A digital water pollution monitoring system based on the Internet of Things

Abstract

The application discloses a kind of digital water pollution monitoring systems based on Internet of Things, including data acquisition module, physical coupling prediction module, physical consistency verification module and decision output module.Data acquisition module collects water quality monitoring data and hydrological monitoring data of target water body;Physical coupling prediction module generates pollution diffusion prediction result and first confidence according to water quality monitoring data;Physical consistency verification module independently calculates pollution transmission theoretical value and second confidence according to hydrological monitoring data;Decision output module calculates the deviation value of two confidences, and executes graded output according to the threshold interval where deviation is located: deviation is small, output fusion report, deviation is medium, weighted fusion is added with artificial review prompt, and deviation is large, intercept prediction result and generate physical conflict alarm.The application improves the reliability and decision transparency of water pollution monitoring system prediction result through double independent calculation and physical consistency verification.

Description

Technical Field

[0001] This invention relates to the field of water pollution monitoring technology, specifically to a digital water pollution monitoring system based on the Internet of Things. Background Technology

[0002] In recent years, with the rapid development of IoT, big data, and AI technologies, water pollution monitoring systems are evolving towards digitalization and intelligence. At the perception layer, water quality monitoring technology based on IoT sensor networks has been widely applied, enabling real-time data acquisition and wireless transmission of target water bodies through the deployment of multi-parameter sensors. At the prediction and analysis layer, various water quality prediction models and intelligent optimization methods have been introduced into the field of water pollution monitoring, improving the efficiency of pollution trend analysis and treatment decisions to some extent. At the information presentation level, some systems have begun to introduce visualization and interpretive functions to help users understand monitoring data and model output results. These technologies provide technical support for water pollution monitoring from different perspectives, but they have developed independently and have not yet formed an organically integrated systemic solution.

[0003] The aforementioned existing technologies still have the following shortcomings:

[0004] 1. Existing AI-based water quality prediction models lack cross-validation of physical laws in their prediction results. When sensor data is abnormal or the model is running outside the coverage of training data, the model may output prediction results that violate basic physical laws. The system cannot identify such reliability issues in predictions, leading to the risk of false alarms or missed alarms.

[0005] 2. Existing systems lack a mechanism for quantitative comparison and hierarchical processing of discrepancies between AI predictions and physical model calculations. When AI predictions conflict with physical laws, the system cannot distinguish between different degrees of consistency and adopt differentiated response strategies, making it difficult for decision-makers to make effective judgments when faced with contradictory information.

[0006] 3. The scheduling decisions in the existing system are mostly in an open-loop mode. That is, after making scheduling decisions based on the current monitoring data, there is a lack of a real-time evaluation and feedback correction mechanism for the scheduling effect. It is impossible to form a closed loop from prediction to decision-making, execution, and then evaluation and feedback correction, which makes it impossible for the system to continuously optimize its performance through continuous operation.

[0007] 4. Existing interpretability methods are mostly ex-post interpretation tools, that is, they passively generate interpretation information after the model completes the prediction. The interpretation content is independent of the model's internal reasoning process. They lack an architectural design that deeply embeds interpretability into the system's decision-making chain, and the form of interpretation output is relatively simple. Summary of the Invention

[0008] (a) Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides a digital water pollution monitoring system based on the Internet of Things, which solves the problems mentioned in the background section.

[0010] (II) Technical Solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: a digital water pollution monitoring system based on the Internet of Things, comprising:

[0012] The data acquisition module is used to collect water quality monitoring data and hydrological monitoring data of the target water body in real time through an Internet of Things sensor network;

[0013] The physical coupling prediction module has a built-in hydrodynamic equation as a physical constraint, which is used to generate water pollution diffusion prediction results and corresponding first confidence levels based on the water quality monitoring data.

[0014] The physical consistency verification module has a built-in hydrodynamic equation independent of the physical coupling prediction module, which is used to independently calculate the theoretical value of pollution transmission and the corresponding second confidence level based on the hydrological monitoring data.

[0015] The decision output module is used to calculate the deviation between the first confidence level and the second confidence level, and to execute the corresponding output strategy based on the threshold range in which the deviation value falls:

[0016] When the deviation value is less than or equal to the first threshold, a comprehensive explanation report that integrates the prediction result and the theoretical value is output.

[0017] When the deviation value is greater than the first threshold and less than or equal to the second threshold, the weighted fusion result is output along with a manual review prompt.

[0018] When the deviation value is greater than the second threshold, the prediction result is intercepted, and only a deterministic conclusion based on the theoretical value is output and a physical conflict alarm is generated, wherein the second threshold is greater than the first threshold.

[0019] The specific process of the physical coupling prediction module embedding the hydrodynamic equation as a physical constraint is as follows: acquire the water quality monitoring data collected in real time by the data acquisition module, embed the hydrodynamic equation into the training objective of the physical coupling prediction model in the form of a loss function term, so that the output of the physical coupling prediction model is constrained by the hydrodynamic equation during the training process, and perform forward inference on the current water quality monitoring data through the trained physical coupling prediction model to generate water pollution diffusion prediction results and corresponding first confidence levels.

[0020] The specific process of the physical consistency verification module independently calculating the theoretical value of pollution transmission using the built-in hydrodynamic equation is as follows: obtain the hydrological monitoring data collected in real time by the data acquisition module, input the hydrological monitoring data into the hydrodynamic equation for forward evolution calculation, and obtain the theoretical value of pollution transmission and the corresponding second confidence level.

