Analysis, prediction and early warning method and system for abnormal water consumption of water meter

Abstract

The invention discloses an abnormal water consumption analysis, prediction and early warning method and system for a water meter, and belongs to the technical field of fault prediction and health management, and the method comprises the following steps: executing parameter configuration which comprises an abnormal configuration rule, a responsible person binding rule and a disposal time limit rule; the method comprises the following steps: collecting metering reading and water meter state data of each remote water meter and inlet flow metering data of each DMA partition in a target period, establishing an incidence relation between the water meters and the DMA partitions, and forming a partition statistical time window; synchronously collecting scene data related to the DMA partition, generating a scene event, and writing the scene event into an abnormal pool according to a DMA identifier and a time window identifier; preliminarily screening the water meter usage according to the exception rule according to a statistical time window to obtain a candidate exception water meter set, converting the candidate exception water meter set into candidate events, and writing the candidate events into an exception pool; the problems that in the prior art, water consumption abnormity explanation is not uniform and not communicated, and water meter abnormity information can only be obtained through passive query are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault prediction and health management, in particular to a water meter abnormal water use analysis, prediction and early warning method and system. BACKGROUND

[0002] In the field of water supply, there are many cases of abnormal water use of water meters, such as construction repair, water use scheduling, DMA leakage, water meter damage, meter reading error of meter readers, and user side leakage, etc.

[0003] If the abnormal water use of the water meter is not analyzed, it is difficult to predict the trend of future abnormal water use of the water meter, and the problem cannot be solved by checking the water meter in time, which may cause unpredictable losses.

[0004] Therefore, more and more enterprises begin to analyze, predict and warn the abnormal water use of the water meter. In the prior art, there are methods of identifying water use property anomaly based on user water use history characteristics and monitoring industrial large user anomaly, etc. The model is constructed by using features such as average value, variance, slope and sequence similarity, to realize the identification of "residential / commercial property mismatch" or "single user water use curve anomaly". Through historical data, the water meter water use situation is analyzed, a fitting curve is made to predict and warn the water meter water use anomaly, which realizes more efficient and more accurate prediction of water use anomaly and clearer warning of water use anomaly.

[0005] For example, the intelligent water monitoring and early warning method based on the Internet of Things disclosed in Chinese patent No. CN119761661B collects water flow data and temperature data through the Internet of Things and uploads them to a data analysis platform. After preprocessing the original data, historical water flow data is introduced, and a dynamic weighted sliding window prediction algorithm is used to obtain water flow prediction values. Based on the actual water flow data and the prediction values, the water flow deviation value is calculated, and the temperature correction coefficient is calculated according to the temperature data. The temperature correction coefficient and the water flow deviation value are fused to obtain a comprehensive error value, and the risk assessment value is further calculated. At the same time, the risk threshold is dynamically adjusted based on the comprehensive error value, and the risk assessment value is compared with the dynamic threshold to determine whether the warning is triggered.

[0006] For example, the specific area enterprise industrial water anomaly monitoring and early warning method disclosed in Chinese patent No. CN114611764B includes: the method combines LSTM and chaos analysis to extract global and local features of enterprise water use and fuse them. On this basis, the prediction models of regional total water use and single enterprise water use are established simultaneously, and the local features and global features are combined to realize the detection of enterprise water use anomaly in a specific area.

[0007] The above-mentioned technology at least has the following technical problems:

[0008] The prior art in water meter abnormal water consumption analysis and prediction warning usually only provides water quantity abnormal result display or query function for single water meter, and the whole still stays at the single point capability level, lacking the covering whole process warning closed loop mechanism. For example, user side hidden leakage, resident household toilet leakage, water heater safety valve dripping or secondary water supply facility slow leakage, etc., often show as continuous small flow at night or baseline lifting all day. Although the existing system can identify the abnormality, due to the lack of work order disposal process and review closing mechanism, the abnormal information often stays in the platform page for a long time, eventually accumulating into high water fee dispute and complaint. For example, DMA partition hidden leakage usually presents a gradual trend of minimum night flow continuously rising and daily water consumption slowly rising. If the closed loop of "trend warning-partition review-site leakage detection-repair-effect tracking" is lacking, it will lead to leakage expansion and uncontrollable loss. For example, short-term concentrated water replenishment, sudden increase and sudden decrease caused by construction repair, pipe network cutting and changing or water restoration, the existing system is difficult to link with planned events and work order information for elimination, which is easy to cause false alarm accumulation, increase the burden of manual screening and reduce the credibility of warning.

[0009] At the same time, for the identified water meter abnormal information, the existing technology relies on meter readers or management personnel to log in the platform for passive query, lacking active push and responsibility binding mechanism, and it is difficult to drive timely disposal. For example, "continuous zero value / discontinuity" caused by communication interruption, data missing or time stamp disorder. If the abnormality only stays in the platform for checking state, the meter reader cannot find it in time without timely login, which often delays the settlement period and causes the increase of re-reading cost and billing disputes. For example, industrial user sudden abnormal large flow (may be pipe burst, valve misoperation or production drainage abnormality), which belongs to high urgency event that needs to link customer service and repair team quickly. But the existing system lacks the ability to distribute and notify by area, responsibility and level, and cannot realize the automatic matching of "abnormality-responsibility person-processing time limit", which is easy to cause disposal delay and loss expansion. In addition, the abnormal information lacks due diligence, overdue upgrade and processing feedback, which leads to the dependence of abnormal disposal process on manual transmission, and the problems of single loss, single leakage and untraceable processing.

[0010] Further, the prior art generally lacks statistical analysis and quantitative control capability of water meter abnormal data, leading to difficulty in forming a measurable operation closed loop for abnormal management. For example, it is difficult for the management side to quickly obtain the number of abnormal water meters in a statistical period, the proportion by type (leakage, meter damage, incorrect reading, communication anomaly, etc.), the distribution of high-incidence communities and high-incidence DMA partitions, and other indicators, making it impossible to reasonably arrange inspection resources, develop leakage management plans, or conduct performance evaluation. For another example, after implementing leakage management or meter replacement, the existing system cannot provide comparative analysis before and after management, cannot quantify the reduction of abnormality, the change of minimum flow at night, or the estimated water recovery, making it difficult to evaluate input and output and to iteratively optimize management strategies. Due to the lack of the above capabilities, the prior art often presents a state of fragmentation of data, personnel, and events: abnormal identification, work order assignment, repair record and review archiving are scattered in different links or different systems, there is a lack of unified abnormal early warning and closed-loop management system, and it is difficult to meet the actual needs of water supply enterprises for rapid discovery, active contact, hierarchical disposal, and continuous optimization of abnormal water use. SUMMARY

[0011] To solve the technical problem that the prior art only provides water quantity abnormal result display or query function for single block water meter, the embodiments of the present application provide a water meter abnormal water use analysis, prediction and early warning method and system. The technical solution is as follows:

[0012] S1: Perform parameter configuration, including configuring partition statistical time window, abnormal rule, responsibility person binding rule and disposal time limit rule; collect the metering reading and water meter state data of each remote water meter in the target period and the inlet flow metering data of each DMA partition, establish the association relationship between the water meter and the DMA partition and form the partition statistical time window; synchronously collect the scene data related to the DMA partition and generate scene events, and write the scene events into the abnormal water pool according to the DMA identifier and the time window identifier.

[0013] S2: According to the abnormal rule, the water consumption is preliminarily screened according to the statistical time window, the candidate abnormal water meter set is obtained, and the candidate events are written into the abnormal water pool.

[0014] S3: For each DMA partition, the inlet cumulative amount and the cumulative amount of water meters in the partition in the corresponding time window are obtained, and the partition imbalance amount is determined according to the difference, and the imbalance type label is generated based on the minimum flow feature of the low disturbance period feature, and the partition imbalance amount and the imbalance type are written into the abnormal water pool as the DMA label.

[0015] S4: performing scene explanation and sorting processing based on a machine learning classification model on the candidate events and scene events in the same DMA partition and matching time window, obtaining scene explainable events and to-be-verified events; calculating explanation priority scores of the candidate events and sorting them from high to low, and sequentially selecting candidate events to form an explanation event set until the remaining unexplained part of the partition imbalance is less than the tolerance or the candidate events are exhausted.

