A store index early warning method based on time sequence anomaly detection

Abstract

The application discloses a store index early warning method based on time sequence anomaly detection, and relates to the technical field of data processing.The application identifies effective state events according to store state event data, determines a stage starting event and a current business life cycle stage in a conflict resolution time unit, converts store business index time sequence data of a target store in the current business life cycle stage into a business age index sequence, and generates a stage baseline interval with a similar store in the same stage segment.The store index comparison object is changed from a historical sequence before a state change to a same-stage baseline in the current business life cycle stage, the business age position, the stage baseline interval and the index deviation state corresponding to the same stage coordinate are used as the basis for early warning triggering, the probability that stage fluctuations during opening, reopening, reconstruction and business change are directly judged as anomalies is reduced, and index sequences continuously deviating from the same-stage baseline in the current stage can form a stage anomaly score and store index early warning information.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for early warning of store indicators based on time-series anomaly detection. Background Technology

[0002] With the continuous collection of operational data from chain stores, key performance indicators such as sales volume, order volume, customer traffic, new members, platform exposure, platform visits, sales volume, and reviews are typically recorded in daily or hourly time-series data. Existing store indicator early warning methods mostly rely on fixed thresholds, year-on-year and month-on-month comparisons, historical averages, dynamic thresholds, or operational risk scores. That is, an early warning baseline is determined based on the target store's historical indicator sequence, historical data from the same period, recent fluctuation range, or preset scoring rules, and an early warning message is generated when the current indicator value or risk score meets the early warning conditions.

[0003] In scenarios where the operating status of a store changes, such as opening a new store, reopening a store, renovating a store, or changing the management of a store, the target store may lack stable and usable historical same-store data, or the existing historical indicator series may correspond to the operating status before the change. In this case, continuing to use historical averages, year-on-year and month-on-month comparisons, or dynamic thresholds over natural time to generate an early warning baseline may easily lead to a mismatch between the early warning baseline and the current operating status of the target store.

[0004] The aforementioned baseline mismatch can cause periodic fluctuations during the opening ramp-up, reopening recovery, renovation recovery, or operational transition observation periods to be identified as abnormal fluctuations, leading to false alarm triggers and frequent changes in alarm levels. When a target store continuously deviates from the normal trajectory during the current operational phase, alarm triggering may be delayed because the outcome indicators have not yet exceeded fixed or historical thresholds. Existing methods typically rely on manually adjusting thresholds, adding store tags, or grouping comparisons by region or store type, which still makes it difficult to determine the time-series alarm baseline corresponding to the current operational phase after changes in store operating status.

[0005] Therefore, how to match the early warning baseline of store indicators with the current stage of the target store's business life cycle after changes in the store's operating status has become a technical problem that needs to be solved. Summary of the Invention

[0006] This application provides a method for early warning of store indicators based on time-series anomaly detection, which solves the problem that the early warning baseline of store indicators does not match the current life cycle stage after the change of business status.

[0007] This invention provides a method for early warning of store metrics based on time-series anomaly detection, comprising: Obtain time-series data of store operation indicators, store attribute data, and store status event data for the target store; Based on the event source, event status, event coverage, event type, event occurrence time, and event effective time in the store status event data, valid status events are identified. From the valid status events, the phase start event is determined. When multiple valid status events exist within the same conflict resolution time unit, the phase start event is determined according to the event priority. The current business life cycle stage is determined based on the phase start event. Using the start time of the stage-starting event as the starting point of the operating age, the time-series data of the store's operating indicators in the current operating life cycle stage of the target store are converted into a series of operating age indicators. The stage start time is taken as the event effective time when the event effective time exists, and as the event occurrence time when the event effective time is missing. Based on the store attribute data and the current business life cycle stage, similar stores in the same stage are identified in the historical store sample library, and a stage baseline interval and a stage baseline source record are generated based on the similar stores in the same stage. The operating age indicator sequence is time-aligned with the stage baseline interval according to the operating age position, and a stage anomaly score is generated based on the deviation status of the aligned indicators. Determine whether the abnormal score of the stage meets the early warning triggering rules. If the early warning triggering rules are met, generate and output the store indicator early warning information.

[0008] In some embodiments, the time-series data of store operation indicators includes at least one of the following operation indicator values: sales revenue, order volume, customer traffic, new member volume, member repurchase volume, platform exposure, platform visits, sales volume, positive reviews, and negative reviews. After obtaining the time-series data of the target store's operating indicators, the different operating indicator values ​​are time-aligned according to the preset time granularity, duplicate records are deduplicated, and missing markers or filler values ​​are generated based on the record status of adjacent operating age positions within the same operating life cycle stage.

[0009] In some embodiments, the time-series data of store operating indicators of the target store during the current stage of its operating life cycle are converted into a series of operating age indicators, including: Convert each point in time within the current business lifecycle stage into its business age position relative to the starting point of the business age; The operating indicator values ​​are arranged according to the operating age position described above; Write the business life cycle stage identifier, business age position, and time granularity identifier into each business indicator value; When a new valid status event occurs at the target store, a new sequence of operating age indicators is generated based on the new stage start event.