[0021] The specific process for the decision output module to perform deviation value calculation and hierarchical output is as follows: Obtain the prediction result and the first confidence level output by the physical coupling prediction module, and obtain the theoretical value and the second confidence level output by the physical consistency verification module. Calculate the deviation value between the first confidence level and the second confidence level according to the following formula: ; in, This represents the first confidence level. This is the second confidence level. The deviation value is then compared sequentially with the first threshold and the second threshold to determine the threshold range in which the deviation value falls. When the deviation value is less than or equal to the first threshold, i.e. When the predicted result is determined to be highly consistent with the theoretical value, the predicted result and the theoretical value are fused to generate a comprehensive explanation report and output, wherein... The first threshold is defined as the threshold value; when the deviation value is greater than the first threshold and less than or equal to the second threshold, i.e. If a moderate deviation is found between the predicted result and the theoretical value, a weighted fusion is performed on the predicted result and the theoretical value. The weighted fusion result, along with a manual review prompt, is then output. The second threshold is used; when the deviation value is greater than the second threshold, i.e. If the predicted result is found to be in serious conflict with the theoretical value, the predicted result is intercepted and output is prohibited. Only the deterministic conclusion based on the theoretical value is output along with the physical conflict alarm.

[0022] It also includes a dynamic interpretation and adaptation module. The specific process of the dynamic interpretation and adaptation module to perform adaptive interpretation output is as follows: obtain the output result of the decision output module, identify the current user's role type, which includes operator role, management decision-maker role and regulatory agency role;

[0023] When identified as an operator, the output is converted into a technically detailed explanatory report containing specific data curves, anomaly point markings, and executable operation steps; when identified as a management decision-maker, the output is converted into a decision-oriented explanatory report containing risk level assessment, pollution trend prediction, and resource allocation recommendations; when identified as a regulatory agency, the output is converted into a compliance-oriented explanatory report containing a legal basis index, monitoring evidence chain, and confidence quantification report, and the converted explanatory report is output to the current user's interactive interface in the corresponding format.

[0024] It also includes a counterfactual inference scheduling module. The specific process of the counterfactual inference scheduling module for performing hypothesis scheduling instruction inference and decision support is as follows: In response to the externally input hypothesis scheduling instruction, which includes the target operation parameters of at least one controllable device of the target water body within a specified time window, the baseline state currently confirmed by the physical consistency verification module is obtained. The baseline state is the state confirmed by the physical consistency verification module after forward evolution calculation of the hydrodynamic equation, which has passed the physical consistency test. The baseline state includes the water quality monitoring data, hydrological monitoring data and the theoretical value at the current moment. The baseline state at the current moment is cloned in the digital twin environment to generate the initial state of inference. Based on the initial state of inference, the target operation parameters in the hypothesis scheduling instruction are used as action input to drive the reinforcement learning agent to perform state inference in the digital twin environment and generate the hypothesis scheduling result.

[0025] The hypothetical scheduling result is compared with the baseline state at each monitoring point using a differential comparison, and the decision deviation value for each monitoring point is calculated according to the following formula: ; in, The total number of monitoring points, For the first The predicted values ​​for each monitoring point under the assumed scheduling result. For the first The theoretical values ​​of each monitoring point under the baseline condition. For the first The decision deviation value of each monitoring point; the overall decision deviation cost is calculated according to the following formula: ; in, For the first Preset weighting coefficients for each monitoring point. The overall decision bias cost; the decision bias value at each monitoring point. The data is output to the interactive interface in the form of a heat map. The heat map is used to display the quantitative difference between the hypothetical scheduling result and the baseline state at each monitoring point. In response to an external confirmation command, the target operation parameters in the hypothetical scheduling command are translated into the actual control command of the controllable device and issued for execution.

[0026] During the state deduction process performed by the reinforcement learning agent in the digital twin environment, the hydrodynamic equation built into the physical consistency verification module serves as a constraint term in the reward function of the reinforcement learning agent, ensuring the physical feasibility of the hypothetical scheduling result.

[0027] The specific data acquisition process of the data acquisition module is as follows: the water quality monitoring data and hydrological monitoring data of the target water body are synchronously acquired through the Internet of Things sensor network at a preset acquisition frequency. The water quality monitoring data includes at least one of pH value, dissolved oxygen concentration, chemical oxygen demand, and ammonia nitrogen concentration. The hydrological monitoring data includes at least one of water flow velocity, water level, riverbed slope, and diffusion coefficient. The acquired data is preprocessed by time alignment and outlier removal. The preprocessed water quality monitoring data is then transmitted to the physical coupling prediction module, and the preprocessed hydrological monitoring data is transmitted to the physical consistency verification module.

[0028] The specific process for the decision output module to output the comprehensive interpretation report is as follows: extract the spatial distribution information and time arrival information of pollution diffusion from the prediction results, extract the spatial distribution information and time arrival information of pollution transmission from the theoretical values, compare the corresponding information of the prediction results with the corresponding information of the theoretical values ​​item by item, generate a comparison result containing consistent item labels and different item labels, embed the deviation value and the threshold interval information of the deviation value into the comparison result, and format the embedded comparison result into the comprehensive interpretation report according to the preset report template and output it.

[0029] The specific process for the decision output module to generate the physical conflict alarm is as follows: In response to the judgment result that the deviation value is greater than the second threshold, the complete deviation data between the prediction result and the theoretical value is obtained, the hydrological monitoring data and hydrodynamic equation calculation steps on which the theoretical value is based are extracted to generate calculation basis description information, the deviation value and the calculation basis description information are integrated into physical conflict alarm content, and the physical conflict alarm content is pushed to the interactive interface in a highlighted form, and the physical conflict alarm content is written to the system log simultaneously.

[0030] It also includes a feedback correction module. The specific process of the feedback correction module in performing decision feedback and model correction is as follows: record the actual monitoring results after each output of the decision output module, calculate the deviation between the actual monitoring results and the prediction results or theoretical values ​​output by the decision output module, and when the deviation is continuously greater than the preset correction threshold, feed the deviation data as a negative feedback signal to the physical coupling prediction module and the physical consistency verification module respectively.

[0031] The physical coupling prediction module adjusts the constraint weights of the hydrodynamic equation as a physical constraint according to the negative feedback signal, and the physical consistency verification module adjusts at least one physical parameter in the hydrodynamic equation according to the negative feedback signal.

