A water pollution monitoring system and method for environmental monitoring

Abstract

The application discloses a water pollution monitoring system and method for environmental monitoring, and relates to the technical field of water pollution monitoring.The system comprises a water sample collection module, a water quality detection module and a data monitoring processing module.The water sample collection module is used for acquiring water samples at a fixed point, time and quantity, and completing water sample pretreatment and sampling record.The water quality detection module is used for receiving pretreated water sample data, performing multi-parameter synchronous detection on the water sample, converting water sample characteristics into electrical signals, and outputting original detection data and water quality index detection results.The data monitoring processing module is used for receiving water quality index detection results, taking each water quality parameter as a node, and constructing a causal propagation graph with the causal influence direction and propagation time delay between parameters as directed edges.The application can effectively distinguish between natural fluctuations in the water environment and systematic deviations caused by pollution, greatly reduce invalid false alarms, accurately lock the pollution propagation range, improve the accuracy of water pollution early warning and judgment and the emergency prevention and control capability, and adapt to the fine intelligent environmental protection management needs of the water environment.

Description

Technical Field

[0001] This invention relates to the field of water pollution monitoring technology, specifically to a water pollution monitoring system and method for environmental monitoring. Background Technology

[0002] The aquatic ecological environment is the core carrier of ecological civilization construction and an important foundation for supporting green and low-carbon development and ensuring ecological security and public health. Water environment monitoring is a core and fundamental link in ecological environmental protection, refined water resource management, and pollution source tracing and control. Against the backdrop of my country's vigorous promotion of ecological civilization construction, the implementation of the dual-carbon strategy, and the rapid development of strategic emerging industries such as energy conservation and environmental protection, high-end intelligent equipment, and next-generation information technology, intelligent, precise, and efficient monitoring of the entire water environment has become a core development trend in the modernization of ecological environment governance. With the continuous advancement of industrialization and urbanization, the types of pollutant emissions into water bodies are becoming increasingly complex, and pollution transmission is exhibiting dynamic and cascading characteristics. The risk of sudden water pollution events is frequent, placing higher demands on the real-time, accurate, and intelligent control capabilities of water environment monitoring. Highly efficient, energy-saving, precise, intelligent, and dynamically adaptable water pollution monitoring technologies have become essential technologies in the field of water environment governance.

[0003] Existing water pollution monitoring systems mostly employ traditional methods based on fixed sampling frequencies, static threshold determination, and independent parameter monitoring. These methods lack overall intelligence and precision, making them ill-suited to the industry's development needs for smart water management, environmental protection, and energy conservation and low-carbon practices. Traditional water sampling methods cannot dynamically adjust parameters based on water quality. High-frequency sampling during stable water quality periods wastes electricity and computing resources, failing to meet energy conservation and environmental protection requirements. Insufficient sampling frequency during abnormal water quality periods can lead to missing key pollution data, and the lack of standardized preprocessing and source tracing mechanisms compromises data reliability. Furthermore, existing technologies often analyze water quality parameters independently, neglecting causal relationships, transmission patterns, and propagation delays between parameters. This makes it impossible to construct dynamic correlation models, accurately depict dynamic pollution changes, and result in weak anomaly prediction capabilities, high false alarm rates, and low accuracy. To address these issues, we propose a water pollution monitoring system and method for environmental monitoring. Summary of the Invention

[0004] To address the aforementioned technical problems, a water pollution monitoring system and method for environmental monitoring are provided. This technical solution solves the problems of difficulty in accurately depicting dynamic changes in pollution, weak anomaly prediction capability, high false alarm rate, and low accuracy.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: a water pollution monitoring system for environmental monitoring, the system comprising:

[0006] The water sampling module is used to acquire water samples at fixed points, at fixed times, and in fixed quantities, and to complete water sample pretreatment and sampling records.

[0007] The water quality testing module is used to receive pre-processed water sample data, perform multi-parameter synchronous testing on the water sample, convert water sample characteristics into electrical signals, and output raw test data and water quality index test results.

[0008] The data monitoring and processing module is used to receive water quality index detection results, construct a causal propagation graph with each water quality parameter as a node and the causal influence direction and propagation delay between parameters as directed edges, and generate dynamic prediction intervals for each parameter based on the causal propagation graph; when the real-time detection data deviates from the dynamic prediction interval, the downstream affected parameter set is determined along the forward propagation path of the deviated parameter in the causal propagation graph, the sampling frequency is increased only for this set, and the sampling adjustment command is fed back to the water sample acquisition module;

[0009] The early warning module reconstructs the expected trajectory of each parameter based on the causal propagation diagram. If the actual detection data deviates systematically from the expected trajectory, it is judged as a pollution event and an early warning is triggered. If the deviation is only manifested as local natural fluctuations, no early warning is triggered.

