An online water quality analyzer operation abnormal mode identification method and system

Abstract

The present application relates to the technical field of water quality online monitoring equipment data processing, and particularly relates to an online water quality analyzer operation abnormal mode identification method and system. The method obtains multi-parameter measurement data, operation state data, action event data and historical baseline data in multiple measurement periods, generates operation correlation data according to the measurement period; forms normal measurement section, cleaning section, calibration section, debugging section and recovery observation section according to the working mode identifier and action event data; generates measurement offset, response lag, state linkage and action feedback features in combination with the historical baseline, and forms an abnormal feature group; matches the abnormal feature group with a preset abnormal mode library and performs conflict resolution to obtain operation abnormal mode identification results, generates data validity markers and maintenance disposal strategies, and updates the identification results or the abnormal mode library using recovery data. The present application can distinguish measurement changes under different working modes and reduce operation abnormal mode confusion.

Description

Technical Field

[0001] This invention relates to the field of data processing technology for online water quality monitoring equipment, and in particular to a method and system for identifying abnormal operating patterns of an online water quality analyzer. Background Technology

[0002] In the field of data processing technology for online water quality monitoring equipment, online water quality analyzers typically collect multi-parameter measurement data through measurement units such as residual chlorine or chlorine dioxide, turbidity, pH, conductivity, and temperature. They also record operational status data in conjunction with sample flow rate, leakage status, sample pressure, communication status, alarm logs, and instrument logs. Existing solutions usually identify anomalies based on exceeding limits for individual measurements, comparing the current status of components with preset statuses, and identifying differences in monitoring data or patterns in output values. While such solutions can complete water quality parameter acquisition, status alarms, and fault indications under normal measurement scenarios, they are prone to limitations during continuous operation of online water quality analyzers, such as asynchronous measurement data and operational status data, unclear data status during cleaning or calibration, and similar behavior across multiple anomaly modes.

[0003] Existing solutions often rely on fixed thresholds, preset state rules, or single monitoring data sequences for processing. When cleaning action events, calibration action events, debugging mode switching events, and measurement value fluctuations occur intertwined, there is a lack of stable connection between measurement data, action event data, and subsequent anomaly identification results. This can easily lead to data from the cleaning, calibration, or debugging stages being mixed with normal measurement data, and can also cause operational anomalies such as flow path abnormalities, measurement cell contamination, electrode adhesion, and sensor drift to be confused during candidate identification.

[0004] For the joint processing of multi-parameter measurement data, operational status data, and action event data under continuous operation of online water quality analyzers, existing technologies still have common shortcomings in areas such as measurement cycle correlation, operating mode differentiation, anomaly feature organization, and candidate anomaly pattern differentiation. Therefore, it is necessary to address the scenario of identifying operational anomaly patterns in online water quality analyzers, specifically how to generate operational anomaly pattern identification results based on multi-parameter measurement data, operational status data, and action event data under continuous operation of the online water quality analyzer, through operating mode segmentation and candidate conflict resolution. Summary of the Invention

[0005] To address the above problems, this invention provides a method and system for identifying abnormal operating patterns of an online water quality analyzer. This method addresses the issue of how to generate abnormal operating pattern identification results based on multi-parameter measurement data, operating status data, and action event data during continuous operation of an online water quality analyzer, through operating mode segmentation and candidate conflict resolution.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] S100: Acquire multi-parameter measurement data, operating status data, action event data, and historical baseline data generated by the online water quality analyzer over multiple measurement cycles. The action event data includes one or more of cleaning action events, calibration action events, and debugging mode switching events. Perform time correlation on the multi-parameter measurement data, the operating status data, and the action event data according to the measurement cycle to generate operating correlation data.

[0008] S200. Based on the working mode identifier and action event data in the operation association data, the working mode is segmented to generate a working mode segmentation result. The working mode segmentation result includes multiple segments such as normal measurement segment, cleaning segment, calibration segment, debugging segment and recovery observation segment.

[0009] S300. Based on the segmentation results of the working mode and the historical baseline data, generate measurement offset features, response hysteresis features, state linkage features and action feedback features. The action feedback features are formed by the changes in multi-parameter measurement data before and after the cleaning segment or calibration segment. Combine the measurement offset features, the response hysteresis features, the state linkage features and the action feedback features into an abnormal feature group.

[0010] S400. Based on the abnormal feature group and the preset abnormal pattern library, perform abnormal pattern candidate matching to generate abnormal pattern candidate results; based on the working mode segmentation results, the state linkage features and the action feedback features, resolve conflicts in the abnormal pattern candidate results to generate operation abnormal pattern identification results.

[0011] S500: Generate data validity markers and maintenance and handling strategies for the corresponding measurement period based on the operation anomaly pattern recognition results; obtain recovery data obtained after the maintenance and handling strategies are executed, and update the preset anomaly pattern library or the operation anomaly pattern recognition results based on the recovery data.

[0012] Furthermore, the multi-parameter measurement data includes multiple values ​​among residual chlorine or chlorine dioxide, turbidity, pH, conductivity, and temperature; the operating status data includes one or more of sample flow rate, leakage status, sample pressure, operating temperature, communication status, alarm log, and instrument log; the historical baseline data includes one or more of normal measurement baseline, measurement fluctuation range, response stabilization time, cleaning recovery baseline, and calibration offset baseline.

[0013] Furthermore, the step of performing time correlation on the multi-parameter measurement data, the operating status data, and the action event data according to the measurement cycle to generate operating correlation data includes: writing multi-parameter measurement data and operating status data within the same measurement cycle into the same operating correlation record based on the measurement cycle identifier; and writing corresponding action event data into an operating correlation record matching the trigger time based on the trigger time of cleaning action events, calibration action events, or debugging mode switching events.

[0014] Furthermore, the segmentation of working modes based on the working mode identifier and action event data in the operational association data includes: dividing the operational association data with the working mode identifier as normal measurement into a normal measurement segment; dividing the continuous measurement cycle formed after the cleaning action event is triggered into a cleaning segment; dividing the continuous measurement cycle formed after the calibration action event is triggered into a calibration segment; dividing the continuous measurement cycle corresponding to the debugging mode switching event into a debugging segment; and dividing the continuous measurement cycle after the cleaning segment, calibration segment, or debugging segment ends into a recovery observation segment.

[0015] Furthermore, the generation of measurement offset features, response hysteresis features, state linkage features, and action feedback features based on the segmentation results of the working mode and the historical baseline data includes: generating measurement offset features based on the difference between the multi-parameter measurement data in the normal measurement segment and the historical baseline data; generating response hysteresis features based on the number of measurement cycles that the measured value takes to enter the stable range; generating state linkage features based on the correspondence between the operating state data and the multi-parameter measurement data within the same measurement cycle; and generating action feedback features based on the changes in measurement offset and response stabilization time before and after the cleaning or calibration segment.

[0016] Furthermore, the preset abnormal mode library includes multiple types of abnormal flow path, leakage, measurement cell contamination, electrode adhesion, sensor drift, compensation abnormality, and communication abnormality; in the preset abnormal mode library, each abnormal operation mode corresponds to one or more combinations of features among measurement offset feature, response hysteresis feature, state linkage feature, and action feedback feature.

[0017] Furthermore, the conflict resolution of the candidate abnormal modes based on the working mode segmentation results, the state linkage features, and the action feedback features includes: when there are multiple candidate abnormal modes among the candidate abnormal modes, marking the candidate abnormal modes corresponding to the debugging segment, cleaning segment, and calibration segment with data validity based on the working mode segmentation results; and ranking the multiple candidate abnormal modes by confidence based on sample flow rate, leakage status, change in measurement offset after cleaning, and change in measurement offset after calibration, thereby generating conflict resolution results.

[0018] Furthermore, the step of generating data validity marking and maintenance handling strategies for the corresponding measurement period based on the operational anomaly pattern identification results includes: marking the data for the corresponding measurement period as valid when the operational anomaly pattern identification result corresponds to a normal measurement segment and does not match an operational anomaly pattern; marking the data for the corresponding measurement period as invalid or pending confirmation when the operational anomaly pattern identification result corresponds to a debugging segment, cleaning segment, or calibration segment; marking the data for the corresponding measurement period as pending confirmation or downweighted output when the operational anomaly pattern identification result corresponds to a recovery observation segment; and generating one or more maintenance handling strategies based on the operational anomaly pattern identification results, such as cleaning, calibration, remote diagnosis, manual maintenance prompts, or alarm level determination.