[0032] (III) Beneficial Effects

[0033] (1) The present invention sets up a dual-path independent computing architecture of physical coupling prediction module and physical consistency verification module, so that the AI ​​prediction results can be output only after being cross-verified by physical laws; when the AI ​​prediction deviates significantly from the calculation results of hydrodynamic equations, the system automatically intercepts the AI ​​output and downgrades it to a physical deterministic conclusion, effectively avoiding the problem of AI outputting prediction results that violate physical laws under the conditions of abnormal sensor data or model mismatch, and greatly reducing the risk of false alarms and false alarms.

[0034] (2) This invention calculates the deviation between the confidence level of AI prediction and the confidence level of the theoretical value of the physical model, and executes a three-level differentiated output strategy according to the threshold range of the deviation. That is, when there is a high degree of consistency, the output is fused; when there is a moderate deviation, the output is weighted and fused with a manual review prompt; when there is a serious conflict, the AI ​​output is blocked and a physical conflict alarm is issued. This hierarchical processing mechanism enables decision-makers to clearly judge the degree of conflict and take corresponding measures when faced with contradictory information, effectively solving the shortcomings of the existing system in lacking deviation quantification comparison and hierarchical processing mechanism.

[0035] (3) The present invention continuously records the deviation between the actual monitoring results and the prediction results through the feedback correction module. When the deviation is greater than the preset correction threshold, the deviation data is fed back to the physical coupling prediction module and the physical consistency verification module as a negative feedback signal, and the physical constraint weight and the physical parameters of the hydrodynamic equation are adjusted respectively. The system continuously self-calibrates during continuous operation, which solves the defects of the existing scheduling decision open-loop operation and inability to continuously optimize its own performance.

[0036] (4) The dynamic interpretation and adaptation module of the present invention can automatically adjust the granularity and presentation of the interpretation output according to the user role type. Operators receive technical detail reports, management decision-makers receive decision-oriented reports, and regulatory agencies receive compliance-oriented reports. At the same time, the comprehensive interpretation report output by the decision output module includes item-by-item comparison of the prediction results and theoretical values, deviation values ​​and threshold range information, and physical conflict alarms come with their own calculation basis descriptions, which solves the shortcomings of existing interpretability methods and model reasoning processes being independent of each other and the interpretation output form being singular. Attached Figure Description

[0037] Figure 1 This is a schematic block diagram of the overall system architecture of the present invention;

[0038] Figure 2 This is a flowchart of the decision output module in this invention;

[0039] Figure 3 This is a flowchart of the counterfactual deduction scheduling module in this invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Please see Figures 1 to 3 As shown, the embodiments of the present invention provide the following technical solutions:

[0042] A digital water pollution monitoring system based on the Internet of Things (IoT) includes a data acquisition module, a physical coupling prediction module, a physical consistency verification module, a decision output module, a dynamic interpretation and adaptation module, a counterfactual inference scheduling module, and a feedback correction module. For details, please refer to [link to relevant documentation]. Figure 1 The following provides a detailed explanation of each module.

[0043] 1. Data acquisition module, used to collect water quality monitoring data and hydrological monitoring data of the target water body in real time through the Internet of Things sensor network;

[0044] The specific data acquisition process of the data acquisition module is as follows: Water quality monitoring data and hydrological monitoring data of the target water body are synchronously collected through the Internet of Things sensor network at a preset acquisition frequency. The water quality monitoring data includes at least one of pH value, dissolved oxygen concentration, chemical oxygen demand, and ammonia nitrogen concentration. The hydrological monitoring data includes at least one of water flow velocity, water level, riverbed slope, and diffusion coefficient. The collected data is preprocessed by time alignment and outlier removal. The preprocessed water quality monitoring data is then transmitted to the physical coupling prediction module, and the preprocessed hydrological monitoring data is transmitted to the physical consistency verification module.

[0045] In one embodiment, the preset acquisition frequency can be set according to actual monitoring needs, such as once every 5 minutes, once every 10 minutes, or once every 30 minutes. Those skilled in the art can flexibly adjust it according to factors such as the pollution sensitivity of the target water body, sensor power consumption, and network bandwidth. This is only an example and not a limitation. Time alignment refers to aligning data collected by different sensors within the same time window according to a unified timestamp to ensure the consistency of different data sources in the time dimension in subsequent calculations. Outlier removal can be carried out using statistical methods, such as identifying and removing abnormal data points that are significantly outside the normal range based on the Laida criterion or interquartile range method, to avoid the adverse effects of occasional sensor failures or communication interference on subsequent predictions and verifications.

[0046] 2. The physical coupling prediction module has built-in hydrodynamic equations as physical constraints, which are used to generate water pollution diffusion prediction results and corresponding first confidence levels based on water quality monitoring data.

[0047] The specific process of the physical coupling prediction module with built-in hydrodynamic equations as physical constraints is as follows: acquire water quality monitoring data collected in real time by the data acquisition module, embed the hydrodynamic equations into the training objective of the physical coupling prediction model in the form of loss function terms, so that the output of the physical coupling prediction model is constrained by the hydrodynamic equations during the training process, and perform forward inference on the current water quality monitoring data through the trained physical coupling prediction model to generate water pollution diffusion prediction results and corresponding first confidence levels.

[0048] Specifically, the physical coupling prediction model can employ a physical information neural network architecture. Its core idea is to use the residuals of the hydrodynamic equations as part of the loss function, jointly guiding model training with the data loss term. This is based on pollutant concentration... Taking prediction as an example, in addition to minimizing the error between the predicted and monitored values ​​during model training, it is also necessary to minimize the residuals of the hydrodynamic equations at several configuration points, so that the model learns a mapping relationship that conforms to physical laws. The aforementioned hydrodynamic equations can be represented by a one-dimensional convection-diffusion equation, specifically: ;in, For pollutant concentration, For time, Spatial coordinates along the direction of water flow, For water flow velocity, The diffusion coefficient is... The reaction rate constant is... The source term is denoted as . The specific form of this equation and the numerical solution method are well known to those skilled in the art, and will not be elaborated upon here.

