A water affair abnormal event early warning system based on data collection and analysis

Abstract

The present application relates to the field of intelligent water management, and particularly relates to a water abnormal event early warning system based on data collection and analysis, which forms time synchronization time series data through multi-source data collection, calculates dynamic risk scores, identifies abnormal events and outputs event type confidence, generates propagation prediction results in combination with pipe network topology, executes dynamic control measures, and receives feedback to update the state through the alarm closed loop, realizes real-time early warning and accurate disposal of water abnormal events, realizes early identification and accurate attribution of short-time negative pressure backflow and high-toxicity small-volume pollution groups through multi-source time synchronization and dynamic risk assessment, gives the influence range and time window in combination with topological propagation prediction, improves the disposal directionality, and updates the state relying on the alarm-control-feedback closed loop, self-optimizes the threshold and strategy, and improves the emergency efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent water management, and more specifically, to an early warning system for abnormal water events based on data acquisition and analysis. Background Technology

[0002] In existing water supply network management, early warning of abnormal events mainly relies on monitoring a single data source such as pressure, flow rate, or water quality. Traditional methods are typically designed for persistent or large-scale abnormal events, such as pipe bursts, pump outages, or long-term contamination, and can provide relatively effective early warning and response measures. However, existing technologies have significant limitations for short-term negative pressure events caused by pressure transients, especially in scenarios with potential cross-connections of highly toxic sources: First, low-frequency sampling and delayed monitoring struggle to capture negative pressure windows on the order of seconds or minutes, making it easy to miss short-term backflow events; second, while the pollutant population may be small in quantity but highly harmful, water volume changes may not be obvious, and water quality anomalies precede flow anomalies, making it difficult to accurately identify traditional leakage or pipe burst logic; third, the propagation process spreads directionally along the pipe network topology, resulting in delayed and spatial wavefront effects in water quality mutations and user complaints, making it difficult for traditional systems to correlate transient hydraulic events with end-point anomalies in time and space; finally, the source of the backflow is doubly unknown, requiring both locating the cross-connection point where the backflow occurs and determining the nature of the backflow pollution (general pollution or highly toxic source), while existing methods typically focus only on single-dimensional evidence and cannot integrate pressure transients, backflow mechanisms, and pollution source characteristics into a single event chain.

[0003] Therefore, existing technologies have limitations in identifying and attributing the entry of highly toxic, small-volume contaminants into the network and their dynamic propagation, including insufficient real-time performance, inadequate evidence integration, and limited source tracing capabilities. There is an urgent need for a comprehensive technical solution that can form a closed loop for identifying and providing early warning of pressure transients, cross-connection backflow, and the formation and propagation of highly toxic contaminants, thereby improving the accuracy and efficiency of public health protection and emergency response. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a water affairs anomaly early warning system based on data acquisition and analysis.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A water-related abnormal event early warning system based on data acquisition and analysis includes: The multi-source data acquisition module is used to collect water supply network data, as well as external event data and user complaint data, to form time-synchronized time-series data; The dynamic risk assessment module is used to calculate dynamic risk scores based on time-synchronized time-series data. The abnormal event identification module is used to identify abnormal events based on dynamic risk scores and output the event type confidence level. The propagation trend prediction module is used to generate propagation prediction results based on pipeline topology information and dynamic risk scores. The dynamic linkage control module is used to perform dynamic control based on the dynamic risk score and the propagation prediction results; The alarm closed-loop feedback module is used to generate alarm and handling information based on event type confidence, propagation prediction results and dynamic control, and to receive handling feedback to update the system status, forming a closed loop.

[0006] Furthermore, the water supply network data includes pressure data, flow data, and water quality data.

[0007] Furthermore, the dynamic risk score is determined by the stress transient intensity characterization, the return flow accessibility characterization, the toxicity suspicion characterization, and the uncertainty characterization.

[0008] Furthermore, the stress transient intensity characterization quantity is obtained based on the stress data; The return flow accessibility characteristic is calculated based on the pipeline topology information and combined with hydraulic conditions characterized by pressure data and / or flow data. The toxicity suspicion index is derived from the anomaly score of water quality data relative to the baseline and is corrected based on the spatiotemporal consistency between external event data and user complaint data. The uncertainty characterization quantity is determined by at least one of the following sources of uncertainty: data reliability source, prediction consistency source, and multi-scheme consistency source.

[0009] Furthermore, the pressure transient intensity characterization quantity is obtained by transient detection of pressure data. The transient detection includes at least one or more of the following: pressure change amplitude, pressure change rate, negative pressure window duration, or transient component energy.

[0010] Furthermore, the calculation of the return flow accessibility characteristic includes at least one of the following: the calculation of the accessibility probability and propagation delay obtained from hydraulic simulation, or the graph accessibility calculation based on flow direction / pressure difference constraints.

[0011] Furthermore, the uncertainty characterization quantity is determined by at least one of the following sources of uncertainty: Data reliability is determined by the missing data rate and / or the proportion of outliers and / or time synchronization deviation. Sources of predictive consistency: Determined by the deviation between the output of the propagation situation prediction module and the corresponding observation data, the deviation including at least one or more of arrival time deviation and / or influence range deviation; Sources of consistency across multiple scenarios: Determined by the degree of dispersion between propagation prediction results obtained from different types of models under the same operating conditions. The degree of dispersion includes at least one or more of the following: propagation path difference, impact range difference, or arrival time variance.

[0012] Furthermore, the event types output by the abnormal event identification module include at least pressure transient events, cross-connection backflow events, and backflow of highly toxic source items.

[0013] Furthermore, the propagation prediction module generates propagation prediction results including propagation path, scope of impact, arrival time, and probability envelope of impact.

[0014] Furthermore, the dynamic control performed by the dynamic linkage control module includes at least sampling resource control, evidence weight control, isolation boundary control, transient suppression control, verification sampling control, and complaint linkage control.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention collects water supply network data and simultaneously gathers external event data and user complaint data to form time-synchronized time-series data. Based on this, a dynamic risk score is calculated, and abnormal events are identified and the event type confidence score is output. Since pressure transients and short-term negative pressures often exhibit characteristics on the order of seconds or minutes, traditional low-frequency monitoring is prone to missing trigger windows. However, this invention utilizes multi-source time synchronization and the continuous evolution of dynamic risk scores to capture abnormal signals in advance before pressure fluctuations evolve into significant water volume anomalies. Combined with the event type confidence score, it distinguishes similar characteristics such as cross-connection backflow, pipe bursts, pump shutdowns, and valve disturbances caused by pressure transients, reducing the possibility of misclassifying highly toxic small-volume pollution clusters as general fluctuations or equipment noise, thereby improving the stability and interpretability of the identification. Based on pipeline topology information and dynamic risk scores, propagation prediction results are generated, enabling early warning to go beyond simply informing the user of an event. Instead, it provides a description of the propagation trend along the hydraulic path, outputting the potential impact range and time window, making intervention more targeted. For highly toxic pollution plumes formed after backflow induced by pressure transients, this invention can establish a correlation between sudden changes in water quality at the terminal level and the spatial-temporal wavefront phenomenon of complaints, linking it to the propagation direction of the upstream triggering source. This avoids the traditional method of judging sensor failure or local disturbance based solely on single-point water quality anomalies. Simultaneously, the propagation prediction results provide a clear sequence of key nodes for subsequent dynamic control, allowing isolation, verification, and review actions to be carried out along the propagation path, thereby reducing the scope of mishandling and improving control benefits. Dynamic control is implemented based on dynamic risk scores and propagation prediction results. Furthermore, alarms and response information are generated based on event type confidence levels, propagation prediction results, and dynamic control, while simultaneously receiving response feedback to update the status, forming a closed loop. This closed loop makes alarms actionable: on the one hand, alarm outputs can carry response suggestions and priorities consistent with the propagation trend, enabling response units to quickly identify high-risk areas and key nodes; on the other hand, response feedback is used to update the status, thereby correcting deviations in event type confidence levels and propagation predictions, prompting subsequent warning thresholds, priorities, and control strategies to continuously converge to a more reasonable configuration. Therefore, in scenarios where highly toxic pollution plumes are small in size but highly hazardous, and the chain of evidence is easily fragmented, this invention can integrate transient triggering, backflow into the network, pollution plume propagation, and peripheral manifestations into a single dynamic chain, shortening the public health risk exposure time and improving the overall efficiency of emergency response. Attached Figure Description