[0010] In some embodiments, identifying similar stores from the same period in a historical store sample database includes: Generate store matching tags based on at least two of the following: the target store's region, store type, business entity attributes, business district attributes, business status, and platform operation status; Candidate stores with corresponding store matching tags and corresponding business life cycle stages are selected from the historical store sample database; Extract time-series segments from the historical operating indicator time-series data of the candidate stores that correspond to the operating age range of the target stores; Time-series segments that meet the segment integrity criteria are identified as segments from the same stage in similar stores.

[0011] In some embodiments, the fragment integrity condition includes: There is a corresponding phase start event, there is a corresponding business life cycle phase identifier, the business indicator records cover the target store's operating age range, and the missing business indicator ratio meets the preset missing ratio condition. After identifying similar stores in the same stage, record the source store identifier, stage start event identifier, business life cycle stage identifier, business age range, and business indicator coverage status for each similar store in the same stage to form a stage baseline source record.

[0012] In some embodiments, generating a stage baseline interval based on the similar stores' same stage segments includes: Based on the operational age and location, the segments of the same stage of multiple similar stores are aligned, and the segments of the same stage of similar stores that meet the segment integrity condition are aggregated. Extract the distribution status, direction of change, and continuous change status of operating indicators at each operating age position; Generate an indicator baseline range based on the distribution status of the operating indicators, and generate a stage change baseline based on the direction of indicator change and continuous change status; The baseline range of the indicator and the baseline of the stage change are combined into a stage baseline interval, and the stage baseline interval is associated with and saved with the stage baseline source record.

[0013] In some embodiments, the stage baseline interval is generated by baseline generation rules or a time-series baseline model; The baseline generation rule is a processing rule that aggregates similar stores in the same stage according to their operating age and location, and outputs the indicator baseline range and stage change baseline. The time-series baseline model takes similar stores in the same stage segment and the target store's operating age range as inputs, and outputs the indicator baseline range and stage change baseline. The training data sources for the time-series baseline model include historical time-series data of store operating indicators, store attribute data, and store status event data. The annotation methods include generating operating life cycle stage identifiers and operating age position identifiers based on the stage start event. The training objectives include fitting the indicator distribution status, indicator change direction, and continuous change status corresponding to the operating age position. After receiving the warning processing results and review markers associated with the store indicator warning information, the sample status in the historical store sample library is updated based on the warning processing results and review markers. The updated sample status will take effect in the next stage baseline version.

[0014] In some embodiments, generating a stage anomaly score based on the aligned metric deviation state includes: Based on the operating age position, the operating indicator values ​​of the target store are compared with the indicator baseline range in the stage baseline interval to obtain the stage deviation status of the single indicator. The continuous deviation status is obtained by determining the single-indicator stage deviation status at the continuous operating age position; The correlation deviation status is obtained based on the sequential changes of multiple operating indicators within the same operating age range; A stage anomaly score is generated based on the single-indicator stage deviation state, continuous deviation state, and associated deviation state.

[0015] In some embodiments, determining whether the stage anomaly score meets the early warning triggering rules includes: When the abnormal score corresponding to the sales volume, order volume or customer traffic reaches the trigger condition of the result indicator, a result deviation warning mark is generated. When the abnormal scores corresponding to the platform exposure, platform visits, new members, core sales, positive reviews or negative reviews reach the trigger conditions of the preceding indicators, but the result indicators do not reach the trigger conditions of the result indicators, a potential transmission warning mark is generated. The warning type is determined based on the deviation warning marker or the potential transmission warning marker.

[0016] In some embodiments, generating and outputting store metric alert information includes: The recipients of the early warning are determined based on the organizational affiliation of the target stores; The warning level is determined based on the stage anomaly score, indicator deviation status, and warning type; Generate store indicator warning information that includes the target store identifier, current operating life cycle stage, stage anomaly score, warning type, warning level, and stage baseline source record identifier; After receiving the warning processing result, the warning processing result, processing time, and review mark are associated with and saved with the corresponding store indicator warning information.

[0017] Through the above technical solution, the present invention can achieve at least the following beneficial effects: This invention identifies valid status events based on store status event data, determines the phase start event and the current operational lifecycle phase within the conflict resolution time unit, converts the time-series data of the target store's operational indicators within the current operational lifecycle phase into an operational age indicator sequence, and then generates a phase baseline interval using similar stores at the same phase. Thus, the comparison object for store indicators changes from historical sequences before status changes to the baseline of the same phase under the current operational lifecycle phase. The operational age position, phase baseline interval, and indicator deviation from the status corresponding to the same phase coordinate are used for early warning triggering, reducing the probability that phase fluctuations during opening, reopening, renovation, and relocation are directly judged as abnormal, and enabling indicator sequences that continuously deviate from the same phase baseline within the current phase to generate phase abnormality scores and store indicator early warning information.

[0018] By performing time alignment, deduplication, and missing marker or filler value generation on the time series data of store operation indicators, different operation indicators can fall into the same time granularity and operation age position, reducing misjudgment of indicator deviation status caused by duplicate records, time misalignment, or unclear missing status.