[0049] The first confidence score is used to quantitatively assess the reliability of the physically coupled prediction model's prediction results. In one embodiment, the first confidence score can be calculated based on a measure of the uncertainty of the model's prediction results, such as obtaining the prediction distribution using methods like Monte Carlo dropout or deep ensemble learning, and then calculating the confidence score. It should be understood that the above confidence score calculation method is merely exemplary, and those skilled in the art can choose an appropriate confidence score calculation method based on the actual model architecture and uncertainty quantification method used; no limitation is made here.

[0050] 3. Physical consistency verification module, which has built-in hydrodynamic equations independent of the physical coupling prediction module, is used to independently calculate the theoretical value of pollution transmission and the corresponding second confidence level based on hydrological monitoring data;

[0051] The specific process for the physical consistency verification module to independently calculate the theoretical value of pollution transmission using the built-in hydrodynamic equation is as follows: acquire the hydrological monitoring data collected in real time by the data acquisition module, input the hydrological monitoring data into the hydrodynamic equation for forward evolution calculation, and obtain the theoretical value of pollution transmission and the corresponding second confidence level.

[0052] Unlike the physical coupling prediction module, the physical consistency verification module does not involve any machine learning or data-driven models. Instead, it relies purely on numerical solutions based on classical hydrodynamic equations. Specifically, it uses hydrological monitoring data such as current flow velocity, water level, riverbed slope, and diffusion coefficient as initial and boundary conditions, substituting them into the hydrodynamic equations for forward time-evolution numerical calculations to obtain the theoretical transport path, arrival time, and concentration distribution of pollutants in the target water body. This hydrodynamic equation is of the same type as the hydrodynamic equation built into the physical coupling prediction module (e.g., both use the aforementioned one-dimensional convection-diffusion equation), but their calculation paths are completely independent: the former uses water quality monitoring data as input and performs predictions through a neural network, while the latter uses hydrological monitoring data as input and solves the problem using pure numerical methods. This provides two independent sources of information for subsequent cross-validation.

[0053] The second confidence level is used to characterize the reliability of the theoretical values ​​calculated by the physical consistency verification module based on the hydrodynamic equations. In one embodiment, the second confidence level can be determined comprehensively based on the quality of the hydrological monitoring data and the convergence of the numerical solution. For example, when the hydrological monitoring data is complete, the sensor accuracy is high, and the numerical solution converges quickly, the second confidence level is higher; otherwise, it is lower. It should be understood that the above confidence level calculation method is only exemplary, and those skilled in the art can choose an appropriate confidence level calculation method according to the actual hydrological monitoring methods and numerical solution methods used, which is not limited here.

[0054] 4. The decision output module calculates the deviation between the first confidence level and the second confidence level, and executes the corresponding output strategy based on the threshold range in which the deviation falls. Please refer to [link / reference] for details. Figure 2 :

[0055] When the deviation value is less than or equal to the first threshold, a comprehensive explanation report of the fused prediction results and theoretical values ​​is output.

[0056] When the deviation value is greater than the first threshold and less than or equal to the second threshold, the weighted fusion result is output along with a manual review prompt.

[0057] When the deviation value is greater than the second threshold, the prediction result is intercepted, and only a deterministic conclusion based on the theoretical value is output and a physical conflict alarm is generated, where the second threshold is greater than the first threshold.

[0058] The specific process for the decision output module to calculate the deviation value and output the graded results is as follows: Obtain the prediction result and first confidence level output by the physical coupling prediction module, and obtain the theoretical value and second confidence level output by the physical consistency verification module. Calculate the deviation value between the first confidence level and the second confidence level using the following formula: (1); In equation (1), As the first confidence level, As the second confidence level, The deviation value is defined by the formula as follows: the deviation value equals the absolute value of the difference between the AI ​​prediction confidence level and the theoretical confidence level of the physical model. A deviation value closer to 0 indicates a greater consistency between the AI ​​prediction and physical laws, while a deviation value closer to 1 indicates a more severe conflict between the two. The deviation value is then compared sequentially with the first and second thresholds to determine the threshold range in which the deviation value falls. When the deviation value is less than or equal to the first threshold, i.e. When the predicted results are found to be highly consistent with the theoretical values, the predicted results and theoretical values ​​are merged to generate a comprehensive interpretation report, which is then output. The first threshold is defined as the threshold value; when the deviation value is greater than the first threshold and less than or equal to the second threshold, i.e. When a moderate deviation is found between the predicted result and the theoretical value, a weighted fusion is performed on the predicted result and the theoretical value. The weighted fusion result, along with a manual review prompt, is then output. The second threshold is used; when the deviation value is greater than the second threshold, that is... If the prediction result is found to be in serious conflict with the theoretical value, the prediction result will be intercepted and output will be prohibited. Only the deterministic conclusion based on the theoretical value will be output along with the physical conflict alarm.

[0059] In one embodiment, the first threshold For example, a value of 0.1 can be used as the second threshold. A value of 0.3 can be used as an example. It should be understood that the above threshold value is merely exemplary; in practical applications, ... and The specific values ​​can be flexibly adjusted based on factors such as the pollution sensitivity of the target water body, the quality of the monitoring data, and the system's tolerance for false alarms and false negatives. For example, for high-risk areas such as drinking water sources, the values ​​can be adjusted accordingly. and The threshold can be set smaller to improve the system's sensitivity and conservatism. Those skilled in the art can adaptively set the above threshold according to the actual application scenario, and no limitation is made here.