[0016] Fig. 1 This is a schematic diagram of a water affairs abnormal event early warning system based on data acquisition and analysis. Fig. 2 A schematic diagram of the event logic chain for the cross-connection reflux of highly toxic pollutant clusters induced by pressure transients entering the network and their dynamic propagation; Fig. 3 This is a schematic diagram illustrating the composition and calculation process of the dynamic risk score in the dynamic risk assessment module. Fig. 4 This is a schematic diagram of the alarm closed-loop feedback process based on propagation trend prediction and dynamic linkage control. Detailed Implementation

[0017] Reference Figs. 1 to 4 A water affairs anomaly early warning system based on data acquisition and analysis, comprising: The multi-source data acquisition module collects water supply network data, external event data, and user complaint data to form time-synchronized time-series data. This module unifies the "internal operating status" and "external triggering clues" required for anomaly early warning into a single data base. Water supply network data reflects the objective operating status of the network, external event data provides clues about the external environment or events that may lead to changes in risk, and user complaint data provides clues about anomalies perceived by users. By forming time-synchronized time-series data, data from different sources, frequencies, and formats are aligned to a unified timeline, ensuring that subsequent dynamic risk assessment, anomaly identification, and propagation trend prediction modules can perform correlation analysis under the same time benchmark, avoiding misjudgments, omissions, or conflicting evidence caused by time mismatches.

[0018] In one specific implementation, the acquisition end acquires and preprocesses water supply network data according to a unified time reference: pressure acquisition points, flow acquisition points, and water quality acquisition points are set up at key nodes and key pipe sections of the water supply network. Pressure data is used to record the continuous change of pressure over time, flow data is used to record the change of water distribution volume in pipe sections or branches, and water quality data is used to record the changes of water quality indicators such as residual chlorine, turbidity, conductivity, and oxidation-reduction potential. The acquisition end timestamps the above data, verifies the units and dimensions, removes outliers and marks missing measurements, and resamples data from different sampling frequencies to a unified time interval to ensure the feasibility of subsequent time sequence alignment. External event data and user complaint data are accessed and structured according to the same time base: External event data includes information on events that may affect water supply operations, such as road construction, power outages and maintenance, abnormal emissions from surrounding enterprises, and sudden weather changes; User complaint data includes user feedback such as odor, turbidity, insufficient water flow, and low pressure. Event elements are extracted from external event data to form event occurrence time, affected area, event category, and intensity level; and address parsing, content element extraction, and time merging are performed on user complaint data to form complaint occurrence time, complaint location, complaint topic, and quantity intensity. The generation and verification of time-synchronized time-series data are performed as follows: Using a unified time interval as an index, pressure data, flow data, water quality data, external event data, and user complaint data are aggregated and aligned according to time windows to form a multi-source feature sequence on the same time axis. Clearly conflicting records or segments marked as low reliability are removed using consistency verification rules. For example, if a region experiences a rapid decline in pressure data and increased fluctuations in flow data within a 10-minute time window, and external event data records a power outage for maintenance in that region, while user complaint data shows low pressure and insufficient water output, then this time window is synchronized and aggregated into a time-series segment with sufficient corroborating evidence for subsequent analysis. If only a single water quality data point deviates briefly while external event data and user complaint data are empty, this time window is still included in the time-synchronized time-series data but retains a missing evidence marker to ensure sufficient disclosure and facilitate subsequent judgment.

[0019] The dynamic risk assessment module calculates dynamic risk scores based on time-synchronized time-series data. This module further transforms the time-synchronized time-series data generated by the multi-source data acquisition module into quantitative indicators suitable for early warning decision-making, namely, dynamic risk scores. The role of dynamic risk scores is to compress complex, multi-dimensional, and potentially contradictory evidence into comparable, rankable, and threshold-judgmentable risk representations, expressing the comprehensive risk level of an abnormal event occurring or escalating at the current moment. Dynamic risk scores provide risk triggering and identification priority criteria for the abnormal event identification module, risk input conditions for propagation trend prediction module, and a unified reference for the dynamic linkage control module in selecting control intensity, control priority, and control strategies.

[0020] In one specific implementation, a calculation framework for dynamic risk scores is constructed, and the input and processing of time-synchronized time-series data are completed: using a unified time interval as an index, pressure data, flow data, water quality data, external event data, and user complaint data are aligned on the same time axis. For each time window, a time-series segment containing multi-source features is formed, and the same set of data quality constraint rules is set for the subsequent calculation of the four types of characterization quantities. Specifically, dimensional consistency verification and sensor drift detection are performed on pressure data, zero drift and abrupt jump screening are performed on flow data, baseline stability assessment and outlier isolation are performed on water quality data, event element integrity verification is performed on external event data, and location parsing and time merging consistency verification are performed on user complaint data. In this step, the combination structure of dynamic risk scores is defined, clarifying that dynamic risk scores are jointly determined by pressure transient intensity characterization quantity, backflow accessibility characterization quantity, toxicity suspicion characterization quantity, and uncertainty characterization quantity. A unified normalization caliber is set for the four types of characterization quantities, so that each characterization quantity outputs continuous values ​​of the same scale for weighted fusion. At the same time, traceable intermediate quantity records are output for each time window to meet the requirements of sufficient disclosure and reproducible experiments. The intensity of pressure transients is characterized by calculations, with transient detection results serving as core evidence. For each time window, transient detection is performed on a continuous sequence of pressure data. Transient detection includes at least one or more of the following: pressure change amplitude, pressure rate of change, negative pressure window duration, or transient component energy. Specifically, the pressure rate of change is first calculated using a sliding window, and candidate abrupt changes exceeding a threshold are identified. Then, the pressure change amplitude is calculated for each candidate abrupt change, and the direction of the abrupt change is determined. Simultaneously, a continuous interval with pressure below a preset negative pressure threshold is searched to obtain the negative pressure window duration. Furthermore, transient components are extracted from the pressure sequence, and transient energy is calculated. The energy is measured, and finally, at least one or more of the above-mentioned detection quantities are scaled and weighted with confidence to obtain the stress transient intensity characterization quantity. An adaptive threshold mechanism is introduced into the calculation of stress change amplitude and stress change rate: the threshold is automatically updated with the statistical quantile of a stable operating period in the past, so that sudden changes in operating conditions and long-term slow drift can be distinguished. At the same time, in order to avoid false triggering caused by single-point sensor anomalies, the pressure data of adjacent spatial nodes are checked for consistency. When a single-point mutation cannot be supported by neighboring nodes, it is marked as a suspected sensor anomaly in the intermediate quantity record and reflected in the subsequent uncertainty characterization quantity. Pressure change rate and abrupt change segment determination and threshold adaptive update: For pressure sequences Calculate the rate of pressure change within the time window The formula is as follows: Segments exceeding the threshold are considered candidate mutation segments. To reduce long-term drift and threshold mismatch under different operating conditions, the threshold is updated using adaptive quantiles.