[0010] Preferably, the data monitoring and processing module constructs the causal propagation graph in the following way: based on historical monitoring data, Granger causality tests are performed on each pair of water quality parameters to determine the causal influence direction between parameters; the propagation entropy value between parameter pairs is calculated, and the propagation entropy value is used as a quantitative index of causal influence intensity and assigned as the weight of the directed edge; the propagation delay attribute of the directed edge is determined according to the time delay of information propagation between parameters.

[0011] Preferably, the data monitoring and processing module generates dynamic prediction intervals for each parameter based on the causal propagation graph in the following way: taking the topological structure of the causal propagation graph as input, a temporal convolutional prediction model is constructed, and the causal influence intensity is incorporated as an attention weight into the convolution calculation process, outputting the predicted values ​​and prediction errors of each parameter; according to the statistical distribution of historical prediction errors within a preset time window, the confidence interval width is dynamically calculated, and a dynamic prediction interval is generated with the predicted value as the center and the confidence interval width as the boundary.

[0012] Preferably, the data monitoring and processing module determines the downstream affected parameter set along the forward propagation path of the deviation parameter in the causal propagation graph by: starting from the deviation parameter, traversing the causal propagation graph along the direction of the directed edge to obtain all reachable downstream parameters as the downstream affected parameter set; calculating the sampling priority weight of each downstream parameter based on the propagation distance and causal influence intensity between each downstream parameter and the deviation parameter; and assigning differentiated sampling frequencies to the downstream affected parameter set based on the sampling priority weight, with higher priority weights resulting in higher sampling frequencies.

[0013] Preferably, the early warning module determines pollution events and local natural fluctuations in the following ways: under conditions of no external pollution input, the expected trajectory of each parameter is reconstructed based on the causal propagation path of each parameter in the causal propagation diagram; the deviation time series between the actual detection data and the expected trajectory is calculated; when multiple parameters simultaneously deviate within the same time window and the direction of deviation conforms to the expected direction of the causal propagation path, it is determined to be a systematic deviation, confirmed as a pollution event, and an early warning is triggered; when the deviation only occurs in a single parameter and the deviation magnitude is within the preset natural fluctuation threshold, it is determined to be a local natural fluctuation, and no early warning is triggered.

[0014] Preferably, the water sample pretreatment involves spatiotemporal alignment of the collected multi-source water quality parameters to unify water quality data from different sampling points and at different detection times into the same spatiotemporal coordinate system; outlier detection and removal are performed on each parameter data based on data quality assessment indicators, and missing data segments are filled using time-series interpolation methods.

[0015] Preferably, the data monitoring and processing module further includes a model adaptive update unit, which is used to periodically monitor the trend of causal influence intensity changes on each side of the causal propagation graph; when the change in causal influence intensity exceeds a preset drift threshold, the model update mechanism is triggered, and the parameters of the temporal convolutional prediction model are updated using incremental learning, so that the dynamic prediction interval adapts to the temporal evolution of the causal relationship between water quality parameters.

[0016] Preferably, the water sample collection module further includes a multi-site collaborative monitoring unit, which is used to establish a spatial causal propagation link between multiple monitoring stations and determine the spatial causal relationship between stations based on the geographical distance between stations and the direction of water flow; when the upstream monitoring station detects a parameter deviation, it automatically triggers the collaborative encrypted sampling of the downstream monitoring station to increase the sampling frequency of the downstream station to be synchronized with that of the upstream station.

[0017] Preferably, the early warning module is also used to classify early warning levels based on the degree of deviation, the number of affected parameters, and the causal propagation range, with different early warning levels corresponding to different response strategies; the early warning information output includes pollution source location estimation and pollution diffusion trend prediction based on causal propagation path inference.

[0018] A water pollution monitoring method for environmental monitoring, comprising the following steps:

[0019] S1. Use a standardized sampling strategy of fixed point, time, and quantity to collect water samples, complete pretreatment operations, and simultaneously record sampling point location, time, water sample parameters, and pretreatment information.

[0020] S2. Receive the pre-processed standard water sample data, convert various water quality characteristics of the water sample into standard electrical signals, and output the original detection data and accurate water quality index results after signal calibration and noise reduction.

[0021] S3. Using water quality parameters as nodes and the causal effects and time delay of parameters as directed edges, construct a parameter causal propagation graph to generate dynamic prediction intervals for each parameter. When real-time data deviates from the interval, rely on the forward propagation path to lock the downstream affected parameters, increase the corresponding sampling frequency, and provide feedback instructions to regulate the sampling module.