[0019] Further, the step of acquiring recovery data obtained after the execution of the maintenance and handling strategy, and updating the preset abnormal mode library or the operation abnormal mode identification result based on the recovery data, includes: acquiring multi-parameter measurement data and operation status data within multiple measurement cycles after cleaning, calibration, or maintenance is completed, and generating recovery data; when the recovery data meets the recovery conditions of the corresponding operation abnormal mode, generating a recovery confirmation result, and writing the recovery confirmation result into the preset abnormal mode library; when the recovery data does not meet the recovery conditions of the corresponding operation abnormal mode, correcting the confidence ranking of the abnormal mode candidate results, and updating the operation abnormal mode identification result based on the corrected confidence ranking.

[0020] This invention also provides a system for identifying abnormal operating patterns of an online water quality analyzer, comprising:

[0021] The data access module is used to acquire multi-parameter measurement data, operating status data, action event data and historical baseline data generated by the online water quality analyzer in multiple measurement cycles;

[0022] The status association module, connected to the data access module, is used to perform time association on the multi-parameter measurement data, the running status data and the action event data according to the measurement cycle, and generate running association data.

[0023] The working mode segmentation module, connected to the state association module, is used to segment the working mode based on the working mode identifier and action event data in the running association data, and generate the working mode segmentation result.

[0024] An anomaly feature generation module, connected to the working mode segmentation module, is used to generate measurement offset features, response hysteresis features, state linkage features, and action feedback features based on the working mode segmentation results and the historical baseline data, and combine them into anomaly feature groups.

[0025] An abnormal pattern recognition module, connected to the abnormal feature generation module, is used to perform abnormal pattern candidate matching based on the abnormal feature group and a preset abnormal pattern library, and generate abnormal pattern candidate results.

[0026] The candidate result resolution module is connected to the abnormal pattern recognition module and is used to resolve conflicts of the abnormal pattern candidate results based on the working mode segmentation results, the state linkage features and the action feedback features, and generate the running abnormal pattern recognition results.

[0027] The result output module is connected to the candidate result resolution module and is used to generate data validity markers and maintenance and handling strategies for the corresponding measurement period based on the operation anomaly pattern recognition results.

[0028] The feedback update module, connected to the result output module, is used to obtain the recovery data obtained after the maintenance and handling strategy is executed, and to update the preset abnormal mode library or the operation abnormal mode identification result based on the recovery data.

[0029] The key innovations of this invention include:

[0030] (1) Establish time correlation between multi-parameter measurement data, operation status data and action event data generated in multiple measurement cycles according to the measurement cycle, and divide the normal measurement segment, cleaning segment, calibration segment, debugging segment and recovery observation segment based on the working mode identifier and action event data, so that subsequent anomaly identification is based on the coupling of operation correlation data and working mode segmentation results.

[0031] (2) Based on the segmented results of the working mode and historical baseline data, the measurement offset characteristics, response hysteresis characteristics, state linkage characteristics and action feedback characteristics are combined into an abnormal feature group. Among them, the action feedback characteristics are formed by the changes of multi-parameter measurement data before and after the cleaning or calibration segment, so that the cleaning action event and the calibration action event participate in the feature organization of the abnormal operation mode.

[0032] (3) Generate candidate results of abnormal modes based on abnormal feature groups and preset abnormal mode library, and resolve conflicts of candidate results of abnormal modes by combining working mode segmentation results, state linkage features and action feedback features to form operation abnormal mode identification results, and then update the preset abnormal mode library or operation abnormal mode identification results based on recovery data.

[0033] The following are its main beneficial effects:

[0034] (1) To address the problem of unclear connection between measurement data, operating status data and action event data in the existing scheme, the cleaning segment, calibration segment, debugging segment and recovery observation segment are separated from the continuous measurement data by measurement cycle association and working mode segmentation, so that the subsequent abnormal feature generation can call the corresponding data according to different working modes.

[0035] (2) To address the problem that existing solutions cannot distinguish the source of anomalies by measuring changes before and after cleaning or calibration, a combination of measurement offset characteristics, response hysteresis characteristics, state linkage characteristics and action feedback characteristics is used to give corresponding characteristic sources to abnormal operation modes such as measurement cell contamination, electrode adhesion, sensor drift, and flow path anomalies.

[0036] (3) To address the problem that candidate results are easily confused when multiple abnormal modes exhibit similar behavior, abnormal mode candidate matching and conflict resolution are used to enable the data status in the debugging, cleaning and calibration stages to participate in the candidate processing, and to enable the sample flow rate, leakage status, changes in measurement offset after cleaning and changes in measurement offset after calibration to participate in the formation of the abnormal mode recognition results. Attached Figure Description

[0037] Figure 1 A flowchart illustrating a method for identifying abnormal operating patterns of an online water quality analyzer, provided in an embodiment of this application.

[0038] Figure 2 This is a structural block diagram of an abnormal operation pattern recognition system for an online water quality analyzer provided in an embodiment of this application. Detailed Implementation

[0039] To enable those skilled in the art to better understand the technical solution, the present invention will be described in detail below with reference to embodiments. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.

[0040] S100: Acquire multi-parameter measurement data, operating status data, action event data, and historical baseline data generated by the online water quality analyzer over multiple measurement cycles. The action event data includes one or more of cleaning action events, calibration action events, and debugging mode switching events. Perform time correlation on the multi-parameter measurement data, the operating status data, and the action event data according to the measurement cycle to generate operating correlation data.

[0041] S200. Based on the working mode identifier and action event data in the operation association data, the working mode is segmented to generate a working mode segmentation result. The working mode segmentation result includes multiple segments such as normal measurement segment, cleaning segment, calibration segment, debugging segment and recovery observation segment.

[0042] S300. Based on the segmentation results of the working mode and the historical baseline data, generate measurement offset features, response hysteresis features, state linkage features and action feedback features. The action feedback features are formed by the changes in multi-parameter measurement data before and after the cleaning segment or calibration segment. Combine the measurement offset features, the response hysteresis features, the state linkage features and the action feedback features into an abnormal feature group.

[0043] S400. Based on the abnormal feature group and the preset abnormal pattern library, perform abnormal pattern candidate matching to generate abnormal pattern candidate results; based on the working mode segmentation results, the state linkage features and the action feedback features, resolve conflicts in the abnormal pattern candidate results to generate operation abnormal pattern identification results.

[0044] S500: Generate data validity markers and maintenance and handling strategies for the corresponding measurement period based on the operation anomaly pattern recognition results; obtain recovery data obtained after the maintenance and handling strategies are executed, and update the preset anomaly pattern library or the operation anomaly pattern recognition results based on the recovery data.

[0045] S100. Acquire multi-parameter measurement data, operating status data, action event data, and historical baseline data generated by the online water quality analyzer over multiple measurement cycles. The action event data includes one or more of cleaning action events, calibration action events, and debugging mode switching events. Perform time correlation on the multi-parameter measurement data, operating status data, and action event data according to the measurement cycle to generate operating correlation data.

[0046] After the online water quality analyzer is put into continuous measurement, the data access module reads the data output by each measurement unit and the instrument control unit. In this embodiment, the online water quality analyzer can be deployed at the outlet of a waterworks, the end of a water supply network, a rural drinking water monitoring point, or a chlorinated water body monitoring point. The local controller of the instrument drives the residual chlorine or chlorine dioxide measurement unit, turbidity measurement unit, pH measurement unit, conductivity measurement unit, and temperature measurement unit to work according to the set measurement cycle. After each multi-parameter measurement is completed, the controller assigns a measurement cycle identifier to that measurement. The measurement cycle identifier can be formed by the measurement start time, measurement end time, and internal cycle number of the instrument, or it can be formed by the sampling batch number already in the instrument log.

