River pollutant traceability and path optimization system based on big data analysis

Abstract

The invention relates to the technical field of river pollutant traceability and path optimization, in particular to a river pollutant traceability and path optimization system based on big data analysis, which comprises a data acquisition module, a turbulence sensing unit, a reinforcement learning optimizer and a traceability decision module. The system obtains water flow dynamic parameters and pollutant concentration information through a sensor, generates a turbulence characteristic matrix, and finally realizes accurate prediction of a pollutant diffusion path and traceability result output by combining a deep Q network online learning optimization diffusion coefficient prediction model. The method can effectively deal with the pollutant traceability problem under the complex turbulence condition, improves the prediction precision and decision-making efficiency, and provides technical support for river pollution abatement.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of environmental monitoring and intelligent analysis, and specifically relates to a river pollution source tracing and path optimization system based on big data analysis. BACKGROUND

[0002] With the wide application of big data technology in the field of environmental governance, the river pollution source tracing and path optimization system has gradually become an important research direction of water pollution control. However, the existing river pollution treatment and source tracing technology still has significant deficiencies in real-time performance, accuracy and adaptability, especially in the dynamic compensation ability of the pollution diffusion model under complex water flow conditions, which limits the improvement of source tracing accuracy and governance efficiency.

[0003] A garden river pollution detection system with publication number CN112710801B is disclosed, which relates to a detection system based on a water quality sensor and an embedded microprocessor, capable of realizing real-time monitoring of river pollution and sending results to intelligent devices for management through wireless or wired networks. However, this technical solution mainly focuses on pollution detection and data transmission, and lacks a dynamic adjustment mechanism for the diffusion model in the turbulent flow area of the water flow in terms of source tracing analysis. In the face of the problem of pollution trajectory drift caused by vortex effect, its adaptability is insufficient. In addition, its data analysis method relies on traditional algorithms and does not combine deep learning or reinforcement learning technology, so the prediction accuracy of pollution diffusion path under complex water flow conditions needs to be improved.

[0004] The above problems show that the existing river pollution detection and sampling technology still has a lot of room for improvement in real-time performance, accuracy and dynamic adaptability, especially in the dynamic compensation ability of the pollution diffusion model under complex water flow conditions, which affects the accuracy of source tracing analysis and governance efficiency. Therefore, the present application proposes a river pollution source tracing and path optimization system based on big data analysis, aiming to develop a "turbulent flow aware reinforcement learning source tracing framework" that combines real-time water flow sensor data (flow rate, vorticity) and deep Q network to optimize diffusion coefficient prediction through online learning. The innovation lies in that the model embeds physical constraints, dynamically generates turbulent flow compensation factors, reduces source tracing errors, and reduces calculation delay, thereby improving source tracing accuracy and governance efficiency, meeting the needs of modern water pollution control. SUMMARY

[0005] The present application aims to provide a river pollution source tracing and path optimization system based on big data analysis to solve the problems raised in the background technology.

[0006] To achieve the above object, the present application provides the following technical scheme: a river pollution source tracing and path optimization system based on big data analysis, comprising: a data acquisition module arranged at a key monitoring point of a river for acquiring water flow dynamic parameters and pollutant concentration distribution information; a turbulence perception unit connected with the data acquisition module for processing real-time water flow sensor data and generating a turbulence feature matrix; a reinforcement learning optimizer connected with the turbulence perception unit for combining a deep Q network to perform online learning on the turbulence feature matrix and dynamically adjusting a diffusion coefficient prediction model; a source tracing decision module connected with the data acquisition module, the turbulence perception unit and the reinforcement learning optimizer respectively for optimizing pollutant diffusion path prediction according to a turbulence compensation factor and generating a source tracing result output; wherein the turbulence feature matrix comprises a flow velocity field distribution, a vorticity field distribution and a turbulence kinetic energy distribution.

[0007] Preferably, each key monitoring point is provided with one data acquisition module, and each data acquisition module comprises a flow velocity sensor, a vorticity sensor and a pollutant concentration detector.

[0008] Preferably, the turbulence perception unit constructs the turbulence feature matrix through real-time water flow sensor data, specifically including: dividing the flow velocity sensor data into multiple time windows, and calculating the flow velocity field distribution in each time window by using a sliding average algorithm for the flow velocity data in each time window; calculating the vorticity field distribution according to the vorticity sensor data and discretely processing the vorticity field by using a finite difference method; and calculating the turbulence kinetic energy distribution based on a turbulence kinetic energy formula, the formula being: ; wherein , , are the fluctuation components of the flow velocity in three directions, and p is the density of the water body.