[0022] S4. Based on the causal propagation diagram, reconstruct the expected trajectory of normal changes in water quality parameters, compare the deviation characteristics between the measured data and the expected trajectory, and determine the pollution event and trigger an early warning when there is a systematic or trend-based deviation; if there is only local instantaneous natural fluctuation, no early warning will be issued, and a pollution judgment will be made.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] This invention abandons the traditional extensive monitoring method of high-frequency sampling across the entire area. Instead, it utilizes a causal propagation diagram of water quality parameters to achieve differentiated adaptive sampling control. Sampling frequency is increased only for downstream monitoring links of abnormal parameters, while regular sampling is maintained during periods of stable water quality. This significantly reduces equipment energy consumption and computing resource consumption, alleviates equipment operating load, extends equipment lifespan, and significantly improves the system's energy-saving and low-carbon operation level. Simultaneously, this invention combines standardized water sample collection, pretreatment, and source tracing recording mechanisms with multi-parameter dynamic prediction intervals and trajectory reconstruction technology to accurately depict the dynamic propagation patterns of water pollution, effectively compensating for the technical shortcomings of traditional static monitoring and independent parameter detection. It can effectively distinguish between natural fluctuations in the water environment and systematic deviations caused by pollution, significantly reducing invalid false alarms, accurately pinpointing the pollution propagation range, improving the accuracy of water pollution early warning and judgment and emergency response capabilities, and adapting to the needs of refined and intelligent environmental governance of the water environment. Attached Figure Description

[0025] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0026] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0027] System Overall Architecture

[0028] The water pollution monitoring system for environmental monitoring includes a water sample acquisition module, a water quality testing module, a data monitoring and processing module, and an early warning module. The overall data flow of the system is as follows: After acquiring and preprocessing water samples at designated locations, the water sample acquisition module transmits the processed water sample data to the water quality testing module; the water quality testing module performs simultaneous multi-parameter testing on the water samples and outputs the water quality index test results to the data monitoring and processing module; the data monitoring and processing module performs dynamic prediction and anomaly detection based on the causal propagation diagram, and feeds back sampling adjustment instructions to the water sample acquisition module when deviations are detected; the early warning module performs counterfactual attribution determination based on the analysis results of the data monitoring and processing module and outputs early warning information.

[0029] The data interaction relationships between the modules are as follows: the output of the water sample acquisition module is connected to the input of the water quality detection module; the output of the water quality detection module is connected to the input of the data monitoring and processing module; and the output of the data monitoring and processing module is connected to the input of the early warning module and the control input of the water sample acquisition module, respectively.

[0030] Water sample collection module

[0031] The water sample acquisition module is used to acquire water samples at fixed points, at fixed times, and in fixed quantities, and to complete water sample pretreatment and sampling records.

[0032] In this embodiment, the water sampling module includes an automatic sampler, a sampling container, a water sample pretreatment unit, and a sampling controller. The automatic sampler acquires water samples at designated monitoring stations according to preset sampling time intervals and sampling volumes. The sampling time interval is determined by a default sampling frequency, which is set according to the water quality risk level of the monitoring station. The specific setting method is shown in Table 1 below:

[0033] Low risk Once every 6 hours upstream non-industrial areas of drinking water sources Medium risk Once every 3 hours General industrial drainage areas High risk Once every hour Downstream of sewage outlets in industrial areas

[0034] Table 1

[0035] The water sample pretreatment unit performs physical pretreatment on the collected water samples, including filtration to remove suspended particulate matter, defoaming, and temperature regulation, to ensure that the water samples meet the sampling requirements of the water quality testing module. The sampling controller records the timestamp, sampling point coordinates, sampling volume, and water sample pretreatment status information for each sampling, forming a sampling record.

[0036] Data quality control in water sample pretreatment

[0037] After completing physical preprocessing, the water sampling module performs the following quality control processing on the collected multi-source water quality parameter data:

[0038] Spatiotemporal alignment processing: The spatiotemporal coordinates of water quality data acquired from different sampling points and at different detection times are unified. The spatial coordinates of all sampling points are transformed to a unified national geodetic coordinate system, and all detection times are transformed to a unified Beijing time standard, so that water quality data from different sources are unified to the same spatiotemporal coordinate system, providing a consistent data foundation for subsequent multi-parameter fusion analysis.

[0039] Outlier Detection and Removal: Outlier detection is performed on each parameter based on data quality assessment indicators. Specifically, the mean and standard deviation within the recent sliding window are calculated for each water quality parameter. Measurements deviating from the mean by more than three times the standard deviation are marked as outliers and removed. Simultaneously, data where parameter values ​​exceed physically reasonable ranges are also marked and removed, such as dissolved oxygen concentrations exceeding saturation solubility.

[0040] Missing data completion: For missing data segments resulting from outlier removal, temporal interpolation methods are used for completion. When the missing time length does not exceed the interval between two adjacent samplings, linear interpolation is used for completion; when the missing time length exceeds the interval between two adjacent samplings, a mean-based imputation method based on historical data from the same period is used for completion. The interpolation source is marked on the completed data for weight adjustment during subsequent causal propagation graph construction.