[0021] Define a "stable operating sample set" A set of historical time windows that meet the following conditions: dynamic risk score Below the stability threshold And the absolute value of the rate of change of pressure The temperature remains below the stable upper limit within this window. During each update, the most recent segment of the sliding window is used. As a statistical sample, let the threshold for the rate of change of pressure be: Let the threshold for the magnitude of the pressure surge be: ;in Represents the quantile operator. To preset high quantile parameters (e.g., using a higher quantile to control false alarms), the sliding window length can be set according to the operation management cycle and saved in the record. The determination of the negative pressure window duration can be based on a given negative pressure threshold. ,when The continuous duration exceeds The time is recorded as the negative pressure window being established; among which It can be determined based on the lower limit of the pressure sensor calibration or the low percentile of a historical stable period. It can be set by the statistical distribution of historical transient events and can be updated adaptively.

[0022] The recirculation accessibility characteristic is calculated and explicitly utilizes network topology information and hydraulic constraints: the recirculation accessibility characteristic is calculated based on network topology information and hydraulic conditions characterized by pressure data and / or flow data, and its calculation includes at least one or more of the following: accessibility probability and propagation delay calculation obtained from hydraulic simulation, or graph accessibility calculation based on flow direction pressure difference constraints. Specifically, the connectivity between nodes and pipe segments is first established based on network topology information, and the flow direction and pressure difference constraints of each pipe segment in the current time window are estimated using pressure data and flow data, forming a set of hydraulic constraints that are updated over time. In one implementation, hydraulic simulation is performed to obtain the accessibility probability and propagation delay calculation of recirculation reaching each node under different source conditions, wherein the accessibility probability is obtained by multi-scenario simulation statistics, and the propagation delay is obtained by particle tracking or characteristic line propagation time accumulation, and the accessibility probability and propagation delay are then calculated. The time delay is mapped to the return flow reachability representation, so that regions with higher reachability probability and shorter propagation delay correspond to higher representation values. In another implementation, in scenarios where complete simulation is not required, a directed reachability graph is constructed based on flow direction and pressure difference constraints. A reachability search is performed on potential return flow source points to target nodes. If a path exists that satisfies the consistency between pressure difference direction and flow direction, it is recorded as reachable. Furthermore, the reachability strength is constructed based on path length, pressure difference margin, and the number of key valve nodes, thereby obtaining the return flow reachability representation. An adaptive switching is adopted between the two implementations: when the data reliability is high and computing resources allow, the reachability probability and propagation delay calculation obtained from hydraulic simulation are given priority to obtain a finer-grained output. When there are many missing data or higher real-time requirements, graph reachability calculation based on flow direction and pressure difference constraints is used to maintain continuous operation, and the selected method is marked in the intermediate quantity record for traceability. The toxicity suspicion index is calculated and corrections are made based on the spatiotemporal consistency between external event data and user complaint data. The toxicity suspicion index is obtained based on the anomaly score of water quality data relative to the baseline, and corrections are made according to the spatiotemporal consistency between external event data and user complaint data. Specifically, a dynamic baseline is first established for each water quality indicator. The dynamic baseline can be obtained from historical data of similar operating conditions, data from the same day and week, or data from stable operating intervals. Anomaly scores of water quality data relative to the baseline are calculated for each time window. These anomaly scores can be composed of standardized residuals, quantile deviations, or multi-indicator fusion anomalies. Spatiotemporal consistency correction is then performed: external event data is analyzed into event occurrence time, affected area, event category, and intensity level; user complaint data is analyzed into complaint occurrence time, complaint location, complaint topic, and quantity intensity. For each time window, the degree of temporal overlap and spatial overlap between external event data and water quality anomalies is calculated. The proximity of external event data and user complaint data with water quality anomalies is calculated, along with the temporal clustering and spatial coverage of these data. When external event data and user complaint data show high consistency with water quality anomalies within the same or adjacent time windows, the water quality anomaly score is adjusted upwards. Conversely, when water quality anomalies occur in isolation and lack support from external event data and user complaint data, the water quality anomaly score is adjusted downwards or labeled as a low-evidence anomaly. This implementation introduces a dual-channel consistency correction mechanism: external event data provides prior triggering clues, and user complaint data provides collective perception verification clues. The two form correction factors and are fused in a non-linear manner to prevent accidental noise from a single source from excessively increasing the toxicity suspicion index. At the same time, a cautious upward adjustment is maintained when external event data is strong but user complaint data is weak, so that subsequent verification sampling methods can be used for further confirmation. Calculation of dynamic baseline and anomaly scores for water quality indicators: For any water quality index Establish a reproducible dynamic baseline With scale One implementation uses a historical sliding window sample set with "the same monitoring point and the same time period". (For example, taking historical samples from several days within the same intraday period) and constructing a baseline using robust statistics: ; The standardized outlier score for this indicator is: ;in To prevent extremely small positive numbers from being divided by zero, To truncate the upper limit.

[0023] The weighted sum of outlier scores from multiple metric fusions can be used: ;in The results are normalized for the outlier scores of each indicator. It can be set and recorded according to the stability of indicators, the reliability of sensors, or the historical discrimination.