[0019] By matching store tags, corresponding business lifecycle stage records, and segment integrity conditions, similar stores at the same stage are identified. This ensures that the stage baseline range is derived from historical segments corresponding to the target store's attributes and business stage, reducing baseline offset caused by generating early warning baselines from historical data before changes in usage status.

[0020] By recording the source of the stage baseline and associating the stage baseline interval with the source store identifier, the stage start event identifier, the operating age range, and the operating indicator coverage status, the store indicator warning information can correspond to the stage sample range and baseline version used when it was generated.

[0021] By generating stage anomaly scores through single-indicator stage deviation status, continuous deviation status, and associated deviation status, early warning triggers can simultaneously reflect single-point out-of-bounds, continuous deviation, and the transmission status of preceding indicators. When the preceding indicator meets the preceding indicator triggering condition but the result indicator does not meet the result indicator triggering condition, a potential transmission-type early warning mark can be generated. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0023] Figure 1This is a flowchart of the store indicator early warning method based on time-series anomaly detection in the embodiment. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0026] To facilitate understanding, the relevant terms and concepts involved in the embodiments of this application will be introduced below: The store operation indicator time-series data is a collection of operation indicator records arranged according to the collection time. Each operation indicator record includes a store identifier, operation indicator identifier, operation indicator value, collection time, and time granularity identifier. Store attribute data is a data set used to describe the basic operating conditions of the target store, including at least one of the following: region, store type, operating entity attribute, business district attribute, operating status, and platform operating status. The historical store sample library is a data set used to store time-series fragments of operation indicators, store attribute data, store status event data, sample status, and sample version identifiers for historical stores at different stages of their operational lifecycle.

[0027] A valid status event is a store status event that, after verification of its source, status, coverage, type, occurrence time, and effective time, can change the store's operational status. The current operational lifecycle stage is the operational status interval of the target store within the current processing cycle, determined by valid status events, including at least one of the following: new opening stage, reopening recovery stage, renovation recovery stage, operational changeover observation stage, and stable operation stage. The current processing cycle is the time interval for executing this store metric early warning calculation. The stage start event is a valid status event used to determine the starting point of the current operational lifecycle stage.

[0028] The phase start time is the time value corresponding to the phase start event. If the event's effective time exists, it is used; otherwise, the event's occurrence time is used. The operating age position is the time position within the phase formed by the indicator collection time relative to the phase start time. The operating age range is a set of time positions within the phase formed by multiple operating age positions. The target store's operating age range is the operating age range of the target store participating in the store indicator early warning calculation within the current processing cycle.

[0029] The phase start event identifier is the event identifier corresponding to the phase start event. The operating age starting point is generated according to the phase start time. The phase end time of the current operating life cycle phase is determined by the phase start time of the next valid state event; if there is no next valid state event, the end time of the current processing cycle is used as the phase end time of the current operating life cycle phase.

[0030] The operating age indicator sequence is a collection of store operating indicator records arranged according to their operating age position. Each operating indicator record includes a store identifier, operating indicator identifier, operating indicator value, collection time, operating age position, current operating life cycle stage, and time granularity identifier. Similar store segments of the same stage are time-series segments of operating indicators from the historical store sample library that have corresponding store matching tags to the target store and are in the corresponding operating life cycle stage. The stage baseline interval is a data range composed of the indicator baseline range formed by similar store segments of the same stage at each operating age position and the stage change baseline. The sample set of operating age segments of the same stage is a set of sample values ​​formed by similar store segments of the same stage under the same operating life cycle stage, the same operating age position, and the same operating indicator identifier.

[0031] A phase baseline version is a set of phase baseline interval records based on the same phase rule version, sample status, and baseline generation time. The phase baseline source record is the data record used to record the sample source on which the phase baseline interval is based. The phase anomaly score is a record of the deviation of the target store from the phase baseline interval at its corresponding operating age position. The warning trigger rule is a judgment rule used to convert phase anomaly scores and indicator deviation status into store indicator warning information. The conflict resolution time unit is a preset time interval used to handle phase start conflicts between multiple valid state events.

[0032] The phase rule version is a rule record used to bind event type, trusted source set, event priority, phase mapping rules, and preset phase change thresholds. The phase mapping rules include preset phase mapping functions, which are used to obtain the post-event business lifecycle stage based on the event type and the business lifecycle stage before the event.

[0033] The indicator baseline range is the data range within the stage baseline interval used to define the normal value boundary of a single operating indicator at the corresponding operating age position. It includes the lower limit of the indicator baseline, the upper limit of the indicator baseline, the operating indicator identifier, the operating age position, and the stage baseline version identifier. The stage change baseline is the data record within the stage baseline interval used to record the direction and continuous change status of the same operating indicator at consecutive operating age positions. It includes the operating indicator identifier, operating age range, direction of change, continuous change status, and allowable range. The sample status is a status field in the historical store sample database used to distinguish whether a sample participated in the generation of the stage baseline interval. It includes normal samples, abnormal samples, samples awaiting review, and excluded samples. The operating indicator distribution status is the central tendency, fluctuation range, and quantile distribution formed by the sample set of operating ages within the same stage at the corresponding operating age position. The indicator change direction is the upward, downward, or stable direction of the same operating indicator between adjacent operating age positions. The continuous change status is the maintenance of the direction of change of the same operating indicator at consecutive operating age positions.