[0016] S5: generating and dispatching active verification tasks for the to-be-verified events according to the responsibility person binding rules, receiving verification results to update the channel reliability and evidence fields, generating early warning information and pushing it to the user end; for the scene explainable events, generating explanation records and writing them into the anomaly pool, and showing the status or reducing the push level on the user end.

[0017] In step S1, the parameter configuration includes configuration of partition statistical time window, anomaly rule, responsibility person binding rule and disposal time limit rule, and the specific content is: the partition statistical time window at least includes a first statistical scale time window, a second statistical scale time window and a feature time window of a low disturbance period, and the first statistical scale is smaller than the second statistical scale, wherein the first statistical scale time window is used for preliminary screening of the first statistical scale anomaly rule, the second statistical scale time window is used for preliminary screening of the second statistical scale anomaly rule, and the feature time window of the low disturbance period is used for extracting the low disturbance period feature of the DMA inlet flow; the anomaly rule includes the first statistical scale anomaly rule and the second statistical scale anomaly rule, wherein the first statistical scale anomaly rule is used to distinguish the use amount anomaly in the first statistical scale time window, and includes use approach range discrimination and use approach comparison discrimination, the second statistical scale anomaly rule is used to distinguish the use amount anomaly in the second statistical scale time window, and is based on the deviation of the target statistical time window use amount from the historical statistical benchmark to distinguish, and writes the deviation direction and deviation amplitude into the anomaly amount field; the responsibility person binding rule is used to automatically bind the candidate events or to-be-verified events to the corresponding responsibility person; the disposal time limit rule is used to set the disposal time limit according to the anomaly level.

[0018] The specific content of the scene data is: the scene data at least includes running organization event data; the running organization event data is used to represent a planned event that has an impact on the water supply working condition or pipe network operation mode or water use boundary condition of the target DMA partition, and a scene event is generated based on the scene data, and at least records the impact range, start and end time and impact direction fields.

[0019] The water meter use amount is preliminarily screened according to the abnormal rule in the statistical time window, and the water meter use amount preliminarily screened according to the abnormal rule further includes meter reading abnormality preliminary screening and water meter state abnormality preliminary screening; the water meter state abnormality preliminary screening is matching discrimination based on water meter state data and state abnormality discrimination conditions, and a candidate event is generated when one or more of offline, under-voltage, empty pipe, dripping or device reporting abnormality is detected; the meter reading abnormality preliminary screening is matching discrimination based on meter reading data formation conditions and meter reading abnormality discrimination conditions, and a candidate event is generated when no meter reading occurs in a plurality of first statistical scale time windows in succession or no valid start point reading / endpoint reading is formed in a preset time window, resulting in that the use amount cannot be calculated; the candidate event records at least a water meter identifier, a DMA identifier, a time window identifier, an abnormality type, an abnormality amount, an abnormality duration, data completeness and a state evidence field; the abnormality type at least includes water quantity abnormality, water meter state abnormality and meter reading abnormality; the water quantity abnormality includes zero use amount abnormality in a plurality of first statistical scale time windows in succession or zero use amount abnormality in a plurality of second statistical scale time windows in succession; the meter reading abnormality includes meter reading abnormality in a plurality of first statistical scale time windows in succession; and the water meter state abnormality includes one or more of offline, under-voltage, empty pipe or device reporting abnormality.

[0020] In step S4, the candidate events and the scene events in the same DMA partition and matching time windows are subjected to scene explanation and sorting processing based on a machine learning classification model, specifically including: obtaining a water meter use amount sequence segment and a DMA inlet flow sequence segment corresponding to the candidate events, calculating a time overlap degree feature of the candidate events and the scene events, a consistency feature of the abnormality morphology of the candidate events and the DMA inlet flow change morphology in the scene period, a matching feature of the scene event type and the abnormality direction of the candidate events, and an explanation related feature obtained based on the abnormality amount, the abnormality duration, the data completeness, the data channel reliability and the evidence conflict situation of the candidate events; inputting the time overlap degree feature, the consistency feature, the matching feature and the explanation related feature into the machine learning classification model supervised and trained by the historical active verification results, and outputting a scene explanation probability and an explanation priority of each candidate event; when the scene explanation probability reaches a preset scene explanation threshold, marking the candidate event as a scene explainable event and generating an explanation record to be written into an abnormality pool, and reducing the push level or only displaying in a traceable state; when the scene explanation probability does not reach the preset scene explanation threshold, marking the candidate event as to be verified and writing it into a to-be-verified set, and the system selects the candidate events from the to-be-verified set in turn from high to low according to the explanation priority to form an explanation event set, and gradually explains the partition imbalance until the remaining unexplained part of the partition imbalance is less than a tolerance or the candidate events are traversed.

[0021] Wherein, the explainable event set is output to step S5 for the generation and dispatch of active verification tasks; after receiving the active verification result returned by step S5, the scene explainability conclusion and the abnormal authenticity conclusion corresponding to the active verification result and their associated identifiers are written back to the abnormal pool to form a closed-loop feedback record.

[0022] Wherein, the collection of metering readings and water meter state data of each remote water meter in the target period and the metering data of each DMA partition entrance in step S1 also includes parsing the abnormal code or state code of different types of terminals, normalizing them into uniform event codes according to the event mapping rule, writing them into the state evidence field, and adjusting the effectiveness mark of the event mapping rule after writing back the verification result.

[0023] Wherein, the specific process of generating and dispatching active verification tasks for the event to be verified in step S5 according to the responsibility binding rule is: first, determine the verification strength level requirement according to the uncertainty; based on the online state and capability support of the water meter, generate a set of feasible verification methods, select the verification method that meets the verification strength level requirement and has the lowest verification cost in the set of feasible verification methods; finally, dispatch the active verification task according to the selected verification method.

[0024] Wherein, the implementation process of uncertainty is: based on the data completeness, data channel reliability, evidence conflict mark and confidence boundary width of the scene explanation probability of the candidate event, the uncertainty of the candidate event is calculated; the confidence boundary width is used to represent the fluctuation range of the scene explanation probability; the data completeness is obtained by analyzing the effective sampling point ratio and / or the number of missing segments and / or the number of timestamp disorders in the statistical time window.

[0025] A water meter abnormal water consumption analysis, prediction and early warning system, comprising: a system configuration and data access module, an abnormal preliminary screening and abnormal pool module, a partition imbalance calculation and DMA tag module, an explanation screening and to-be-verified generation module, an active verification dispatch and push closed-loop module;

[0026] Wherein, the system configuration and data access module is used to perform parameter configuration, including configuring the partition statistical time window, the abnormal rule, the responsibility binding rule and the disposal time limit rule, collecting the metering readings and water meter state data of each remote water meter in the target period and the metering data of each DMA partition entrance, establishing the association between the water meter and the DMA partition and forming the partition statistical time window, synchronously collecting the scene data related to the DMA partition and generating the scene event, and writing the scene event into the abnormal pool according to the DMA identifier and the time window identifier;

[0027] The abnormal preliminary screening and abnormal pool module is used to preliminarily screen the water consumption according to the abnormal rule in the statistical time window to obtain a set of candidate abnormal water meters, and convert the set of candidate abnormal water meters into candidate events and write them into the abnormal pool.

[0028] The partition imbalance calculation and DMA label module is used for obtaining the entry cumulative amount and the partition internal water meter cumulative amount summary in a corresponding time window for each DMA partition, and determining the partition imbalance amount according to the difference, and generating an imbalance type label based on the minimum flow feature of the low disturbance period feature, and writing the partition imbalance amount and the imbalance type into the abnormal water pool as the DMA label;

[0029] The explanation screening and to-be-verified generation module is used for performing scene explanation and sorting processing based on a machine learning classification model on the candidate events and the scene events of the same DMA partition and the same time window, obtaining scene explainable events and to-be-verified events, calculating an explanation priority score of the candidate events, and sorting the candidate events from high to low according to the explanation priority score, and sequentially selecting the candidate events to form an explanation event set until the remaining unexplained part of the partition imbalance amount is less than a tolerance or the candidate events are traversed completely.