[0024] Calculate the uncertainty characterization and complete the fusion output of the dynamic risk score: The uncertainty characterization is determined by at least one of the following uncertainty sources, including data reliability source, prediction consistency source, and multi-scheme consistency source; wherein the data reliability source is determined by the missing measurement rate and / or outlier ratio and / or time synchronization deviation; the prediction consistency source is determined by the deviation between the output of the propagation situation prediction module and the corresponding observation data, and the deviation includes at least one or more of arrival time deviation and / or impact range deviation; the multi-scheme consistency source is determined by the degree of dispersion between the propagation prediction results obtained by different types of models under the same working conditions, and the degree of dispersion includes at least one or more of propagation path difference, impact range difference, or arrival time variance; Specifically, the missing rate and outlier ratio of pressure, flow, and water quality data are statistically analyzed for each time window. Combined with the time synchronization deviation during the multi-source alignment process, an uncertainty score for the source of data reliability is formed. Then, the output results of the propagation trend prediction module are compared with the corresponding observation data. When the actual observed abnormal diffusion boundary, set of affected nodes, or arrival time is inconsistent with the prediction, the arrival time deviation and the deviation of the affected area are calculated to form an uncertainty score for the source of prediction consistency. At the same time, a multi-scheme consistency source is introduced: under the same hydraulic conditions, different types of propagation prediction methods are used to generate propagation prediction results. For example, hydraulic simulation path deduction and graph reachability deduction are output in parallel. When the two show significant discrepancies in propagation path, affected area, or arrival time, the propagation path difference, affected area difference, or arrival time variance are calculated to form an uncertainty score for the source of multi-scheme consistency. Normalizing and aggregating at least one of the aforementioned uncertainty scores yields an uncertainty characterization, which is then used to fuse dynamic risk scores. In one implementation, the dynamic risk score is composed of a weighted sum of pressure transient intensity characterization, return accessibility characterization, and toxicity suspicion characterization, along with a suppression term for the uncertainty characterization. This ensures that the overall risk output is more cautious when uncertainty is higher, thereby reducing false alarms. In another implementation, the uncertainty characterization is used to adjust the weights of the three types of risk evidence. This reduces the weight of the pressure transient intensity characterization when data reliability is low, reduces the weight of the return accessibility characterization when prediction consistency is poor, and reduces the upward adjustment of the toxicity suspicion characterization when multi-scheme consistency is poor. For example, within a time window in a certain region, if the pressure data shows a significant increase in the magnitude of pressure mutations and continuous exceedances of the pressure change rate, the transient component energy increases, resulting in a higher pressure transient intensity characterization. Based on the pipeline topology information and the flow direction pressure difference constraints estimated by the flow data, there are reachable paths from suspected cross-connection points to multiple branches. The graph reachability calculation gives a high reachability intensity, and the return flow reachability characterization is increased. Water quality data shows an abnormal increase in residual chlorine and turbidity relative to the baseline, and external event data records maintenance switching near the area. User complaint data shows odor and turbidity concentrated in adjacent time windows. The spatiotemporal consistency correction factor further increases the toxicity suspicion characterization. However, at the same time, if there is a certain missing rate in this time window and the output of the propagation trend prediction module deviates from the subsequently observed impact range, the uncertainty characterization increases and the dynamic risk score is suppressed or its weight is adjusted. This ensures that the final dynamic risk score can reflect the high risk caused by the resonance of multiple evidences, and can also maintain caution when the quality of evidence is insufficient or there are prediction biases, meeting the requirements of full disclosure, traceability, and feasibility.

[0025] To ensure that the pressure transient intensity characterization, reflux accessibility characterization, toxicity suspicion characterization, and uncertainty characterization can be weighted and fused at the same scale, for any given time window, the above four types of characterization are first normalized. Let any original quantity be... Its normalized output is Then, truncated linear normalization can be used: ;in, and It can be given by the quantiles of historical stable operating samples or the upper and lower limits of the design, and updated by sliding window during system operation; This indicates that the input will be truncated to a range. Inside.

[0026] In one implementation, dynamic risk score It consists of a weighted sum of three types of risk evidence and an uncertainty suppression term: ;in, This is the normalized value of the transient intensity characterization quantity of pressure. This is the normalized value of the reflux accessibility characteristic. This is the normalized value of the toxicity suspicion index. This is the normalized value of the uncertainty characteristic; The weights are non-negative and satisfy the following conditions: Example: Take , , , The reasons are as follows: The focus is on identifying the "transient negative pressure backflow—topology propagation" link, therefore, return flow accessibility and pressure transients are considered primary evidence (with a higher combined weighting). Toxicity suspicion is used for confirmation and qualitative analysis, but its weight is slightly reduced due to fluctuations in sensor readings and external cues. A certain proportion of the uncertainty suppression term is retained to automatically converge risk output and reduce false alarms in cases of missing data, prediction bias, or divergence among multiple scenarios. In another implementation, the uncertainty characterization is used to reduce the confidence level of risk evidence. Both of the above implementations retain all intermediate variables. Weight parameters are used for tracing records.

[0027] Example of obtaining a characterization of pressure transient intensity: pressure sequence within a time window Perform transient detection. First, calculate the rate of pressure change. Candidate mutation segments are identified using an adaptive threshold; the magnitude of pressure mutations is calculated within each candidate mutation segment. Meanwhile, the statistical pressure is below the negative pressure threshold. The duration of the continuous period is obtained from the duration of the negative pressure window. Furthermore, the transient components are obtained by high-pass filtering or wavelet decomposition of the pressure sequence. and with transient component energy Characterizing transient intensity. The above detected quantities are normalized to... , , , The transient stress intensity characterization quantity was obtained by weighted fusion according to confidence. ;in Furthermore, it can be set and recorded according to sensor stability or historical discrimination. To avoid single-point false triggering, consistency checks can be performed on pressure mutations at neighboring nodes; when a single-point mutation cannot be supported by neighboring nodes, this time window is marked as low reliability and reflected in subsequent uncertainty characterization. In a typical example, assume... , , , The reasons are as follows: The magnitude of pressure surges is given a high weight because it directly reflects the severity of system pressure changes and is often a major indicator of pollution or accidents. Weights are assigned to the rate of change of pressure because its changes can quickly indicate anomalies in the system. The duration of the negative pressure window is assigned as an important indicator reflecting the safety of the pipeline network. The transient component energy is given a lower weight because it is more closely related to details over a short period of time and has a relatively smaller impact.

[0028] Example of obtaining the return flow reachability representation: Based on the pipeline network topology, construct a node-pipe segment connectivity graph, and use pressure / flow to estimate the flow direction and pressure difference constraints of each pipe segment in the current time window to form a directed constraint graph.

[0029] Method 1 (Hydraulic Simulation): For the set of potential backflow sources... Perform multi-scenario simulations to obtain the return flow reaching the target node. reachability probability With propagation delay Mapping these two to return flow reachability: ;in for The normalized value, This is the normalized value for latency (the shorter the latency, the larger the value). An example allocation is... , The reason is as follows: the probability of reflux reachability is the core indicator for determining whether contamination can reach the target node, and it plays a dominant role, therefore it is given a high weight. The time sensitivity of the impact of propagation delay on risk can reflect the urgency of rapid response, but it is relatively weaker than core accessibility, therefore it is assigned... The weighting of these factors ensures accessibility priority while also taking into account time factors, thus improving the accuracy of overall risk characterization.

[0030] Method 2 (Graph Reachability): In scenarios where a full simulation is not required, starting from the source... To the node Search for paths that satisfy the "consistent flow direction / pressure difference constraints," and if such a path exists, mark it as reachable; then use the path length as the criterion. Pressure margin Number of key valve nodes achievable strength ; then normalize to get The two methods can be adaptively switched according to data reliability and real-time requirements, and method markers are retained for traceability.