[0047] Multi-parameter measurement data comes from the detection equipment of the online water quality analyzer, and may specifically include multiple measurements such as residual chlorine or chlorine dioxide, turbidity, pH, conductivity, and temperature. Operational status data comes from components that reflect sample supply, environmental, and communication status during instrument operation, such as sample flow rate, leakage status, sample pressure, operating temperature, communication status, alarm logs, and instrument logs. Action event data comes from cleaning control records, calibration control records, and debugging mode switching records. Cleaning action events are typically triggered by turbidity measurement cell cleaning, residual chlorine electrode cleaning, or flow path cleaning; calibration action events are triggered by pH calibration, residual chlorine calibration, or calibration of other measurement units; and debugging mode switching events are generated when the local touchscreen, remote monitoring platform, or maintenance operation enters debugging mode.

[0048] After receiving the above data, the status association module first reads the measurement cycle identifier from the multi-parameter measurement data, and then searches for the operating status data within the same measurement cycle. If the operating status data has a cycle identifier consistent with the measurement cycle, it is directly written into the same operating association record; if the operating status data only has a collection timestamp, it is merged according to whether the collection timestamp falls within the start and end time range of the measurement cycle. For continuous acquisition states such as sample flow rate, sample pressure, and operating temperature, the average value, final value, or status change identifier within the measurement cycle can be selected to enter the operating association record; for event-type states such as leakage status and alarm logs, they can be written into the corresponding measurement cycle according to the event trigger time, and the event persistence status can be retained for subsequent use in the generation of working mode segmentation and status linkage features.

[0049] The association of action event data employs a matching method between event trigger time and measurement cycle range. When a cleaning action event, calibration action event, or debugging mode switching event occurs within a specific measurement cycle, the status association module writes the action event into the corresponding operation association record for that measurement cycle. When an action event spans multiple measurement cycles, the starting measurement cycle records the action start identifier, the continuous measurement cycle records the action continuity identifier, and the ending measurement cycle records the action end identifier. If an action event occurs between two measurement cycles, it is assigned to the subsequent measurement cycle that will generate a measurement response, based on the time interval between the event trigger time and the adjacent measurement cycle. The original trigger time is retained in the operation association record to prevent the boundaries of subsequent cleaning or calibration segments from being advanced or delayed.

[0050] Historical baseline data is formed from historical normal measurement data, historical cleaning records, historical calibration records, and maintenance confirmation records. Historical baseline data is not simply used as a fixed threshold, but is written into the operational correlation data as a reference during the status correlation phase. Specifically, the status correlation module selects the corresponding normal measurement baseline, measurement fluctuation range, response stabilization time, cleaning recovery baseline, and calibration offset baseline based on the current measurement point, instrument number, measurement parameter type, and operating mode. If a newly installed instrument lacks sufficient historical data, it can reference the factory configuration baseline or a baseline from a similar monitoring point; this referenced baseline is distinguished in the operational correlation record by the baseline source identifier, and subsequent abnormal feature generation can differentiate feature weights according to the baseline source.

[0051] During data acquisition, if a measurement parameter is missing within a certain measurement cycle, but the operational status data and action event data are complete, the status association module retains the operational association record for that measurement cycle and writes a missing parameter identifier at the location of the missing parameter. If the measurement cycle identifier appears repeatedly, the order is distinguished according to the measurement end time in the instrument log, and the original record corresponding to the repeated cycle identifier is retained; subsequent working mode segmentation is based on time continuity. If a communication failure results in data not being uploaded for multiple consecutive measurement cycles, the data access module reads the local cached data after communication is restored and fills in the corresponding operational association record according to the local measurement time, avoiding the remote reception time replacing the actual measurement time.

[0052] The operational correlation data includes measurement cycle identifiers, multi-parameter measurement data, operational status data, action event data, and historical baseline references. This operational correlation data is not simply a data set; rather, it places measurement results, sample supply status, instrument status, and action events within the same measurement cycle into the same processing unit. Subsequently, S200 only reads the operating mode identifier, action event data, and temporal continuity from the operational correlation data for segmentation. S300 then reads the measured values, status values, and historical baselines from the same operational correlation data to generate corresponding anomaly characteristics.

[0053] S200. Based on the working mode identifier and action event data in the operation association data, the working mode is segmented to generate a working mode segmentation result. The working mode segmentation result includes multiple segments such as normal measurement segment, cleaning segment, calibration segment, debugging segment, and recovery observation segment.

[0054] After receiving the operational correlation data generated by S100, the operating mode segmentation module does not directly identify anomalies in the measured values. Instead, it first identifies the instrument's operating mode for each measurement cycle. During cleaning, calibration, debugging, and recovery processes, online water quality analyzers generate data changes different from normal water sample measurements. If these data are directly entered into the anomaly mode candidate matching, short-term fluctuations caused by instrument actions may be mistaken for operational anomalies. This step segments the operational correlation data according to the actual instrument operation process through operating mode segmentation. Subsequently, S300 uses different feature extraction boundaries within different data segments.

[0055] During the segmentation process, the working mode segmentation module first reads the working mode identifier from the associated running data. When the working mode identifier is "normal measurement," the measurement cycle is temporarily assigned to the normal measurement segment; when the working mode identifier is "debugging," the measurement cycle is assigned to the debugging segment. If both a normal measurement identifier and a debugging mode switching event exist within the same measurement cycle, the debugging mode switching event takes priority, and the cycle and its continuous measurement cycles during that period are assigned to the debugging segment, while the original normal measurement identifier is retained as a state conflict record. This state conflict record is not included in the measurement offset feature calculation of S300, but it can be used as the basis for invalidating debugging data during S400 conflict resolution.

[0056] For the division of cleaning segments, the working mode segmentation module uses the trigger and end times of the cleaning action event as boundaries. The continuous measurement cycle formed after the cleaning action event is triggered is defined as a cleaning segment; if the cleaning action event only has a trigger time but lacks an end time, the cleaning completion status, flow recovery status, or subsequent normal measurement indicator in the instrument log is used as the end boundary. For calibration segments, the working mode segmentation module uses the same processing method, dividing the continuous measurement cycle formed after the calibration action event is triggered into a calibration segment. If the cleaning action event and the calibration action event overlap in time, the segmentation result retains the overlap relationship between the two, and the changes before and after cleaning and before and after calibration are calculated separately when generating subsequent action feedback characteristics.

[0057] The recovery observation period is formed by the continuous measurement cycle following the completion of the cleaning, calibration, or debugging phases. The length of the recovery observation period can be determined based on the response stabilization time in historical baseline data or the observation cycle configured for the instrument. In one implementation, if a parameter has not yet entered a stable range within the continuous measurement cycle following the cleaning phase, the recovery observation period extends to the measurement cycle in which that parameter enters a stable range; if multiple parameters have different stabilization times, the longer stabilization time among the parameters involved in generating the abnormal features is used as the boundary of the recovery observation period. In this way, data still in the recovery process after the cleaning or calibration operation is not directly classified as a normal measurement segment.

[0058] The operating mode segmentation also handles situations involving missing status and conflicting actions. If the operating mode identifier is missing, but the action event data indicates the existence of a cleaning or calibration action event, the segmentation module establishes a cleaning or calibration segment based on the action event data and supplements the missing operating mode identifier as an action derivation identifier. If the action event data is missing, but the instrument log contains records of cleaning completion, calibration completion, or debugging exit, the segmentation module uses the log time to deduce the previous continuous measurement interval and records the source of the deduction in the segmentation results. If a leak alarm and a cleaning action event occur simultaneously in the same cycle, that cycle is still classified as a cleaning segment, but the leak status is retained in the segmentation results. Subsequently, the S400 uses this leak status to handle candidate conflicts between flow path anomalies and measurement cell contamination during conflict resolution.

[0059] The working mode segmentation results include the data segment type, start and end boundaries of the data segment, source of the action event, and recovery observation boundary for each measurement cycle. This result is directly passed to S300. S300 primarily uses the normal measurement segment when generating measurement offset features, uses the cleaning segment, calibration segment, and their preceding and following adjacent data segments when generating action feedback features, and uses continuous measurement cycles within the recovery observation segment when generating response hysteresis features. The working mode segmentation results also participate in candidate abnormal mode conflict resolution in S400 and data validity marking in S500.