[0034] Example 1: like Figure 1 As shown, this embodiment employs a store indicator early warning method based on time-series anomaly detection, including: Step S1: Obtain time-series data of store operation indicators, store attribute data, and store status event data of the target store. The store status event data includes event identifier, store identifier, event type, event occurrence time, event effective time, event source, event status, and event coverage. Step S2: Establish a store status event sequence according to the event effective time. If the event effective time is missing, arrange them according to the event occurrence time. Step S3: Identify valid status events based on the event source, event status, event coverage, event type, event occurrence time, and event effective time in the store status event data; determine the phase start event from the valid status events; and when there are multiple valid status events in the same conflict resolution time unit, determine the phase start event according to the event priority; and determine the current business life cycle stage based on the phase start event. Step S4: Take the start time of the phase start event as the starting point of the operating age. If the effective time of the event exists, take the effective time of the event. If the effective time of the event is missing, take the occurrence time of the event. Convert the time series data of the store operating indicators of the target store in the current operating life cycle stage into a series of operating age indicators. Step S5: Based on store attribute data and the current business lifecycle stage, identify similar stores in the same stage from the historical store sample library, and generate stage baseline intervals and stage baseline source records based on similar stores in the same stage. Step S6: Align the operating age indicator sequence with the stage baseline interval according to the operating age position, and generate stage anomaly scores based on the deviation status of the aligned indicators. Step S7: Determine whether the stage abnormal score meets the early warning triggering rules. If the early warning triggering rules are met, generate and output store indicator early warning information based on the stage abnormal score and indicator deviation status.

[0035] In one implementation, the event type includes at least one of the following: opening event, reopening event, renovation event, change of operation event, and stable operation confirmation event. When multiple valid status events exist within the same conflict resolution time unit, the starting event for the phase is determined according to the event priority recorded in the phase rule version; when multiple records exist for the same event type, the records to be retained are determined according to the preset sorting of the event source in the corresponding trusted source set and the event effective time, and the merged records are saved as merged source records.

[0036] Specifically, the preset set of operating status events, event priorities, and stage mapping rules are all saved with the stage rule version. After the stage rule version is updated, the already generated operating age indicator sequences and store indicator warning information continue to be associated with the stage rule version used when they were generated.

[0037] The event source is the business system or manual review source identifier that generated the store status event. Event status records whether the store status event is currently effective, including effective, revoked, and pending review. Event coverage records the set of store objects corresponding to the store status event. The preset operational status event set is a set of event types that can trigger operational lifecycle stage judgments; its values ​​are stored in association with the stage rule version. Event priority is a sorting field used to handle stage start conflicts between multiple valid status events, and it remains fixed within the same stage rule version.

[0038] In one implementation, the time-series data of store operating indicators includes at least one of the following operating indicator values: sales revenue, order volume, customer traffic, new member volume, member repurchase volume, platform exposure, platform visits, sales volume, positive reviews, and negative reviews. After obtaining the time-series data of the target store's operating indicators, the different operating indicator values ​​are time-aligned according to a preset time granularity, duplicate records are deduplicated, and missing markers or filler values ​​are generated based on the record status of adjacent operating age positions within the same operating life cycle stage.

[0039] The preset time granularity includes daily or hourly time granularity. Data generation status refers to the data generation conditions corresponding to the operational indicators within the collection period, including store operation status, platform listing status, indicator collection status, and data reporting status. Missing data flags include missing data for stores not operating, missing data for platforms not generating data, missing data for data collection anomalies, and missing data for data reporting delays. Imputation values ​​are generated based on operational indicator records at adjacent operational age positions within the same current operational lifecycle stage, or operational indicator records at corresponding operational age positions in the same segment of similar stores, and are associated with and stored with the imputation source identifier. The imputation source identifier records that the imputation value originates from operational indicator records at adjacent operational age positions or operational indicator records in the same segment of similar stores.

[0040] In one implementation, the time-series data of store operation indicators of the target store in the current operation life cycle stage is converted into an operation age indicator sequence, including: converting each time point in the current operation life cycle stage into an operation age position relative to the starting point of the operation age; arranging each operation indicator value according to the operation age position; writing an operation life cycle stage identifier, an operation age position and a time granularity identifier into each operation indicator value; and generating a new operation age indicator sequence based on the new stage starting event when a new valid state event occurs in the target store.

[0041] The unit for the operational age position is consistent with the preset time granularity. The operational age indicator sequence uses store identifier, current operational life cycle stage, operational age position, and operational indicator identifier as search fields. After the target store enters a new current operational life cycle stage, the new stage start event is used as the new operational age starting point, and the new operational age indicator sequence is saved separately from the operational age indicator sequence of the previous current operational life cycle stage.