[0041] Multi-site collaborative monitoring unit

[0042] The water sampling module also includes a multi-site collaborative monitoring unit, which is used to establish spatial causal propagation links between multiple monitoring sites and realize cross-site collaborative sampling control.

[0043] Multi-site collaborative monitoring units determine spatial causal relationships between stations based on their geographic distance and water flow direction. Specifically, based on the hydrological model of the river or water body, the upstream and downstream relationships and water flow transmission time between each monitoring station are determined, establishing spatial causal propagation links between stations. These spatial causal propagation links use monitoring stations as nodes and water flow direction and transmission time as directed edges.

[0044] When the data monitoring and processing module of the upstream monitoring station detects that a parameter deviates from the dynamic prediction range, it automatically triggers collaborative encrypted sampling at the downstream monitoring station. Specifically, the upstream station sends the deviation information to the sampling controller of the downstream station via the communication link. The sampling controller of the downstream station then increases its sampling frequency to synchronize with the current sampling frequency of the upstream station until the parameter at the upstream station returns to within the dynamic prediction range. The trigger condition for collaborative encrypted sampling is that the deviation of the parameter detected by the upstream station exceeds a preset proportion of the boundary of the dynamic prediction range.

[0045] Water quality testing module

[0046] The water quality testing module is used to receive pre-processed water sample data, perform multi-parameter synchronous testing on the water sample, convert water sample characteristics into electrical signals, and output raw test data and water quality index test results.

[0047] In this embodiment, the water quality detection module uses a multi-parameter water quality analyzer, which can simultaneously detect the following water quality parameters as shown in Table 2 below:

[0048] physical parameters Water temperature, turbidity, conductivity Electrode method Analog electrical signal Chemical parameters pH value, dissolved oxygen (DO), chemical oxygen demand (COD) Electrochemical method / spectroscopy Digital signals Nutrient parameters <![CDATA[Ammonia nitrogen (NH3-N), total phosphorus (TP), total nitrogen (TN)]]> Spectrophotometry Digital signals Heavy metal parameters Lead (Pb), Mercury (Hg), Cadmium (Cd) Anodic stripping voltammetry Analog electrical signal

[0049] Table 2

[0050] The water quality testing module outputs the test results of each parameter in a unified data format, including: parameter identifier, test value, test unit, test timestamp, and test confidence level. The test confidence level is determined by a combination of the instrument's self-test results and the signal-to-noise ratio.

[0051] Multi-parameter synchronous detection requires that all parameters be detected within the same sampling period, and the detection time deviation between parameters shall not exceed 5% of the sampling period. When a parameter cannot output a valid detection value due to instrument failure or calibration reasons, the detection value of that parameter is marked as missing, but this does not affect the normal output of other parameters.

[0052] Data monitoring and processing module

[0053] The data monitoring and processing module is the core processing unit of the system. It is used to receive water quality index detection results, construct causal propagation diagrams, generate dynamic prediction intervals, perform causal path-oriented sampling, and trigger adaptive model updates.

[0054] Causal propagation graph construction

[0055] The data monitoring and processing module constructs a causal propagation graph, using each water quality parameter as a node and the causal influence direction and propagation delay between parameters as directed edges. The construction process of the causal propagation graph includes the following steps:

[0056] Determining the direction of causal influence

[0057] Based on historical monitoring data, Granger causality tests were performed pairwise on each water quality parameter to determine the direction of causal influence between parameters. Let the time series of water quality parameter X be {x(t)} and the time series of water quality parameter Y be {y(t)}, and the following autoregressive model be constructed for testing:

[0058] Unconstrained model:

[0059]

[0060] Constrained models:

[0061]

[0062] Where p is the lag order of the model, determined by the AIC criterion. The F-statistic is calculated by comparing the sums of squared residuals of the two models using the F-test:

[0063]

[0064] in For the sum of squared residuals of the unconstrained model, Let N be the sum of squared residuals of the constrained model, and N be the sample length. When the F-statistic exceeds the critical value at a given significance level, parameter X is determined to be a Granger cause of parameter Y, and a directed edge from X to Y is established in the causal propagation graph.

[0065] Quantification of the strength of causal influence

[0066] The transfer entropy between parameter pairs is calculated, and this transfer entropy value is used as a quantification index of the strength of causal influence, assigning it as the weight of the directed edge. Transfer entropy measures the amount of additional information that the historical information of parameter X provides for predicting the future state of parameter Y, given the historical information of parameter Y itself.

[0067] Let the discretized time series of parameters X and Y be {x(t)} and {y(t)}, respectively. The formula for calculating the transfer entropy is:

[0068]

[0069] in For the historical embedding vector of parameter Y, Let X be the historical embedding vector of parameter X, and k be the embedding dimension.

[0070] A larger transitive entropy value indicates a stronger causal influence of parameter X on parameter Y. Normalizing the transitive entropy value to... The interval is used as the weight of the directed edge in the causal propagation graph. .

