A three-source intelligent linkage water pollution accident early warning method and system

Abstract

This invention discloses a three-source intelligent linkage water pollution accident early warning method and system. This invention relates to the field of water pollution accident early warning technology. The pollution monitoring area is divided into three spatial levels: plant level, pipeline level, and river level. Multi-source water quality data is collected, and a change line is plotted using the sliding window moving average method. A mutation signal is generated based on the joint mutation detection of the first and second derivatives. The pollution state identification model outputs a pollution degree value and triggers a graded early warning. The advantages of this invention are: it establishes for the first time an automatic mapping mechanism from early warning level to control instructions. The pollution degree value D output by the pollution state identification model is used as the control level trigger signal. Through a preset mapping relationship, the control area is automatically matched, and a multi-level gate control instruction sequence is generated, completely eliminating the response delay caused by manual judgment and operation, and realizing fully automatic linkage from anomaly detection to gate control response.

Description

Technical Field

[0001] This invention relates to the field of water pollution accident early warning technology, specifically to a three-source intelligent linkage water pollution accident early warning method and system. Background Technology

[0002] With the rapid advancement of industrialization and urbanization, problems such as industrial wastewater discharge, municipal pipeline leakage, and agricultural non-point source pollution have become increasingly prominent. The frequency of sudden water pollution accidents is on the rise. These accidents are characterized by their suddenness, rapid spread, and wide impact. Once they occur, they will not only disrupt the ecological balance of water bodies, leading to the death of aquatic organisms and water quality deterioration, but may also pollute drinking water sources, threaten public health, and cause serious ecological losses and social impacts. Common early warning methods and systems for sudden water pollution incidents often rely on manual analysis of monitoring data. They cannot automatically match prevention and control measures based on the degree of pollution. The pollution status identification results need to be manually interpreted and converted before the gate control command is issued. This process is cumbersome, the response is slow, and human error can easily lead to delays in prevention and control. At the same time, there is a lack of standardized automatic mapping between early warning levels and prevention and control commands, making it difficult to form a coherent linkage between regional control and gate control operations. The overall response efficiency is low and the accuracy is insufficient. It is impossible to achieve integrated closed-loop handling from anomaly detection to gate control execution. To address this, we propose a three-source intelligent linkage method and system for early warning of water pollution incidents. Summary of the Invention

[0003] The purpose of this invention is to provide a three-source intelligent linkage water pollution accident early warning method and system.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a three-source intelligent linkage water pollution accident early warning method, wherein the sudden water pollution accident early warning method includes the following steps: S100, Data Acquisition and Preprocessing: Obtain the water area range for early warning of sudden water pollution accidents, divide the pollution monitoring area within the water area into three levels according to spatial hierarchy: plant level, pipeline level and river level, and detect each level to obtain three-source information; Real-time data on water flow velocity, direction and depth at different cross sections of the river are collected. A two-dimensional water flow field simulation model is constructed based on fluid dynamics equations. Real-time water flow data and topographic data are input into the two-dimensional water flow field simulation model to obtain the water flow status at each monitoring point in the water area. S200, Dynamic Baseline Drawing and Abrupt Change Detection: Denoising is performed on the three-source information, physical parameter values ​​are collected and time series analysis is performed, and the sliding window moving average method is used to smooth each physical parameter. The change line corresponding to each physical parameter is drawn, and the change line is continuously updated in real time during system operation. The first and second derivatives of the change line of each physical parameter are calculated based on the standard time interval, and the first derivative abrupt change threshold and the second derivative abrupt change threshold are set. A mutation signal is determined to have occurred when the absolute value of the first derivative calculated in real time is not less than the first derivative mutation threshold. When the absolute value of the second derivative calculated in real time is not less than the second derivative mutation threshold, a mutation signal is also determined to have been generated. S300: Linking pollution identification, tiered early warning, and differentiated prevention and control: A pollution status identification model is established. When a sudden change signal is generated, the sudden change signal is input into the pollution status identification model. The pollution status identification model outputs the pollution degree value and the pollution status classification result. The pollution status classification includes normal status, suspected pollution status and confirmed pollution status. When the pollution status classification result is suspected pollution status, the early warning framework issues a yellow warning and automatically triggers the first level of prevention and control, activating the emergency pool activation command in the plant. When the pollution status classification result is confirmed as pollution status, the early warning framework issues a red warning and maps the pollution level value to a prevention and control risk coefficient. Based on the prevention and control risk coefficient, it matches the preset multi-level prevention and control areas, automatically generates a multi-level gate control instruction sequence, and activates the gate control facilities in the corresponding prevention and control areas in sequence according to the multi-level gate control instruction sequence. S400, source tracing and dynamic adjustment of prevention and control boundaries: When the early warning framework issues an early warning, the accident tracing unit retrieves the change line and corresponding three-source information from the continuously stored historical change line data, which are located L hours before the warning time. It then analyzes the source of the accident by combining the warning time and water flow velocity, where L is a positive integer. After the mobile sensing device completes the source tracing and positioning, it feeds back the determined pollution source location coordinates to the prevention and control system in real time. The prevention and control system dynamically adjusts the gate control boundaries of each level of prevention and control area according to the pollution source location coordinates, forming a complete closed loop from anomaly detection, graded early warning, differentiated prevention and control to source tracing feedback.