[0060] S300. Based on the segmentation results of the working mode and the historical baseline data, generate measurement offset characteristics, response hysteresis characteristics, state linkage characteristics, and action feedback characteristics. The action feedback characteristics are formed by changes in multi-parameter measurement data before and after the cleaning or calibration segment. Combine the measurement offset characteristics, the response hysteresis characteristics, the state linkage characteristics, and the action feedback characteristics into an abnormal characteristic group.

[0061] After receiving the operating mode segmentation results generated by S200, the anomaly feature generation module first selects the measurement cycle to participate in feature generation based on the data segment type. The normal measurement segment is used to generate measurement offset and status linkage features; the recovery observation segment is used to generate response hysteresis features; and the continuous measurement cycles before and after the cleaning or calibration segment are used to generate action feedback features. This is because the technical meaning of measured value changes differs under different operating modes; continuous offset in the normal measurement segment is more suitable as a clue to drift or water quality changes, while recovery changes before and after the cleaning or calibration segment are more suitable for reflecting pool contamination, electrode adhesion, or sensor drift.

[0062] The measurement offset feature is formed by the difference between multi-parameter measurement data in the normal measurement segment and historical baseline data. The anomaly feature generation module reads the parameter type, measurement point, and historical baseline reference for the current measurement cycle, and compares multiple measurements of residual chlorine or chlorine dioxide, turbidity, pH, conductivity, and temperature with the corresponding normal measurement baseline. If the historical baseline data includes the measurement fluctuation range, the measurement offset feature records not only the offset amount but also the offset direction and the number of measurement cycles in which the offset continues. If a parameter is missing, that parameter is not included in the measurement offset feature calculation for this cycle, but its missing identifier is retained in the anomaly feature group for S400 to determine communication anomalies or measurement unit anomaly candidate modes.

[0063] The response hysteresis characteristic reflects the number of measurement cycles required for a measured value to transition from a changing state to a stable range. During the recovery observation period or consecutive measurement cycles following a sudden change in the measured value, the anomaly feature generation module reads the response stabilization time or measurement fluctuation range from historical baseline data and compares the amplitude of the continuous measured value changes with the stable range. If the measured value continuously deviates from the stable range over multiple measurement cycles, the number of hysteresis cycles for the corresponding parameter is recorded. If a short-term fluctuation occurs after the cleaning or calibration period but subsequently returns to the stable range, the hysteresis characteristic is recorded as recovery hysteresis. The S400 will then combine this with motion feedback characteristics, and will not directly use it as the sole basis for sensor drift candidate analysis.

[0064] The status linkage feature is formed by the correspondence between the operating status data and multi-parameter measurement data within the same measurement cycle. The anomaly feature generation module reads sample flow rate, leakage status, sample pressure, operating temperature, communication status, alarm log, and instrument log, and associates them with the multi-parameter measurement data in the corresponding measurement cycle. For example, when the sample flow rate is lower than the instrument's configured effective injection range and multiple measurement values ​​show simultaneous response lag, the status linkage feature records that the flow rate status and multi-parameter lag occur in the same cycle; when a leakage status is triggered and the measurement value shows an abnormal jump, the status linkage feature records that the leakage status and measurement anomaly occur in the same cycle; when the communication status is abnormal and multi-parameter measurement data is continuously missing, the status linkage feature records that the communication anomaly and data missing occur in the same cycle. The above status linkage features are used by the S400 to distinguish between instrument operating abnormalities and actual changes in water quality.

[0065] Action feedback features are formed by changes in multi-parameter measurement data before and after the cleaning or calibration phase. For cleaning action events, the anomaly feature generation module selects the adjacent normal measurement segment before the start of the cleaning phase as the pre-action data and the recovery observation segment after the end of the cleaning phase as the post-action data, calculating the changes in measurement offset and response stabilization time before and after cleaning. If the turbidity measurement value continuously deviates from the historical baseline before cleaning, and the offset decreases and the response stabilization time shortens after cleaning, the action feedback feature is recorded as a cleaning recovery feature; if the measurement offset does not change accordingly after cleaning, it is recorded as a cleaning non-recovery feature. For calibration action events, the module generates calibration offset recovery features based on the changes in measurement offset before and after the calibration phase, used to subsequently distinguish between sensor drift and measurement cell contamination.

[0066] In one implementation, the anomaly feature generation module further separates cases where the cleaning and calibration segments overlap. If the calibration action immediately follows the cleaning action, the cleaning action feedback feature is first calculated using data before the cleaning action and data after the cleaning action and before the calibration action. Then, the calibration action feedback feature is calculated using data before the calibration action and data from the recovery observation segment after the calibration action. If there is a missing available measurement period in between, the action feedback feature is marked as pending confirmation and does not participate in the dominant calculation of the confidence ranking in this round, but it is retained in the anomaly feature group for supplementation when the S500 recovers the data feedback update.

[0067] The anomaly feature group is formed by combining measurement offset features, response hysteresis features, state linkage features, and action feedback features. This anomaly feature group retains the source data segment, measurement period range, and parameter type for each feature, rather than just a single numerical result. When performing anomaly pattern candidate matching, the S400 reads the feature combinations in the anomaly feature group and matches them with a preset anomaly pattern library. When multiple candidate anomaly patterns exist, the S400 also focuses on reading state linkage features and action feedback features for subsequent conflict resolution.

[0068] S400. Based on the abnormal feature group and the preset abnormal pattern library, perform abnormal pattern candidate matching to generate abnormal pattern candidate results; based on the working mode segmentation results, the state linkage features and the action feedback features, resolve conflicts in the abnormal pattern candidate results to generate operation abnormal mode identification results:

[0069] After receiving the abnormal feature group generated by S300, the abnormal pattern recognition module performs abnormal pattern candidate matching with a preset abnormal pattern library. The preset abnormal pattern library can be configured with various operational abnormality modes, including flow path abnormality, leakage abnormality, measurement cell contamination, electrode adhesion, sensor drift, compensation abnormality, and communication abnormality. Each operational abnormality mode does not correspond to a single feature, but rather to one or more combinations of features such as measurement offset features, response hysteresis features, status linkage features, and action feedback features. For example, measurement cell contamination is usually related to turbidity measurement offset and changes in offset after cleaning; electrode adhesion is usually related to residual chlorine or pH response hysteresis and changes in response stabilization time after cleaning; flow path abnormalities are usually related to sample flow rate or sample pressure status and synchronous hysteresis of multiple measurement parameters.

[0070] During the candidate matching process, the anomaly pattern recognition module first enters the corresponding anomaly pattern subset according to the parameter type in the anomaly feature group, and then performs matching according to feature combinations. If the same anomaly feature group matches the feature combination of only one running anomaly pattern, a single anomaly pattern candidate result is generated; if the measurement offset feature, state linkage feature, and action feedback feature point to different anomaly patterns, multiple candidate anomaly patterns are generated, and the matching criteria for each candidate anomaly pattern are retained. The matching criteria include the data segment type involved in the matching, the corresponding measurement cycle range, the triggered feature combination, and the action feedback result, which are then further processed by the candidate result resolution module.

[0071] When resolving conflicts among candidate results for abnormal modes, the candidate result resolution module prioritizes reading the results from the working mode segment. If a candidate abnormal mode is triggered solely by changes in measured values ​​during the debugging segment, the data corresponding to that candidate abnormal mode is first marked as valid and is not considered a primary abnormal mode under normal measurement segments. If a candidate abnormal mode is triggered by action process data during the cleaning or calibration segment, the module processes it in conjunction with action feedback characteristics. Short-term turbidity fluctuations within the cleaning segment are not directly used as the primary basis for candidate contamination in the measurement cell; only the recovery observation segment after the cleaning segment ends is used to confirm whether the cleaning feedback matches the candidate mode.

[0072] The status linkage feature is used to handle multiple candidate anomaly patterns with similar measurement performance. If both flow path anomaly and measurement cell contamination exist in the candidate anomaly pattern results, the candidate result resolution module reads the synchronous changes in sample flow rate, sample pressure, and multiple measurement parameters. When sample flow rate anomaly and response hysteresis of multiple parameters occur within the same measurement cycle, the confidence level of flow path anomaly is prioritized over measurement cell contamination. If both leakage anomaly and communication anomaly exist in the candidate anomaly pattern results, the module reads the leakage status, communication status, and data missing pattern. When the leakage status is triggered and the measurement value changes abruptly, leakage anomaly is prioritized in the operational anomaly pattern identification results, while communication anomaly is only retained as a parallel candidate when data transmission is interrupted or instrument logs are missing.