[0071] Determining the propagation delay

[0072] The propagation delay attribute of directed edges is determined based on the time delay of information propagation between parameters. Specifically, the cross-correlation function between the time series of parameter X and the time series of parameter Y is calculated:

[0073]

[0074] in and Let X and Y be the means of parameters X and Y, respectively. and These are the standard deviations of parameters X and Y, respectively. The time delay is what causes the cross-correlation function to reach its maximum value. This is the propagation delay of parameter X to parameter Y. The propagation delay attribute of the directed edge is assigned.

[0075] Through the above three steps, a causal propagation graph is constructed, with water quality parameters as nodes, causal influence directions as directed edge directions, normalized propagation entropy values ​​as directed edge weights, and optimal cross-correlation delays as directed edge propagation delays. Where V is the set of parameter nodes, E is the set of directed edges, and W is the weight matrix of the edges. Let be the time delay matrix of the edge.

[0076] Dynamic prediction interval generation

[0077] The data monitoring and processing module generates dynamic prediction intervals for each parameter based on the causal propagation graph. Specifically, it constructs a temporal convolutional prediction model using the topology of the causal propagation graph as input, incorporates the causal influence strength as attention weights into the convolution calculation process, and outputs the predicted values ​​and prediction errors of each parameter.

[0078] The temporal convolutional prediction model employs a causal convolutional architecture. Its input layer receives the historical observation sequence of the target parameter itself and the historical observation sequences of all its upstream parameters (parameters that have directed edges pointing to the target parameter in the causal propagation graph). When processing the input of the upstream parameters, the convolutional layer uses the weights of the corresponding edges (i.e., the normalized value of the propagation entropy) as attention weights to perform a weighted summation of the convolutional features of the upstream parameters.

[0079] Specifically, let the predicted value of the target parameter Y at time t be... Its upstream parameter set is The calculation process for attention-weighted convolution is as follows:

[0080]

[0081] in For activation function, For causal convolution operations, upstream parameters Considering the input sequence after propagation delay, For the historical input of the target parameters themselves, This represents the corresponding causal influence strength weight.

[0082] The model outputs predicted values ​​of the target parameters. and prediction error estimation The prediction error estimate is provided by the model's output layer through a soft activation function, ensuring that the output value is positive.

[0083] The confidence interval width is dynamically calculated based on the predicted value and the estimated prediction error. According to the historical prediction error statistical distribution within a preset time window (in this embodiment, the most recent N=100 time steps), the confidence interval width is calculated as follows:

[0084]

[0085] in The standard normal distribution is at a confidence level of The quantiles below (in this embodiment) , ), The rolling root mean square error within the preset time window:

[0086]

[0087] With predicted values Centered on, confidence interval width Generate dynamic prediction intervals using the boundaries. .because The prediction range is dynamically updated based on the actual prediction error, and can adaptively reflect the degree of fluctuation in the current water quality status.

[0088] Causal path-oriented sampling

[0089] When real-time detection data deviates from the dynamic prediction range, the data monitoring and processing module determines the set of downstream affected parameters along the forward propagation path of the deviation parameters in the causal propagation diagram, and performs differentiated sampling frequency adjustment on the set.

[0090] Specifically, with deviation parameters Starting from the root, perform a breadth-first search (BFS) traversal along the directed edges of the causal propagation graph to obtain all reachable downstream parameters as the set of downstream affected parameters. During the traversal, record each downstream parameter. With deviation parameters Shortest propagation path length between (Measured by the number of directed edges) and cumulative propagation delay on the path.

[0091] Based on the propagation distance and causal influence strength between each downstream parameter and the deviation parameter, calculate the sampling priority weight of each downstream parameter:

[0092]

[0093] in From the causal propagation diagram arrive The maximum value of the normalized entropy transmitted through all paths. The shortest path length. The attenuation coefficient (taken in this embodiment) ).

[0094] Differentiated sampling frequencies are assigned to the downstream affected parameter set based on sampling priority weights:

[0095]

[0096] in This is the default sampling frequency. The maximum allowable sampling frequency (in this embodiment) ), For parameters The sampling priority weight is assigned. The higher the priority weight, the greater the increase in the sampling frequency of downstream parameters, thus realizing targeted encrypted sampling along the causal propagation path and avoiding invalid encryption of unaffected parameters.

[0097] The data monitoring and processing module sends the sampling adjustment command to the sampling controller of the water sample acquisition module. The sampling controller executes the sampling task according to the new sampling frequency. When the deviation parameter returns to the dynamic prediction range and continues for a preset observation time (3 sampling cycles in this embodiment), the sampling frequency is gradually reduced to the default value.

[0098] Model Adaptive Update Unit

[0099] The data monitoring and processing module also includes a model adaptive update unit, which monitors the stability of the causal propagation graph and triggers model updates when causal relationships change significantly.