[0042] In one implementation, identifying similar store segments of the same stage in a historical store sample database includes: generating store matching tags based on at least two of the target store's region, store type, operating entity attributes, business district attributes, business status, and platform operation status; filtering candidate stores from the historical store sample database that have corresponding store matching tags and corresponding business lifecycle stage records; extracting time-series segments from the historical business indicator time-series data of the candidate stores that correspond to the target store's business age range; and identifying time-series segments that meet the segment integrity condition as similar store segments of the same stage.

[0043] Store matching tags include at least two of the following: region tags, store type tags, business entity tags, business district tags, operating status tags, and platform operation status tags. The corresponding business lifecycle stage is a historical operating status range with the same stage type as the target store's current business lifecycle stage. Candidate stores are ranked according to the number of matching tags, the completeness of records for the corresponding business lifecycle stage, and the coverage of operating indicators. When the number of similar store segments at the same stage formed by candidate stores is lower than the minimum sample size requirement, the candidate range is expanded according to the hierarchical relationship of region tags, business district tags, store type tags, and business entity tags.

[0044] The number of matching tags for a store is the number of tag items where the candidate store and the target store have the same tag value. The corresponding business lifecycle stage record must include at least the business lifecycle stage identifier, the stage start event identifier, the stage start time, and the stage end time. The minimum sample size requirement is an implementation parameter used within the stage baseline version to determine whether the number of similar stores in the same stage meets the baseline generation requirements. Its statistical scope includes the number of segments that can participate in aggregation within the same business lifecycle stage, the same business indicator identifier, and the same business age range.

[0045] In one implementation, the fragment integrity conditions include: the existence of a corresponding stage start event, the existence of a corresponding business lifecycle stage identifier, business indicator records covering the target store's operating age range, and the business indicator missing ratio meeting a preset missing ratio condition. After determining similar stores in the same stage, the source store identifier, stage start event identifier, business lifecycle stage identifier, operating age range, and business indicator coverage status are recorded for each similar store's segment in the same stage to form the stage baseline source record. The preset missing ratio condition is an implementation parameter saved within the stage baseline version, and the business indicator missing ratio is determined according to the ratio of the number of missing business indicator records within the operating age range to the number of required business indicator records.

[0046] The baseline source record for each phase includes the source store identifier, phase start event identifier, business lifecycle phase identifier, business age range, business metric coverage status, sample version identifier, and baseline generation time. Business metric coverage status records the completeness of each business metric within the business age range for similar stores in the same phase segment. The sample version identifier distinguishes sample sets formed at different update times from the historical store sample library.

[0047] The source records of the stage baseline are stored in a one-to-one or many-to-one relationship with the stage baseline intervals. When multiple similar stores generate the same stage baseline interval for the same stage segment, the source records of the stage baseline are summarized and indexed according to the stage baseline version identifier and the operating age range. The stage baseline version identifier is a record identifier used to distinguish stage baseline intervals formed under different sample states, different stage rule versions, or different baseline generation times.

[0048] In one implementation, generating a stage anomaly score based on the aligned indicator deviation status includes: comparing the operating indicator values ​​of the target store with the indicator baseline range in the stage baseline interval according to the operating age position to obtain the single indicator stage deviation status; obtaining the continuous deviation status based on the single indicator stage deviation status at the continuous operating age position; obtaining the associated deviation status based on the sequential changes of multiple operating indicators within the same operating age range; and generating a stage anomaly score based on the single indicator stage deviation status, the continuous deviation status, and the associated deviation status.

[0049] Indicator deviation status includes single-indicator stage deviation status, continuous deviation status, and correlated deviation status. Single-indicator stage deviation status records the positional relationship of a single operating indicator value relative to the indicator baseline range. Continuous deviation status records the persistence of deviation of the same operating indicator across consecutive operating age positions. Correlated deviation status records the sequential changes of multiple operating indicators within the same operating age range. Stage anomaly score records include store identifier, operating indicator identifier, operating age range, single-indicator stage deviation status, continuous deviation status, correlated deviation status, and score generation time.

[0050] Single-indicator stage deviation states include below the lower limit of the indicator baseline, within the range of the indicator baseline, and above the upper limit of the indicator baseline. Continuous deviation states are generated based on the number of times the same operating indicator deviates in the same direction at each consecutive operating age position. Related deviation states are generated according to the order of deviation of the preceding and resulting indicators within the same operating age range. Stage anomaly scores range from 0 to 100, where 0 indicates no stage anomaly record is formed, and higher values ​​indicate a greater degree of deviation from the stage baseline range. When generating stage anomaly scores, single-indicator stage deviation states, continuous deviation states, and related deviation states are converted into corresponding scores and synthesized according to their respective score weights; the score weights are saved as implementation parameters along with the stage baseline version.