[0073] Action feedback features are used to handle conflicts between candidate anomalies of contamination, adhesion, and drift. If the measurement offset changes significantly after cleaning and the response settling time shortens, the candidate anomalies of measurement cell contamination or electrode adhesion related to the cleaning action are moved forward in the ranking. If the measurement offset change after calibration conforms to the calibration offset baseline, the candidate anomalies of sensor drift are moved forward in the ranking. If no corresponding recovery change occurs after cleaning and calibration, the candidate result resolution module lowers the confidence ranking of contamination and drift candidate anomalies and retains candidate results in the direction of flow path anomalies, hardware anomalies, or communication anomalies. The above ranking is not a simple numerical ranking of all candidate results, but rather a conflict resolution based on the logical sequence of working modes, state linkages, and action feedback.

[0074] When multiple candidate anomaly patterns still exhibit a parallel relationship after conflict resolution, the operational anomaly pattern identification result can retain the primary anomaly pattern and accompanying anomaly patterns. The primary anomaly pattern corresponds to the candidate anomaly pattern with the highest confidence ranking and supported by state linkage or action feedback; accompanying anomaly patterns record candidate anomaly patterns that appear simultaneously with the primary anomaly pattern but whose evidence sources are incomplete. In one implementation, if data from the debugging, cleaning, or calibration phases participates in candidate matching, the operational anomaly pattern identification result also records the data segment source for the corresponding measurement period, for use by S500 when generating data validity markers. The operational anomaly pattern identification result output by S400 includes the anomaly pattern type, the corresponding measurement period range, the candidate ranking result, and the conflict resolution basis; this output is directly fed into S500.

[0075] S500: Generate data validity markers and maintenance handling strategies for the corresponding measurement period based on the operational anomaly pattern recognition results; obtain recovery data after the maintenance handling strategy is executed, and update the preset anomaly pattern library or the operational anomaly pattern recognition results based on the recovery data.

[0076] After receiving the operational anomaly pattern identification results generated by S400, the output module first combines the working mode segmentation results to generate data validity markers for the corresponding measurement period. If the operational anomaly pattern identification result corresponds to a normal measurement segment and does not match an operational anomaly pattern, the data for the corresponding measurement period is marked as valid. If the operational anomaly pattern identification result corresponds to a debugging segment, cleaning segment, or calibration segment, the data for the corresponding measurement period is marked as invalid or pending confirmation; this marker prevents these data from directly entering the formal water quality output, but they can still be used as action feedback features or recovery data for subsequent identification. If the operational anomaly pattern identification result corresponds to a recovery observation segment, the data for the corresponding measurement period is marked as pending confirmation or deweighted output, and its validity status is updated after subsequent recovery data verification.

[0077] The maintenance and handling strategy is generated jointly from the anomaly pattern type, conflict resolution results, and data validity markers in the anomaly pattern identification results. For measurement cell contamination or electrode adhesion, the result output module generates a cleaning strategy and associates the cleaning object with the corresponding measurement unit; for sensor drift, a calibration strategy is generated; for flow path anomalies or leakage anomalies, remote diagnostic or manual maintenance prompts are generated, and the measurement cycle range corresponding to sample flow rate, sample pressure, or leakage status is recorded; for communication anomalies, a communication status verification strategy is generated; for alarm level determination, the result output module generates an alarm level field corresponding to the measurement cycle based on the main anomaly pattern and data validity markers in the anomaly pattern identification results. The above maintenance and handling strategies can be executed by the local controller or sent to the maintenance terminal through a remote monitoring platform.

[0078] The feedback update module reads the recovery data after the maintenance and handling strategy is executed. The recovery data comes from multi-parameter measurement data and operational status data across multiple measurement cycles following cleaning, calibration, or maintenance. It uses the measurement cycle correlation method from S100 to form the correlation data for the recovery phase. If, after the cleaning strategy is executed, the measurement offset of the corresponding parameter in the recovery data returns to the allowable measurement fluctuation range of the historical baseline, and the response stabilization time matches the cleaning recovery baseline, a recovery confirmation result is generated and written into the preset abnormal mode library, corresponding to the feature combination for measurement cell contamination or electrode adhesion. If, after the calibration strategy is executed, the change in measurement offset matches the calibration offset baseline, the recovery confirmation result is written into the feature combination corresponding to sensor drift.

[0079] When the recovered data does not meet the recovery conditions for the corresponding abnormal operating mode, the feedback update module does not directly delete the original abnormal operating mode identification result, but instead corrects the confidence ranking of the candidate abnormal mode results. Specifically, if no change in measurement offset or response settling time occurs after cleaning, the module lowers the ranking of candidate abnormalities such as measurement cell contamination or electrode adhesion; if no offset recovery occurs after calibration, the module lowers the ranking of candidate abnormalities such as sensor drift and re-calls the candidate abnormal modes retained in S400 to re-rank flow path abnormalities, compensation abnormalities, or communication abnormalities. The corrected confidence ranking is used to update the abnormal operating mode identification result, and the updated result and the original identification result are both retained in the instrument log for reference in subsequent measurement cycles.

[0080] In one implementation, the recovered data is also used to update the data validity markers. If the recovery confirmation result indicates that the previous abnormal operation mode has been cleared, calibrated, or maintained, the feedback update module marks the measurement cycles in the recovery observation segment that meet the recovery conditions as valid; measurement cycles that do not meet the recovery conditions remain pending confirmation or are output with reduced weight. If the recovered data indicates that the anomaly still exists, the result output module regenerates the maintenance handling strategy according to the corrected abnormal operation mode identification results and associates this strategy with the previous strategy. Thus, the output of S500 not only ends the current round of identification but also serves as the source of historical baseline data, recovered data, and anomaly mode library update results in the next round of measurement cycle, participating in the subsequent processing of S100 and S400.

[0081] Example 2: Figure 2 A structural block diagram of an abnormal operation pattern recognition system for an online water quality analyzer according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0082] The data access module 01 is used to acquire multi-parameter measurement data, operating status data, action event data, and historical baseline data generated by the online water quality analyzer over multiple measurement cycles. This data access module is connected to the online water quality analyzer's detection equipment, instrument control unit, and instrument log storage unit. After the instrument enters continuous measurement mode, it reads multiple values ​​from the following according to the measurement cycle: residual chlorine or chlorine dioxide measurement value, turbidity measurement value, pH measurement value, conductivity measurement value, and temperature measurement value. Simultaneously, it reads sample flow rate, leakage status, sample pressure, operating temperature, communication status, alarm log, and instrument log. When receiving measurement results output by the detection equipment, the data access module uses the measurement cycle identifier generated internally by the instrument as the acquisition boundary. When the measurement cycle identifier is missing, it reads the measurement start time and measurement end time, generates a temporary cycle identifier according to the measurement cycle of the instrument control unit, and retains this temporary cycle identifier along with the original acquisition time.

[0083] The action event data is generated by the instrument control unit and originates from cleaning action events, calibration action events, and debugging mode switching events. When the data access module reads the action event data, it does not store it separately as a regular operation log. Instead, it retains the event trigger time, action type, and action duration, enabling subsequent status association modules to categorize the action event data into the corresponding measurement cycle. For data retransmitted after a communication interruption, the data access module prioritizes using the instrument's local acquisition time and does not use the remote platform's reception time to replace the measurement time. If a single measurement parameter is missing within a measurement cycle, it retains the other measurement data and operating status data corresponding to that measurement cycle and sets a missing status, allowing subsequent modules to identify whether the missing data is caused by acquisition anomalies, communication anomalies, or non-measurement operating modes.

[0084] Historical baseline data consists of one or more of the following: normal measurement baseline, measurement fluctuation range, response stabilization time, cleaned and recovered baseline, and calibration offset baseline. The data access module retrieves the corresponding historical baseline data based on the instrument number, measurement point, and measurement parameter type, and establishes a reference relationship between the historical baseline data and the current acquisition batch. If the instrument is currently in the initial installation or operational phase with insufficient historical data, the data access module reads the factory-configured baseline or a baseline from a similar point, and retains the baseline source status. This status does not directly generate an operational anomaly pattern recognition result but is transmitted along with the data to the status association module and the anomaly feature generation module.