[0100] The model adaptive update unit periodically (every 24 hours in this embodiment) monitors the changing trend of the causal influence intensity of each edge in the causal propagation graph. Specifically, it calculates the relative rate of change between the propagation entropy value of each edge in the latest monitoring window and the average propagation entropy value in the historical sliding windows:

[0101]

[0102] When the magnitude of change in the intensity of causal influence Exceeding the preset drift threshold (in this embodiment, it is taken as...) When this occurs, the model update mechanism is triggered.

[0103] The model update employs an incremental learning approach, specifically: based on the existing temporal convolutional prediction model, data from the latest monitoring window is used as training samples, and gradient descent is fine-tuned for a limited number of epochs (5 epochs in this embodiment). During fine-tuning, the learning rate is set to 1 / 10 of the initial training learning rate to prevent drastic changes in model parameters.

[0104] After incremental learning is completed, the propagation entropy and propagation delay of each edge in the causal propagation graph are recalculated, and the weight matrix W and the delay matrix ΔT are updated to make the dynamic prediction interval adapt to the time evolution of the causal relationship between water quality parameters.

[0105] Early warning module

[0106] The early warning module is used to perform counterfactual attribution inference based on the causal propagation graph, determine the nature of the deviation event, and output early warning information.

[0107] Counterfactual attribution inference

[0108] Under conditions of no external pollution input, the early warning module reconstructs the expected trajectory of each parameter based on the causal propagation path of each parameter in the causal propagation diagram. The core idea of ​​counterfactual attribution inference is: assuming there is no external pollution source input in the current period, based solely on the natural causal relationship and historical trends between water quality parameters, what kind of change trajectory should each parameter exhibit?

[0109] Specifically, the process of reconstructing the expected trajectory is as follows: using the observed values ​​of each parameter within a complete sampling period prior to the deviation as initial conditions, and within the framework of the causal propagation graph, the forward prediction capability of the temporal convolutional prediction model is utilized to extrapolate the expected change trajectory of each parameter step by step over time. During the extrapolation process, it is assumed that there is no external pollution input (i.e., no abrupt changes beyond the range of historical natural fluctuations are considered), and predictions are made solely based on the causal propagation relationships between parameters.

[0110] Let the time of deviation be The expected trajectory of parameter Y is subscript It represents a counterfactual prediction.

[0111] Calculate the time series of the deviation between the actual detection data and the expected trajectory:

[0112]

[0113] Determining Pollution Events and Natural Fluctuations

[0114] The early warning module determines pollution events and local natural fluctuations based on the characteristics of the deviation time series.

[0115] Criteria for determining systematic deviation: A systematic deviation is identified and a contamination event is confirmed when all of the following conditions are met:

[0116] Multi-parameter synchronous deviation: In a causal propagation graph, multiple parameters with causal relationships exist within the same time window. Inner (in this embodiment, it is taken as) Each sampling period (in which the parameters deviate) produces a deviation, and the number of deviation parameters is no less than 2.

[0117] Deviation direction consistency: The deviation direction of each parameter conforms to the expected direction of the causal propagation path. Specifically, if the deviation of parameter X leads to a positive expected change in parameter Y (determined by the directed edges and the sign of the propagation entropy in the causal propagation graph), then the actual deviation direction of parameter Y should also be positive, and vice versa. The deviation direction consistency index is defined as:

[0118]

[0119] when At that time, it is determined that the deviation direction is consistent.

[0120] Significance of deviation: At least one parameter deviates by more than twice the width of its dynamic prediction interval.

[0121] When all three conditions are met, it is determined to be a systemic deviation, confirmed as a pollution event, and an early warning is triggered.

[0122] Criteria for determining local natural fluctuations: When the following conditions are met, it is determined to be a local natural fluctuation, and no warning is triggered:

[0123] The deviation occurs only in a single parameter, and there are no synchronous deviation parameters that are causally related to it.

[0124] The deviation is within the preset natural fluctuation threshold (in this embodiment, the deviation is taken as not exceeding 1.5 times the width of the dynamic prediction interval).

[0125] The deviation duration does not exceed the preset fluctuation duration threshold (2 sampling periods in this embodiment).

[0126] Early warning level classification and response strategy

[0127] The early warning module classifies early warning levels based on the degree of deviation, the number of affected parameters, and the scope of causal propagation. Different early warning levels correspond to different response strategies, as shown in Table 3 below:

[0128] Level 1 (Blue) The deviation is 1.5-2 times CI. 2-3 Propagation distance ≤ 2 hops Encrypt sampling at twice the frequency, record deviation events, and continuously observe for 30 minutes. Level 2 (Yellow) The deviation is 2-3 times the CI. 3-5 Propagation distance 3-4 hops Encrypt sampling is performed at the maximum frequency. Inspection personnel are notified to verify the information on-site and initiate pollution source tracing. Level 3 (Red) Deviation > 3 times CI ≥6 Propagation distance > 4 hops or across sites Immediately activate the emergency monitoring plan, simultaneously increase sampling density at all sites, and report to the environmental protection authorities.