[0051] In one implementation, determining whether a stage abnormal score meets the early warning triggering rules includes: generating a result deviation early warning flag when the stage abnormal score corresponding to sales volume, order volume, or customer traffic reaches the result indicator triggering condition; generating a potential transmission early warning flag when the stage abnormal score corresponding to platform exposure, platform visits, new members, core sales, positive reviews, or negative reviews reaches the pre-indicator triggering condition, but the result indicator does not reach the result indicator triggering condition; and determining the early warning type based on the result deviation early warning flag or the potential transmission early warning flag.

[0052] Outcome metrics are operational indicators that directly reflect store performance, including sales revenue, order volume, and customer traffic. Leading metrics are operational indicators that change before outcome metrics and are used to form a judgment, including platform exposure, platform visits, new members, verified sales volume, positive reviews, and negative reviews. Warning types include warnings for insufficient new store ramp-up, delayed recovery from renovations, abnormal operational observations, abnormal platform conversion rates, and abnormal member recovery. Warning types are determined according to the current stage of the business lifecycle, the category of operational indicator that deviated, and the state of the deviation.

[0053] Both the outcome indicator trigger conditions and the prerequisite indicator trigger conditions include a trigger score threshold, a triggering operating age range, and a number of persistent deviation positions. The trigger score threshold is an implementation parameter saved within the stage rule version, determined based on historical sample quantiles or validation sets. When determining the warning trigger rule, the abnormal score of the stage is first confirmed according to the operating age position, and then the warning flag is determined by combining it with the number of persistent deviation positions; when multiple warning flags are formed simultaneously within the same operating age range, the store indicator warning information is generated according to the warning flag with the higher warning level.

[0054] In one implementation, generating and outputting store indicator early warning information includes: determining the early warning recipients based on the organizational affiliation of the target stores; The warning level is determined based on the stage abnormal score, indicator deviation status, and warning type; store indicator warning information is generated, which includes the target store identifier, current business life cycle stage, stage abnormal score, warning type, warning level, and stage baseline source record identifier; after receiving the warning processing result, the warning processing result, processing time, and review mark are associated and saved with the corresponding store indicator warning information.

[0055] Organizational affiliation records include the affiliation records between the target store and headquarters, regional organizations, branches, supervisors, and operating entities. Store indicator early warning information includes the target store identifier, organizational affiliation identifier, current operating lifecycle stage, stage start event identifier, early warning type, early warning level, stage anomaly score, indicator deviation status, stage baseline source record identifier, generation time, and recipient identifier.

[0056] The early warning processing result is a record of the response taken by the early warning recipient to the store indicator early warning information, including confirmed, processed, delayed, and no further action required. The review tag is used to record whether the early warning processing result participates in the sample status update process of the historical store sample library. After the early warning processing result and review tag are associated and saved with the store indicator early warning information, the corresponding sample status is updated according to the review tag; the sample status update result takes effect when the next phase baseline version is generated.

[0057] In a preferred embodiment of Example 1, when identifying valid status events that change the store's operating status in the store status event sequence, the event source, event effective time, event coverage, and event status are verified for each store status event in the store status event sequence.

[0058] Each store status event is recorded as The event identifier, event type, event source, event occurrence time, event effective time, event coverage, event status, and event priority are written according to the event occurrence time. The event type includes at least one of the following: opening event, reopening event, renovation event, change of operation event, and stable operation confirmation event. The renovation event is used to represent a change of operation status between the start and completion of renovation. The change of operation event is used to represent a change of operation status among the following: change of operating entity, change of main product category, and change of platform operation status.

[0059] Store status events The method for determining basic validity is as follows: , in, Indicates store status events The basic effective marker; This function represents a condition that takes the value 1 when the condition within the parentheses is true and 0 when the condition is false. Indicates store status events The source of the incident; Indicates event type The corresponding set of trusted sources; Indicates store status events Event types; Indicates store status events The event's effective time; when the event's effective time is missing. Retrieve store status events The time when the event occurred; Indicates store status events The time when the event occurred; Indicates event type The corresponding forward allowance for the event occurrence time; Indicates event type The corresponding allowable duration after the event occurs; Indicates store status events The set of events covering the scope of events; This represents the collection of store objects corresponding to the target store. Indicates store status events The event status flag, when the value is 0, indicates that the store status event is in effect.

[0060] exist When the value is 1, the store status event will be... Input the phase mapping rules to determine whether the store status event causes a change in the target store's operational lifecycle phase.

[0061] The phase mapping rules generate the post-event operational lifecycle phase based on the event type and the operational lifecycle phase before the event. The intensity of change is determined as follows: , in, Indicates store status events The intensity of phase changes; This indicates the preset stage mapping function; Indicates by event type and the business life cycle stage before the event occurred The operational lifecycle stage following the occurrence of the mapped event; Indicates store status events The stage of the target store's business lifecycle prior to the incident; Indicates event type The corresponding event priority value indicates that the higher the priority value, the greater the degree of control that event type has over the transition of business life cycle stages.