[0085] The state association module 02, connected to the data access module, is used to perform time association on the multi-parameter measurement data, the operating status data, and the action event data according to the measurement cycle, generating operating association data. After receiving the data output by the data access module, the state association module first reads the measurement cycle identifier, and then merges the multi-parameter measurement data and operating status data within the same measurement cycle into the same operating association record. For continuous states such as sample flow rate, sample pressure, and operating temperature, the state association module extracts the state changes within the corresponding time period based on the start and end times of the measurement cycle. For event-type states such as leakage status, alarm logs, and communication status, the state association module writes the corresponding operating association record according to the overlap between the event trigger time and the measurement cycle.

[0086] The temporal correlation of action event data is completed by the state correlation module based on the trigger time and duration. When a cleaning action event, calibration action event, or debugging mode switching event occurs within a certain measurement cycle, the state correlation module writes the corresponding action event data into that measurement cycle. When an action event spans multiple measurement cycles, the start state, duration state, and end state of the action are respectively assigned to the corresponding consecutive measurement cycles. If the trigger time of an action event is between two measurement cycles, the state correlation module assigns the action event to the subsequent measurement cycle in which the measurement change occurs, based on the principle of subsequent measurement response attribution, while retaining the original trigger time for the working mode segmentation module to divide the cleaning segment, calibration segment, or recovery observation segment.

[0087] When there is a time offset between multi-parameter measurement data and operational status data, the status association module first checks the measurement cycle identifier of the instrument control unit, and then checks the local acquisition time. If the two are inconsistent, the original time is retained and a time inconsistency status is generated. This time inconsistency status is transmitted to subsequent modules along with the operational association data. If duplicate measurement records occur in the same measurement cycle, the order is determined by the measurement end time in the instrument log. Duplicate records do not overwrite the original records, but are retained as duplicate statuses within the same cycle. After the operational association data is generated, the status association module provides the measurement cycle identifier, operational mode identifier, action event data, and operational status data to the working mode segmentation module, enabling subsequent modules to identify normal measurement segments, cleaning segments, calibration segments, debugging segments, and recovery observation segments based on the same measurement cycle boundary.

[0088] The working mode segmentation module 03, connected to the state association module, is used to segment the working mode based on the working mode identifier and action event data in the running association data, and generate a working mode segmentation result. After reading the running association data, the working mode segmentation module first identifies whether the current measurement cycle belongs to the normal measurement state, debugging state, or action execution state according to the working mode identifier. Running association data with the working mode identifier indicating normal measurement is assigned to the normal measurement segment; continuous measurement cycles with debugging mode switching events are assigned to the debugging segment. If the working mode identifier indicates normal measurement, but there is a debugging mode switching event in the same running association record, the working mode segmentation module prioritizes the debugging mode switching event, assigns the measurement cycle to the debugging segment, and retains the state conflict mark. The subsequent result output module then marks the data validity of the corresponding measurement cycle accordingly.

[0089] For cleaning action events, the working mode segmentation module uses the trigger time of the cleaning action event as the starting point of the cleaning segment; if a cleaning completion status exists in the associated data, the measurement cycle corresponding to the cleaning completion status is used as the ending point of the cleaning segment. For calibration action events, the working mode segmentation module uses the same processing method to divide the calibration segment. If a cleaning action event or calibration action event lacks a completion status, the working mode segmentation module reads the action completion information or subsequent working mode identifier from the instrument log, and uses the first normal measurement cycle after the action is completed as the action completion boundary. If the action event abnormally coincides with a leakage status or communication status, the working mode segmentation result retains the abnormal status and does not change the division boundary of the cleaning segment or calibration segment.

[0090] The recovery observation segment is formed by continuous measurement cycles following the completion of the cleaning, calibration, or debugging segments. The operating mode segmentation module reads the response settling time from historical baseline data and determines the length of the recovery observation segment based on the recovery times of different measurement parameters. When the recovery times of multiple parameters are inconsistent, the longer recovery time among the parameters involved in anomaly feature generation is used to determine the boundary of the recovery observation segment. The operating mode segmentation result includes the start and end measurement cycles, data segment type, and action event source for each data segment. This result is transmitted to the anomaly feature generation module, which uses this information to distinguish between measurement offsets within normal measurement segments, response hysteresis within recovery observation segments, and action feedback before and after the cleaning or calibration segments.

[0091] The anomaly feature generation module 04, connected to the working mode segmentation module, is used to generate measurement offset features, response hysteresis features, status linkage features, and action feedback features based on the working mode segmentation results and the historical baseline data, and combine them into anomaly feature groups. After receiving the working mode segmentation results, the anomaly feature generation module calls the measurement values, operating status data, and historical baseline data from the running associated data according to the data segment type. The normal measurement segment is used to generate measurement offset features, the recovery observation segment is used to generate response hysteresis features, and the continuous measurement cycles before and after the cleaning or calibration segment are used to generate action feedback features. The anomaly feature generation module does not directly use short-term measurement fluctuations in the cleaning and calibration segments as measurement offset features, but instead incorporates them together with the data before and after the action into the calculation of action feedback features.

[0092] The measurement offset feature is formed by the difference between multi-parameter measurement data in the normal measurement segment and historical baseline data. The anomaly feature generation module reads the normal measurement baseline and measurement fluctuation range, and records the offset direction, the duration of the offset measurement period, and the source of the corresponding data segment according to the parameter type. The response hysteresis feature is formed by the number of measurement periods required for the measured value to enter the stable interval. After the cleaning or calibration segment ends, if the measured value in the recovery observation segment continues to fail to enter the stable interval corresponding to the historical baseline data, the number of hysteresis periods for that parameter is recorded. If a certain measurement parameter is missing in the current measurement period, the anomaly feature generation module does not calculate the measurement offset feature based on the missing parameter, but retains the missing state, which is called by the anomaly pattern recognition module when there is a communication anomaly or a measurement unit anomaly candidate matching.

[0093] The status linkage feature is formed by the correspondence between operating status data and multi-parameter measurement data within the same measurement cycle. The anomaly feature generation module reads sample flow rate, leakage status, sample pressure, operating temperature, communication status, alarm log, and instrument log, and correlates them with changes in measurement values ​​within the same cycle. For example, when an abnormal sample flow rate and response lag of multiple measurement parameters occur simultaneously within the same continuous measurement cycle, the status linkage feature retains the correspondence between the flow rate status and the multi-parameter lag; when a leakage status is triggered and accompanied by a sudden change in measurement value, the status linkage feature retains the correspondence between the leakage status and the measurement anomaly; when a communication status anomaly is accompanied by continuous data loss, the status linkage feature retains the correspondence between the communication status and the data loss.

[0094] Action feedback features are formed by changes in multi-parameter measurement data before and after the cleaning or calibration phase. The anomaly feature generation module selects adjacent normal measurement segments before the start of the cleaning phase as pre-action data and recovery observation segments after the end of the cleaning phase as post-action data, calculating the change in measurement offset and response settling time after cleaning. For calibration action events, measurement data before and after the calibration phase are selected, and the change in measurement offset after calibration is calculated. If the cleaning and calibration actions occur adjacently, the anomaly feature generation module calculates the cleaning action feedback features and calibration action feedback features separately to avoid mixing cleaning recovery changes and calibration offset changes into the same feature. After the anomaly feature group is generated, the anomaly feature generation module transmits the measurement offset features, response hysteresis features, state linkage features, and action feedback features to the anomaly pattern recognition module as input for anomaly pattern candidate matching.

[0095] An anomaly pattern recognition module 05, connected to the anomaly feature generation module, is used to perform anomaly pattern candidate matching based on the anomaly feature group and a preset anomaly pattern library to generate anomaly pattern candidate results. After receiving the anomaly feature group, the anomaly pattern recognition module reads the feature combination corresponding to the abnormal operating mode of the online water quality analyzer from the preset anomaly pattern library. The preset anomaly pattern library includes multiple abnormal operating modes such as flow path anomaly, leakage anomaly, measurement cell contamination, electrode adhesion, sensor drift, compensation anomaly, and communication anomaly. Each abnormal operating mode corresponds to one or more combinations of measurement offset features, response hysteresis features, state linkage features, and action feedback features.