[0129] Table 3

[0130] Where CI is the dynamic prediction interval width of the corresponding parameter.

[0131] The warning message output includes the following information:

[0132] A list of deviation parameters, along with the deviation magnitude and direction for each parameter.

[0133] Pollution source location estimation based on causal propagation path backtracking: In the causal propagation graph, starting from all deviation parameters, backtracking along the opposite direction of the directed edges, the intersection of all backtracking paths represents the most probable pollution source parameters. Combining this with the spatial location information of monitoring stations, the pollution source parameters are mapped to monitoring points in physical space, providing an estimate of the pollution source location.

[0134] Pollution diffusion trend prediction: Based on the forward propagation path from the pollution source parameters in the causal propagation diagram, predict which parameter paths the pollution will spread downstream, and give the expected arrival time and expected deviation of each downstream parameter.

[0135] Example

[0136] The following is a specific application scenario example.

[0137] A river monitoring section in a certain city has three monitoring stations (station A is located at the drainage outlet of the upstream industrial area, station B is located in the midstream residential area, and station C is located at the downstream drinking water intake). The monitoring parameters include five parameters: pH value, dissolved oxygen (DO), ammonia nitrogen (NH3-N), chemical oxygen demand (COD), and turbidity.

[0138] The data monitoring and processing module constructs a causal propagation diagram based on historical monitoring data from the past 30 days. Through Granger causality testing and propagation entropy calculation, the following main causal relationships are obtained: COD at station A → DO at station A (weight 0.82, time delay 30 min), NH3-N at station A → NH3-N at station B (weight 0.71, time delay 2 h, inter-station propagation), COD at station A → COD at station B (weight 0.65, time delay 2 h), turbidity at station B → turbidity at station C (weight 0.58, time delay 3 h).

[0139] When the COD at site A deviates from the dynamic prediction range during a certain sampling period, the data monitoring and processing module traverses the causal propagation graph forward to determine the downstream affected parameter set as {DO (site A), COD (site B), NH3-N (site B), turbidity (site C)}, and allocates differentiated sampling frequencies according to sampling priority weights. Simultaneously, the early warning module performs counterfactual attribution inference. If the DO at site A decreases synchronously and in the expected direction (COD increase leads to DO decrease), and the COD at site B deviates in the same direction after 2 hours, it is determined to be a systemic deviation (industrial pollution event), triggering a level-two early warning. If only the turbidity at site A deviates briefly and recovers within one period, it is determined to be a local natural fluctuation (rainfall causing sediment disturbance), and no early warning is triggered.

[0140] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A water pollution monitoring system for environmental monitoring, characterized in that, The system includes: The water sampling module is used to acquire water samples at fixed points, at fixed times, and in fixed quantities, and to complete water sample pretreatment and sampling records. The water quality testing module is used to receive pre-processed water sample data, perform multi-parameter synchronous testing on the water sample, convert water sample characteristics into electrical signals, and output raw test data and water quality index test results. The data monitoring and processing module is used to receive water quality index detection results, construct a causal propagation graph with each water quality parameter as a node and the causal influence direction and propagation delay between parameters as directed edges, and generate dynamic prediction intervals for each parameter based on the causal propagation graph; when the real-time detection data deviates from the dynamic prediction interval, the downstream affected parameter set is determined along the forward propagation path of the deviated parameter in the causal propagation graph, the sampling frequency is increased only for this set, and the sampling adjustment command is fed back to the water sample acquisition module; The specific method by which the data monitoring and processing module generates dynamic prediction intervals for each parameter based on the causal propagation graph is as follows: taking the topological structure of the causal propagation graph as input, a temporal convolutional prediction model is constructed, and the causal influence intensity is incorporated as an attention weight into the convolutional calculation process, outputting the predicted values ​​and prediction errors of each parameter; according to the statistical distribution of historical prediction errors within a preset time window, the confidence interval width is dynamically calculated, and a dynamic prediction interval is generated with the predicted value as the center and the confidence interval width as the boundary. The early warning module reconstructs the expected trajectory of each parameter based on the causal propagation diagram. If the actual detection data deviates systematically from the expected trajectory, it is judged as a pollution event and an early warning is triggered. If the deviation is only manifested as local natural fluctuations, no early warning is triggered. The specific method by which the early warning module determines pollution events and local natural fluctuations is as follows: under the condition of no external pollution input, the expected trajectory of each parameter is reconstructed based on the causal propagation path of each parameter in the causal propagation diagram. The process of reconstructing the expected trajectory is as follows: using the observed values ​​of each parameter within a complete sampling period before the deviation occurs as initial conditions, within the framework of the causal propagation graph, the forward prediction capability of the temporal convolutional prediction model is used to deduce the expected change trajectory of each parameter step by step. During the deduction process, it is assumed that there is no external pollution input, and the prediction is made only based on the causal propagation relationship between parameters.