[0031] Example of obtaining toxicity suspicion characterization parameters: for each water quality indicator Establish dynamic baselines With scale (This data can be obtained from historical data of the same operating conditions / same day / week / stable operating intervals), and a standardized anomaly score is calculated. The abnormal scores of multiple water quality indicators are weighted and merged into a total water quality abnormality score. ;in and , This represents the result after normalizing the abnormal scores of each indicator. Subsequently, spatiotemporal consistency correction is performed: external events are analyzed into occurrence time, affected area, category, and intensity; user complaints are analyzed into complaint time, location, theme, and quantity intensity; and consistency factors such as "temporal overlap / clustering" and "spatial proximity / coverage" are calculated for each time window. When external events and complaints support water quality anomalies, [the following is applied to the analysis]. Adjust the gating; when water quality is abnormally isolated and lacks support, adjust the gating accordingly. Downgrade or label as a low-evidence anomaly. In a reproducible implementation, an external event correction factor can be used. Complaint Correction Factor Saturated nonlinear fusion was performed, and the corrected toxicity suspicion level was obtained: ;in To avoid excessive boosting of single-source noise, a saturated nonlinear function is used. To enable calibrated parameters and record them. Finally, for... Normalization yields the toxicity suspicion characterization quantile. Example: Take , , The reasons are as follows: External events (such as construction, pipe bursts, and water outages) usually have a more specific triggering characteristic, hence the need for an event correction factor. Assign higher weight User complaints Susceptible to subjective and delayed influences, its weight as supplementary evidence should be slightly reduced. . Controlling the "increase" and taking a moderate value can appropriately raise the risk when there is consistent support, while avoiding excessive amplification and false alarms caused by a single event / complaint.

[0032] Example of obtaining external event and complaint correction factors: In one embodiment, external event correction factors... This can be obtained by parsing external event data (such as construction, pipeline maintenance, pipe bursts, water outage scheduling, etc.). Each event record includes the event type, occurrence time, affected area, and impact intensity. For the target time window... The set of events that are spatially adjacent to the target node or overlap with the affected pipe segment within the time window is statistically analyzed. And assign weights according to event type. (This can be determined based on historical risk correlation). The formula for calculating the external event correction factor is as follows: ; in The time overlap factor (the degree of overlap between the event and the current time window, with values ​​ranging from 1 to 10). arrive ), The spatial proximity factor (the coverage of the event's affected area and nodes, ranging from $0$ to $1$) ensures... For example: a construction event has a weight of 0.7, and a pipe burst event has a weight of 1.0; if a construction event affects 50% of the nodes and the time window is fully covered, then the contribution of this event is... .

[0033] A similar complaint correction factor can be obtained from user complaint data. Each complaint record includes the complaint time, location, subject, number, and severity score. For time windows... Statistically count the set of complaints that are close to the target node. And empower according to the severity of the complaint. The calculation formula is as follows: ; in and Same as above, ensuring normalization For example: If a complaint pertains to the topic of unusual water odor (weight 0.8), occurs within the coverage area of ​​a nearby pipe network, and the time windows overlap, then the contribution of this complaint is... .

[0034] Example of obtaining uncertainty characterization: In one embodiment, uncertainty characterization is determined by at least one source of uncertainty, including a source of data reliability, a source of prediction consistency, and a source of multi-scheme consistency.

[0035] Data reliability source: Statistical analysis of the missing test rate in each time window. Outlier ratio Time synchronization deviation ,structure Example: Take , , The reasons are as follows: the missing detection rate directly weakens the computability and continuity of features, and is often the primary source of uncertainty, hence it has the highest weight; the outlier ratio reflects sensor drift / noise, and has a significant impact on short windows, hence it has the second highest weight; synchronization deviation can mostly be mitigated by alignment and interpolation, and is only significant in multi-source fusion, hence its relatively low weight, balancing robustness and usability.

[0036] Sources of prediction consistency: Comparing propagation predictions with subsequent observations to calculate arrival time deviations. Deviation from the scope of influence ,structure Example: Take , The reasons are as follows: arrival time deviation affects the timeliness of early warning and response, and is more sensitive to "whether to respond in advance", so it is higher; the influence range deviation reflects the spatial diffusion error, but it can usually be gradually corrected with supplementary observations, and the immediacy of single-node judgment is slightly weaker, so it is appropriately downweighted to achieve "time priority and range consideration".

[0037] Sources of consistency across multiple scenarios: The dispersion measures are used to analyze the path differences, range differences, and arrival time variances output by different propagation prediction methods under the same operating conditions, and a framework is constructed. Example: Take , , The reasons are as follows: once the path differences diverge, it often means that the judgment of flow direction / valve status is inconsistent, which has the greatest impact on source tracing and scheduling; the range difference determines the coverage affected, and the importance is similar but slightly less; the variance of arrival time reflects more the differences in model details and clock errors, which can usually be corrected by observation, so the weight is relatively low.

[0038] The above uncertainty scores are normalized and aggregated to obtain... ; and will Used as a suppression term or dynamic weighting in dynamic risk score fusion, it makes the risk output more prudent when uncertainty is high, thereby reducing false alarms and improving system robustness. Example: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] , , The reasons are as follows: Data reliability is the root cause of uncertainty at the "input end". If the number of missing or abnormal data is high, other uncertainty assessments will also be distorted, hence it is the highest; Prediction consistency can reflect the deviation between the model and the observation and directly guide risk suppression, so it is the second highest; Multi-scheme consistency has limited distinguishing power when there are insufficient methods or models from the same source, so it is given a lower weight as a supplementary item to avoid over-amplifying model divergence.

[0039] The anomaly event identification module is used to identify anomalies based on dynamic risk scores and output event type confidence levels. This module further transforms "risk status" into "manageable events." The core of event-based identification is to link continuously changing risk scores with multi-source evidence to determine whether an anomaly event has occurred, categorize the abnormal behavior into a specific event type, and output the event type confidence level to express the reliability of the event type determination. The event type confidence level supports the alarm closed-loop feedback module in generating more interpretable and traceable alarm and handling information. It also supports the dynamic linkage control module in selecting different control strategies, control intensities, or verification paths at different confidence levels, thereby avoiding over-handling when evidence is insufficient and avoiding delayed handling when risk is high and confidence is high.