[0005] As a further aspect of the present invention, it also includes source tracing task generation and path planning: the cloud platform continuously runs the water quality prediction model and regularly updates the predicted concentration spatial distribution field according to a preset update cycle; After the early warning framework issues an early warning, based on the predicted concentration spatial distribution field and risk assessment results, the A-satellite path planning algorithm is run to generate a search path waypoint sequence, and the search path waypoint sequence and the predicted concentration spatial distribution field are sent to the mobile sensing device. The water quality prediction model takes historical monitoring data, meteorological data, and hydrological data from three sources as input, and the predicted pollutant concentration values ​​of each grid in the water area as output. The grids in the spatial distribution field of predicted concentration where the predicted pollutant concentration value is higher than the concentration warning value are identified as high-risk areas, and a search path waypoint sequence is generated from the current location of the mobile sensing device through each high-risk area.

[0006] As a further aspect of the present invention, it also includes: Intelligent Search and Bayesian Update: The mobile sensing device navigates according to the search path waypoint sequence, prioritizing areas where the risk assessment value is higher than the preset risk threshold, and simultaneously collects water quality data and uploads it to the cloud during the navigation process; The water area is divided into several grids, and each grid is assigned a prior probability based on the normalized concentration prediction values ​​of each grid in the spatial distribution field of the predicted concentration. For each set of measured data, a likelihood function value is constructed based on the deviation between the measured concentration value and the predicted concentration value of the grid. The prior probability of each grid is multiplied by the likelihood function value and normalized to obtain the posterior probability of each grid. The posterior probability is then used as the prior probability for the next update. Calculate the probability centroid coordinates of the posterior probability distribution, and dynamically adjust the heading and speed of the mobile sensing device based on the probability centroid coordinates; Precise positioning: Calculate the information entropy of the posterior probability distribution, and determine that the posterior probability has converged when the information entropy is lower than a preset convergence threshold; The mobile sensing device aligns its heading with the grid with the highest probability value in the posterior probability distribution. Within the area where the grid is located, switch to concentration gradient tracking mode. The mobile sensing device collects concentration values ​​along multiple sampling directions, calculates the concentration difference in each direction, selects the direction with a positive concentration difference and the first value in the ranking as the next navigation direction, gradually approaches the extreme position of the increasing concentration gradient, determines the location coordinates of the pollution source and uploads them, and generates a pollution hotspot map. When collaborative positioning is required, the scheduling slave device and master device form a spatial triangular configuration. Each device collects concentration values, and the location of the concentration extreme point is determined by calculating the second derivative of the concentration field. The location of the extreme point is then determined as the coordinate of the pollution source location.

[0007] As a further aspect of the present invention, it also includes model feedback optimization: the measured concentration data and corresponding location coordinates collected during this source tracing process, as well as the finally determined pollution source location coordinates, are fed back to the water quality prediction model. An incremental learning algorithm is used to form new training samples from the above data, and local parameters are updated based on the existing parameters of the water quality prediction model, without the need to retrain all historical training data. When the updated water quality prediction model generates a new spatial distribution field of predicted concentration in the next update cycle, it reflects the new information obtained from this source tracing, so that the accuracy of the spatial distribution field of predicted concentration gradually improves with the increase of the number of source tracings.

[0008] As a further aspect of the present invention: the mapping relationship between the pollution level value and the prevention and control risk coefficient is as follows: ; in, The pollution level value output by the pollution state identification model is the pollution level value. , This represents the numerical threshold between suspected and confirmed contamination states. To control the risk factor, ; The preset matching rule for the classification of prevention and control areas is: when At that time, the first-level prevention and control zone is activated, a first-level gate control command sequence is generated, the surrounding pipeline network cutoff gates are closed, and the emergency accident pool within the plant is dispatched to receive incoming water; when At that time, the second-level prevention and control area is activated, and a second-level gate control command sequence is generated on top of the first-level gate control, further closing the main pipe cut-off gate at the intersection of the park's drainage network and the river; when At that time, the third-level prevention and control area is activated, a third gate control command sequence is generated, and the river control gate and downstream interception gate are activated simultaneously to implement basin-wide prevention and control; among which, and All are preset positive real numbers, and satisfy the following conditions: Users have editing capabilities and Numerical permissions.

[0009] As a further aspect of the present invention: the method by which the prevention and control system dynamically adjusts the gate control boundaries of each level of prevention and control area according to the location coordinates of the pollution source is as follows: Calculate the Euclidean distance between each gate control facility in the prevention and control system and the location coordinates of the pollution source, compare the Euclidean distance with the preset graded distance threshold, redefine the boundary range of each level of prevention and control area, and reissue the opening and closing command to the gate control facilities within the boundary range; When the source coordinates deviate from the initial estimated position at the time the warning is issued, the boundary is recalculated in real time, so that the control range gradually tightens as the accuracy of pollution source location improves.