2. A water pollution monitoring system for environmental monitoring according to claim 1, characterized in that, The specific method for the data monitoring and processing module to construct the causal propagation graph is as follows: based on historical monitoring data, Granger causality tests are performed on each pair of water quality parameters to determine the causal influence direction between parameters; the propagation entropy value between parameter pairs is calculated, and the propagation entropy value is used as a quantitative index of causal influence intensity and assigned as the weight of the directed edge. The propagation delay attribute of directed edges is determined based on the time delay of information propagation between parameters.

3. A water pollution monitoring system for environmental monitoring according to claim 1, characterized in that, The data monitoring and processing module determines the downstream affected parameter set along the forward propagation path of the deviation parameter in the causal propagation graph, specifically including: starting from the deviation parameter, traversing the causal propagation graph along the direction of the directed edge to obtain all reachable downstream parameters as the downstream affected parameter set; calculating the sampling priority weight of each downstream parameter based on the propagation distance and causal influence intensity between each downstream parameter and the deviation parameter; and assigning differentiated sampling frequencies to the downstream affected parameter set based on the sampling priority weight, with higher priority weights resulting in higher sampling frequencies.

4. A water pollution monitoring system for environmental monitoring according to claim 1, characterized in that, The early warning module calculates the time series of deviations between the actual detection data and the expected trajectory; when multiple parameters simultaneously deviate within the same time window and the direction of deviation conforms to the expected direction of the causal propagation path, it is determined to be a systematic deviation, confirmed as a pollution event, and an early warning is triggered. When a deviation occurs only in a single parameter and the deviation magnitude is within the preset natural fluctuation threshold, it is determined to be a local natural fluctuation and no warning is triggered.

5. A water pollution monitoring system for environmental monitoring according to claim 1, characterized in that: The water sample preprocessing involves spatiotemporal alignment of the collected multi-source water quality parameters to unify water quality data from different sampling points and at different detection times into the same spatiotemporal coordinate system; outlier detection and removal are performed on each parameter data based on data quality assessment indicators, and missing data segments are filled using time-series interpolation methods.

6. A water pollution monitoring system for environmental monitoring according to claim 1, characterized in that: The data monitoring and processing module also includes a model adaptive update unit, which is used to periodically monitor the trend of causal influence intensity changes on each side of the causal propagation graph. When the change in causal influence intensity exceeds the preset drift threshold, the model update mechanism is triggered, and the parameters of the temporal convolutional prediction model are updated using incremental learning, so that the dynamic prediction interval adapts to the temporal evolution of causal relationships between water quality parameters.

7. A water pollution monitoring system for environmental monitoring according to claim 1, characterized in that: The water sample collection module also includes a multi-site collaborative monitoring unit, which is used to establish a spatial causal propagation link between multiple monitoring sites and determine the spatial causal relationship between sites based on the geographical distance between sites and the direction of water flow. When an upstream monitoring station detects a parameter deviation, it automatically triggers a coordinated encrypted sampling process at the downstream monitoring station, increasing the sampling frequency of the downstream station to be synchronized with that of the upstream station.

8. A water pollution monitoring system for environmental monitoring according to claim 1, characterized in that: The early warning module is also used to classify early warning levels based on the degree of deviation, the number of affected parameters, and the causal propagation range. Different early warning levels correspond to different response strategies. The early warning information output includes pollution source location estimation and pollution diffusion trend prediction based on the causal propagation path.

9. A water pollution monitoring method for environmental monitoring, applied to a water pollution monitoring system for environmental monitoring as described in any one of claims 1 to 8, characterized in that, The specific steps are as follows: S1. Use a standardized sampling strategy of fixed point, time, and quantity to collect water samples, complete pretreatment operations, and simultaneously record sampling point location, time, water sample parameters, and pretreatment information. S2. Receive the pre-processed standard water sample data, convert various water quality characteristics of the water sample into standard electrical signals, and output the original detection data and accurate water quality index results after signal calibration and noise reduction. S3. Using water quality parameters as nodes and the causal effects and time delay of parameters as directed edges, construct a parameter causal propagation graph to generate dynamic prediction intervals for each parameter. When real-time data deviates from the interval, rely on the forward propagation path to lock the downstream affected parameters, increase the corresponding sampling frequency, and provide feedback instructions to regulate the sampling module. S4. Based on the causal propagation diagram, reconstruct the expected trajectory of normal changes in water quality parameters, compare the deviation characteristics between the measured data and the expected trajectory, and determine the pollution event and trigger an early warning when there is a systematic or trend deviation. No warning is issued for localized, instantaneous natural fluctuations; instead, a pollution assessment is performed.

Images