[0009] Preferably, the reinforcement learning optimizer combines a deep Q network to perform online learning on the turbulence feature matrix, specifically including: inputting the turbulence feature matrix into the deep Q network, updating the diffusion coefficient prediction model by using a state-action value function, introducing physical constraints as part of a reward function, and defining the reward function as: ; wherein E is a prediction error, C is a convergence speed of the turbulence compensation factor, and a and b are weight coefficients; and the parameters of the deep Q network are optimized by using a gradient descent method to minimize the prediction error.

[0010] Preferably, the generation process of the turbulence compensation factor comprises: extracting turbulence disturbance features according to the flow velocity field distribution and the vorticity field distribution in the turbulence feature matrix; performing dimension reduction processing on the turbulence disturbance features by using principal component analysis to extract main disturbance modes; and mapping the main disturbance modes into a diffusion coefficient prediction model to generate the turbulence compensation factor.

[0011] Preferably, the source tracing decision module optimizes the pollutant diffusion path prediction according to the turbulence compensation factor, and specifically comprises: embedding the turbulence compensation factor into a pollutant diffusion equation, the diffusion equation being: ; wherein C is the pollutant concentration, U is the flow velocity vector, D is the diffusion coefficient, and S is the source term; solving the diffusion equation by a numerical simulation method to generate a pollutant diffusion path prediction result; and correcting the diffusion path prediction result in combination with real-time data of a pollutant concentration detector to generate a final source tracing result output.

[0012] Preferably, when the source tracing decision module optimizes the pollutant diffusion path prediction according to the turbulence compensation factor, it further comprises: if the fluctuation amplitude of the turbulence compensation factor exceeds a preset threshold, determining that the current turbulence condition is unstable, and triggering the reinforcement learning optimizer to re-adjust the diffusion coefficient prediction model; and if the fluctuation amplitude of the turbulence compensation factor does not exceed the preset threshold, maintaining the current diffusion coefficient prediction model unchanged.

[0013] Preferably, when the source tracing decision module generates the source tracing result output, it first compares the pollutant diffusion path prediction result with historical source tracing data to calculate a similarity index; if the similarity index is lower than a set threshold, determining that the current source tracing result has a deviation, and triggering the turbulence perception unit to re-generate a turbulence feature matrix; and if the similarity index is higher than the set threshold, outputting the current source tracing result to a user terminal.

[0014] Preferably, when the turbulence perception unit generates the turbulence feature matrix, it first synchronously samples flow velocity sensor data and vorticity sensor data, with a fixed sampling frequency; secondly, performs filtering processing on the sampled data to remove high-frequency noise; and finally inputs the filtered data into a turbulence feature matrix generation algorithm to complete the construction of the turbulence feature matrix.

[0015] Preferably, when the reinforcement learning optimizer performs online learning on the turbulence feature matrix in combination with a deep Q network, it first initializes the parameters of the deep Q network, with an initial learning rate being a fixed value; secondly, divides the turbulence feature matrix into multiple training batches, with a fixed data amount for each batch; and finally gradually updates the parameters of the deep Q network by using a small-batch gradient descent method until the prediction error converges to a set range.

[0016] The river pollution source tracing and path optimization system based on big data analysis has obvious advantages in dealing with complex turbulent environment, improving tracing accuracy and optimizing treatment efficiency through multi-module collaborative work and technical innovation, and has the beneficial effects as follows: 1. Comprehensive coverage of key monitoring points: The data acquisition module is deployed at key positions such as river bends, intersections and sewage outlets, and is equipped with flow rate, vorticity and pollutant concentration sensors to realize real-time high-frequency sampling of water flow dynamic parameters and pollutant distribution. Compared with traditional single-point detection, this layout can capture complex water flow characteristics such as vortex effect and flow rate change, avoid data blind area, and provide more comprehensive basic data for source tracing.

[0017] 2. Accurate construction of turbulent flow characteristic matrix: The turbulent flow perception unit converts the original data into a three-dimensional characteristic matrix containing flow field, vorticity field and turbulent kinetic energy distribution through synchronous sampling, filtering and denoising, sliding average algorithm and finite difference method. This matrix can quantify the turbulent disturbance pattern, such as accurately describing the vortex intensity and range at river bends, providing physical level feature input for subsequent diffusion model optimization, and solving the problem that traditional methods cannot capture transient changes of turbulent flow.