[0040] In one specific implementation, candidate event triggering and segmentation are performed based on dynamic risk scores: the dynamic risk scores are continuously monitored in a unified time window. When the dynamic risk score exceeds a preset trigger threshold or shows a continuous upward trend in adjacent time windows, it is marked as a candidate abnormal segment. The start time, end time, and peak time of the candidate abnormal segment are determined. At the same time, an evidence container is established for each candidate abnormal segment to collect structured elements corresponding to pressure data, flow data, water quality data, external event data, and user complaint data that are synchronized with the time of that segment, thereby ensuring that the input for subsequent event-based identification is traceable and reproducible. Based on the evidence container, event type discrimination and type fingerprint formation are performed: For each candidate anomaly segment, the transient detection characteristics of the pressure transient intensity characterization quantity are preferentially calculated as the discrimination criterion for pressure transient events. When the pressure change amplitude and pressure change rate are significantly significant simultaneously in a short period of time and accompanied by anomalies in the duration of the negative pressure window or the energy of the transient component, a type fingerprint of the pressure transient event is formed. Furthermore, combined with the reachability path, reachability intensity, or propagation delay characteristics corresponding to the return flow reachability characterization quantity, when there is a return flow path that satisfies the flow direction pressure difference constraint and is consistent with the hydraulic conditions characterized by the pressure data and / or flow data within the candidate anomaly segment, a type fingerprint of the cross-connected return flow event is formed. Furthermore, when the anomaly score of the water quality data relative to the baseline is significantly increased within the candidate anomaly segment and the consistency between the external event data and the user complaint data is enhanced in time and space, and the return flow reachability characterization quantity indicates that the pollution can reach the sensitive node within a short propagation delay, a type fingerprint of the return flow high toxicity source event is formed. "Highly toxic source item" refers to a pollution source item that, when introduced into the water supply network, may lead to a significant increase in public health risks. The criteria for its determination are not limited by the "specific name of the toxic substance" but by the "hazard level index". The hazard level index can be determined by at least one source: (1) the online comprehensive water quality risk index exceeds the high-risk threshold (e.g., obtained by fusing abnormal scores of multiple indicators such as residual chlorine, turbidity, conductivity, and ORP); (2) the results of verification sampling (on-site / laboratory) indicate that there is a pollution risk that requires the initiation of a high-level emergency response; (3) external event data indicate that there is exposure to a highly toxic source item (such as chemical leaks, hazardous chemical transportation accidents, alarms at key sensitive points, etc.). When (1) is met and one of (2) or (3) is met, the event type is marked as "return of highly toxic source item event"; when only (1) is met and (2) / (3) is lacking, the event type is marked as "return of suspected highly toxic source item event", and further confirmation is triggered by verification sampling control.

[0041] The event type confidence level is calculated based on a multi-type competition mechanism: Confidence scoring functions are constructed for pressure transient events, cross-connected backflow events, and backflow high-toxicity source events. Each scoring function integrates at least the peak and persistence of dynamic risk scores, the matching degree of the corresponding type fingerprint, and the support strength of internal and external event data and user complaint data within the evidence container. A confidence reduction term is introduced for candidate anomaly segments with high uncertainty representation to prevent outputting artificially high confidence levels when evidence quality is insufficient. Subsequently, the three types of confidence levels are normalized to obtain an event type confidence vector. A strategy combining mutual exclusion and concurrency constraints is adopted: when the confidence level of a pressure transient event is significantly higher than other types and no synchronous anomalies are observed in the water quality data, the pressure transient event is output as the primary type. When the confidence levels of both cross-connected backflow events and backflow high-toxicity source events are high, the cross-connected backflow event is allowed to be output as the primary type, with the backflow high-toxicity source event labeled as an accompanying type, to reflect the composite situation where backflow paths and water quality anomalies occur simultaneously. The process involves encapsulating and recording event-based results: Candidate abnormal segments are encapsulated as event objects. Each event object includes at least the event occurrence time, peak time, event type, event type confidence level, and a summary of supporting evidence. This ensures the event output is interpretable and facilitates subsequent processing. For example, if, within two consecutive time windows in a certain area, the dynamic risk score continuously increases and exceeds the trigger threshold, pressure data shows a significant short-term pressure mutation amplitude and a rapid increase in the pressure change rate, and water quality data shows a synchronous increase in turbidity and conductivity relative to the baseline, external event data records maintenance switching near the area, user complaint data reflects odor and turbidity in the same time period, and the return accessibility indicator shows a reachable path from the suspected cross-connection point to multiple user nodes, then the event-based identification outputs high confidence levels for both the cross-connection return event and the return high-toxicity source event. The cross-connection return event is identified as the primary type, and the return high-toxicity source event is identified as the secondary type, thereby achieving refined identification and executable expression of abnormal events.

[0042] The propagation trend prediction module generates propagation prediction results based on pipeline topology information and dynamic risk scores. This module answers the questions: "If an abnormal event exists or is occurring, how might it propagate along the pipeline network, where will it propagate to, when will it arrive, and what will be its impact?" Pipeline topology information describes the pipeline network structure and node connections, providing structural constraints for propagation simulation. Dynamic risk scores provide risk input conditions for propagation simulation, enabling predictions to not only rely on static structures but also reflect the risk situation changing over time. The propagation prediction results provide spatial and temporal guidance for the alarm closed-loop feedback module and key information for the dynamic linkage control module, such as control location selection, boundary setting, and resource scheduling sequence, thus further grounding "early warning" into an executable level of "how to organize and handle the situation at the pipeline network level."

[0043] In one specific implementation, a propagation path analysis framework based on pipeline network topology information is constructed: A complete pipeline network topology map is built based on the pipe segments, nodes, and their connections within the water supply network, incorporating pressure data, flow data, and return flow accessibility metrics for each pipe segment into the map structure. On this basis, the impact sources of potential abnormal events are identified using dynamic risk scores, determining the preliminary location and propagation direction of the event source points. Using pipeline network topology information and hydraulic constraints, a preliminary deduction of the propagation path of water flow in the pipeline network is performed. The purpose of this step is to determine possible propagation paths through a thorough analysis of the pipeline network structure and operating status, and to assess the propagation risk intensity of different paths by combining the risk level information provided by the dynamic risk scores. A refined simulation of the propagation path and calculation of the impact range are performed: Hydraulic simulation methods are used to simulate water flow behavior along the propagation path. Based on the pressure distribution, flow velocity, and water quality parameters of the pipe sections, the propagation direction and path of the abnormal event are inferred. By setting a propagation delay model, the differences in water flow rate in different areas of the pipe network are considered, thereby accurately predicting the possible arrival time of the abnormal event. Simultaneously, combined with the impact range model, the impact range of the abnormal event is calculated based on the characteristics of the event source point, the pressure distribution of the pipe network, and flow data, further determining the probability and area of ​​impact for each pipe section and node. This process ensures that the calculation of the propagation path and impact range is not limited to local pipe sections but can comprehensively and accurately reflect the propagation situation of the entire pipe network. The propagation results are corrected by combining external event data and user complaint data: Based on the preliminary calculation results of the propagation path and impact range, external event data (such as maintenance, construction, weather changes, etc.) and user complaint data (such as low pressure, abnormal water quality, etc.) are introduced to correct the propagation prediction results. By calculating the spatiotemporal impact of external events on the pipeline network, the propagation prediction results are corrected in time and space, so that the prediction not only relies on internal pipeline network data, but also takes into account the impact of external disturbances on the propagation path. For example, when construction and abnormal weather occur simultaneously in a certain area, it may cause changes in water flow direction, thereby affecting the prediction results. The system should make appropriate adjustments based on these external events to improve the prediction accuracy. The system generates propagation prediction results and outputs an impact probability envelope: Based on the aforementioned path simulation, delay calculation, and impact range estimation, combined with propagation data corrected by external events and user feedback, it outputs the propagation path, impact range, arrival time, and impact probability envelope. The impact probability envelope describes the probability interval of each region being affected within a specific time window. This result can provide decision-makers with real-time early warnings, helping them prioritize high-risk areas. For example, if, during a certain period, the backflow propagation path extends along a pipe segment and affects downstream areas, the impact probability envelope will show a high probability of that area being affected, allowing relevant departments to prioritize resource allocation for emergency response. This process not only ensures the accuracy of propagation trend prediction but also enhances the early warning system's adaptability to various interference factors, thus providing reliable decision support for handling abnormal events.