[0096] The anomaly pattern recognition module first filters candidate anomaly patterns according to the measurement parameter type and data segment source, and then performs candidate matching between the feature combinations in the anomaly feature group and the preset anomaly pattern library. If the turbidity measurement offset feature and the change in measurement offset after cleaning occur together, the measurement cell contamination is included in the candidate anomaly pattern results; if the residual chlorine or pH response hysteresis feature and the change in response stabilization time after cleaning occur together, the electrode adhesion is included in the candidate anomaly pattern results; if the sample flow rate anomaly and the response hysteresis of multiple parameters occur within the same measurement cycle range, the flow path anomaly is included in the candidate anomaly pattern results. During the candidate matching process, the anomaly pattern recognition module retains the matching source of each candidate anomaly pattern, including feature type, corresponding measurement cycle, and data segment type.

[0097] If the feature combinations in the abnormal feature group match multiple operational abnormal modes, the abnormal pattern recognition module does not directly output the final operational abnormal mode. Instead, it generates multiple abnormal pattern candidate results and outputs them to the candidate result resolution module. If the abnormal feature group does not match any of the feature combinations in the preset abnormal pattern library, the abnormal pattern recognition module generates an unmatched state and retains the abnormal feature group that triggered the unmatched state. This unmatched state is processed by the candidate result resolution module in conjunction with the working mode segmentation results, or by the feedback update module updating the preset abnormal pattern library after the recovery data appears. The abnormal pattern candidate results include the candidate abnormal pattern type, the candidate basis, and the initial candidate sorting state, for the candidate result resolution module to continue processing.

[0098] The candidate result resolution module 06, connected to the abnormal pattern recognition module, is used to resolve conflicts in the candidate abnormal pattern results based on the working mode segmentation results, the state linkage features, and the action feedback features, generating an operational abnormal pattern recognition result. After receiving the candidate abnormal pattern results, the candidate result resolution module first reads the working mode segmentation results. If a candidate abnormal pattern is triggered only by changes in measured values ​​within the debugging segment, the candidate result resolution module marks the measurement cycle corresponding to that candidate abnormal pattern with data validity and restricts it from being output as the primary abnormal pattern. If the candidate abnormal pattern is triggered by action process data within the cleaning or calibration segment, the module prioritizes reading the recovery observation segment data after the cleaning or calibration segment ends, and uses the action feedback features to determine whether the candidate abnormal pattern has a recovery basis after the action.

[0099] The status linkage feature is used to distinguish multiple candidate anomaly patterns with similar performance. If both flow path anomaly and measurement cell contamination exist in the candidate anomaly pattern results, the candidate result resolution module reads the correspondence between sample flow rate, sample pressure, and response hysteresis of multiple measurement parameters; when sample flow rate anomaly and multi-parameter hysteresis occur in the same period, the confidence level of flow path anomaly is prioritized. If both leakage anomaly and communication anomaly exist in the candidate anomaly pattern results, the candidate result resolution module reads the leakage status, communication status, and measurement data missing status; when leakage status is triggered and accompanied by a sudden change in measurement value, leakage anomaly is included in the operational anomaly pattern identification results, while communication anomaly is retained as a candidate during measurement periods where data transmission is interrupted or instrument logs are missing.

[0100] Action feedback features are used to handle candidate conflicts between measurement cell contamination, electrode adhesion, and sensor drift. If the changes in measurement offset and response settling time after cleaning conform to the cleaning recovery baseline, the candidate result resolution module prioritizes measurement cell contamination or electrode adhesion. If the changes in measurement offset after calibration conform to the calibration offset baseline, sensor drift is prioritized. If no corresponding recovery changes are formed after cleaning and calibration, the candidate result resolution module reduces the ranking of contamination, adhesion, and drift candidate anomalies, while retaining flow path anomalies, compensation anomalies, or communication anomalies as subsequent identification targets. After the anomaly pattern recognition results are generated, the candidate result resolution module provides the result output module with the anomaly pattern type, corresponding measurement cycle, conflict resolution result, and data segment source.

[0101] The result output module 07, connected to the candidate result resolution module, is used to generate data validity markers and maintenance and handling strategies for the corresponding measurement period based on the operation anomaly pattern identification results. After receiving the operation anomaly pattern identification results, the result output module reads the anomaly pattern type, corresponding measurement period, and data segment source. For measurement periods in normal measurement segments that do not match an operation anomaly pattern, the result output module generates a valid marker; for measurement periods corresponding to debugging, cleaning, or calibration segments, it generates an invalid marker or a marker to be confirmed; for measurement periods in recovery observation segments, it generates a marker to be confirmed or a downweighted output marker based on the operation anomaly pattern identification results. The data validity markers are saved corresponding to the measurement periods and serve as the basis for subsequent recovery data updates and remote diagnostic calls.

[0102] The maintenance and handling strategy is formed based on the results of the abnormal operation pattern identification. For contamination of the measuring cell, the result output module generates a cleaning strategy and directs the cleaning target to the corresponding measuring cell; for electrode adhesion, an electrode cleaning strategy is generated; for sensor drift, a calibration strategy is generated; for flow path anomalies, leakage anomalies, or communication anomalies, remote diagnostics, manual maintenance prompts, or alarm level determination are generated. If there are a primary anomaly pattern and a secondary anomaly pattern in the abnormal operation pattern identification results, the result output module prioritizes outputting the maintenance and handling strategy corresponding to the primary anomaly pattern and retains the pending confirmation status corresponding to the secondary anomaly pattern for the feedback update module to process after the recovery data appears.

[0103] When a mismatch is found in the abnormal pattern recognition results, the result output module does not generate a specific cleaning or calibration strategy. Instead, it generates a remote diagnostic or manual maintenance prompt and retains the source measurement cycle of the abnormal feature group. The result output module transmits the data validity mark and maintenance handling strategy to the feedback update module. If the maintenance handling strategy requires local execution by the instrument, the result output module sends the corresponding control object and action type to the instrument control unit; if the maintenance handling strategy requires manual processing, the result output module transmits the abnormal pattern recognition results, the corresponding measurement cycle, and the data validity mark to the remote monitoring platform.

[0104] The feedback update module 08, connected to the result output module, is used to acquire the recovery data obtained after the maintenance and handling strategy is executed, and to update the preset abnormal mode library or the operation abnormal mode identification result based on the recovery data. After receiving the maintenance and handling strategy output by the result output module, the feedback update module waits for the execution feedback after the corresponding cleaning, calibration, remote diagnosis, or manual maintenance is completed, and reads the multi-parameter measurement data and operation status data within multiple measurement cycles after the execution is completed. The recovery data is formed using the measurement cycle association method of the status association module, and includes the changes in measurement values, changes in operation status, and the boundaries of the recovery observation segment after the maintenance and handling strategy is executed.

[0105] The recovered data is compared with the recovery conditions for the corresponding abnormal operating mode. If, after the cleaning strategy is executed, the changes in measurement offset and response settling time in the recovered data meet the recovery conditions corresponding to measurement cell contamination or electrode adhesion, the feedback update module generates a recovery confirmation result and writes the recovery confirmation result into the corresponding feature combination in the preset abnormal mode library. If, after the calibration strategy is executed, the changes in measurement offset in the recovered data meet the recovery conditions corresponding to sensor drift, the feedback update module writes the calibration recovery result into the preset abnormal mode library. If the recovered data does not meet the recovery conditions for the current abnormal operating mode, the feedback update module calls the abnormal mode candidate results retained by the candidate result resolution module, corrects the confidence ranking of the candidate abnormal modes, and updates the operating abnormal mode identification result based on the corrected ranking.