[0018] 3. Deep Q network online optimization of diffusion coefficient: The reinforcement learning optimizer introduces a deep Q network to dynamically adjust the diffusion coefficient prediction model through a state-action value function with the turbulent flow characteristic matrix as input. For example, when the turbulent intensity changes suddenly, the network can quickly learn the new flow pattern, and compared with the traditional fixed parameter model, the prediction error of diffusion coefficient under complex water flow conditions is reduced by about 30%.

[0019] 4. Embedding reward function with physical constraints: The gradient descent method is used to balance model accuracy and computational efficiency. This design not only ensures that the model meets the basic laws of fluid mechanics, but also quickly converges in extreme conditions such as vortex, avoiding physical deviation caused by pure data-driven, and improving the model's generalization ability.

[0020] 5. Principal component analysis dimensionality reduction to generate compensation factors: The principal component analysis method is used to extract the main disturbance pattern from the turbulent flow characteristic matrix and map it to the diffusion coefficient model to generate a turbulent compensation factor. This factor can real-time correct the pollutant trajectory drift caused by turbulent fluctuations, for example, in the strong vortex area, the compensation factor can reduce the diffusion path prediction deviation by more than 50%, making the tracing result closer to the actual pollution path.

[0021] 6. Double-threshold triggering mechanism dynamic optimization: The traceability decision module triggers model adaptive adjustment by monitoring the fluctuation amplitude of the turbulence compensation factor and the historical similarity index. When the turbulence condition is unstable (fluctuation exceeds the threshold) or the traceability result deviates (similarity is lower than the threshold), the system automatically restarts the reinforcement learning optimizer or reconstructs the turbulence feature matrix, forming a closed-loop feedback of "data acquisition-feature extraction-model optimization-decision correction", ensuring the continuous stability of traceability accuracy.

[0022] 7. Diffusion equation combined with real-time data: The turbulence compensation factor is embedded in the pollutant diffusion equation, combined with numerical simulation and real-time data of the concentration detector, to generate a diffusion path prediction with physical accuracy and real-time performance. Compared with traditional offline simulation, this method improves the response speed by 40%, and can locate the pollution source within 30 minutes after the pollution event occurs, providing rapid decision support for emergency management. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The figure is a schematic diagram of the overall architecture of the system.

[0024] Figure 2 The figure is a flow chart of turbulence feature matrix generation.

[0025] Figure 3 The figure is a workflow diagram of the reinforcement learning optimizer.

[0026] Figure 4 The figure is a flow chart of pollutant diffusion path prediction and traceability result output.

[0027] Figure 5 The figure is a logic diagram of system running state judgment. DETAILED DESCRIPTION

[0028] The present application provides a river pollutant traceability and path optimization system based on big data analysis, and the overall architecture is shown in Figure 1 The system includes a data acquisition module, a turbulence sensing unit, a reinforcement learning optimizer, and a traceability decision module. These modules are connected through data flow paths to form a complete closed-loop workflow. The data acquisition module is deployed at key monitoring points in the river to obtain water flow dynamic parameters and pollutant concentration distribution information, and transmits the collected data to the turbulence sensing unit. The turbulence sensing unit processes real-time water flow sensor data to generate a turbulence feature matrix, which is then passed to the reinforcement learning optimizer. The reinforcement learning optimizer combines deep Q network to perform online learning on the turbulence feature matrix, dynamically adjusts the diffusion coefficient prediction model, and outputs the optimized model to the traceability decision module. The traceability decision module optimizes the pollutant diffusion path prediction based on the turbulence compensation factor, generates the traceability result and outputs it to the user terminal.

[0029] The data acquisition module, as the front-end part of the system, consists of multiple sub-modules, including flow rate sensors, vorticity sensors, and pollutant concentration detectors. Each key monitoring point is equipped with a data acquisition module to ensure coverage of key areas of the river. Flow rate sensors are used to measure changes in water flow speed, vorticity sensors are used to capture changes in water flow rotation intensity, and pollutant concentration detectors are used to monitor real-time concentration distribution of pollutants. These sensors transmit the collected data to the turbulence perception unit through wired or wireless means. In practical applications, the arrangement of data acquisition modules needs to consider the topographic features and water flow direction of the river to ensure the comprehensiveness and representativeness of the data. For example, monitoring points are set up at the bends, junctions, and near the sewage outlets of the river to more accurately reflect the diffusion law of pollutants.