[0030] The active verification dispatching and pushing closed loop module is used for generating and dispatching an active verification task according to a responsibility person binding rule, receiving a verification result to write back the abnormal water pool to update the channel reliability and the evidence field, generating an early warning information and pushing to a user end, generating an explanation record and writing into the abnormal water pool for the scene explainable events, and showing a state or reducing a pushing level on the user end.

[0031] The technical scheme provided by the embodiment of the application brings at least the following beneficial effects:

[0032] 1. The abnormal rules are screened, the abnormal water meters obtained through the screening are converted into candidate events containing a water meter identifier, a time window, an abnormal type, an abnormal amount, continuous information, data integrity and a state evidence, and written into an abnormal water pool, and the DMA partition label is bound for the candidate events and the partition imbalance and the night minimum flow imbalance type label is generated, so that the abnormal information is associated with each other, the abnormal information is upgraded from "recognition-showing" to "attributable and focusable abnormal candidate set output", the alignment and association of the entry amount, the partition summary amount and the candidate events are completed in the same time window, and the linkage positioning and explanation of the DMA imbalance and the single table abnormality are realized.

[0033] 2. Through the explainable event balancing and active verification closed loop, in the same DMA partition and time window matching candidate event range, the explanation contribution is generated according to the abnormal amount, persistence, data integrity and channel reliability, and the evidence conflict and verification cost are sorted and selected to form an explanation event set, and at the same time, the active verification task of remote encrypted duplicate copy, remote resampling or on-site verification is automatically generated and dispatched for the explanation event set, and is pushed to the user end according to the responsibility person binding rule and the disposal time limit, so as to realize the "priority executable, responsibility to person, time limit processing, traceable review" of abnormal disposal, and further realize the timely discovery and rapid disposal of the communication interruption missing number, the industrial sudden large flow and the user side hidden leakage.

[0034] 3. Through the verification result write-back and statistical quantification control mechanism, the conclusion obtained by duplicate copy, resampling or on-site verification is written back to the abnormal pool, the evidence field, channel reliability and state evidence validity are updated synchronously, and the DMA partition, abnormal type, abnormal level, processing state, disposal time length, overdue situation and repeated abnormality are summarized and counted in the management end, supporting the comparison and analysis before and after the governance, so as to realize the continuous optimization of early warning rule and data quality and the measurable management of abnormal governance, and further realize the quantitative basis of leakage management, inspection resource scheduling and performance evaluation, effectively solving the problems of lack of statistical analysis and governance effect evaluation ability and difficulty in forming operation closed loop in the prior art.

[0035] 4. Through the scene explanation and sorting processing based on machine learning classification model for the same DMA partition and time window matching candidate event and scene event, whether the event can be explained is no longer dependent on the window overlap, but is determined by time overlap, direction matching, shape consistency and evidence quality, so that in the scenes such as planned flushing, repair cutting and changing, water recovery, user opening / communication change, explanation records can be automatically generated and degraded display; at the same time, events with low explanation probability and wide confidence boundary (large fluctuation range) are kept in verified priority, avoiding real leakage / hidden leakage events from being misjudged as "explained" and degraded, and ensuring that the explanation and dispatch of partition imbalance are closer to the real cause. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating labor.

[0037] Figure 1 A water meter abnormal water analysis and early warning method flow chart provided by the embodiment of the present application;

[0038] Figure 2A structural schematic diagram of a water meter abnormal water analysis prediction early warning system provided by an embodiment of the present application is provided.

[0039] Figure 3 A rule configuration interface schematic diagram provided by an embodiment of the present application is provided.

[0040] Figure 4 A DMA partition management map interface schematic diagram provided by an embodiment of the present application is provided.

[0041] Figure 5 A DMA partition statistical analysis interface schematic diagram provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0042] Hereinafter, some terms in the present application are explained and described. It should be noted that these explanations are for the convenience of understanding by those skilled in the art, and do not constitute a limitation on the scope of protection required by the present application.

[0043] At least one involved in an embodiment of the present application includes one or more; wherein the plurality refers to greater than or equal to two. In addition, it should be understood that in the description of the present specification, the terms "first", "second", "third" and the like are used only for the purpose of distinguishing the description, and cannot be understood as indicating or implying relative importance, nor can it be understood as indicating or implying order. For example, the first device and the second device do not represent the importance or order of the two, but only distinguish the description. In an embodiment of the present application, "and / or" is only used to describe the relationship between the two, which means that there are three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents that the front and rear associated objects are a "or" relationship.

[0044] The orientation terms mentioned in the embodiments of the present application, such as "up", "down", "left", "right", "in", "out" and the like, are only the direction of the drawing, therefore, the orientation terms used are for better and clearer description and understanding of the embodiments of the present application, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, therefore, it cannot be understood as a limitation on the embodiments of the present application.

[0045] References to "one embodiment," "in some examples," or "some embodiments" as described in the embodiments of this application mean that one or more embodiments of this specification include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in some examples," "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0046] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0047] Embodiment 1 of the present invention: In this embodiment, the water supply company has built a centralized meter reading platform and a DMA (Digital Metering and Control) zone metering system. The remote water meters can report cumulative metering readings and status codes, and inlet flow metering devices are installed at the DMA zone inlets. The system provides a rule configuration page, an anomaly warning and handling page, and an anomaly summary page in the background, and automatically performs anomaly screening, interpretation screening, and task assignment through scheduled tasks to form a closed loop for anomaly handling.

[0048] In this embodiment, the first statistical scale is selected as one day, and the second statistical scale is selected as one month. The corresponding first statistical scale time window is a daily statistical time window, the first statistical scale anomaly rule is a daily anomaly rule, the second statistical scale time window is a monthly statistical time window, and the second statistical scale anomaly rule is a monthly anomaly rule; wherein the nighttime minimum flow time window is selected as the characteristic time window for the low disturbance period.

[0049] like Figure 1 As shown, Figure 1 A flowchart of a method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters, provided in this application embodiment; the method includes the following steps:

[0050] S1: The system executes parameter configuration on the management terminal, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the rule configuration interface provided in an embodiment of this application. The following rules are configured on the management side:

[0051] Day anomaly rule: Use range judgment: Set upper and lower limit values of daily consumption according to water meter use and water meter caliber; for example, configure the rule for commercial DN20 caliber, configure daily consumption range as 3.50-12.60 tons; trigger day anomaly when daily consumption is not in the range; use range comparison judgment: Take the daily consumption distribution of water meters with the same use + caliber in the same DMA or in the same area as the comparison benchmark, calculate the quantile or Z-score of the water meter in the distribution; trigger day anomaly when the quantile is higher than the preset high quantile threshold or the Z-score exceeds the threshold, which is used to identify 'not exceeding the hard threshold but significantly higher than the same type'.

[0052] Month anomaly rule: The monthly consumption in the monthly statistical time window is the difference between the final consumption and the initial consumption, and the deviation amplitude is calculated. The benchmark value of the deviation amplitude can be the average of the monthly consumption of the previous three months. When the deviation amplitude exceeds the preset threshold (such as yellow anomaly threshold, red anomaly threshold), it is marked as an anomaly, and the deviation direction and amplitude are written into the anomaly amount field; for example, configure the rule: when the monthly consumption is 0-10 tons, do not trigger the month anomaly; when the monthly consumption is 11-100 tons and exceeds the average consumption of the past 3 months by 30%, trigger yellow anomaly, and exceeds 50% to trigger red anomaly.

[0053] Responsible person binding rule: Based on the meter reader / team field in the water meter account book, or based on the responsibility area relationship of DMA partition and personnel, automatically bind the responsible person. For example, through role management and area management to establish 'DMA / water meter-responsible person' mapping relationship.

[0054] Disposal time limit rule: Set disposal time limit according to anomaly level, for example, blue 48 hours, yellow 24 hours, red 4 hours.