[0106] The feedback update module also uses the recovery confirmation results to update the data validity flags. Measurement cycles in the recovery observation segment are updated to valid after the recovery conditions are met; measurement cycles that do not meet the recovery conditions remain pending confirmation or are output with reduced weight. If recovery data is not generated after the maintenance and handling strategy is executed, the feedback update module retains the current operation anomaly pattern identification results and returns an execution feedback missing status to the result output module. This status does not directly change the preset anomaly pattern library, but continues to be used as a condition for obtaining recovery data in the next measurement cycle. Thus, the feedback update module provides the maintenance and handling results, recovery data, and updated operation anomaly pattern identification results of the current round to subsequent measurement cycles, enabling the system to maintain the connection between measurement cycle correlation, anomaly identification, and recovery update during continuous operation.

Claims

1. A method for identifying abnormal operating patterns of an online water quality analyzer, characterized in that, include: S100: Acquire multi-parameter measurement data, operating status data, action event data, and historical baseline data generated by the online water quality analyzer over multiple measurement cycles. The action event data includes one or more of cleaning action events, calibration action events, and debugging mode switching events. The multi-parameter measurement data, the operational status data, and the action event data are correlated over time according to the measurement cycle to generate operational correlation data. S200. Based on the working mode identifier and action event data in the operation association data, the working mode is segmented to generate a working mode segmentation result. The working mode segmentation result includes multiple segments such as normal measurement segment, cleaning segment, calibration segment, debugging segment and recovery observation segment. S300. Based on the segmentation results of the working mode and the historical baseline data, generate measurement offset features, response hysteresis features, state linkage features and action feedback features. The action feedback features are formed by the changes in multi-parameter measurement data before and after the cleaning segment or calibration segment. Combine the measurement offset features, the response hysteresis features, the state linkage features and the action feedback features into an abnormal feature group. S400. Based on the abnormal feature group and the preset abnormal pattern library, perform abnormal pattern candidate matching to generate abnormal pattern candidate results; based on the working mode segmentation results, the state linkage features and the action feedback features, resolve conflicts in the abnormal pattern candidate results to generate operation abnormal pattern identification results. S500: Based on the abnormal operation pattern identification results, generate data validity markings and maintenance and handling strategies for the corresponding measurement period; Obtain the recovery data obtained after the maintenance and handling strategy is executed, and update the preset abnormal mode library or the operation abnormal mode identification result based on the recovery data.

2. The method according to claim 1, characterized in that, The multi-parameter measurement data includes multiple values ​​of residual chlorine or chlorine dioxide, turbidity, pH, conductivity, and temperature; the operating status data includes one or more of sample flow rate, leakage status, sample pressure, operating temperature, communication status, alarm log, and instrument log; the historical baseline data includes one or more of normal measurement baseline, measurement fluctuation range, response stabilization time, cleaning recovery baseline, and calibration offset baseline.

3. The method according to claim 1, characterized in that, The step of performing time correlation on the multi-parameter measurement data, the operating status data, and the action event data according to the measurement cycle to generate operating correlation data includes: writing multi-parameter measurement data and operating status data within the same measurement cycle into the same operating correlation record based on the measurement cycle identifier; and writing corresponding action event data into the operating correlation record matching the trigger time based on the trigger time of cleaning action events, calibration action events, or debugging mode switching events.

4. The method according to claim 1, characterized in that, The segmentation of working modes based on the working mode identifier and action event data in the operational association data includes: dividing the operational association data with the working mode identifier as normal measurement into a normal measurement segment; dividing the continuous measurement cycle formed after the cleaning action event is triggered into a cleaning segment; dividing the continuous measurement cycle formed after the calibration action event is triggered into a calibration segment; dividing the continuous measurement cycle corresponding to the debugging mode switching event into a debugging segment; and dividing the continuous measurement cycle after the cleaning segment, calibration segment, or debugging segment ends into a recovery observation segment.

5. The method according to claim 1, characterized in that, The generation of measurement offset features, response hysteresis features, state linkage features, and action feedback features based on the segmentation results of the working mode and the historical baseline data includes: generating measurement offset features based on the difference between multi-parameter measurement data in the normal measurement segment and the historical baseline data; generating response hysteresis features based on the number of measurement cycles that the measured value takes to enter the stable range; generating state linkage features based on the correspondence between the operating status data and the multi-parameter measurement data within the same measurement cycle; and generating action feedback features based on the changes in measurement offset and response stabilization time before and after the cleaning or calibration segment.

6. The method according to claim 1, characterized in that, The preset abnormal mode library includes multiple types of abnormal flow path, leakage, measurement cell contamination, electrode adhesion, sensor drift, compensation abnormality, and communication abnormality. In the preset abnormal mode library, each abnormal operation mode corresponds to one or more combinations of features such as measurement offset feature, response hysteresis feature, status linkage feature, and action feedback feature.

7. The method according to claim 1, characterized in that, The conflict resolution of the candidate abnormal modes based on the working mode segmentation results, the state linkage features, and the action feedback features includes: when there are multiple candidate abnormal modes among the candidate abnormal modes, marking the candidate abnormal modes corresponding to the debugging segment, cleaning segment, and calibration segment with data validity based on the working mode segmentation results; and ranking the multiple candidate abnormal modes by confidence based on sample flow rate, leakage status, change in measurement offset after cleaning, and change in measurement offset after calibration, thereby generating conflict resolution results.

8. The method according to claim 1, characterized in that, The step of generating data validity marking and maintenance handling strategies for the corresponding measurement period based on the operation anomaly pattern identification results includes: marking the data of the corresponding measurement period as valid when the operation anomaly pattern identification result corresponds to a normal measurement segment and does not match an operation anomaly pattern; marking the data of the corresponding measurement period as invalid or pending confirmation when the operation anomaly pattern identification result corresponds to a debugging segment, cleaning segment, or calibration segment; marking the data of the corresponding measurement period as pending confirmation or downweighted output when the operation anomaly pattern identification result corresponds to a recovery observation segment; and generating one or more maintenance handling strategies based on the operation anomaly pattern identification results, such as cleaning, calibration, remote diagnosis, manual maintenance prompts, or alarm level determination.

9. The method according to claim 1, characterized in that, The step of acquiring recovery data obtained after the execution of the maintenance and handling strategy, and updating the preset abnormal mode library or the operation abnormal mode identification result based on the recovery data, includes: acquiring multi-parameter measurement data and operation status data within multiple measurement cycles after cleaning, calibration, or maintenance is completed, and generating recovery data; when the recovery data meets the recovery conditions of the corresponding operation abnormal mode, generating a recovery confirmation result and writing the recovery confirmation result into the preset abnormal mode library; when the recovery data does not meet the recovery conditions of the corresponding operation abnormal mode, correcting the confidence ranking of the abnormal mode candidate results, and updating the operation abnormal mode identification result based on the corrected confidence ranking.

10. An operational anomaly pattern recognition system for an online water quality analyzer, applied to the method described in any one of claims 1 to 9, characterized in that, include: The data access module is used to acquire multi-parameter measurement data, operating status data, action event data and historical baseline data generated by the online water quality analyzer in multiple measurement cycles; The status association module, connected to the data access module, is used to perform time association on the multi-parameter measurement data, the running status data and the action event data according to the measurement cycle, and generate running association data. The working mode segmentation module, connected to the state association module, is used to segment the working mode based on the working mode identifier and action event data in the running association data, and generate the working mode segmentation result. An anomaly feature generation module, connected to the working mode segmentation module, is used to generate measurement offset features, response hysteresis features, state linkage features, and action feedback features based on the working mode segmentation results and the historical baseline data, and combine them into anomaly feature groups. An abnormal pattern recognition module, connected to the abnormal feature generation module, is used to perform abnormal pattern candidate matching based on the abnormal feature group and a preset abnormal pattern library to generate abnormal pattern candidate results. The candidate result resolution module is connected to the abnormal pattern recognition module and is used to resolve conflicts of the abnormal pattern candidate results based on the working mode segmentation results, the state linkage features and the action feedback features, and generate the running abnormal pattern recognition results. The result output module is connected to the candidate result resolution module and is used to generate data validity markers and maintenance and handling strategies for the corresponding measurement period based on the operation anomaly pattern recognition results. The feedback update module, connected to the result output module, is used to obtain the recovery data obtained after the maintenance and handling strategy is executed, and to update the preset abnormal mode library or the operation abnormal mode identification result based on the recovery data.

Images