[0010] As a further aspect of the present invention: the pollution state identification model includes an input layer, a hidden layer, and an output layer; The input layer is used to receive the fused feature vectors of various physical parameters and real-time water flow status data; The hidden layer uses a genetic algorithm to globally optimize the initial weights and topology of the backpropagation neural network; The output layer outputs a pollution level value. And according to the preset numerical threshold and Perform state classification, where , and All values ​​are positive real numbers, and users have the authority to edit the stated value thresholds; when When this occurs, it is considered a normal state; when At that time, it was determined to be a suspected contamination state; when At that time, it is determined to be a confirmed pollution state.

[0011] As a further aspect of the present invention: after the range of the three sources is determined, the water level and flow rate in the water storage tank and rainwater tank in the plant level are collected, the conductivity, temperature and flow velocity of the water at different nodes of the park drainage network in the pipeline level are detected, and the turbidity and dissolved oxygen of the water at the river inlet in the river level are detected. The standard time interval is determined as follows: When the user does not set a standard time interval, extract the change lines of each physical parameter in the previous X hours, where X is a positive integer. Extract the peak and trough values ​​of each change line, calculate the difference between the peak and trough values ​​of each change line, obtain the set of peak-trough differences, and count the number M of peak-trough differences. Calculate the average of all peak-trough differences, and combine this with the collection time F of the three-source information for the day to obtain the standard time interval. The calculation formula is: ; in, For the first The peak-to-valley difference of the changing line. The time for collecting information from the three sources on that day. The value range is greater than 0 and does not exceed 24. The index is the sequence number of the peak-valley difference; When the user does not edit the mutation threshold, the peak response value set of the absolute value of the first derivative and the peak response value set of the absolute value of the second derivative within the time window of the accident occurrence in the historical water pollution accident records are extracted. The mean and standard deviation of each set are calculated respectively. The mutation threshold of the first derivative and the mutation threshold of the second derivative are calculated by subtracting the product of the sensitivity coefficient and the standard deviation from the mean. The sensitivity coefficient is set to 1 by default.

[0012] A three-source intelligent linkage water pollution accident early warning system is provided, and the early warning system is applicable to a three-source intelligent linkage water pollution accident early warning method.

[0013] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows: 1. This invention establishes for the first time an automatic mapping mechanism from early warning level to control instructions. The pollution level value D output by the pollution status identification model is used as the trigger signal for the control level. Through the preset mapping relationship, the control area is automatically matched and a multi-level gate control instruction sequence is generated, which completely eliminates the response delay caused by manual judgment and manual operation, and realizes fully automatic linkage from anomaly detection to gate control response. 2. This invention solves the problem of one-size-fits-all prevention and control in the prior art by adopting a differentiated prevention and control strategy. The yellow alert triggers a lightweight in-plant emergency pool response, and the red alert is subdivided into three prevention and control levels according to the severity of pollution and activated level by level. This ensures the effectiveness of prevention and control while avoiding excessive consumption of prevention and control resources. 3. This invention achieves a unified closed loop of four links: early warning, prevention and control, source tracing, and feedback by feeding back the source tracing coordinates of the mobile sensing device to the prevention and control system in real time to dynamically adjust the gate control boundary and feeding back the source tracing measured data to the water quality prediction model for incremental learning. This overcomes the defects of the existing technology in which the subsystems are independent and the data cannot be shared. 4. This invention collects differentiated data from three sources: plant level, pipeline level, and river level, to accurately monitor water quality changes at each pollution stage. Based on a joint mutation detection mechanism of first and second derivatives, it covers two water quality anomaly scenarios: numerical mutation and trend mutation, forming a complete process of detection, analysis, identification, early warning, and prevention. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention. Detailed Implementation