[0055] The system reads the water meter account book and completes the DMA partition basic data maintenance, such as Figure 4 , as shown in Figure 4 DMA partition management map interface schematic diagram provided by the embodiments of the present application. The system displays the DMA partition boundary in a map way on the partition management page and provides a partition list; the administrator maintains DMA identification, partition range and entry metering point information on the page, and aligns the water meter attribution relationship in the account book with the DMA identification (for example, attributes water meters M001-M010 to DMA-03). At the same time, as shown in Figure 4As shown in the ranking list on the right side, the system can rank the relevant indicators by the partition leakage rate to form a partition-level ranking result, which is used for subsequent abnormal screening and priority treatment guidance. In addition, the system generates a time window for statistics and alignment, and records the time window in the form of "time window identifier + start and end time + type" for subsequent candidate event and DMA tag unified reference, ensuring that the system uses the same time window identifier and the same start and end time for cumulative difference statistics of DMA inlet metering and each water meter in the partition, so that the inlet metering statistical result and the partition internal summary statistical result correspond to the same statistical period, thereby avoiding the imbalance of the metering deviation caused by the inconsistent start and end time.

[0056] In the target period, the system collects from the set copy platform: the cumulative metering reading of each remote water meter such as the start point reading and the end point reading, from which the time window consumption is obtained; and the state data of each water meter such as offline, under-voltage, empty pipe, dripping, reading anomaly, and manufacturer state code. The system also collects the DMA partition inlet flow metering data, which at least includes: the inlet cumulative metering reading of the DMA inlet metering device in the statistical time window (used to obtain the inlet cumulative metering increment by difference between the start point cumulative reading and the end point cumulative reading) and the inlet flow sequence formed in the statistical time window (used to represent the sequence segment of the instantaneous flow of the inlet changing with time), to form the consumption statistical result and the evidence data that can be aligned in the same statistical time window. At the same time, the system collects the operation organization event data from the repair work order system, the dispatching system, the water stop recovery record and the flushing plan, and generates the scene event based on the operation organization event data. The generation process is: the work order / plan record in the operation organization event data is parsed into a structured event according to the preset event mapping rule, at least the event type field (one or more of repair cutting / plan flushing / water stop recovery / distribution supply change) is extracted and generated, the influence range field (DMA identifier or covered water meter set), the start time field and the end time field; after generating the scene event, write "DMA identifier + time window identifier" into the abnormal pool for subsequent explanation matching. The abnormal pool is a data structure for storing candidate events, scene events, DMA tags and their associated relationships, which can be implemented by a database table, a message queue or a cache set, and its record at least contains water meter identifier, DMA identifier, time window identifier and event field. For example, DMA-03 has a planned flushing scene event SE-01 (10:00-12:00) on 2025-11-17.

[0057] S2: The system performs preliminary screening according to daily statistical time windows and monthly statistical time windows respectively: daily anomaly preliminary screening (take DAY_20251117 as an example) The system calculates the daily consumption of the water meter in DMA-03, and distinguishes according to the "use + caliber" rule: the daily consumption of water meter M001 (residential DN15) is 8.5 tons, which exceeds the upper limit of residential DN15 of 2.5 tons, triggering daily anomaly; the daily consumption of water meter M002 (commercial DN20) is 7 tons, which is within the range of 3.50-12.60 tons, and does not trigger anomaly; the daily consumption of water meter M003 (industrial DN50) is 80 tons, which is within the range of 50-300 tons, and does not trigger anomaly; monthly anomaly preliminary screening (take MON_202511 as an example) The system compares the monthly consumption of each water meter with the average consumption of the past three months at the monthly transfer, and generates a monthly candidate event if the deviation exceeds the threshold.

[0058] In addition to the amount of use according to the time window, the system matches and distinguishes the state anomaly and meter reading anomaly of the water meter, such as "the meter reader misread the number" or "the water meter dial is dirty, causing remote identification error", which belongs to "meter reading anomaly", such as offline, under-voltage, empty pipe, dripping or device reporting anomaly, which belongs to water meter state anomaly. When any of the judgment conditions is met, a candidate event is generated and written into the abnormal pool. For example: M007 lacks terminal reading for 3 consecutive days → E-002 meter reading anomaly; M006 hits EMPTY_PIPE → E-003 state anomaly. Each candidate event at least contains: water meter identification, wherein the water meter identification is used to uniquely mark the target remote water meter, facilitating the positioning, dispatching and writing back of the abnormal event, DMA identification, wherein the DMA identification is used to mark which DMA the water meter belongs to, time window identification, wherein the time window identification is used to mark the label of "which statistical period this anomaly occurs in"; anomaly type such as daily consumption out of range, monthly deviation, negative value, continuous zero consumption, offline missing number, etc. is the "classification label" of which category / trigger reason the anomaly belongs to; anomaly amount such as excess amount, deviation amplitude, negative value amplitude; anomaly duration such as consecutive days / continuous months / repeat number of the same anomaly, indicating how long this anomaly lasts / repeat how many times; data integrity: the proportion of valid sampling points in the time window, the number of missing numbers, the number of time stamp disorders; state evidence field: unified state event code, original manufacturer code, key fragment description. For example: write candidate event E-001 for M001: DMA=DMA-03, window=DAY_20251117, type=day consumption out of range, anomaly amount=6.0 tons excess, duration=continuous 2 days high, integrity=96%, state evidence="dripping / small flow continues".

[0059] S3: The system calculates the alignment of each DMA partition in the same statistical time window: the entry cumulative amount: the cumulative amount of the DMA entry table in the time window; the difference between the end of the table and the beginning of the table in the same statistical time window, the amount of water in the time increment; the partition internal summary amount: the sum of the cumulative amount of all water meters in the partition in the time window; the partition imbalance amount: the difference between the entry cumulative amount and the partition internal summary amount. The system visualizes the alignment calculation results on the statistical analysis page, as shown in FIG. 8. Figure 5 Figure 5 The DMA partition statistical analysis interface provided by the embodiment of the present application is shown in FIG. 8. After selecting the DMA partition and the target time range, the page displays the trend comparison of the entry metering of the partition and the partition internal summary metering, and outputs the corresponding statistical results; the system forms a linkage display with the candidate event list in the same time window on the page, thereby supporting the search and disposal from the "partition imbalance" to the "single meter positioning". At the same time, the system extracts the entry flow characteristics (night average / minimal flow and stability) in the minimum flow time window (for example, 02:00-04:00) at night, generates imbalance type labels, such as "suspected dark leakage type (night minimal flow continuously high)", "short-time disturbance type (short-time sudden increase and sudden decrease)", etc., the system records the partition imbalance amount and the imbalance type as the partition label, and writes the "partition imbalance amount + imbalance type label" as the DMA label into the abnormal water pool. For example, in DAY_20251117, the entry amount of DMA-03 is 110 tons, the partition summary is 95.5 tons, the partition imbalance amount is 14.5 tons, and the night minimal flow is high, and the label is "suspected dark leakage type". In the abnormal early warning processing page, the system aggregates and displays the DMA label of "partition imbalance amount 14.5 tons + suspected dark leakage type" with DMA-03 and DAY_20251117 as the primary keys, and displays the candidate event list (including E-001, etc.) in the same page, thereby realizing the one-key linkage positioning from the partition imbalance to the single meter candidate event.

[0060] S4: The system obtains the water meter consumption sequence segment corresponding to each candidate event, the DMA entry flow sequence segment in the same time window, and the scene event set of the DMA in the time window from the abnormal water pool and the original sequence; the system forms a candidate pair with the candidate event and the scene event, respectively calculates a feature set for scene explanation, the feature set at least includes time overlap degree feature, abnormal direction matching feature, abnormal morphology consistency feature, and evidence quality feature, the evidence quality feature is obtained from the abnormal amount, abnormal duration, data integrity, data channel reliability, and evidence conflict label of the candidate event; the feature set is input into the machine learning classification model supervised and trained by the historical active verification result, and the scene explanation probability of each candidate pair is output; the candidate pair with the maximum scene explanation probability is taken as the hit scene event of the same candidate event, and the scene explanation probability of the candidate event is obtained.