[0060] The specific process for the decision output module to output a comprehensive interpretation report is as follows: extract the spatial distribution information and time arrival information of pollution diffusion from the prediction results, extract the spatial distribution information and time arrival information of pollution transmission from the theoretical values, compare the corresponding information of the prediction results with the corresponding information of the theoretical values ​​item by item, generate a comparison result containing consistent item annotations and difference item annotations, embed the deviation value and the threshold interval information of the deviation value into the comparison result, and format the embedded comparison result into a comprehensive interpretation report according to the preset report template and output it.

[0061] In one embodiment, the comprehensive interpretation report may include, but is not limited to, the following: a comparison between predicted and theoretical values ​​of pollution source location and intensity; a comparison between predicted and theoretical values ​​of pollutant diffusion paths and ranges; a comparison between predicted and theoretical values ​​of the estimated time for pollutants to reach downstream sensitive points; annotations of the consistent and differing items; and deviation values ​​and their corresponding threshold ranges. Pre-defined report templates can be predefined according to different application scenarios. For example, the comparison results can be presented in a combination of tables and text descriptions, or a map overlay can be used to simultaneously display the spatial distribution of AI prediction results and physical model theoretical values ​​on a GIS map of the target water body, allowing users to intuitively compare the similarities and differences between the two.

[0062] The specific process for the decision output module to generate physical conflict alarms is as follows: In response to the judgment result that the deviation value is greater than the second threshold, the complete deviation data between the prediction result and the theoretical value is obtained, the hydrological monitoring data and hydrodynamic equation calculation steps on which the theoretical value is based are extracted to generate calculation basis description information, the deviation value and the calculation basis description information are integrated into physical conflict alarm content, and the physical conflict alarm content is pushed to the interactive interface in a highlighted form, and the physical conflict alarm content is written to the system log simultaneously.

[0063] In one embodiment, the highlighting method may include, but is not limited to, presenting physical conflict alarm content in the interactive interface through methods such as red highlighting, pop-up prompts, and audible and visual alarms; the calculation basis description information shall at least include: specific values ​​of hydrological monitoring data such as current water flow velocity, water level, riverbed slope, and diffusion coefficient, the type of hydrodynamic equation used, and the key steps and results of numerical solution. By presenting the above calculation basis together, users can independently verify the correctness of the physical model calculation results, thereby enhancing the objectivity and credibility of the alarm.

[0064] 5. Dynamic Interpretation and Adaptation Module: The specific process of the dynamic interpretation and adaptation module to perform adaptive interpretation output is as follows: obtain the output results of the decision output module, identify the current user's role type, which includes operator role, management decision-maker role, and regulatory agency role;

[0065] When identified as an operator, the output is converted into a technically detailed explanatory report containing specific data curves, anomaly point markings, and executable operation steps; when identified as a management decision-maker, the output is converted into a decision-oriented explanatory report containing risk level assessment, pollution trend prediction, and resource allocation recommendations; when identified as a regulatory agency, the output is converted into a compliance-oriented explanatory report containing a legal basis index, monitoring evidence chain, and confidence quantification report, and the converted explanatory report is output to the current user's interactive interface in the corresponding format.

[0066] User role type identification can be achieved in various ways, such as: pre-setting corresponding role permission identifiers in the system database based on the user's login account; or automatically identifying based on the type of terminal the user is currently accessing. The above identification methods are merely illustrative, and those skilled in the art can choose appropriate user role identification methods according to the actual system architecture, which are not limited here.

[0067] 6. Counterfactual Inference Scheduling Module: The specific process of the counterfactual inference scheduling module in executing hypothetical scheduling instruction inference and decision support is as follows: Responding to an externally input hypothetical scheduling instruction, which contains target operation parameters for at least one controllable device in the target water body within a specified time window, the module obtains the current baseline state confirmed by the physical consistency verification module. The baseline state is the state confirmed by the physical consistency verification module to have passed the physical consistency check after forward evolution calculation using hydrodynamic equations. The baseline state includes current water quality monitoring data, hydrological monitoring data, and theoretical values. The module clones the current baseline state in the digital twin environment to generate the initial inference state. Based on the initial inference state, the target operation parameters in the hypothetical scheduling instruction are used as action inputs to drive the reinforcement learning agent to perform state inference in the digital twin environment, generating the hypothetical scheduling result. For details, please refer to [link to relevant documentation]. Figure 3 .

[0068] Controllable equipment may include, but is not limited to: controllable water conservancy facilities involved in water pollution monitoring and treatment scenarios such as sluice gates, pumping stations, valves, and inlets of sewage treatment facilities; target operating parameters may include, but are not limited to: the on or off status of the equipment, the percentage of opening, the operating power, the flow regulation value, etc.; the specified time window refers to the time interval during which the scheduling instruction is assumed to take effect, for example, within the next 2 hours.

[0069] A digital twin environment is a virtual mapping of a target water body, containing a built-in physical model based on the aforementioned hydrodynamic equations. It can simulate the temporal evolution of the water body's state under different input conditions. Reinforcement learning agents employ reinforcement learning algorithms (such as deep Q-networks, proximal policy optimization, or the Actor-Critic algorithm) to perform decision-making learning within the digital twin environment. Their goal is to optimize a specified scheduling objective while satisfying physical constraints. The basic framework of reinforcement learning is as follows: the agent interacts with the environment, selects an action based on the current state, the environment returns a reward signal, and the agent transitions to the next state. This is well-known to those skilled in the art, and the specific training details of the algorithms will not be elaborated here.

[0070] The hypothetical scheduling result is compared with the baseline state at each monitoring point using a differential comparison. The decision deviation value for each monitoring point is calculated using the following formula: (2); In equation (2), The total number of monitoring points, For the first Predicted values ​​for each monitoring point under the assumed scheduling outcome. For the first The theoretical concentration values ​​of each monitoring point under baseline conditions. For the first The decision deviation value of each monitoring point; the overall decision deviation cost is calculated according to the following formula: The meaning of this formula is: for each monitoring point, calculate the absolute difference between the predicted value after implementing the hypothetical scheduling and the baseline theoretical value without implementing the scheduling. The larger the difference, the greater the impact of the hypothetical scheduling on the point.