[0015] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0016] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other; This invention discloses a three-source intelligent linkage water pollution accident early warning method and system. Along the Qingchuan River, an important drinking water source in the suburbs of a certain city, there are chemical industrial parks, food processing plants, and small-scale breeding bases. The river not only supplies drinking water to residents along its banks but also provides irrigation water for downstream towns. One early morning, the wastewater treatment facility of a chemical plant in the chemical industrial park suddenly malfunctioned due to equipment aging. Wastewater containing high concentrations of heavy metals and toxic organic matter was directly seeped into the ground without treatment and slowly seeped into the Qingchuan River tributary through soil fissures. At the same time, the manure storage pond of the upstream breeding base collapsed due to heavy rain, and a large amount of nitrogen and phosphorus-containing sewage flowed into the main stream through the rainwater pipe network. Under the superposition of the two pollutions, the water quality of the Qingchuan River deteriorated rapidly. Therefore, in order to effectively solve the above problems, this application proposes a three-source intelligent linkage water pollution accident early warning method, as shown in the attached figures of the specification. Figure 1 As shown; Example 1: Data Acquisition and Preprocessing To obtain the water area for early warning of sudden water pollution incidents in the Qingchuan River Basin, the pollution monitoring area within the water area is divided into three levels according to spatial hierarchy: plant level, pipeline level, and river level. The plant-level coverage includes water storage tanks and rainwater tanks of various enterprises in the chemical industrial park; the pipeline-level coverage includes key nodes of the park's drainage pipeline network; and the river-level coverage includes all sections and inlets of the Qingchuan River. Information from the three sources is obtained by detecting each level. The system collects water level and flow rate data from the plant-level intermediate water storage tank and rainwater tank; detects the conductivity, temperature, and flow velocity of water at different nodes of the park's drainage network at the pipe network level; and detects the turbidity and dissolved oxygen of the water at the river inlet at the river level. Simultaneously, real-time data on water flow velocity, flow direction, and water depth at different cross sections of the river are collected. A two-dimensional water flow field simulation model is constructed based on fluid dynamics equations. Real-time water flow data and topographic data are input into the two-dimensional water flow field simulation model to obtain the water flow status at each monitoring point in the water area. Example 2: Dynamic baseline mapping and mutation detection: The three-source information is denoised, and the physical parameter values ​​are collected and time series analysis is performed. The physical parameter values ​​include the liquid level and flow rate at the plant level, the conductivity, temperature and flow velocity at the pipeline level, and the turbidity and dissolved oxygen at the river level. For each physical parameter, the time series data of the parameter is smoothed by using the sliding window moving average method with time as the horizontal axis and the measured value of the parameter as the vertical axis. The resulting continuous curve is the change line, and each physical parameter corresponds to an independent change line. The calculation method for the sliding window moving average method is as follows: Set the sliding window length to... Each sampling point, for time... A certain physical parameter value The corresponding change line value The calculation formula is: ; in, The sliding window length is the number of consecutive sampling points used in the averaging calculation. It is a positive integer, and the user has the ability to edit. Numerical permissions, when the user has not set them. hour, The default value is taken as the previous value. 10% of the total number of sampling points within the hour, rounded up. For a moment The measured value of the physical parameter after noise reduction. For the current moment, For sampling time index; The change line is continuously updated in real time during system operation; that is, the baseline value at the current moment is recalculated every time a new sampling data point is acquired. This allows the variation line to reflect the fluctuation trend and pattern of the physical parameter over time under normal conditions; Based on standard time intervals Calculate the first derivative of the piecewise linear curve for the variation of each physical parameter. and second derivative The specific calculation formula is as follows: ; ; in, For a moment The change in the broken line value, For a moment The change in the broken line value, For a moment The first derivative value, For a moment The first derivative value, Standard time interval; The standard time interval is determined as follows: when the user does not set a standard time interval, the previous time interval is extracted. Linear graphs showing the changes in various physical parameters over hours ( (For positive integers), extract the peak and trough values ​​of each change line, and calculate the peak value of each change line. Valley value The difference between Obtain the set of peak-valley differences and count the number of peak-valley differences. Calculate the average of all peak-valley differences, and combine this with the data collection time from the three sources on that day. The standard time interval is obtained by calculation using the following formula: ; ; in, For standard time intervals, For the first The peak-to-valley difference of the changing line. Peak value, The valley value, The number of peak-to-valley differences. The time of collection of the three sources of information on that day (the value range is greater than 0 and does not exceed 24). The index is the sequence number of the peak-valley difference; Set a threshold for abrupt changes in the first derivative. and the threshold for the mutation of the second derivative Users have permission to edit the mutation threshold. When the user does not edit the mutation threshold, the set of peak response values ​​of the absolute value of the first derivative within the time window of the accident occurrence in historical water pollution accident records is extracted, and the mean of this set is calculated. with standard deviation : ; ; ; in, The number of historical water pollution incidents recorded. This is the sequence number of the accident record. For the first Peak response value of the absolute value of the first derivative in this accident. The mean of all first-order derivative peak response values. Let the standard deviation be the peak response values ​​of all first derivatives. This is the sensitivity coefficient, with a default value of 1. When the calculation result is less than 0, take ; Second derivative mutation threshold Calculate using the same method as for the first derivative, extracting historical data. The set of peak response values ​​of the absolute value of the second derivative within the corresponding time window of the secondary water pollution incident