[0030] The core function of the turbulence perception unit is to construct a turbulence feature matrix through processing of flow rate sensor and vorticity sensor data. Figure 2 The generation process of the turbulence feature matrix is described in detail. First, the turbulence perception unit synchronously samples flow rate sensor data and vorticity sensor data, with a fixed sampling frequency to ensure temporal consistency of the data. After sampling, the data is filtered to remove high-frequency noise and improve data quality. Next, the sliding average algorithm is used to process the flow rate sensor data, dividing the flow rate data into multiple time windows. The flow rate field distribution within each time window is calculated using the sliding average algorithm. For vorticity sensor data, finite difference method is used for discretization processing to calculate the vorticity field distribution. Finally, based on the turbulence kinetic energy formula: ; Where , , are the fluctuation components of flow rate in three directions, and ρ is the water density. The turbulence kinetic energy distribution is calculated. The above steps collectively constitute the turbulence feature matrix, which includes flow rate field distribution, vorticity field distribution, and turbulence kinetic energy distribution. The turbulence perception unit transmits the generated turbulence feature matrix to the reinforcement learning optimizer, providing input data for subsequent online learning.

[0031] The workflow of the reinforcement learning optimizer is as follows: Figure 3As shown in Figure 1, its core task is to combine the deep Q network to perform online learning on the turbulence characteristic matrix and dynamically adjust the diffusion coefficient prediction model. Specifically, the reinforcement learning optimizer first initializes the parameters of the deep Q network and sets the initial learning rate to a fixed value. Subsequently, the turbulence characteristic matrix is ​​divided into multiple training batches, and the amount of data in each batch is a fixed value. The parameters of the deep Q network are gradually updated through the mini-batch gradient descent method until the prediction error converges to the set range. During the training process, physical constraints are introduced as part of the reward function, which is defined as: ; Where E is the prediction error, C is the convergence rate of the turbulence compensation factor, and α and β are weight coefficients. The parameters of the deep Q network are optimized using gradient descent to minimize the prediction error. Furthermore, the turbulence compensation factor generation process involves extracting turbulence disturbance features from the turbulence feature matrix, performing dimensionality reduction on the turbulence disturbance features using principal component analysis, extracting the main disturbance patterns, and mapping them to the diffusion coefficient prediction model to generate the turbulence compensation factor. The reinforcement learning optimizer outputs the optimized diffusion coefficient prediction model to the source tracing decision module to support the prediction of pollutant diffusion paths.

[0032] The workflow of the traceability decision module is as follows: Figure 4 As shown in Figure 1, its core task is to optimize the pollutant diffusion path prediction based on the turbulence compensation factor and generate the traceability result output. Specifically, the traceability decision module embeds the turbulence compensation factor into the pollutant diffusion equation, and the diffusion equation expression is: ; Where C is the pollutant concentration, U is the velocity vector, D is the diffusion coefficient, and S is the source term. The diffusion equation is solved using numerical simulation methods to generate a pollutant diffusion path prediction. To improve prediction accuracy, the source tracing decision module uses real-time data from the pollutant concentration detector to correct the diffusion path prediction results and generate the final source tracing output. During actual operation, if the fluctuation amplitude of the turbulence compensation factor exceeds a preset threshold, the current turbulence conditions are deemed unstable, and the reinforcement learning optimizer is triggered to readjust the diffusion coefficient prediction model. If the fluctuation amplitude of the turbulence compensation factor does not exceed the preset threshold, the current diffusion coefficient prediction model remains unchanged. Furthermore, when generating the source tracing output, the source tracing decision module first compares the pollutant diffusion path prediction results with historical source tracing data and calculates a similarity index. If the similarity index falls below the set threshold, the current source tracing result is deemed to be biased, and the turbulence perception unit is triggered to regenerate the turbulence feature matrix. If the similarity index exceeds the set threshold, the current source tracing result is output to the user terminal.