[0061] ​When the scene explanation probability of the candidate event reaches the preset scene explanation threshold, the candidate event is marked as a scene explainable event and written into the hit scene event identification and explanation record; when the scene explanation probability of the candidate event does not reach the preset scene explanation threshold, the candidate event is marked as to be verified and written into the to-be-verified set. The system calculates an explanation priority score based on the abnormal quantity, abnormal duration, data completeness, data channel reliability, evidence conflict label and scene explanation probability of the candidate event, and selects candidate events in descending order of explanation priority score to form an explanation event set, which is used to gradually balance and reduce the partition imbalance quantity until the remaining unexplained part of the partition imbalance quantity is less than the tolerance or the candidate event is traversed completely; the explanation event set is output to step S5 for active verification task generation and dispatch.

[0062] S5: The system automatically dispatches active verification tasks to the to-be-verified event set according to the responsibility person binding rule. Before dispatching specific verification methods, the system first calculates the uncertainty U of each to-be-verified event, which is determined by the data completeness C, data channel reliability R, evidence conflict label K and confidence boundary width W of the scene explanation probability of the candidate event, wherein: C is calculated based on the effective sampling point ratio, the number of missing sections and the number of timestamp disorder times in the time window, and the more missing and disorder, the lower C; R is maintained by the consistency result of historical remote copying and on-site verification through an exponential sliding method, and the more reliable the channel, the higher R; K is obtained by checking the consistency of the state evidence field, the metering reading / usage sequence in the same time window, the DMA inlet flow sequence and the hit scene event influence direction and start and end time, and K=1 when there is obvious contradiction, otherwise K=0; W is determined by the upper and lower quantile difference of the probability sequence obtained by executing multiple probability inferences on the same candidate event under the same characteristic input, and the larger W indicates that the scene explanation probability is less stable. The system follows a preset monotonic rule, such as U=(1-C)+(1-R)+W+K, so that the uncertainty U monotonically increases when C is lower, R is lower, K is larger and W is larger, and when the data completeness C is lower than the preset threshold, the uncertainty component related to C is amplified accordingly, thereby realizing automatic up-regulation of U.

[0063] The introduction of "uncertainty based on data completeness, data channel reliability, evidence conflict marking, and scene explanation probability confidence boundary width" in this embodiment is to solve the two prominent problems brought by the traditional solution of directly assigning orders based on "abnormal amount / degree of deviation": one is that unreliable data is treated as "very certain abnormality", leading to a large number of false positives and invalid on-site verification; the other is that the scene explanation probability given by the model is high on the surface, but in fact it fluctuates greatly, and is prone to unstable behavior such as "judging explanation today and verification tomorrow". By integrating C, R, K, and W into uncertainty U, and constraining it to "C is lower, R is lower, K is larger, and W is larger, U monotonically increases", the system can make a fine classification of the "credibility" and "verification cost needed" of each candidate event before assigning orders: on the one hand, when the data is missing seriously or the timestamp is disorderly (C is low), the channel history consistency is poor (R is low), or the state evidence is obviously contradictory to the entrance flow and scene event (K=1), even if the abnormal amount is large, the system will first raise U and preferentially arrange remote re-sampling and remote re-copy, avoiding direct conclusion "real water leakage", thereby reducing false positives and invalid attendance caused by "bad data"; on the other hand, when the scene explanation probability value itself is high but the confidence boundary width W is large, the system will identify the "model judgment instability" and preferentially enter verification as a high-uncertainty event, rather than simply relying on the one-time result for absolute trust. Through this multi-factor integrated uncertainty measurement, compared with the traditional method based on a single threshold or fixed rule, this embodiment can control the false positive rate while more reasonably allocating verification resources such as remote encrypted re-copy, remote re-sampling, and on-site verification of different cost levels, achieving a dynamic balance of "high-risk high-uncertainty events preferentially verified, low-risk low-uncertainty events weakened push", and engineering the accuracy, cost, and interpretability.

[0064] The system divides the verification demand into at least three levels according to the uncertainty U: when U is low, it is considered to require only the first level of verification strength, when U is in the middle interval, it is considered to require the second level of verification strength, and when U is high, it is considered to require the third level of verification strength. For each event to be verified, the system generates a set of feasible verification methods F for the event in combination with the online state of the water meter, whether it supports remote encrypted re-copy, whether it supports remote re-sampling, and the accessibility of on-site resources, etc. F includes one or more of remote encrypted re-copy, remote re-sampling, and on-site verification; wherein the first cost level corresponds to remote encrypted re-copy, the second cost level corresponds to remote re-sampling, and the third cost level corresponds to on-site verification.

[0065] After generating the set of feasible verification methods, the system determines the final verification method according to the following rules: first, when the data integrity C is lower than the preset threshold, remote resampling is determined as the first priority verification method for the event; if remote resampling is included in the set F, remote resampling is preferentially selected; if remote resampling is not in the set F, or the number of consecutive remote resampling failures reaches the preset threshold, on-site verification is upgraded. Second, in the case where the data integrity does not decrease seriously, the system preferentially selects the method with lower cost level and meeting the corresponding verification strength requirement in the set F according to the level of uncertainty U: when U is in the first level, remote encrypted duplicate copying is preferentially selected, and if remote encrypted duplicate copying is not feasible, remote resampling or on-site verification is selected; when U is in the second level, remote resampling is preferentially selected, and if remote resampling is not feasible, on-site verification is selected; when U is in the third level, on-site verification is preferentially selected, and if necessary, remote encrypted duplicate copying or remote resampling can be tried once as an auxiliary means, but the on-site verification result is the final basis. Through the above rules, the decision-making process of "determining the verification level requirement according to the uncertainty U, and then selecting the verification method from the feasible set according to the cost from low to high" is realized.

[0066] After determining the specific verification method, the system generates pre-warning and work order information including fields such as "abnormal object, abnormal type, abnormal level, recommended verification method, disposal suggestion, deadline, and person in charge", and pushes it to the user end through system messages, short messages, or App: strong push and follow-up are performed on the set of events to be verified; for scene explainable events, only explanation records are written into the abnormal pool, and "has been explained / traceable" is displayed on the user end or the push level is reduced to reduce false positives and repeated orders. After the person in charge completes remote encrypted duplicate copying, remote resampling, or on-site verification, the system writes the verification result back to the abnormal pool, updates the channel reliability R, the evidence field, and the event state (unprocessed→processing→processed / closed); multiple time windows of the same abnormal type of the same water meter unprocessed pre-warning can be marked for processing synchronously according to the rules to avoid repeated work orders, and form quantifiable indexes such as abnormal quantity, disposal time, overdue rate, and repeated abnormal rate in subsequent statistics.

[0067] Embodiment two of the application: based on embodiment one, the step S4 of embodiment one is further described in this embodiment.

[0068] The system performs the following processing on the candidate events and scene events that match the same DMA partition and time window: for each candidate event, the system obtains the usage sequence segment of the water meter within the target time window and its adjacent buffer interval from the set copy platform, obtains the entry flow sequence segment within the same time range from the DMA entry metering device, and obtains the scene event that has been written from the abnormal water pool. Among them, the usage sequence segment and the entry flow sequence segment are resampled according to the unified time granularity and aligned with the same timestamp. The system calculates a set of scene explanation features for each "candidate event-scene event" combination, including at least the following four types: , wherein the candidate event abnormal segment interval , is the scene event interval, wherein is the abnormal start time, is the abnormal end time, is the scene start time, is the scene end time; the abnormal morphology consistency feature reflecting whether the "water meter abnormal morphology" and the "entry morphology" are the same: taking the candidate water meter usage change sequence and the entry flow change sequence in the overlapping interval respectively, calculating the correlation coefficient as the morphology consistency: , wherein is the water meter usage change sequence, is the entry flow change sequence, and the Pearson correlation or cosine similarity can be used in implementation; the abnormal direction matching feature reflecting whether the directions are consistent: the candidate event abnormal direction is encoded as , the scene event impact direction is encoded as , and the direction consistency indicator is defined as , is an indicator function used to represent whether the condition is met: 1 if the condition is met, otherwise 0; the explanation-related evidence feature reflecting "whether this candidate value is worth explaining / verifying": the candidate event has the following fields: abnormal amount A, abnormal duration D, data integrity C, channel reliability R, and evidence conflict marker K. In order to facilitate model input, the system normalizes them as follows: , wherein is obtained by the system according to the quantile normalization of the same DMA scale historical distribution, which can be directly calculated. Among them, the evidence conflict marker K is obtained by performing consistency checking between the state evidence field of the candidate event, the metering reading / usage sequence within the same time window, the DMA entry flow sequence, and the impact direction and start and end time of the hit scene event; when there is evidence contradiction in the consistency checking, K=1, otherwise K=0.

