution intensity, propagation timeliness, and path structure is introduced, making the pollution source identification process more scientific, systematic, and quantitative. By fusing and judging the scoring results with remote sensing images, knowledge graphs, and node concentration temporal information, the accuracy and reliability of pollution source location are effectively improved, avoiding misjudgment or omission that may be caused by a single tracing method, and providing a highly credible data basis for subsequent pollution diffusion prediction and early warning.

[0088] S4: According to the identification result of the pollution source location, perform timeliness prediction and trigger a pollution warning response in combination with the set threshold.

[0089] After completing the confirmation of the pollution source location and the construction of the propagation path, the system performs timeliness prediction on the diffusion process of pollutants in the water body based on the dynamic atlas evolution result driven by the dynamic boundary condition parameters. Specifically, it includes: using the location of the pollution source node s, the pollutant emission intensity, and the corresponding historical propagation path set Combined with the attributes of each edge in the path (including water flow rate flow water level height ) and the pollutant concentration at the node to deduce the arrival time of pollutants at each downstream node and the concentration evolution trend. The above parameters all come from the fusion prediction output and the atlas time series update process of the dynamic boundary condition parameters in the previous stage, ensuring that the pollutant diffusion simulation is consistent with the overall prediction logic. The system calculates the predicted peak value of the pollutant concentration at each downstream node i in the hydrological connectivity graph at the future time step t + ΔT Subsequently, compare this predicted value with the multi-level safety thresholds θ 1 2 3

[0090] For the predicted peak value of the pollutant concentration at the future time step t + ΔT of node i Its risk level formula is expressed as:

[0091]

[0092] Among them, θ 1 , θ 2 , θ 3 represent the multi-level safety thresholds corresponding to the pollutant type. L i represents the warning level (0 = safe, 1 = light risk, 2 = medium risk, 3 = high risk), which is used for multi-channel alarm linkage.

[0093] When L i = 0, the system does not trigger an alarm, only records data for model learning and posterior verification, and updates the spatio-temporal evolution trend of pollutants.

[0094] When L i = 1, the system enters the monitoring and warning state, marks the yellow warning area on the platform interface. Synchronously push a prompt message to the manager's App, and start the local frequency enhancement monitoring of the edge node.

[0095] When L i ​​​​When L = 2, the system triggers an orange warning, sends a text message / voice reminder, and notifies the relevant environmental protection supervision departments. At the same time, it is recommended to conduct manual sampling verification in the downstream area of the pollution path.

[0096] When L i When L = 3, the system activates a red high-priority alarm, links to on-site alarm devices (such as warning lights and broadcasts); retrieves the emergency response suggestion library and pushes it to the decision-making system. At the same time, mark the high-risk pollution propagation path and continuously refresh the prediction window.

[0097] Furthermore, by comparing the future pollution concentrations at each key node with multiple-level safety thresholds, hierarchical warning triggering and response linkage are realized. This mechanism can not only identify in advance the risk areas where exceeding the standard may occur, but also automatically match corresponding response strategies according to the differences in risk levels (such as platform prompts, text message notifications, on-site alarms, emergency plan push, etc.), so as to achieve early detection, early warning, and early intervention of pollution incidents, and effectively improve the system's dynamic perception ability and intelligent response ability to sudden water pollution.

[0098] Embodiment 2 is an embodiment of the present invention, which provides a water ecological pollution diffusion prediction system based on big data, including:

[0099] A data module that constructs a sensing network for the water environment, obtains real-time water environment data, preprocesses it using edge computing nodes, and synchronously transmits it to the central platform.

[0100] A pollutant boundary condition parameter module that establishes a dynamic prediction model for pollutant boundary condition parameters based on historical data and real-time water environment data.

[0101] A pollution diffusion simulation module that constructs a pollutant propagation map according to the dynamic boundary condition parameters, analyzes the temporal and spatial distribution trends of pollutant concentrations, and identifies the source location of pollutants.

[0102] A warning module that, according to the source identification result, performs pollution diffusion path simulation and time effect prediction, and triggers a pollution warning response in combination with the set threshold.