[0071] (3); In equation (3), For the first Preset weighting coefficients for each monitoring point. Cost of overall decision-making bias; weighting coefficient This is used to reflect the importance of different monitoring points in the overall decision-making assessment. In one embodiment, monitoring points located downstream of drinking water intakes, in ecologically sensitive areas, or near densely populated areas can be assigned higher weighting coefficients, while monitoring points located in general industrial water use areas or non-sensitive areas can be assigned lower weighting coefficients. The above weighting method can be flexibly adjusted according to the actual application scenario and the importance of the protection target, and is not limited here.

[0072] The decision deviation value of each monitoring point The results are output to the interactive interface in the form of a heatmap. The heatmap is used to display the quantitative differences between the hypothetical scheduling results and the baseline state at each monitoring point. Specifically, the heatmap can use color depth or color gradation to represent the magnitude of the difference at each monitoring point. The larger the difference, the darker the color, and the smaller the difference, the lighter the color, so that users can intuitively identify the areas most significantly affected by the hypothetical scheduling. At the same time, the specific values ​​of each monitoring point and the overall deviation cost can be overlaid on the heatmap to allow users to comprehensively evaluate the potential impact of the hypothetical scheduling.

[0073] In response to external confirmation commands, the system translates the target operating parameters in the hypothetical scheduling commands into actual control commands for the controllable equipment and issues them for execution. External confirmation commands are affirmative commands issued by the user through the interactive interface after reviewing the heatmap and decision deviation cost information. Upon receiving such a command, the system converts the previously hypothetical scheduling parameters into actual control signals executable by the equipment and issues them to the corresponding controllable equipment for execution via the industrial control network.

[0074] During the state inference process performed by the reinforcement learning agent in the digital twin environment, the hydrodynamic equation built into the physical consistency verification module serves as a constraint term for the reward function of the reinforcement learning agent, which is used to ensure the physical feasibility of the hypothetical scheduling result.

[0075] Specifically, the total reward of the reinforcement learning agent The function consists of two parts: task-oriented reward and physically constrained reward. (4); In equation (4), This is the current state. Actions taken by the agent Task-oriented rewards This is a physical constraint reward item.

[0076] (5); In equation (5), It is a hydrodynamic differential operator (consistent with the hydrodynamic equations built into the physical consistency verification module). In the current state and actions The predicted physics field values ​​from the pushdown performance. For source terms, This is the physical constraint weighting coefficient (a positive number), used to control the influence of physical constraints on the total reward. The specific value can be adjusted based on the expected importance of physical constraints in the total reward. This formula means that the greater the deviation of the agent's deduced result from the hydrodynamic equations, the higher the reward. The larger the negative value, the heavier the penalty, thus guiding the agent to prioritize scheduling actions that conform to physical laws.

[0077] Among them, the task-oriented reward is designed based on task objectives such as pollutant interception rate and energy saving; the physical constraint reward is designed as follows: the greater the deviation of the result deduced by the agent from the hydrodynamic equation, the greater the negative reward is given, thereby guiding the agent to prioritize scheduling actions that conform to physical laws.

[0078] 7. Feedback Correction Module: The specific process of the feedback correction module for decision feedback and model correction is as follows: record the actual monitoring results after each output of the decision output module, calculate the deviation between the actual monitoring results and the predicted results or theoretical values ​​output by the decision output module, and when the deviation continues to exceed the preset correction threshold, feed the deviation data as a negative feedback signal to the physical coupling prediction module and the physical consistency verification module respectively. Specifically, single deviation It can be represented as: (6); In equation (6), For a moment The actual monitoring value, For a moment The predicted or theoretical value. To determine whether the deviation continues to exceed the preset correction threshold, a cumulative deviation determination method within a time window can be used: when the cumulative deviation within the most recent time window exceeds the preset correction threshold, feedback correction is triggered. In one embodiment, the time window can be exemplarily set to the most recent 24 hours, the most recent 7 days, or the most recent 30 days, and the specific value can be flexibly determined according to the hydrological change cycle of the target water body and the sampling frequency of the monitoring data; the preset correction threshold can be exemplarily determined according to the statistical distribution of historical monitoring data, for example, set to 2 or 3 times the standard deviation of the historical average deviation. It should be understood that the above values ​​are only exemplary, and those skilled in the art can make adaptive adjustments according to the actual data characteristics, which are not limited here.

[0079] The physical coupling prediction module adjusts the constraint weights of the hydrodynamic equations as physical constraints based on the negative feedback signal. The constraint weights can be adjusted in the following ways: (7); In equation (7), The physical constraint weights before adjustment. The adjusted physical constraint weights, The learning rate (used to control the step size of each adjustment). This represents the gradient of the total loss function with respect to the physical constraint weights. The meaning of this formula is: update the weights based on the sensitivity of the current total loss to the constraint weights, gradually adjusting the model to a better state.

[0080] The physical consistency verification module adjusts at least one physical parameter in the hydrodynamic equations based on the negative feedback signal. This parameter is the diffusion coefficient. Taking the adjustment as an example, the following update method can be adopted: (8); In equation (8), The diffusion coefficient before adjustment. The adjusted diffusion coefficient is... To update the step size, This represents the current deviation. The formula means that the diffusion coefficient is adjusted forward or backward based on the magnitude of the current prediction deviation, so that the output of the physical model gradually approximates the actual monitoring results.

[0081] The following example illustrates the monitoring and handling process of a complete pollution incident.

[0082] The data acquisition module continuously collects water quality and hydrological monitoring data from various monitoring points at a preset acquisition frequency through an IoT sensor network deployed in the target water body. The acquired data undergoes time-series alignment and outlier removal preprocessing. The preprocessed water quality monitoring data is then routed to the physical coupling prediction module, while the hydrological monitoring data is routed to the physical consistency verification module.