[0069] The system concatenates the above feature set into a model input vector: and input the machine learning classification model trained by the historical active verification result supervision, and output the probability that the candidate event is a scene explanation: To represent the fluctuation range of the scene explanation probability, the system performs multiple probability inferences on the same candidate event under the same input feature vector to obtain a probability sequence, wherein the multiple probability inferences are implemented in any of the following ways: an integrated classification model containing multiple sub-models is used to output probabilities respectively; or a random inactivation / random disturbance is enabled for the same classification model to output probabilities repeatedly. The system calculates a lower quantile value and an upper quantile value from the probability sequence, and calculates a confidence boundary width as the difference between the upper quantile value and the lower quantile value, wherein the greater the confidence boundary width W is, the more unstable the scene explanation probability of the candidate event is, and the system accordingly increases the subsequent uncertainty evaluation and verification strength.

[0070] Meanwhile, the system generates an explanation priority score for ranking, wherein the explanation priority score is used to select events that "need to continue explanation imbalance / need to be verified preferentially", for example, a simplified linear combination is taken: wherein is a configuration parameter or is fitted from historical data; is used to represent that the event has a large abnormal amount, lasts for a long time, and is not easily explained by the scene; incomplete data, unreliable channels or conflicting evidence will reduce the priority.

[0071] When (scene explanation threshold), the system marks the candidate event as a scene explainable event and writes it into an explanation record; when , it is marked as to be verified and written into a to-be-verified set. The system sorts the to-be-verified set from high to low, initializes the remaining unexplained part of the partition imbalance amount: wherein L is the difference between the entry cumulative amount and the summary amount in the partition. Then, the to-be-verified with a high ranking is selected in turn, and its abnormal amount field is used to update and reduce the remaining unexplained part: wherein is an explanation conversion coefficient (used to handle the "single-table abnormal amount and DMA imbalance amount difference in caliber", which can be 1 or calibrated according to history). When (tolerance) or the to-be-verified is traversed, the selection is stopped, and an explanation event set is obtained.

[0072] Embodiment three of the present application: Based on embodiment two, the explanation event set is output to S5 for dispatching and self-learning gain after active verification write-back, so that the subsequent scene explanation probability and explanation priority score are continuously corrected with the real disposal result, the false positive accumulation is reduced, and the long-term accuracy is improved. ​​​

[0073] The system outputs the interpretation event set obtained in Embodiment Two to Step S5 as candidate input for active verification tasks, wherein the interpretation event set at least contains: water meter identification, DMA identification, time window identification, abnormal type, abnormal amount, abnormal duration, , , to Step S5 as candidate input for active verification tasks; wherein the system prioritizes the generation of active verification tasks for events with high and low . After the person in charge completes remote copying / remote resampling / field verification, the system receives the verification conclusion and writes back the abnormal pool, forming a supervision label, which at least includes: : whether the verification conclusion is "explainable by the scene"; : whether the verification conclusion is "real abnormality"; disposal method, disposal time consumption, supplementary evidence such as photos, text records, etc. For each data channel such as the reporting link maintenance reliability R of a certain manufacturer and a certain model, the index sliding update is performed according to the verification result: ( consistent with the system judgment), wherein is a smoothing coefficient (for example, 0.8~0.95). This update can be directly calculated from the write-back result without external unavailable data. The system writes the (x, ) formed by each write-back as a new supervision sample into the training sample library, and performs incremental training or retraining on the machine learning classification model at a preset period (for example, daily / weekly), thereby updating the output of =Model(x). At the same time, the update of the channel reliability R will directly affect the deduction term of , so that the subsequent sorting and dispatching are closer to the real disposal result.

[0074] Embodiment Four of the Invention: Based on Embodiment Two, this embodiment provides a closed-loop feedback utilization method for enhancing the stability of scene interpretation and sorting processing in Step S4. Unlike Embodiment Three, this embodiment does not require frequent retraining of the machine learning model directly, but converts the disposal write-back result of Step S5 into "calibration parameter set" and "channel credibility maintenance amount", which are used to indirectly correct the scene interpretation probability and interpretation priority of subsequent candidate events, thereby realizing adaptive optimization with accumulated running data.

[0075] In this embodiment, the system outputs the set of interpretation events obtained in Embodiment Two to Step S5 for active verification task generation and dispatch. After the person in charge completes remote encrypted duplication, remote re-sampling, or on-site verification, the system receives and writes back the treatment results. The treatment results form a closed-loop feedback record and are written into the abnormal pool or the feedback sample library. The closed-loop feedback record includes: candidate event identification (including water meter identification, DMA identification, time window identification, and abnormal type), treatment method, treatment time consumption, and supplementary evidence field, and further includes two types of interpretation-related verification conclusions: one is the verification conclusion of whether the candidate event can be explained by the scene event, and the other is the verification conclusion of whether the candidate event belongs to a real anomaly.

[0076] Based on the closed-loop feedback record, the system maintains a set of calibration parameters for Step S4, which at least includes the following two types of parameters:

[0077] Probability output calibration parameters: The system performs statistics on “scene explanation probability segmentation interval-actual hit ratio” according to a preset statistical period (for example, daily or weekly), obtains the calibration coefficient of each segmentation interval, and uses it to calibrate the scene explanation probability output in subsequent Step S4. This makes the candidate events in the same probability interval maintain consistent explanation hit level in the long-term operation, thereby reducing the misjudgment caused by the drift of probability output with data distribution changes.

[0078] Ranking factor revision parameters: The system statistics the proportion of “explanation priority ranking in front but actually not established / treatment invalid” and the proportion of “ranking behind but actually real anomaly” according to the dimensions of abnormal type, manufacturer channel, or DMA partition, and accordingly adaptively revises the calculation factor of the explanation priority ranking, so that the screening of the subsequent interpretation event set is closer to the real treatment result, reducing the accumulation of false positives and invalid work orders.

[0079] In addition, for the credibility maintenance related to the data link, the system in this embodiment maintains a channel reliability index R ∈ [0, 1] for different data channels (for example, a certain manufacturer's certain type of water meter reporting link, a certain concentrator link). The system updates R smoothly according to the consistency of “remote review results and on-site verification results” in the closed-loop feedback record; when the consistency is high for a long time, R is increased, and when the consistency is low for a long time or time stamp disorder or missing anomaly frequently occurs, R is decreased. R will be used as an input for Step S4 interpretation-related features and Step S5 uncertainty calculation, so that the subsequent scene explanation and verification dispatch strategy are consistent with the real write-back results.

[0080] In this way, under the premise of not changing the overall process of embodiment one and embodiment two, an enhanced mechanism of "closed loop write back-parameter calibration-strategy self adaptation" is introduced: the system does not need to explicitly retrain the model every time, but can also use the treatment write back result to continuously correct the scene explanation probability output and explanation priority calculation, so that the explanation screening is more stable, the order allocation is more convergent, and the long-term false alarm rate gradually decreases.

[0081] Embodiment five of the present application: on the basis of embodiment one, the user opening state and communication state change information in the marketing system / communication platform are further accessed as business state change event data to generate business state scene events, which are used to explain or gate the candidate events of zero consumption, discontinuity, abnormal recovery, etc. related to the single table.