was used to calculate the mean. with standard deviation : ; ; ; in, For the first Peak response value of the absolute value of the second derivative in this accident. The mean of all second-order derivative peak response values. Let be the standard deviation of all peak response values ​​of the second derivative, when When the calculation result is less than 0, take ; During the real-time operation of the system, when the absolute value of the first derivative... Not less than , or the absolute value of the second derivative Not less than At that time, the physical parameter is determined at time [time]. The generated mutation signal, along with the corresponding physical parameter type, three-source attribution information, and mutation time, is transmitted to the pollution status identification model for further analysis. The first derivative is used to detect abrupt changes in parameter values, and the second derivative is used to detect abrupt inflection points in parameter change trends. The two are used together to cover two water quality anomaly scenarios: numerical mutation and trend mutation, avoiding missed detection by a single derivative. Example 3: Linkage between pollution identification, tiered early warning, and differentiated prevention and control: A pollution state identification model is established, comprising an input layer, a hidden layer, and an output layer. The input layer receives fused feature vectors of various physical parameters and real-time water flow state data. The hidden layer uses a genetic algorithm to globally optimize the initial weights and topology of the backpropagation neural network. The output layer outputs a numerical value indicating the degree of water pollution. ( ), and establish pollution status classification, setting numerical thresholds. (The boundary between normal and suspected contamination) and (The boundary between suspected and confirmed pollution), among which , and All are positive real numbers: when When the condition is normal, no warning is issued. when When the situation is determined to be a suspected pollution state, the early warning framework issues a yellow warning and automatically triggers the first-level prevention and control - activating the emergency pool in the plant, introducing the suspected polluted water into the emergency accident pool to prevent further discharge of sewage. when When the pollution status is confirmed, the early warning framework issues a red alert and generates a multi-level gate control instruction sequence according to the following steps: Will Mapped to risk coefficient for prevention and control : ; in, The pollution level value output by the pollution status identification model. This represents the numerical threshold between suspected and confirmed contamination states. , A higher value indicates a higher degree of pollution severity. according to With preset grading threshold , Size relationship ( (User-editable) Matches control areas and generates corresponding gate control command sequences: when At that time, the first-level prevention and control area is activated, and the first-level gate control command sequence is generated: close the pipe network cutoff gate around the pollution source, and dispatch the emergency accident pool in the plant to receive the incoming water; when At that time, the second-level prevention and control area is activated, and a second-level gate control command sequence is generated on the basis of the first-level gate control: further close the main pipe cut-off gate at the intersection of the park's drainage network and the river; when At that time, the third-level prevention and control area is activated, and the third gate control command sequence is generated: the river control gate and the downstream interception gate are started simultaneously to implement the prevention and control of the entire basin; Taking this scenario as an example: after the mutation signal is input into the pollution state recognition model, the model outputs a pollution level value. ,set up , ,because If the pollution status is confirmed, the early warning framework issues a red alert and calculates the prevention and control risk coefficient. ,set up , ,because Activate the second-level prevention and control area, generate an overlay command sequence containing the first-level and second-level gate control operations, and send it to each gate control facility; Example 4: Source tracing task generation and path planning: The cloud platform continuously runs the water quality prediction model, taking historical monitoring data, meteorological data and hydrological data from three sources as input, and the predicted value of pollutant concentration in each grid in the water area as output. According to the preset update cycle (the update cycle is set to 30 minutes in this embodiment), the platform predicts the pollutant concentration in the water area and generates a predicted concentration spatial distribution field covering the water area. After the early warning framework issues an alert, the cloud platform conducts a risk assessment based on the predicted concentration spatial distribution field, identifies areas where the predicted concentration value is higher than the concentration warning value as high-risk areas, generates a source tracing task, and adopts the A-Star path planning algorithm. On the rasterized water map, the current location of the mobile sensing device is used as the starting point, and each high-risk area is the priority search target. The cost function of the A-Star path planning algorithm consists of two parts: path distance cost and concentration risk cost. The path distance cost represents the actual distance traveled by the mobile sensing device from the current node to the target node, and the concentration risk cost represents the ratio of the predicted concentration value to the concentration warning value at the target node. The higher the predicted concentration value, the lower the concentration risk cost, thereby guiding the mobile sensing device to prioritize searching in high-concentration areas. The search path waypoint sequence and the predicted concentration spatial distribution field are then sent to the mobile sensing device. The accident tracing unit synchronously extracts the change data of each physical parameter and the corresponding three-source information (L is a positive integer) from the historical change line data that has been continuously stored, and backtracks them to the L hours before the warning time. Combined with the warning time and water flow velocity, it conducts a preliminary directional analysis of the accident source and assists in path planning to determine the priority search area. Example 5: Intelligent Search and Bayesian Update The mobile sensing device navigates according to the search path waypoint sequence, prioritizing areas where the risk assessment value is higher than the preset risk threshold, and simultaneously collects water quality data and uploads it to the cloud during the navigation process; The water area is divided into Let the nth grid be the nth grid. The coordinates of each grid are Each grid cell is assigned an initial prior probability based on the normalization of the predicted concentration spatial distribution field. : ; in, To predict the concentration spatial distribution field of the first The predicted pollutant concentration values ​​for each grid cell. The total number of grid cells. Indicates the first