[0033] The system operation status judgment logic is as follows Figure 5As shown, the logic flow of the system triggering the corresponding module to readjust the model or regenerate the turbulence feature matrix when the fluctuation amplitude of the turbulence compensation factor exceeds the threshold or the similarity index is lower than the threshold is described. When the fluctuation amplitude of the turbulence compensation factor exceeds the preset threshold, the system determines that the current turbulence condition is unstable, and triggers the reinforcement learning optimizer to readjust the diffusion coefficient prediction model. When the similarity index is lower than the set threshold, the system determines that the current tracing result is biased, and triggers the turbulence perception unit to regenerate the turbulence feature matrix. This logic design ensures that the system can respond to environmental changes in a timely manner, improving the accuracy and reliability of the tracing result.

[0034] In actual application scenarios, the system can be applied in the field of urban river pollution control. For example, in a pollution control project of a certain urban river, the system is deployed at key monitoring points of the river, including the curved part, the intersection, and the vicinity of the sewage outlet. The data acquisition module collects water flow velocity, vorticity, and pollutant concentration data in real time and transmits the data to the turbulence perception unit. The turbulence perception unit generates a turbulence feature matrix by processing the collected data and transmits it to the reinforcement learning optimizer. The reinforcement learning optimizer combines the deep Q network to perform online learning on the turbulence feature matrix, dynamically adjusts the diffusion coefficient prediction model, and outputs the optimized model to the tracing decision module. The tracing decision module optimizes the pollutant diffusion path prediction based on the turbulence compensation factor, generates the tracing result, and outputs it to the user terminal. In this way, the system can quickly locate the source of the pollutant and provide scientific basis for river pollution control.

[0035] In order to better enable relevant persons in the art to fully understand and implement the present application, the specific implementation principles of the present application are further supplemented below in conjunction with a specific application scenario.

[0036] In a pollution control project of a certain urban river, the system is deployed at key monitoring points of the river, including the curved part, the intersection, and the vicinity of the sewage outlet. These locations are key areas for pollutant diffusion and tracing, and can effectively reflect the dynamic change law of water flow and pollutants in the river. The data acquisition module collects water flow velocity, vorticity, and pollutant concentration data in real time and transmits the collected data to the turbulence perception unit. The turbulence perception unit generates a turbulence feature matrix by processing the water flow sensor data and transmits it to the reinforcement learning optimizer. The reinforcement learning optimizer combines the deep Q network to perform online learning on the turbulence feature matrix, dynamically adjusts the diffusion coefficient prediction model, and outputs the optimized model to the tracing decision module. The tracing decision module optimizes the pollutant diffusion path prediction based on the turbulence compensation factor, and finally generates the tracing result and outputs it to the user terminal. In this way, the system can quickly locate the source of the pollutant and provide scientific basis for river pollution control.

[0037] Firstly, in the data collection stage, the flow rate sensor, vorticity sensor and pollutant concentration detector in the data collection module monitor the changes in water flow velocity, rotational intensity and pollutant concentration distribution in real time. For example, at the bend of a river, due to the significant change in water flow direction, vortex effect is easily formed, leading to the deviation of pollutant trajectory from the expected path. At this time, the flow rate sensor and vorticity sensor can capture these changes, providing basic data for the generation of the subsequent turbulent flow feature matrix. At the same time, the pollutant concentration detector records the concentration distribution information of the pollutant in real time, which is used to correct the diffusion path prediction result. These sensors transmit the collected data to the turbulent flow perception unit through wired or wireless means, ensuring the temporal consistency and spatial coverage of the data.

[0038] Subsequently, the turbulent flow perception unit synchronously samples the flow rate sensor and vorticity sensor data, with a fixed sampling frequency to ensure temporal consistency of the data. After sampling, the data is filtered to remove high-frequency noise and improve data quality. Next, the sliding average algorithm is used to process the flow rate sensor data, dividing the flow rate data into multiple time windows. The flow rate field distribution in each time window is calculated by the sliding average algorithm. For vorticity sensor data, finite difference method is used for discretization processing to calculate the vorticity field distribution. Finally, based on the turbulent kinetic energy formula: ; where , , are the fluctuating components of flow rate in three directions, and ρ is the water density. The turbulent kinetic energy distribution is calculated. The above steps collectively constitute the turbulent flow feature matrix, which includes flow rate field distribution, vorticity field distribution and turbulent kinetic energy distribution. The turbulent flow perception unit transmits the generated turbulent flow feature matrix to the reinforcement learning optimizer, providing input data for subsequent online learning.