[0103] Embodiment 3, an embodiment of the present invention, is different from the previous two embodiments in that:

[0104] If the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0105] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a predefined sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0106] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), optical fiber devices, and portable compact disc read-only memories (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or processing it in other suitable ways as necessary, and then storing it in a computer memory.

[0107] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0108] Embodiment 4, an embodiment of the present invention, provides a method and system for predicting the diffusion of water ecological pollution based on big data. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0109] In this embodiment, the A river basin in the suburbs of a certain city is used as the test area. The total length of the river course is about 23.5 kilometers, with 4 water inlets, 2 industrial discharge points and 1 drinking water source protection area distributed. In order to verify the accuracy and response ability of the method for predicting the diffusion of water ecological pollution based on big data proposed by the present invention, a complete experimental system is constructed and deployed at key sections and potential pollution high-incidence areas in this basin.

[0110] First, 12 fixed sensing nodes are arranged along the river, including conventional water quality sensors (pH, conductivity, dissolved oxygen, COD, ammonia nitrogen, total phosphorus, total nitrogen, temperature) and flow velocity sensors. Each sensing node is configured with an edge computing unit to perform local preliminary processing on the data and filter out anomalies. Two unmanned boats equipped with water quality sensors and camera modules are additionally arranged to make up for the blind spots of static nodes, and perform high-frequency cruise monitoring especially in areas close to tributaries and industrial discharge outlets. The system sampling frequency is set to once every 5 minutes.

[0111] The received data is de-noised and data-cleaned by the edge nodes and transmitted to the central platform. In the construction of the prediction model, the monitoring data within the past 72 hours is used as the historical training sequence to form a sliding window input sequence of 12 time steps each time. The linear regression sub-model and the GRU deep learning sub-model are respectively used to predict the trends of pollutant concentration and flow velocity, and the outputs are fused. The weight α is dynamically adjusted by the prediction deviation. The model is adaptively adjusted once an hour to support online incremental training.

[0112] At 08:00 on the 3rd day, the system detected that the ammonia nitrogen concentration in a tributary upstream (node N3) suddenly rose from 0.86 mg / L to 2.35 mg / L and spread to the downstream node N6 within the next 1 hour. Combining indicators such as the pollution path, the rate of change of the concentration peak value, and the water flow direction, the system determined that the pollution source tracing node was an industrial discharge point upstream of N3. It retrieved remote sensing images to verify the existence of suspicious water body color differences in this area, which coincided with the location of a certain printing and dyeing factory registered in the knowledge graph. The credibility score of the pollution source reached 0.91, and the pollution source was finally confirmed.

[0113] Subsequently, the system made diffusion predictions for 5 key nodes downstream of the pollution path. During the simulation time steps from 08:10 to 10:00, the predicted concentration peak value of node N6 was 2.14 mg / L, exceeding the set threshold of 1.50 mg / L in this area. The system automatically triggered the orange warning mechanism, highlighted the risk area on the platform interface, and simultaneously pushed the warning information to the mobile phones of the management personnel. It also recommended immediately starting the operation of switching to the backup water source downstream to reduce the risk impact.

[0114] The average ammonia nitrogen concentration of the upstream tributary N3 in the 24 hours before the event trigger was 0.82 mg / L, and the value at the abnormal increase moment was 2.35 mg / L. The proportion of false alarm data eliminated by the edge node preprocessing was 3.67%.

[0115] After the model fusion prediction, the predicted concentration of node N4 at 08:20 was 1.82 mg / L, and the measured value was 1.79 mg / L.

[0116] The predicted pollution arrival time of node N6 was 09:05, and the actual concentration exceeded the standard at 09:03, with a deviation of less than 2 minutes.

[0117] The predicted peak value of node N6 was 2.14 mg / L, and the actual measured value was 2.10 mg / L.

[0118] The mean square error of the GRU model's single prediction was 0.084, that of the linear model was 0.131, and that of the fusion model was 0.056.

[0119] The prediction deviation of the traditional model was 11.3%, the prediction deviation of the model of the present invention was 3.7%, and the error reduction reached 67.3%.

[0120] The advance amount of the system warning trigger was 58 minutes, and it was 15 minutes later than the traditional scheme in terms of delayed alarm.