[0082] Taking water meter M004 as an example, the marketing system shows that the user is in the "unopened state". When the system finds that the water consumption of the water meter is 0 in the target day statistical time window, no water consumption anomaly candidate event is generated; when the system finds that the user is not opened / connected but the water consumption is not 0, a candidate event is still generated and a reminder is pushed (pay special attention to the risk of "misconnection / table position misbinding"). The system generates a warning from the "to-be-verified event set" and pushes it to the user end; the pushing method at least includes a system message or a short message. At the same time, the display is filtered according to personnel permissions: the meter reader can only see the warnings assigned to himself; the meter reader team leader can see the warnings of the meter readers in his management range; higher authority roles can view the full amount. When the warning is close to the disposal time limit and is not handled, the system performs a reminder; if it is overdue, it is upgraded and pushed to higher levels such as team leaders / area responsible persons, and the number of reminders and the overdue time are recorded as subsequent statistical indicators. When the same type of abnormality of the same water meter repeatedly generates warnings in multiple time windows, the system adopts a "disposal synchronization" strategy: as long as any of the warnings is disposed and written back, the remaining unprocessed warnings of the same type are automatically marked as handled to avoid repeated order allocation and accumulation.

[0083] Embodiment six of the present application: in this embodiment, the system has accessed two types of remote water meters of different manufacturers (manufacturer A and manufacturer B) and the collection platform; both types of water meters report readings and state codes according to a fixed sampling period, but their state code meanings and codes are inconsistent. The system maintains an "event mapping rule table" in the background, which is used to normalize the state codes reported by different manufacturers into unified event codes, and writes the normalized results into the state evidence field of the candidate event, and dynamically marks and adjusts the effectiveness of the mapping rule after subsequent verification and write back, to improve the subsequent discrimination reliability.

[0084] Taking the DMA-03 water meter M006 as an example, Manufacturer A reports a status code "0x11" indicating "empty pipe," while Manufacturer B reports a status code "E07" indicating "no water / empty pipe." The system maps both to the event code "EMPTY_PIPE" and simultaneously records "Unified event code = EMPTY_PIPE, original manufacturer code = 0x11 / E07, occurrence time segment = continuous from 02:10 to 06:30 on the same day" in the status evidence field of the candidate event. This process corresponds to the requirement of "parse and normalize the codes reported by different manufacturers and write them into the status evidence field." For each statistical time window, such as DAY_20251117, the system calculates the data completeness: Expected number of sampling points: 96 points should be obtained per day according to the configured sampling period (e.g., once every 15 minutes); Actual number of sampling points: count the number of valid points within the window; Number of missing segments: merge consecutive missing points into a missing segment count; Timestamp disorder times: count the number of times "timestamps of the same water meter are reversed / repeated." For example, if M006 should have 96 data points on a given day, but only 72 are actually present (two missing segments and one timestamp error), then the data integrity is low. When the data integrity is below a threshold, the uncertainty of the candidate event is increased, and remote resampling is triggered first instead of being directly identified as a real data leak, in order to avoid "false anomalies caused by missing data".

[0085] For example, M006 falls under the category of "missing data + empty control evidence," and the system directly dispatches a second-cost level remote re-collection. If "EMPTY_PIPE" continues to appear after re-collection and the abnormal usage direction is consistent with the evidence, it is upgraded to on-site verification. In this embodiment, each "data channel" can be understood as a maintenance reliability indicator for a certain manufacturer / concentrator / communication link: if the remote re-collection is consistent with the on-site verification conclusion and the reading error is within the allowable range, the reliability of the channel is increased; if the remote re-collection is inconsistent with the on-site verification multiple times, or if timestamp disorder occurs frequently, the reliability of the channel is decreased; if a status code is found to have a meaning inconsistent with the mapping after multiple verifications, the validity of the "event mapping rule" is marked as low confidence, and the administrator is prompted to revise it. Through the above write-back, the system achieves "increasing accuracy with use": the same type of anomaly will select the appropriate verification method more quickly in subsequent cycles, reducing ineffective manual attendance.

[0086] Embodiment 7 of the present invention: Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of a water meter abnormal water usage analysis, prediction, and early warning system provided in an embodiment of this application. The system includes a system configuration and data access module, an abnormal initial screening and abnormal water pool module, a partition imbalance calculation and DMA tag module, an interpretation filtering and test-to-verify generation module, and an active verification dispatch and push closed-loop module;

[0087] The system configuration and data access module is configured to perform parameter configuration, including configuring partition statistical time window, abnormality rule, responsibility person binding rule and disposal time limit rule, collecting metering reading and water meter state data of each remote water meter in a target period and entrance flow metering data of each DMA partition, establishing the association between the water meter and the DMA partition and forming the partition statistical time window, synchronously collecting scene data related to the DMA partition and generating scene events, and writing the scene events into an abnormality pool according to DMA identifier and time window identifier;

[0088] The abnormality preliminary screening and abnormality pool module is configured to preliminarily screen the water meter consumption according to the abnormality rule according to the statistical time window, obtain a candidate abnormality water meter set, and convert the candidate abnormality water meter set into candidate events and write the candidate events into the abnormality pool;

[0089] The partition imbalance calculation and DMA label module is configured to obtain entrance cumulative metering and partition internal water meter cumulative metering summary in a corresponding time window for each DMA partition, determine a partition imbalance according to the difference, generate an imbalance type label based on the minimum flow feature of a low disturbance period feature, and write the partition imbalance and the imbalance type into the abnormality pool as a DMA label;

[0090] The explanation screening and to-be-verified generation module is configured to perform scene explanation and sorting processing based on a machine learning classification model on candidate events and scene events that are in the same DMA partition and match the time window, obtain scene explainable events and to-be-verified events, calculate explanation priority points of the candidate events, sort the candidate events from high to low according to the explanation priority points, select the candidate events to form an explainable event set in sequence until the remaining unexplained part of the partition imbalance is less than a tolerance or the candidate events are traversed completely, and generate and dispatch an active verification task according to the responsibility person binding rule.

[0091] The active verification dispatching and push closed loop module is configured to receive a verification result, write the verification result back to the abnormality pool to update channel reliability and evidence fields, generate an early warning information and push the early warning information to a user end, generate an explanation record and write the explanation record into the abnormality pool for a scene explainable event, and show a state or reduce a push level on the user end.

Claims

1. A method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters, characterized in that, Includes the following steps: S1: Execute parameter configuration, which includes configuring partition statistical time window, anomaly rules, responsible person binding rules and handling time limit rules, collecting meter readings and water meter status data of each remote water meter and the inlet flow metering data of each DMA partition within the target period, establishing the association between water meters and DMA partitions and forming partition statistical time window, synchronously collecting scene data related to DMA partitions and generating scene events, and writing scene events into the abnormal water pool according to DMA identifier and time window identifier; S2: Based on the statistical time window and the aforementioned anomaly rules, the water meter usage is initially screened to obtain a set of candidate abnormal water meters, and these are converted into candidate events and written into the abnormal water pool. S3: For each DMA partition, obtain the sum of the inlet cumulative amount and the water meter cumulative amount in the partition within the corresponding time window, and determine the partition imbalance amount based on the difference. At the same time, generate an imbalance type label based on the minimum flow characteristic of the low disturbance period. Write the partition imbalance amount and imbalance type as DMA labels into the abnormal water pool. S4: Perform scene interpretation and sorting processing based on machine learning classification model on candidate events and scene events in the same DMA partition with matching time windows to obtain scene interpretable events and events to be verified. Calculate the interpretation priority score for candidate events and sort them from high to low according to the interpretation priority score. Select candidate events in turn to form an interpretation event set until the remaining uninterpreted part of the partition is less than the tolerance or the candidate events have been traversed. S5: Generate and dispatch active verification tasks for events to be verified according to the responsible person binding rules, receive verification results and write them back to the anomaly pool to update channel reliability and evidence fields, generate early warning information and push it to the user terminal, and for events that can be explained in the scenario, generate explanation records and write them to the anomaly pool, and display the status or reduce the push level on the user terminal.