[0062] when When the change threshold is greater than or equal to the preset threshold, the store status event will be... As a candidate valid state event; when When the number of events is less than the preset stage change threshold, the store status event is retained as a normal status record and not as a stage start event for the business lifecycle stage. The preset stage change threshold is calibrated offline based on the event priority distribution of confirmed stage start events in the historical store sample database. Its value lies between the lowest and highest event priorities that can trigger stage switching, and remains consistent within the same store type, the same business lifecycle stage, and the same stage rule version. The stage rule version is used to bind the trusted source set, event priority, stage mapping rules, and preset stage change threshold. If the number of confirmed stage start events used for calibration is less than the preset number requirement, the preset stage change threshold of the previous frozen stage rule version is used. If there is no previous frozen stage rule version, only store status events whose event priority reaches the highest configured level, passes the basic validity judgment, and is determined by the preset stage mapping function to cause a change in the business lifecycle stage are considered as candidate valid status events, and the remaining store status events are retained as normal status records. The preset number requirement, the highest configured level, and the preset maximum merging duration are all saved with the stage rule version.

[0063] For similar store status events that are repeatedly reported within a short period of time, they are merged according to event type, event coverage, and event effective time.

[0064] For any benchmark store status event that has passed the basic validity determination Its set of recurring events is: , in, Indicates store status events A set of recurring events formed based on a benchmark; This represents the baseline store status event used to determine short-term recurring events; Indicates store status events Event types; Indicates store status events The set of events covering the scope of events; Indicates store status events The effective time of the event; Indicates event type The corresponding time window for merging recurring events.

[0065] For a set of repeated events During merging, the store status event with the highest priority is retained; when event priorities are the same, the store status event with the higher preset ranking in the corresponding trusted source set is retained; when the preset ranking of event sources is the same, the store status event with the earlier effective time is retained, and the remaining store status events are recorded as merge source records. The merging time window for duplicate events of each event type is independently defined, with a value range from one time granularity to the preset maximum merging duration.

[0066] For the same conflict resolution time unit When multiple candidate valid state events exist, conflict resolution is performed based on event priority, stage change intensity, and the time deviation between the event's effective time and occurrence time to obtain valid state events for stage switching; conflict resolution time unit. Divided according to calendar days or preset time granularity: , in, Represents the conflict resolution time unit Valid state events retained after the resolution of conflicts in the Inner Canon; This function represents the lexicographical maximum selection based on the order of the variables within the parentheses. Represents the conflict resolution time unit The set of candidate valid state events retained after merging short-term repetitive events in the Inner Canon; Indicates the time unit number used for conflict resolution; , , and The aforementioned definition will be used.

[0067] If, after selecting the candidate valid state event based on lexicographical order (event priority, intensity of stage change, and time deviation between event effective time and event occurrence time), multiple candidate valid state events remain in tandem, the candidate valid state event with the earlier effective time will be retained. If the event effective times are still the same, a candidate valid state event will be selected based on the lexicographical order of its event identifier. The remaining candidate valid state events are then written into the conflict resolution source record. The conflict resolution source record is only used to trace candidate valid state events resolved within the same conflict resolution time unit and does not change... The corresponding phase start event identifier.

[0068] After the above processing, store status events that are marked as effective with a base value of 1, whose stage change intensity reaches the preset stage change threshold, and which are retained after repeated event merging and conflict resolution within the same conflict resolution time unit are identified as effective status events that change the store's operating status.

[0069] The start time of a valid status event is used as the start time of the corresponding business lifecycle stage, and the event identifier of the valid status event is used as the stage start event identifier. The stage start time is taken as the event effective time if it exists, and as the event occurrence time if it is missing. If a store status event with a cancelled event status is received after a stage start event has been generated, and has the same event identifier as the stage start event or has a cancelled event status pointing to the stage start event identifier, the stage start event is set to a pending review status. In the pending review status, the previously confirmed business lifecycle stage is maintained as the conservative stage result, and the use of the event to generate a new business age starting point is suspended until a confirmation event is received from the same event source or from an event source with a higher preset ranking in the corresponding trusted source set.

[0070] Example 2: Based on Example 1, this example provides a specific method for generating the baseline range of indicators according to the distribution status of operating indicators in the store indicator early warning method based on time-series anomaly detection; In one implementation, the stage baseline interval is generated by a baseline generation rule or a time-series baseline model. The baseline generation rule is a processing rule that aggregates similar stores in the same stage according to their operating age position and outputs the indicator baseline range and the stage change baseline. The time-series baseline model takes similar stores in the same stage and the operating age range of the target store as inputs and outputs the indicator baseline range and the stage change baseline. The training data sources for the time-series baseline model include historical store operating indicator time-series data, store attribute data, and store status event data. The annotation method includes generating operating life cycle stage identifiers and operating age position identifiers based on the stage start event. The training objective includes fitting the indicator distribution state, indicator change direction, and continuous change state corresponding to the operating age position. After receiving the warning processing results and review tags associated with the store indicator warning information, the sample status in the historical store sample library is updated based on the warning processing results and review tags. The updated sample status will take effect in the next phase of the baseline version.