[0083] After receiving water quality monitoring data, the physical coupling prediction module performs forward extrapolation using a trained physical coupling prediction model to generate water pollution diffusion prediction results and corresponding first confidence levels. Simultaneously, the physical consistency verification module receives hydrological monitoring data, inputs it into hydrodynamic equations for forward evolution calculations, and obtains theoretical pollution transport values ​​and corresponding second confidence levels. These two modules execute in parallel using the same type of hydrodynamic equations but completely independent calculation paths, without interfering with each other.

[0084] The decision output module obtains the first confidence level and the second confidence level, and then... The deviation between the two values ​​is calculated. Then, the deviation is compared with a first threshold and a second threshold: when the deviation is less than or equal to the first threshold, a high degree of consistency is determined, and a comprehensive explanation report of the fused prediction result and the theoretical value is output; when the deviation is greater than the first threshold but less than or equal to the second threshold, a moderate deviation is determined, and a weighted fusion result is output with a manual review prompt; when the deviation is greater than the second threshold, a serious conflict is determined, the system automatically intercepts the prediction result, outputs only a deterministic conclusion based on the theoretical value, and generates a physical conflict alarm.

[0085] The dynamic interpretation and adaptation module obtains the output results from the decision output module, automatically identifies the current user's role type, and converts the output results into an interpretation report in the corresponding format based on the identification results, which is then pushed to the user interaction interface.

[0086] When a user wishes to assess the potential impact of a hypothetical scheduling scheme, they can input hypothetical scheduling commands into the counterfactual scheduling module via an interactive interface. The counterfactual scheduling module obtains the baseline state currently confirmed by the physical consistency verification module, clones this baseline state in the digital twin environment, executes the simulation, and generates a heatmap by performing a differential comparison of the simulation results with the baseline state at each monitoring point. This heatmap allows the user to assess the cost of decision deviation and confirm whether to execute the plan. After user confirmation, the counterfactual scheduling module translates the target operating parameters in the hypothetical scheduling command into actual control commands for the controllable equipment and issues them for execution.

[0087] After the decision output module outputs its results, the feedback correction module continuously records the actual monitoring results at each monitoring point. When the deviation between the actual monitoring results and the predicted or theoretical values ​​continues to exceed a preset correction threshold, the feedback correction module sends the deviation data as a negative feedback signal to both the physical coupling prediction module and the physical consistency verification module. The physical coupling prediction module adjusts the constraint weights of the physical constraints accordingly, and the physical consistency verification module adjusts the physical parameters in the hydrodynamic equations, thereby enabling continuous self-calibration and performance evolution of the system during continuous operation.

[0088] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0090] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A digital water pollution monitoring system based on the Internet of Things, characterized in that, include: The data acquisition module is used to collect water quality monitoring data and hydrological monitoring data of the target water body in real time through an Internet of Things sensor network; The physical coupling prediction module has a built-in hydrodynamic equation as a physical constraint, which is used to generate water pollution diffusion prediction results and corresponding first confidence levels based on the water quality monitoring data. The physical consistency verification module has a built-in hydrodynamic equation independent of the physical coupling prediction module, which is used to independently calculate the theoretical value of pollution transmission and the corresponding second confidence level based on the hydrological monitoring data. The decision output module is used to calculate the deviation between the first confidence level and the second confidence level, and to execute the corresponding output strategy based on the threshold range in which the deviation value falls: When the deviation value is less than or equal to the first threshold, a comprehensive explanation report that integrates the prediction result and the theoretical value is output. When the deviation value is greater than the first threshold and less than or equal to the second threshold, the weighted fusion result is output along with a manual review prompt. When the deviation value is greater than the second threshold, the prediction result is intercepted, and only a deterministic conclusion based on the theoretical value is output and a physical conflict alarm is generated, wherein the second threshold is greater than the first threshold.

2. The IoT-based digital water pollution monitoring system according to claim 1, characterized in that, The specific process of the physical coupling prediction module embedding the hydrodynamic equation as a physical constraint is as follows: acquire the water quality monitoring data collected in real time by the data acquisition module, embed the hydrodynamic equation into the training objective of the physical coupling prediction model in the form of a loss function term, so that the output of the physical coupling prediction model is constrained by the hydrodynamic equation during the training process, and perform forward inference on the current water quality monitoring data through the trained physical coupling prediction model to generate water pollution diffusion prediction results and corresponding first confidence levels. The specific process of the physical consistency verification module independently calculating the theoretical value of pollution transmission using the built-in hydrodynamic equation is as follows: obtain the hydrological monitoring data collected in real time by the data acquisition module, input the hydrological monitoring data into the hydrodynamic equation for forward evolution calculation, and obtain the theoretical value of pollution transmission and the corresponding second confidence level.

3. The IoT-based digital water pollution monitoring system according to claim 1, characterized in that, The specific process for the decision output module to perform deviation value calculation and hierarchical output is as follows: Obtain the prediction result and the first confidence level output by the physical coupling prediction module, and obtain the theoretical value and the second confidence level output by the physical consistency verification module. Calculate the deviation value between the first confidence level and the second confidence level according to the following formula: ； in, This represents the first confidence level. This is the second confidence level. The deviation value is then compared sequentially with the first threshold and the second threshold to determine the threshold range in which the deviation value falls. When the deviation value is less than or equal to the first threshold, i.e. When the predicted result is determined to be highly consistent with the theoretical value, the predicted result and the theoretical value are fused to generate a comprehensive explanation report and output, wherein... The first threshold is defined as the threshold value; when the deviation value is greater than the first threshold and less than or equal to the second threshold, i.e. If a moderate deviation is found between the predicted result and the theoretical value, a weighted fusion is performed on the predicted result and the theoretical value. The weighted fusion result, along with a manual review prompt, is then output. The second threshold is used; when the deviation value is greater than the second threshold, i.e. If the predicted result is found to be in serious conflict with the theoretical value, the predicted result is intercepted and output is prohibited. Only the deterministic conclusion based on the theoretical value is output along with the physical conflict alarm.