[0044] The impact probability envelope describes the upper and lower bounds of the probability that each node or pipe segment will be affected by contamination within different time windows in a given prediction time domain. (Based on nodes...) In the time window "Internal pollutant concentration exceeds threshold" "For an influencing event, the probability of influence is defined as..." ; In one implementation, multi-scenario sampling is performed for uncertain operating conditions, and hydraulic / water quality simulations are invoked to obtain the results. The results show that the uncertain operating conditions include at least water demand, valve status, return source point, and source strength; let the... Sub-scenario simulation within a time window internal nodes The concentration results are The probability of influence can then be estimated from the sample frequency. ; in This is the indicator function. Further, it is the output influence probability envelope, within the same time window. Below Or for the corresponding Statistical analysis was conducted based on preset quantiles. and Constructing upper and lower bounds of probability ; in Represents the quantile operator. for The lower bound corresponding to the quantile for The upper bound corresponding to the quantile is thus formed into a node. In the time window The probability envelope of the impact within the pipe segment. The probability envelope of the impact on the pipe segment can be defined and calculated in the same way as that of the node, or the probability envelope of the impact on the key nodes on the pipe segment can be obtained according to a preset rule (such as taking the maximum value or weighted aggregation).

[0045] For scenarios with higher real-time requirements, graph reachability results and propagation delay distribution can be used to simulate propagation delays without requiring a complete hydraulic / water quality simulation. Approximate construction is performed: First, based on the flow direction / pressure difference constraints, the distance from the suspected source point to the node is obtained. The reachability determination and its propagation delay statistics are then used to determine the "reachability strength" and the "arrival time falling within the time window". The probability of the approximate impact probability is obtained by combining the probabilities of the two scenarios, and the type of method used (e.g., "multi-scenario simulation method" or "graph reachable approximation method") is marked in the output for traceability and interpretation.

[0046] The dynamic linkage and control module is used to execute dynamic control measures based on dynamic risk scores and propagation prediction results. This module transforms risk assessment and propagation prediction into dynamic control actions, achieving a closed-loop linkage of "risk-prediction-control". The dynamic risk score characterizes the current strength and trend of risk, while the propagation prediction results characterize the potential impact path and scope. Together, they determine whether control measures are initiated, their priority, scope, intensity, and sequence. More proactive control measures are taken promptly when risk increases or the possibility of propagation rises; more conservative and verification-oriented control measures are adopted when risk decreases or evidence is insufficient; and control strategies are adjusted in real time when the risk and propagation situation changes, thereby improving the system's response speed, resource utilization efficiency, and targeted approach to abnormal events.

[0047] In one specific implementation, sampling resources are adjusted based on dynamic risk scores and propagation prediction results: when the dynamic risk score exceeds a preset threshold and the propagation prediction results indicate that the abnormal event may expand in a short period of time, sampling resource adjustment is initiated. The core purpose of sampling resource adjustment is to increase the data sampling frequency or expand the sampling points in high-risk areas to improve the real-time performance and accuracy of risk identification. Specifically, when the dynamic risk score of a certain area reaches a high-risk level, the adjustment strategy will adjust the sampling frequency, changing the data sampling from timed sampling to real-time sampling, or adding additional sampling points in key pipe sections and nodes to enhance data density and spatial coverage. At the same time, the selection of sampling points is based on the propagation prediction results, focusing on pipe sections that may become the source of the event and their upstream and downstream key nodes, ensuring that the collected sample data has high representativeness and completeness, and providing accurate data support for subsequent event identification and handling. Evidence weighting is adjusted based on dynamic risk scores and propagation prediction results: In the event identification process, the weight of evidence determines the influence of various types of evidence in the final decision. Evidence weighting can be effective when a source of evidence (such as pressure data, flow data, water quality data, etc.) is unstable or has low credibility. By monitoring dynamic risk scores in real time and combining this with the risk areas shown in the propagation prediction results, the evidence weights are dynamically adjusted. For example, if pressure data in a certain area fluctuates significantly and the risk level of that area is high, the system will automatically reduce the weight of pressure data in that area. Conversely, when external event data (such as user complaints, weather changes, etc.) has a strong correlation with abnormal water quality data, the weight of the relevant evidence will be increased. This adjustment mechanism helps the system prioritize higher-quality evidence when there is conflict among multiple sources of evidence, thereby reducing misjudgments caused by data inconsistencies. The system implements isolation boundary control and transient suppression control: Isolation boundary control primarily targets abnormal diffusion in localized areas. It sets isolation boundaries within the pipe network to prevent the uncontrolled spread of pollutants or risks. When propagation predictions indicate a high risk of pollutant diffusion in a certain area, the system automatically generates and controls the isolation boundary of the pipe network based on the predicted propagation path and impact range. For example, in the event of a sudden change in water quality, dynamic linkage control immediately adjusts valve positions, shutting off the water supply to the affected pipe section to prevent the risk from spreading to other areas. Simultaneously, transient suppression control addresses the risks posed by instantaneous pressure fluctuations or rapid changes in water flow. When the system detects a sudden pressure change or a drastic change in flow, it automatically controls the flow or pressure of the relevant pipe section according to transient suppression strategies, reducing the risk of potential damage to the pipe network structure or water pollution. For example, by adjusting flow rate and pressure, it mitigates instantaneous fluctuations in water flow, avoiding negative pressure areas or extreme conditions of excessively high pressure, thereby protecting the pipe network.

[0048] Verification sampling control and complaint-linked control are implemented: Verification sampling control is used to strengthen verification sampling in a specific risk area to ensure the effectiveness of emergency measures taken in high-risk areas. Verification sampling control will be activated when the system's propagation prediction and risk assessment results for a specific area indicate potential errors. The system will increase the number of sampling points for focused verification to confirm the effectiveness and accuracy of emergency measures. For example, when a suspected cross-connection backflow event occurs in a localized part of the pipeline network, the system will increase the number of sampling points and the sampling frequency in that area to verify the authenticity of the backflow and the extent of pollution spread. Complaint-linked control is based on user complaint data for feedback control. When users submit numerous complaints regarding water quality, quantity, and pressure, the system will combine complaint information with dynamic risk scores to confirm whether the complained area is in a risk zone and prioritize control measures according to the risk level. For example, when user complaint data centrally reflects abnormal water quality in a specific area, the system will prioritize analyzing the pipeline network in that area and implement corresponding control measures, such as adjusting the water quality monitoring frequency or initiating corresponding water quality improvement measures.