Each grid represents an event at the location of a pollution source. For grid index variables; For each set of measured data acquired, a Bayesian update is performed to correct the probability distribution of the pollution source location. This update is applied when the mobile sensing device is at the location... Collected measured concentration values At that time, for each grid Constructing the likelihood function value : ; in, Assuming the pollution source is located in the grid At the measurement position Predicted concentration values ​​at [location] For sensor standard deviation measurement, The likelihood function is a natural exponential function. Based on the Gaussian distribution assumption, the closer the measured concentration value is to the predicted concentration value, the larger the likelihood function value is, indicating that the corresponding grid is more likely to be the location of the pollution source. The updated posterior probability is: ; in, For the first After the second Bayesian update The posterior probability of each grid cell For the first After the first update The probability value of each grid cell. The index of the number of Bayesian updates is used, and the denominator is the normalization constant. This ensures that the sum of the posterior probabilities of all grid cells is 1. The posterior probabilities are then... As the prior probability for the next update; Calculate the centroid coordinates of the updated posterior probability distribution. : ; ; in, Let x be the x-coordinate of the probability centroid. The ordinate of the probability centroid is... and The first The horizontal and vertical coordinates of the grid center, and the probability centroid coordinates represent the expected estimate of the pollution source location. The mobile sensing device dynamically adjusts its course based on the probability centroid coordinates, and adjusts its speed based on the distance between the probability centroid and the current location. Example 6, Precise Positioning: Calculate the information entropy of the posterior probability distribution : ; in, The information entropy of the posterior probability distribution. For the first After the first update The posterior probability of each grid cell It is the natural logarithm function. The total number of grid cells, information entropy The smaller the value, the more concentrated the posterior probability distribution, meaning the lower the uncertainty of the pollution source location; When information entropy Below the preset convergence threshold ( When the value is a positive real number and the user has editing permissions, the posterior probability is determined to converge, the heading of the mobile sensing device is aligned with the grid with the first probability value in the posterior probability distribution, and the precise positioning mode is entered. Within the area of ​​the grid, switch to concentration gradient tracking mode: The mobile sensing device moves one step distance along multiple preset sampling directions (e.g., east, south, west, north, northeast, southeast, southwest, and northwest, a total of eight directions) at the current position and collects concentration values. Calculate the concentration difference in each direction, select the direction with a positive concentration difference and the first value in the ranking as the next navigation direction, and continue to move one step along this direction. Repeat the above process to gradually approach the extreme position of the increasing concentration gradient. When collaborative positioning is required to improve positioning accuracy, the scheduling slave device and master device form a spatial triangular configuration. Each device collects the concentration value at its location, determines the location of the concentration extreme point by calculating the second derivative of the concentration field, determines the location of the extreme point as the coordinate of the pollution source, uploads the location coordinates and generates a pollution hotspot map. In this embodiment, the information entropy is calculated after eight consecutive Bayesian updates. Below the preset convergence threshold Once the probability convergence is determined, the master device switches to concentration gradient tracking mode. After concentration sampling and gradient tracking along eight directions, the slave devices are coordinated to form a triangular configuration for collaborative positioning. Finally, it is determined that the pollution source is located at the intersection of the chemical industrial park drainage network and the Qingchuan River tributary, with coordinates (117.25°E, 31.82°N). Example 7: Source tracing feedback and dynamic adjustment of prevention and control boundaries: Mobile sensing devices determine the location coordinates of pollution sources Then, the coordinates are fed back to the prevention and control system in real time, and the prevention and control system calculates the Euclidean distance between the coordinates of each gate control facility and the location of the pollution source. : ; in, For the first The location coordinates of each gate control facility To determine the location coordinates of the pollution source, the prevention and control system compares the Euclidean distance with the preset graded distance threshold, redefines the boundary range of each level of prevention and control area, and reissues the opening and closing command to the gate control facilities within the boundary range. When the source coordinates deviate from the initial estimated position at the time of the warning, the boundary recalculation is triggered in real time, so that the prevention and control range gradually tightens as the pollution source positioning accuracy improves, and the precise allocation of prevention and control resources is achieved. Example 8, Model Feedback Optimization: The measured concentration data and corresponding sampling location coordinates collected by the mobile sensing devices during this source tracing process, along with the finally determined pollution source location coordinates, will be fed back to the water quality prediction model on the cloud platform. The incremental learning algorithm is used to combine the measured concentration data of this source tracing with the corresponding location coordinates to form new training samples. Based on the existing parameters of the water quality prediction model, local parameters are updated without the need to retrain all historical training data. This incremental update method incorporates new observational information while maintaining the model's existing knowledge, allowing the model parameters to gradually approximate the true concentration distribution pattern. When the updated water quality prediction model generates a new predicted concentration spatial distribution field in the next update cycle, it reflects the new information obtained from this source tracing. As the number of source tracings increases, the accuracy of the predicted concentration spatial distribution field gradually improves, thereby simultaneously improving the efficiency of path planning and the convergence speed of Bayesian updates in the next source tracing task. While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Any variations and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the scope of the invention, fall within the protection scope defined by the claims of the present invention.