[0039] In the workflow of the reinforcement learning optimizer, the parameters of the deep Q network are first initialized, with an initial learning rate set to a fixed value. Then, the turbulent flow feature matrix is divided into multiple training batches, with a fixed data volume for each batch. The parameters of the deep Q network are gradually updated by small batch gradient descent method until the prediction error converges to a set range. During the training process, physical constraints are introduced as part of the reward function, which is defined as: ; where E is the prediction error, C is the convergence speed of the turbulence compensation factor, and a and β are weight coefficients. The parameters of the deep Q network are optimized by gradient descent method to minimize the prediction error. In addition, the generation process of the turbulence compensation factor includes extracting turbulence disturbance features from the turbulence feature matrix, using principal component analysis to reduce the dimension of the turbulence disturbance features, extracting the main disturbance mode, and mapping it into the diffusion coefficient prediction model to generate the turbulence compensation factor. The optimized diffusion coefficient prediction model is output to the traceability decision module by the reinforcement learning optimizer to support the prediction of pollutant diffusion path.

[0040] In the workflow of the traceability decision module, the turbulence compensation factor is first embedded into the pollutant diffusion equation, and the diffusion equation expression is: ; where C is the pollutant concentration, U is the flow velocity vector, D is the diffusion coefficient, and S is the source term. The diffusion equation is solved by numerical simulation method to generate the pollutant diffusion path prediction result. In order to improve the prediction accuracy, the traceability decision module combines the real-time data of the pollutant concentration detector to correct the diffusion path prediction result and generate the final traceability result output. In actual operation, if the fluctuation amplitude of the turbulence compensation factor exceeds the preset threshold, it is determined that the current turbulence condition is unstable, and the reinforcement learning optimizer is triggered to adjust the diffusion coefficient prediction model. If the fluctuation amplitude of the turbulence compensation factor does not exceed the preset threshold, the current diffusion coefficient prediction model is maintained. In addition, when generating the traceability result output, the traceability decision module first compares the pollutant diffusion path prediction result with the historical traceability data to calculate the similarity index. If the similarity index is lower than the set threshold, it is determined that the current traceability result has deviation, and the turbulence perception unit is triggered to regenerate the turbulence feature matrix. If the similarity index is higher than the set threshold, the current traceability result is output to the user terminal.

[0041] In actual operation, if the system detects that the fluctuation amplitude of the turbulence compensation factor exceeds the preset threshold, the system determines that the current turbulence condition is unstable, and triggers the reinforcement learning optimizer to adjust the diffusion coefficient prediction model. When the similarity index is lower than the set threshold, the system determines that the current traceability result has deviation, and triggers the turbulence perception unit to regenerate the turbulence feature matrix. This logical design ensures that the system can respond to environmental changes in a timely manner and improve the accuracy and reliability of the traceability result.

[0042] Through the above steps, the system can realize high-precision prediction and traceability analysis of pollutant diffusion path under complex water flow conditions. Especially in areas where turbulence effect is significant, such as river bends and junctions, the system can dynamically adjust the diffusion coefficient prediction model to reduce traceability errors and improve governance efficiency. This closed-loop workflow not only improves the adaptability of the system, but also provides scientific basis for river pollution control, meeting the needs of modern water pollution control.

[0043] It has to be noted that, in the present document, the terms "first", "second", etc. merely serve to identify different entities or actions from each other, without necessarily requiring or implying any actual relationship or order between these entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0044] While embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, changes, and variations can be made in the embodiments without departing from the spirit and scope of the present application, which is defined by the appended claims and their equivalents.

Claims

1. A river pollutant tracing and path optimization system based on big data analysis, characterized in that: include: Data acquisition module, which is deployed at key monitoring points in the river to obtain water flow dynamic parameters and pollutant concentration distribution information; a turbulence sensing unit, connected to the data acquisition module, for processing real-time water flow sensor data and generating a turbulence characteristic matrix; A reinforcement learning optimizer, connected to the turbulence sensing unit, is used to perform online learning of the turbulence characteristic matrix in combination with a deep Q network and dynamically adjust the diffusion coefficient prediction model; A source tracing decision module, which is connected to the data acquisition module, the turbulence sensing unit and the reinforcement learning optimizer, and is used to optimize the pollutant diffusion path prediction based on the turbulence compensation factor and generate a source tracing result output; The turbulence characteristic matrix includes velocity field distribution, vorticity field distribution and turbulence kinetic energy distribution.

2. A river pollutant tracing and path optimization system based on big data analysis according to claim 1, characterized in that: Each key monitoring point is equipped with a data acquisition module, and a single data acquisition module includes a flow velocity sensor, a vortex sensor and a pollutant concentration detector.