[0121] The overall operation time delay was controlled within 2.65 seconds per iteration, compared with about 9.84 seconds for the traditional model iteration duration.

[0122] Through the above experimental process and data, it can be concluded that the method of the present invention shows significantly better performance and responsiveness than the existing technology in pollutant diffusion prediction. First, in terms of prediction accuracy, the present invention constructs a fused dynamic prediction model, uses linear regression to capture the basic trend of pollutant changes, and uses GRU to learn complex nonlinear fluctuation characteristics in time series, showing stronger generalization ability and adaptability in real environments. The fusion mechanism introduces a dynamic weighting factor, which is adjusted in real time according to the model deviation, significantly reducing the risk of deviation of a single model under specific pollution forms. The data shows that the prediction mean square error of the fusion model is 0.056, which is more than 67% lower than the traditional model, and the prediction error converges significantly.

[0123] Secondly, in terms of pollution source identification, the present invention constructs a dynamic graph and introduces a graph tracing algorithm, which not only considers the frequency of path overlap, but also combines the time sequence of pollutant concentration peaks, water flow direction, and auxiliary verification of remote sensing images, making the location of pollution sources more scientific. In this implementation, the credibility score of pollution source identification is as high as 0.91, which is much higher than the traditional method based on concentration mutation point estimation (the score is generally less than 0.75), significantly enhancing the credibility of pollution source judgment.

[0124] Third, in terms of early warning response, the present invention uses a dynamic classification mechanism for pollution levels to trigger an early warning 58 minutes in advance before the pollution prediction result reaches the set threshold, and simultaneously pushes it to the user terminal and on-site emergency equipment, forming a closed-loop process from identification to response. Compared with the passive disposal solution caused by delayed alarms in traditional models, this solution can gain more emergency response windows for decision makers.

[0125] In addition, the edge computing mechanism of the present invention also shows great practical advantages. During the experiment, the edge node controlled the delay of processing raw data within 3 seconds, and was able to complete false alarm elimination, abnormal identification and preliminary judgment in the first time, which gained a lot of processing space for back-end modeling. However, the traditional method of centralized processing of all data leads to significant processing delays and is not suitable for high-frequency monitoring scenarios.

[0126] In summary, this embodiment has been deployed in real scenarios and verified with complete data streams, proving that the pollution diffusion prediction method proposed in the present invention has significant advantages in many aspects, such as prediction accuracy, warning timeliness, pollution source location accuracy, and system real-time performance, and has broad prospects for promotion and application in various types of water ecosystems, reflecting the novelty and substantial progress of the method in technical means.

[0127] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the spread of water ecological pollution based on big data, characterized in that: include: Build a sensor network for the water environment, obtain real-time water environment data, pre-process the data using edge computing nodes, generate pre-processed real-time water environment data and synchronously transmit it to the central platform; Based on historical data and pre-processed real-time water environment data, a dynamic prediction model is established to output dynamic boundary condition parameters; Based on the dynamic boundary condition parameters, a pollutant propagation map is constructed to analyze the temporal and spatial distribution trend of pollutant concentrations and identify the location of pollution sources; Based on the results of pollution source location identification, timeliness prediction is performed, and pollution warning response is triggered in combination with set thresholds.

2. The method for predicting the spread of water ecological pollution based on big data according to claim 1, characterized in that: The real-time water environment data includes water quality monitoring data, hydrodynamic related data, meteorological background data, remote sensing and image data, biological and ecological indicator data, and meta-information data; The preprocessing includes performing noise filtering and outlier detection, eliminating sensor false alarm data, using logical rules and model-driven mechanisms to fill in missing data, and completing data cleaning operations to achieve data normalization and standardization; The data from different sensors are formatted in a unified manner and aligned in time and space dimensions.