Claims

1. A three-source intelligent linkage water pollution accident early warning method, characterized in that, The method for early warning of sudden water pollution accidents includes the following steps: S100, Data Acquisition and Preprocessing: Obtain the water area range for early warning of sudden water pollution accidents, divide the pollution monitoring area within the water area into three levels according to spatial hierarchy: plant level, pipeline level and river level, and detect each level to obtain three-source information; Real-time data on water flow velocity, direction and depth at different cross sections of the river are collected. A two-dimensional water flow field simulation model is constructed based on fluid dynamics equations. Real-time water flow data and topographic data are input into the two-dimensional water flow field simulation model to obtain the water flow status at each monitoring point in the water area. S200, Dynamic Baseline Drawing and Abrupt Change Detection: Denoising is performed on the three-source information, physical parameter values ​​are collected and time series analysis is performed, and the sliding window moving average method is used to smooth each physical parameter. The change line corresponding to each physical parameter is drawn, and the change line is continuously updated in real time during system operation. The first and second derivatives of the change line of each physical parameter are calculated based on the standard time interval, and the first derivative abrupt change threshold and the second derivative abrupt change threshold are set. A mutation signal is determined to have occurred when the absolute value of the first derivative calculated in real time is not less than the first derivative mutation threshold. When the absolute value of the second derivative calculated in real time is not less than the second derivative mutation threshold, a mutation signal is also determined to have been generated. S300: Linking pollution identification, tiered early warning, and differentiated prevention and control: A pollution status identification model is established. When a sudden change signal is generated, the sudden change signal is input into the pollution status identification model. The pollution status identification model outputs the pollution degree value and the pollution status classification result. The pollution status classification includes normal status, suspected pollution status and confirmed pollution status. When the pollution status classification result is suspected pollution status, the early warning framework issues a yellow warning and automatically triggers the first level of prevention and control, activating the emergency pool activation command in the plant. When the pollution status classification result is confirmed as pollution status, the early warning framework issues a red warning and maps the pollution level value to a prevention and control risk coefficient. Based on the prevention and control risk coefficient, it matches the preset multi-level prevention and control areas, automatically generates a multi-level gate control instruction sequence, and activates the gate control facilities in the corresponding prevention and control areas in sequence according to the multi-level gate control instruction sequence. S400, source tracing and dynamic adjustment of prevention and control boundaries: When the early warning framework issues an early warning, the accident tracing unit retrieves the change line and corresponding three-source information from the continuously stored historical change line data, which are located L hours before the warning time. It then analyzes the source of the accident by combining the warning time and water flow velocity, where L is a positive integer. After the mobile sensing device completes the source tracing and positioning, it feeds back the determined pollution source location coordinates to the prevention and control system in real time. The prevention and control system dynamically adjusts the gate control boundaries of each level of prevention and control area according to the pollution source location coordinates, forming a complete closed loop from anomaly detection, graded early warning, differentiated prevention and control to source tracing feedback.

2. The three-source intelligent linkage water pollution accident early warning method according to claim 1, characterized in that, It also includes source tracing task generation and path planning: the cloud platform continuously runs the water quality prediction model and regularly updates the spatial distribution field of predicted concentrations according to the preset update cycle; After the early warning framework issues an early warning, based on the predicted concentration spatial distribution field and risk assessment results, the A-satellite path planning algorithm is run to generate a search path waypoint sequence, and the search path waypoint sequence and the predicted concentration spatial distribution field are sent to the mobile sensing device. The water quality prediction model takes historical monitoring data, meteorological data, and hydrological data from three sources as input, and the predicted pollutant concentration values ​​of each grid in the water area as output. The grids in the spatial distribution field of predicted concentration where the predicted pollutant concentration value is higher than the concentration warning value are identified as high-risk areas, and a search path waypoint sequence is generated from the current location of the mobile sensing device through each high-risk area.

3. The three-source intelligent linkage water pollution accident early warning method according to claim 2, characterized in that: Also includes: Intelligent Search and Bayesian Update: The mobile sensing device navigates according to the search path waypoint sequence, prioritizing areas where the risk assessment value is higher than the preset risk threshold, and simultaneously collects water quality data and uploads it to the cloud during the navigation process; The water area is divided into several grids, and each grid is assigned a prior probability based on the normalized concentration prediction values ​​of each grid in the spatial distribution field of the predicted concentration. For each set of measured data, a likelihood function value is constructed based on the deviation between the measured concentration value and the predicted concentration value of the grid. The prior probability of each grid is multiplied by the likelihood function value and normalized to obtain the posterior probability of each grid. The posterior probability is then used as the prior probability for the next update. Calculate the probability centroid coordinates of the posterior probability distribution, and dynamically adjust the heading and speed of the mobile sensing device based on the probability centroid coordinates; Precise positioning: Calculate the information entropy of the posterior probability distribution, and determine that the posterior probability has converged when the information entropy is lower than a preset convergence threshold; The mobile sensing device aligns its heading with the grid with the highest probability value in the posterior probability distribution. Within the area where the grid is located, switch to concentration gradient tracking mode. The mobile sensing device collects concentration values ​​along multiple sampling directions, calculates the concentration difference in each direction, selects the direction with a positive concentration difference and the first value in the ranking as the next navigation direction, gradually approaches the extreme position of the increasing concentration gradient, determines the location coordinates of the pollution source and uploads them, and generates a pollution hotspot map. When collaborative positioning is required, the scheduling slave device and master device form a spatial triangular configuration. Each device collects concentration values, and the location of the concentration extreme point is determined by calculating the second derivative of the concentration field. The location of the extreme point is then determined as the coordinate of the pollution source location.