2. The method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters as described in claim 1, characterized in that: The parameter configuration in step S1 includes configuring the partitioned statistical time window, anomaly rules, responsible person binding rules, and handling time limit rules. Specifically, the partitioned statistical time window includes at least a first statistical scale time window, a second statistical scale time window, and a feature time window for low-disturbance periods. The first statistical scale is smaller than the second statistical scale. The first statistical scale time window is used for the initial screening of anomaly rules at the first statistical scale, the second statistical scale time window is used for the initial screening of anomaly rules at the second statistical scale, and the feature time window for low-disturbance periods is used to extract the low-disturbance period features of DMA inbound traffic. The anomaly rules include a first statistical scale anomaly rule and a second statistical scale anomaly rule. The first statistical scale anomaly rule is used to identify usage anomalies within a first statistical scale time window, and includes usage scope discrimination and usage scope comparison discrimination. The second statistical scale anomaly rule is used to identify usage anomalies within a second statistical scale time window, and is based on the deviation between the usage in the target statistical time window and the historical statistical benchmark, and writes the deviation direction and deviation magnitude into the anomaly quantity field. The responsible person binding rule is used to automatically bind candidate events or events to be verified to the corresponding responsible persons; The processing time limit rules are used to set processing time limits according to the level of abnormality.

3. The method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters as described in claim 1, characterized in that: The specific content of the scenario data is as follows: the scenario data includes at least the data on operational organization events; The operational event data is used to characterize planned events that affect the water supply conditions or pipeline operation mode or water use boundary conditions of the target DMA zone. Scenario events are generated based on the scenario data, and at least the fields of impact range, start and end time, and impact direction are recorded.

4. The method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters as described in claim 1, characterized in that: The preliminary screening of water meter usage based on the statistical time window and the abnormal rules also includes preliminary screening of meter reading abnormalities and preliminary screening of water meter status abnormalities. The initial screening of water meter status anomalies is based on matching the water meter status data with the status anomaly discrimination conditions. When one or more of the following are detected: offline, undervoltage, empty pipe, dripping, or equipment reported anomalies, candidate events are generated. The initial screening of meter reading anomalies is based on matching and judging the meter reading data formation situation with the meter reading anomaly discrimination conditions. When multiple consecutive first statistical scale time windows are not read, or when no valid starting point reading / ending point reading is formed within the preset time window, resulting in the usage being uncalculated, a candidate event is generated. The candidate events shall at least record the water meter identifier, DMA identifier, time window identifier, anomaly type, anomaly quantity, anomaly duration, data integrity, and status evidence fields; The abnormality types include at least water volume abnormality, water meter status abnormality, and meter reading abnormality. The water volume anomaly includes multiple consecutive zero usage anomalies in the first statistical scale time window or multiple consecutive zero usage anomalies in the second statistical scale time window; The meter reading anomalies include anomalies caused by multiple consecutive time windows of the first statistical scale where meter reading was not performed. The abnormal water meter status includes one or more of the following: offline, low voltage, empty pipe, or equipment reported abnormality.

5. The method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters as described in claim 1, characterized in that: In step S4, candidate events and scene events with matching time windows in the same DMA partition are subjected to scene interpretation and ranking processing based on a machine learning classification model. Specifically, this includes: obtaining water meter usage sequence fragments and DMA inlet flow sequence fragments corresponding to the candidate events; calculating the time overlap characteristics between the candidate events and scene events, the consistency characteristics between the abnormal patterns of the candidate events and the DMA inlet flow change patterns within the scene period, the matching characteristics between the scene event type and the abnormal direction of the candidate events, and interpretation-related characteristics obtained based on the abnormal quantity, abnormal duration, data integrity, data channel reliability, and evidence conflict of the candidate events. The time overlap feature, the consistency feature, the matching feature, and the explanation-related feature are input into a machine learning classification model trained under the supervision of historical active verification results, and the scene explanation probability and explanation priority score of each candidate event are output. When the probability of scene explanation reaches the preset scene explanation threshold, the candidate event is marked as a scene explainable event and an explanation record is generated and written to the abnormal pool. The push level is reduced or it is only displayed in a traceable state. When the scene explanation probability does not reach the preset scene explanation threshold, the candidate event is marked as to be verified and written into the verification set. The system selects candidate events from the verification set in descending order of explanation priority to form an explanation event set. The partition mismeasurement is explained step by step until the remaining unexplained part of the partition mismeasurement is less than the tolerance or the candidate events have been traversed.

6. The method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters as described in claim 5, characterized in that: The set of explained events is output to step S5 for the generation and dispatch of active verification tasks; After receiving the active verification result returned in step S5, the scenario interpretability conclusion and anomaly authenticity conclusion corresponding to the active verification result and their associated identifiers are written back to the anomaly pool to form a closed-loop feedback record.

7. The method for analyzing, predicting, and warning of abnormal water usage in water meters as described in claim 1, characterized in that: The step S1, which involves collecting the metering readings and status data of each remote water meter within the target period, as well as the inlet flow metering data of each DMA partition, also includes parsing the abnormal codes or status codes of different types of terminals, normalizing them into unified event codes according to the event mapping rules and writing them into the status evidence field, and adjusting the validity flag of the event mapping rules after writing back the verification results.

8. The method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters as described in claim 1, characterized in that: The specific process of generating and dispatching active verification tasks for events to be verified in step S5 according to the responsible person binding rule is as follows: For events to be verified, first determine the verification intensity level requirement based on the uncertainty. Then, based on the online status and capability support of the water meter, a set of feasible verification methods is generated, and the verification method that meets the verification intensity level requirement and has the lowest verification cost level is selected from the set of feasible verification methods. Finally, active verification tasks are dispatched based on the selected verification method.

9. The method for analyzing, predicting, and issuing early warnings of abnormal water usage in water meters as described in claim 8, characterized in that: The uncertainty is calculated as follows: the uncertainty of the candidate event is calculated based on the data integrity, data channel reliability, evidence conflict markers, and the confidence boundary width of the scenario interpretation probability. The confidence boundary width is used to characterize the fluctuation range of the scenario interpretation probability; The data completeness is obtained by analyzing the proportion of valid sampling points within the statistical time window and / or the number of missing segments and / or the number of times timestamps are disordered.

10. A water meter abnormal water usage analysis, prediction, and early warning system, characterized in that, include: System configuration and data access module, anomaly screening and anomaly pool module, partition imbalance calculation and DMA tag module, interpretation screening and verification generation module, active verification distribution and push closed loop module; The system configuration and data access module is used to execute parameter configuration, which includes configuring partition statistical time windows, anomaly rules, responsible person binding rules and handling time limit rules, collecting meter readings and water meter status data of each remote water meter and inlet flow metering data of each DMA partition within the target period, establishing the association between water meters and DMA partitions and forming partition statistical time windows, synchronously collecting scene data related to DMA partitions and generating scene events, and writing scene events into the abnormal water pool according to DMA identifier and time window identifier; The anomaly screening and anomaly pool module is used to perform an initial screening of water meter usage according to the anomaly rules based on a statistical time window, obtain a set of candidate abnormal water meters, and convert them into candidate events and write them into the anomaly pool. The partition imbalance calculation and DMA tag module is used to obtain the sum of the inlet cumulative amount and the water meter cumulative amount in the partition for each DMA partition within the corresponding time window, and determine the partition imbalance amount based on the difference. At the same time, it generates an imbalance type tag based on the minimum flow characteristic of the low disturbance period and writes the partition imbalance amount and imbalance type as DMA tags into the abnormal water pool. The explanation filtering and verification generation module is used to perform scene explanation and sorting processing based on machine learning classification model on candidate events and scene events in the same DMA partition and with matching time windows, to obtain scene explainable events and verification events, calculate explanation priority scores for candidate events and sort them from high to low according to explanation priority scores, and select candidate events in turn to form an explanation event set until the remaining unexplained part of the partition is less than the tolerance or the candidate events have been traversed. The active verification dispatch and push closed-loop module is used to generate and dispatch active verification tasks for events to be verified according to the responsible person binding rules, receive verification results and write them back to the anomaly pool to update the channel reliability and evidence fields, generate early warning information and push it to the user terminal, and for events that can be explained in the scenario, generate explanation records and write them to the anomaly pool, and display the status or reduce the push level on the user terminal.

Images