3. The method for predicting the spread of water ecological pollution based on big data according to claim 2, characterized in that: The establishment of the dynamic prediction model includes: setting a time step, using pre-processed historical data and real-time data of the past n time steps, and organizing a water environment data sequence of n consecutive time steps through a sliding window method as an input of the dynamic prediction model; The dynamic prediction model adopts a fusion prediction framework, which is composed of a parallel linear regression model and a deep learning model; The linear regression model is used to capture the linear variation trend of water environment parameters and output prediction results based on the water environment data sequence of the past n time steps; The deep learning model uses the GRU network structure to capture the nonlinear dynamic characteristics of water environment parameters, predicts water environment data sequences, and supports online incremental training to adapt to real-time data changes; Performing weighted fusion on the prediction output of the linear regression model and the deep learning model to generate dynamic boundary condition parameter output; The dynamic boundary condition parameters include: water flow rate, flow rate, pollutant concentration and water level at the boundary.

4. The method for predicting the spread of water ecological pollution based on big data according to claim 3 is characterized by: The dynamic prediction model also includes, when new real-time water environment data arrives, updating the sliding window input, using recursive least squares method to update the linear regression model coefficients and performing incremental training based on real-time data to update the deep learning model parameters.

5. The method for predicting the spread of water ecological pollution based on big data according to claim 4 is characterized in that: Constructing the pollutant propagation map includes generating a hydrological connectivity map based on dynamic boundary condition parameters, combined with water body geographic information and watershed structure, wherein the nodes in the hydrological connectivity map represent key water monitoring points, tributary confluences and drainage outlets, and the edges represent the flow relationship between water bodies; The dynamic boundary condition parameters are used to dynamically assign values ​​to the edges in the hydrological connectivity graph, determine the flow direction, flow velocity and flow properties of the water body, and the pollutant concentration properties of the nodes are also determined by the dynamic boundary condition parameters and the real-time collected data; Generate pollutant propagation maps based on updated dynamic boundary condition parameters at predetermined time steps, and continuously form a dynamic map sequence; The dynamic graph sequence is subjected to spatiotemporal aggregation analysis to extract the change trajectory of pollutant concentration at each node, determine the abnormal growth and diffusion direction of pollutant concentration in the local area, and form a pollution diffusion path.

6. The method for predicting the spread of water ecological pollution based on big data according to claim 5 is characterized by: The identification of the pollution source location includes: using a graph tracing algorithm to reversely trace the pollution diffusion path, identifying the upstream node combination with the highest path overlap frequency as the pollution source candidate area, and combining the time series information of the concentration peak of each node and the water flow direction for judgment; The corresponding remote sensing image data of the candidate pollution source area is retrieved for auxiliary verification, and the known sewage discharge facility information in the area is compared with the pre-stored environmental knowledge graph. The credibility of the pollution source identification results is calculated based on the path continuity, temporal rationality and hydrodynamic consistency, and the location of the pollution source is finally confirmed.

7. The method for predicting the spread of water ecological pollution based on big data according to claim 6 is characterized by: The execution time prediction includes: using the geographical location of the pollution source to obtain the pollutant emission intensity and historical diffusion behavior; Based on the pollution diffusion path, the expected arrival time of pollutants, the peak concentration of pollutants and the duration of pollutant diffusion are predicted; by comparing the peak concentration of pollutants predicted at each node in the hydrological connectivity diagram with the preset safety threshold in real time, it is determined whether there is a risk of exceeding the peak concentration of pollutants in the next time step; When the pollutant concentration peak exceeds the preset threshold at a preset time step in the future, warning information is automatically generated; the warning information includes the pollution diffusion path, the expected exceeding time, the predicted concentration peak and the corresponding risk level.

8. A water ecological pollution diffusion prediction system based on big data using the method according to any one of claims 1 to 7, characterized in that: The data module builds a sensor network for the water environment, obtains real-time water environment data, pre-processes it using edge computing nodes, and transmits it synchronously to the central platform; The pollutant boundary condition parameter module establishes a dynamic prediction model for pollutant boundary condition parameters based on historical data and real-time water environment data; The pollution diffusion simulation module constructs a pollutant diffusion map based on the dynamic boundary condition parameters, analyzes the temporal and spatial distribution trend of pollutant concentrations and identifies the location of pollution sources; The early warning module performs pollution diffusion path simulation and time prediction based on the pollution source identification results, and triggers pollution early warning response in combination with the set threshold.

Images