[0049] The alarm closed-loop feedback module generates alarm and handling information based on event type confidence, propagation prediction results, and dynamic control, and receives handling feedback to update the system status, forming a closed loop. This module transforms internal system analysis results into executable and interpretable outputs, and feeds handling results back to the system for a closed-loop update. Alarm and handling information is driven by event type confidence, propagation prediction results, and dynamic control: event type confidence indicates "what kind of event it is and how confident is the assessment," propagation prediction results indicate "which areas and time windows might be affected," and dynamic control indicates "what control actions the system has suggested or already implemented." Simultaneously, receiving handling feedback rewrites the effects of on-site handling, verification results, or operational adjustments back into the system status, enabling subsequent dynamic risk assessment, abnormal event identification, and propagation trend prediction modules to continue analysis based on the updated status. This avoids a disconnect between alarms and handling, ensuring the early warning system operates continuously and self-consistently, forming a true closed-loop management system.

[0050] In one specific implementation, candidate alarms are generated and initially ranked based on event type confidence: After receiving the event objects from the event-based identification output, the event type confidence is used as the core indicator to prioritize each event. A preliminary alarm list is then generated by combining the event's occurrence time, peak time, and supporting evidence summary. During the ranking process, high-confidence events are given priority in the alarm release sequence, while low-confidence events are marked with a suspicious flag for subsequent verification sampling and dynamic linkage control. For example, when a pressure transient event is detected in a certain area and the confidence exceeds a threshold, while the confidence of the high-toxicity source event is moderate but accompanied by a high concentration of user complaint data, the events in this area will be prioritized for alarm purposes. The alarm content is enriched by combining propagation prediction results and impact probability envelopes: For initial alarm events, refined alarm content is generated using propagation path, impact range, arrival time, and impact probability envelope information, clearly identifying the affected areas and time windows, and providing risk propagation trends and potentially affected key nodes. For example, in cross-connection backflow events, the affected pipe sections and downstream nodes are marked using propagation prediction results, and the probability envelopes are combined to provide probability ranges for the degree of risk in different areas, allowing response personnel to prioritize and allocate resources according to the level of risk. The system combines dynamic linkage control with alarm handling strategies and recording actions: After an alarm is generated, based on the event type, confidence level, and propagation trend, it triggers sampling resource control, isolation boundary control, transient suppression control, or complaint-linked control, and attaches the control measures taken and their effectiveness indicators to the alarm record. This process ensures that alarms are not only information notifications but also linked to actual control actions, thus forming an executable closed loop of "early warning—handling—verification". For example, in a water quality anomaly event, the system automatically adjusts the density of local sampling points, closes key valves to form an isolation boundary, and simultaneously records the control time and a list of affected nodes in the alarm information; The system receives and updates status feedback to form a closed loop: After responding to an alarm, the handling unit sends back the verification results, sampling and testing results, and user feedback. The system compares and verifies the feedback information with the initial event information and propagation prediction results, adjusts the event type confidence level, dynamic risk score, and propagation prediction model parameters, updates the internal status, and optimizes the trigger threshold and priority allocation in subsequent alarm generation. For example, in the aforementioned water quality anomaly event, if the verification sampling shows that the pollution has not spread along the predicted path, the system will adjust the propagation probability model of the relevant pipe section and reduce the alarm priority of similar events, thereby achieving closed-loop optimization and making subsequent alarms more accurate and executable. Through this implementation method, the alarm closed-loop feedback not only provides information notification functions but also closely integrates analysis, handling, and verification to achieve real-time, traceable, and optimizable closed-loop management.

[0051] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A water affairs anomaly event early warning system based on data acquisition and analysis, characterized in that, include: The multi-source data acquisition module is used to collect water supply network data, as well as external event data and user complaint data, to form time-synchronized time-series data; The dynamic risk assessment module is used to calculate dynamic risk scores based on time-synchronized time-series data. The abnormal event identification module is used to identify abnormal events based on dynamic risk scores and output the event type confidence level. The propagation trend prediction module is used to generate propagation prediction results based on pipeline topology information and dynamic risk scores. The dynamic linkage control module is used to perform dynamic control based on the dynamic risk score and the propagation prediction results; The alarm closed-loop feedback module is used to generate alarm and handling information based on event type confidence, propagation prediction results and dynamic control, and to receive handling feedback to update the system status, forming a closed loop.

2. The water affairs abnormal event early warning system based on data acquisition and analysis according to claim 1, characterized in that, Water supply network data includes pressure data, flow data, and water quality data.

3. The water affairs abnormal event early warning system based on data acquisition and analysis according to claim 1, characterized in that, The dynamic risk score is determined by the stress transient intensity characterization, reflux accessibility characterization, toxicity suspicion characterization, and uncertainty characterization.

4. The water affairs abnormal event early warning system based on data acquisition and analysis according to claim 3, characterized in that, The transient intensity characterization of pressure is obtained based on pressure data; The return flow accessibility characteristic is calculated based on the pipeline topology information and combined with hydraulic conditions characterized by pressure data and / or flow data. The toxicity suspicion index is derived from the anomaly score of water quality data relative to the baseline and is corrected based on the spatiotemporal consistency between external event data and user complaint data. The uncertainty characterization quantity is determined by at least one of the following sources of uncertainty: data reliability source, prediction consistency source, and multi-scheme consistency source.

5. A water affairs anomaly early warning system based on data acquisition and analysis according to claim 4, characterized in that, The intensity of pressure transients is characterized by transient detection of pressure data. Transient detection includes at least one or more of the following: pressure change amplitude, pressure change rate, negative pressure window duration, or transient component energy.

6. A water affairs anomaly early warning system based on data acquisition and analysis according to claim 4, characterized in that, The calculation of the return flow accessibility characteristics includes at least one or more of the following: the accessibility probability and propagation delay calculation obtained from hydraulic simulation, or the graph accessibility calculation based on flow direction / pressure difference constraints.

7. A water affairs anomaly early warning system based on data acquisition and analysis according to claim 4, characterized in that, The uncertainty characterization quantity is determined by at least one of the following sources of uncertainty: Data reliability is determined by the missing data rate and / or the proportion of outliers and / or time synchronization deviation. Sources of predictive consistency: Determined by the deviation between the output of the propagation situation prediction module and the corresponding observation data, the deviation including at least one or more of arrival time deviation and / or influence range deviation; Sources of consistency across multiple scenarios: Determined by the degree of dispersion between propagation prediction results obtained from different types of models under the same operating conditions. The degree of dispersion includes at least one or more of the following: propagation path difference, impact range difference, or arrival time variance.

8. A water affairs anomaly early warning system based on data acquisition and analysis according to claim 1, characterized in that, The abnormal event identification module outputs at least the following event types: pressure transient event, cross-connection backflow event, and backflow of highly toxic source item event.

9. A water affairs anomaly early warning system based on data acquisition and analysis according to claim 1, characterized in that, The propagation prediction module generates propagation prediction results including propagation path, scope of impact, arrival time, and probability envelope of impact.

10. A water affairs anomaly early warning system based on data acquisition and analysis according to claim 1, characterized in that, The dynamic control measures executed by the dynamic linkage control module include at least sampling resource control, evidence weight control, isolation boundary control, transient suppression control, verification sampling control, and complaint linkage control.

Images