4. The IoT-based digital water pollution monitoring system according to claim 1, characterized in that, It also includes a dynamic interpretation and adaptation module. The specific process of the dynamic interpretation and adaptation module to perform adaptive interpretation output is as follows: obtain the output result of the decision output module, identify the current user's role type, which includes operator role, management decision-maker role and regulatory agency role; When identified as an operator, the output is converted into a technically detailed explanatory report containing specific data curves, anomaly point markings, and executable operation steps; when identified as a management decision-maker, the output is converted into a decision-oriented explanatory report containing risk level assessment, pollution trend prediction, and resource allocation recommendations; when identified as a regulatory agency, the output is converted into a compliance-oriented explanatory report containing a legal basis index, monitoring evidence chain, and confidence quantification report, and the converted explanatory report is output to the current user's interactive interface in the corresponding format.

5. The Internet of Things-based digital water pollution monitoring system according to claim 1, characterized in that, It also includes a counterfactual inference scheduling module. The specific process of the counterfactual inference scheduling module for performing hypothesis scheduling instruction inference and decision support is as follows: In response to the externally input hypothesis scheduling instruction, which includes the target operation parameters of at least one controllable device of the target water body within a specified time window, the baseline state currently confirmed by the physical consistency verification module is obtained. The baseline state is the state confirmed by the physical consistency verification module after forward evolution calculation of the hydrodynamic equation, which has passed the physical consistency test. The baseline state includes the water quality monitoring data, hydrological monitoring data and the theoretical value at the current moment. The baseline state at the current moment is cloned in the digital twin environment to generate the initial state of inference. Based on the initial state of inference, the target operation parameters in the hypothesis scheduling instruction are used as action input to drive the reinforcement learning agent to perform state inference in the digital twin environment and generate the hypothesis scheduling result. The hypothetical scheduling result is compared with the baseline state at each monitoring point using a differential comparison, and the decision deviation value for each monitoring point is calculated according to the following formula: ； in, The total number of monitoring points, For the first The predicted values ​​for each monitoring point under the assumed scheduling result. For the first The theoretical values ​​of each monitoring point under the baseline condition. For the first The decision deviation value of each monitoring point; the overall decision deviation cost is calculated according to the following formula: ； in, For the first Preset weighting coefficients for each monitoring point. The overall decision bias cost; the decision bias value at each monitoring point. The data is output to the interactive interface in the form of a heat map. The heat map is used to display the quantitative difference between the hypothetical scheduling result and the baseline state at each monitoring point. In response to an external confirmation command, the target operation parameters in the hypothetical scheduling command are translated into the actual control command of the controllable device and issued for execution.

6. The Internet of Things-based digital water pollution monitoring system according to claim 5, characterized in that, During the state deduction process performed by the reinforcement learning agent in the digital twin environment, the hydrodynamic equation built into the physical consistency verification module serves as a constraint term for the reward function of the reinforcement learning agent, ensuring the physical feasibility of the hypothetical scheduling result.

7. The Internet of Things-based digital water pollution monitoring system according to claim 1, characterized in that, The specific data acquisition process of the data acquisition module is as follows: the water quality monitoring data and hydrological monitoring data of the target water body are synchronously acquired through the Internet of Things sensor network at a preset acquisition frequency. The water quality monitoring data includes at least one of pH value, dissolved oxygen concentration, chemical oxygen demand, and ammonia nitrogen concentration. The hydrological monitoring data includes at least one of water flow velocity, water level, riverbed slope, and diffusion coefficient. The acquired data is preprocessed by time alignment and outlier removal. The preprocessed water quality monitoring data is then transmitted to the physical coupling prediction module, and the preprocessed hydrological monitoring data is transmitted to the physical consistency verification module.

8. The Internet of Things-based digital water pollution monitoring system according to claim 1, characterized in that, The specific process for the decision output module to output the comprehensive interpretation report is as follows: extract the spatial distribution information and time arrival information of pollution diffusion from the prediction results, extract the spatial distribution information and time arrival information of pollution transmission from the theoretical values, compare the corresponding information of the prediction results with the corresponding information of the theoretical values ​​item by item, generate a comparison result containing consistent item labels and different item labels, embed the deviation value and the threshold interval information of the deviation value into the comparison result, and format the embedded comparison result into the comprehensive interpretation report according to the preset report template and output it.

9. The Internet of Things-based digital water pollution monitoring system according to claim 1, characterized in that, The specific process for the decision output module to generate the physical conflict alarm is as follows: In response to the judgment result that the deviation value is greater than the second threshold, the complete deviation data between the prediction result and the theoretical value is obtained, the hydrological monitoring data and hydrodynamic equation calculation steps on which the theoretical value is based are extracted to generate calculation basis description information, the deviation value and the calculation basis description information are integrated into physical conflict alarm content, and the physical conflict alarm content is pushed to the interactive interface in a highlighted form, and the physical conflict alarm content is written to the system log simultaneously.

10. The Internet of Things-based digital water pollution monitoring system according to claim 1, characterized in that, It also includes a feedback correction module. The specific process of the feedback correction module in performing decision feedback and model correction is as follows: record the actual monitoring results after each output of the decision output module, calculate the deviation between the actual monitoring results and the prediction results or theoretical values ​​output by the decision output module, and when the deviation is continuously greater than the preset correction threshold, feed the deviation data as a negative feedback signal to the physical coupling prediction module and the physical consistency verification module respectively. The physical coupling prediction module adjusts the constraint weights of the hydrodynamic equation as a physical constraint according to the negative feedback signal, and the physical consistency verification module adjusts at least one physical parameter in the hydrodynamic equation according to the negative feedback signal.

Images