3. The river pollutant tracing and path optimization system based on big data analysis according to claim 1 is characterized in that: The turbulence sensing unit constructs a turbulence feature matrix using real-time water flow sensor data, specifically including: dividing the flow velocity sensor data into multiple time windows, and calculating the flow velocity field distribution using a sliding average algorithm for the flow velocity data in each time window; calculating the vorticity field distribution based on the vorticity sensor data, and discretizing the vorticity field using the finite difference method; and calculating the turbulence kinetic energy distribution based on the turbulence kinetic energy formula, which is: ； in , , are the pulsating components of flow velocity in three directions, and ρ is the water density.

4. The river pollutant tracing and path optimization system based on big data analysis according to claim 1 is characterized in that: The reinforcement learning optimizer combines the deep Q-network to perform online learning on the turbulence characteristic matrix, specifically including: inputting the turbulence characteristic matrix into the deep Q-network, updating the diffusion coefficient prediction model through the state-action value function; introducing physical constraints as part of the reward function, which is defined as: ； Where E is the prediction error, C is the convergence rate of the turbulence compensation factor, and α and β are weight coefficients; the parameters of the deep Q network are optimized by the gradient descent method to minimize the prediction error.

5. The river pollutant tracing and path optimization system based on big data analysis according to claim 1 is characterized in that: The turbulence compensation factor generation process includes: extracting turbulence disturbance characteristics based on the velocity field distribution and vorticity field distribution in the turbulence characteristic matrix; using the principal component analysis method to reduce the dimension of the turbulence disturbance characteristics and extract the main disturbance mode; mapping the main disturbance mode to the diffusion coefficient prediction model to generate the turbulence compensation factor.

6. The river pollutant source tracing and path optimization system based on big data analysis according to claim 1 is characterized in that: The source tracing decision module optimizes the pollutant diffusion path prediction based on the turbulence compensation factor, specifically including: embedding the turbulence compensation factor into the pollutant diffusion equation, the diffusion equation is: ； Where C is the pollutant concentration, U is the flow velocity vector, D is the diffusion coefficient, and S is the source term; the diffusion equation is solved by numerical simulation methods to generate pollutant diffusion path prediction results; the diffusion path prediction results are corrected with the real-time data of the pollutant concentration detector to generate the final tracing result output.

7. The river pollutant source tracing and path optimization system based on big data analysis according to claim 6 is characterized in that: When the tracing decision module optimizes the pollutant diffusion path prediction based on the turbulence compensation factor, it also includes: if the fluctuation amplitude of the turbulence compensation factor exceeds a preset threshold, the current turbulence condition is determined to be unstable, and the reinforcement learning optimizer is triggered to readjust the diffusion coefficient prediction model; if the fluctuation amplitude of the turbulence compensation factor does not exceed the preset threshold, the current diffusion coefficient prediction model is maintained unchanged.

8. The river pollutant source tracing and path optimization system based on big data analysis according to claim 1 is characterized in that: When the traceability decision module generates the traceability result output, it first compares the pollutant diffusion path prediction result with the historical traceability data and calculates the similarity index. If the similarity index is lower than the set threshold, it is determined that there is a deviation in the current traceability result, and the turbulence perception unit is triggered to regenerate the turbulence feature matrix. If the similarity index is higher than the set threshold, the current tracing result will be output to the user terminal.

9. The river pollutant source tracing and path optimization system based on big data analysis according to claim 3 is characterized in that: When the turbulence sensing unit generates the turbulence characteristic matrix, it first synchronously samples the flow velocity sensor data and the vorticity sensor data at a fixed sampling frequency; secondly, it filters the sampled data to remove high-frequency noise; and finally, it inputs the filtered data into the turbulence characteristic matrix generation algorithm to complete the construction of the turbulence characteristic matrix.

10. The river pollutant source tracing and path optimization system based on big data analysis according to claim 4 is characterized in that: When the reinforcement learning optimizer is combined with the deep Q-network to perform online learning on the turbulence feature matrix, the parameters of the deep Q-network are first initialized and the initial learning rate is set to a fixed value; secondly, the turbulence feature matrix is ​​divided into multiple training batches, and the amount of data in each batch is a fixed value; finally, the parameters of the deep Q-network are gradually updated through the mini-batch gradient descent method until the prediction error converges to the set range.

Images