4. The three-source intelligent linkage water pollution accident early warning method according to claim 2, characterized in that, It also includes model feedback optimization: the measured concentration data and corresponding location coordinates collected during this source tracing process, as well as the finally determined pollution source location coordinates, are fed back to the water quality prediction model. The incremental learning algorithm is used to form new training samples from the above data, and local parameters are updated based on the existing parameters of the water quality prediction model without having to retrain all historical training data. When the updated water quality prediction model generates a new spatial distribution field of predicted concentration in the next update cycle, it reflects the new information obtained from this source tracing, so that the accuracy of the spatial distribution field of predicted concentration gradually improves with the increase of the number of source tracings.

5. The three-source intelligent linkage water pollution accident early warning method according to claim 1, characterized in that, The mapping relationship between the pollution level and the prevention and control risk coefficient is as follows: ； wherein, a pollution degree value output by the pollution state identification model, , a numerical threshold between the suspected pollution state and the confirmed pollution state, a prevention and control risk coefficient, ; The preset matching rule for the classification of prevention and control areas is: when At that time, the first-level prevention and control zone is activated, a first-level gate control command sequence is generated, the surrounding pipeline network is shut off, and the emergency accident pool within the plant is dispatched to receive the incoming water. At that time, the second-level prevention and control zone is activated, and a second-level gate control command sequence is generated on top of the first-level gate control, further closing the main pipe cut-off gate at the intersection of the park's drainage network and the river. At that time, the third-level prevention and control area is activated, a third gate control command sequence is generated, and the river control gate and downstream interception gate are activated simultaneously to implement basin-wide prevention and control. and All are preset positive real numbers, and satisfy the following conditions: Users have editing capabilities and Numerical permissions.

6. The method for early warning of water pollution accidents using a three-source intelligent linkage system according to claim 1, characterized in that, The method by which the prevention and control system dynamically adjusts the control boundaries of each level of prevention and control area based on the location coordinates of the pollution source is as follows: Calculate the Euclidean distance between each gate control facility in the prevention and control system and the location coordinates of the pollution source, compare the Euclidean distance with the preset graded distance threshold, redefine the boundary range of each level of prevention and control area, and reissue the opening and closing command to the gate control facilities within the boundary range; When the source coordinates deviate from the initial estimated position at the time the warning is issued, the boundary is recalculated in real time, so that the control range gradually tightens as the accuracy of pollution source location improves.

7. The method for early warning of water pollution accidents using a three-source intelligent linkage system according to claim 1, characterized in that: The pollution status identification model includes an input layer, a hidden layer, and an output layer; The input layer is used to receive the fused feature vectors of various physical parameters and real-time water flow status data; The hidden layer uses a genetic algorithm to globally optimize the initial weights and topology of the backpropagation neural network; The output layer outputs a pollution level value. And according to the preset numerical threshold and Perform state classification, where , and All values ​​are positive real numbers, and users have the authority to edit the stated value thresholds; when When this occurs, it is considered a normal state; when At that time, it was determined to be a suspected contamination state; when At that time, it is determined to be a confirmed pollution state.

8. The method for early warning of water pollution accidents using a three-source intelligent linkage system according to claim 1, characterized in that: After the scope of the three sources is determined, the water level and flow rate in the water storage tank and rainwater tank in the plant level are collected, the conductivity, temperature and flow velocity of the water at different nodes of the park drainage network in the pipeline level are detected, and the turbidity and dissolved oxygen of the water at the river inlet in the river level are detected. The standard time interval is determined as follows: When the user does not set a standard time interval, extract the change lines of each physical parameter in the previous X hours, where X is a positive integer. Extract the peak and trough values ​​of each change line, calculate the difference between the peak and trough values ​​of each change line, obtain the set of peak-trough differences, and count the number M of peak-trough differences. Calculate the average of all peak-trough differences, and combine this with the collection time F of the three-source information for the day to obtain the standard time interval. The calculation formula is: ； in, For the first The peak-to-valley difference of the changing line. The time of collection of information from the three sources on that day. The value range is greater than 0 and does not exceed 24. The index is the sequence number of the peak-valley difference; When the user does not edit the mutation threshold, the peak response value set of the absolute value of the first derivative and the peak response value set of the absolute value of the second derivative within the time window of the accident occurrence in the historical water pollution accident records are extracted. The mean and standard deviation of each set are calculated respectively. The mutation threshold of the first derivative and the mutation threshold of the second derivative are calculated by subtracting the product of the sensitivity coefficient and the standard deviation from the mean. The sensitivity coefficient is set to 1 by default.

9. A three-source intelligent linkage water pollution accident early warning system, characterized in that: The aforementioned early warning system is applicable to the three-source intelligent linkage water pollution accident early warning method described in any one of claims 1-8.

Images