[0071] Based on the above implementation, the output fields of the baseline generation rules and the time-series baseline model are consistent, both including the indicator baseline range, the stage change baseline, the stage baseline version identifier, and the stage baseline source record identifier. When the baseline generation rules are adopted, similar stores in the same stage are aggregated according to their operating age and location, and the above output fields are generated directly; when the time-series baseline model is adopted, the above output fields are generated based on the same input fields. Only one of the two will be effective within the same stage baseline version.

[0072] In a preferred embodiment of Example 2, when generating the baseline range of indicators based on the distribution status of operating indicators, similar stores at the same stage are aggregated according to the stage of the operating life cycle, the position of the operating age, and the operating indicator identifier.

[0073] For any operational indicator identifier under the current operational lifecycle stage, only sample values ​​that meet the fragment integrity requirement and have valid operational indicator records at the corresponding operational age position are retained to form a sample set of operational ages at the same stage: , in, Indicates the operating age position and operating indicators The corresponding sample set of business ages at the same stage; Indicates the operational age position relative to the starting point of the phase; Indicates operational performance indicators; This indicates segments from similar stores at the same stage. In terms of operating age position Operating indicator labels below The index value; The segment number represents a segment from similar stores at the same stage; Indicates the stage of the business life cycle. Below are the operating indicator labels A collection of similar stores from the same period; This indicates the current stage of the target store's operational lifecycle. This indicates segments from similar stores at the same stage. The fragment integrity flag, with a value of 1 indicating that the fragment integrity condition is met; This indicates segments from similar stores at the same stage. In terms of operating age position Operating indicator labels below The record overwrite flag, with a value of 1, indicates the existence of valid operating indicator records.

[0074] After forming a sample set of operating ages within the same stage, before extracting central tendency and fluctuation range, the sample size of the sample set is validated. The sample size is calculated based on the number of sample values ​​in the sample set. If the sample size falls below the minimum requirement corresponding to the operating life cycle stage and operating indicator, the upper-level matching scope is expanded in the order of same store type, same business district attribute, and same regional level. Historical store segments that still belong to the same operating life cycle stage and have corresponding operating indicator records are then added to the expanded sample set to re-form the sample set. Then, based on the newly formed sample set of operating age at the same stage, the central tendency, robust volatility, and quantile values ​​are extracted. If the sample size is still lower than the minimum requirement after incremental supplementation, the quantile statistics of the operating age position are no longer used to generate the original indicator baseline range independently. When there is an already generated indicator baseline range for an adjacent operating age position, the already generated indicator baseline range of the adjacent operating age position is directly used as the conservative indicator baseline range for that operating age position, and a low sample baseline position mark is written for that operating age position. If there is also no already generated indicator baseline range for an adjacent operating age position, a low sample baseline position mark is written for that operating age position, and operating age positions with low sample baseline position marks do not generate indicator baseline ranges, do not trigger single-point early warning judgments independently, and do not participate in the generation of stage abnormal scores before the adjacent operating age positions generate effective indicator baseline ranges.

[0075] The minimum sample size requirement is an implementation parameter stored within the stage baseline version. Its value is determined based on the business lifecycle stage, business indicator identifier, and time granularity, and is formed through historical sample quantiles or validation set integration. Sample size verification only counts samples with a normal status and a record coverage mark that is valid. The sample size verification result includes meeting the sample size requirement and writing a low sample baseline position mark, used to determine whether the business age position generates an indicator baseline range.

[0076] Central tendency and range of variation were extracted from a sample set of businesses operating at the same age. Central tendency was represented by the median, and range of variation was represented by the scale-corrected median absolute deviation. , , in, Indicates the operating age position and operating indicators The corresponding central tendency value; This indicates the operation of taking the median value of the sample within the parentheses; Indicates the operating age position and operating indicators The corresponding robust volatility value; , , , and Following the previous definition, the value 1.4826 represents the fixed correction constant used to convert the median absolute deviation into a robust oscillation scale.

[0077] For a sample set of operating ages at the same stage that meets the minimum sample size requirement after sample size verification, the original indicator baseline range is generated based on its quantile distribution and robust volatility value: , , in, Indicates the operating age position and operating indicators The corresponding lower limit of the original indicator baseline; Indicator of operating indicators The corresponding lower boundary of the indicator value; This represents a sample set of business operations at the same stage of age. Take the quantile level as quantile values; Indicates the stage of the business life cycle. and operating indicators The corresponding baseline tail position parameters; Indicates the stage of the business life cycle. and operating indicators The corresponding fluctuation expansion coefficient; Indicates the operating age position and operating indicators The corresponding upper limit of the original indicator baseline; Indicator of operating indicators The corresponding minimum baseline width; This represents a sample set of business operations at the same stage of age. Take the quantile level as quantile values; , , , and The aforementioned definition will be used.

[0078] The lower boundary of the indicator values ​​is determined based on the operational indicator identifier; for sales revenue, order volume, customer traffic, new member acquisition, member repurchase, platform exposure, platform visits, sales volume, positive reviews, and negative reviews, the lower boundary of the indicator values ​​is zero. The baseline tail radix parameter is offline calibrated based on the same period segment of samples in the historical store sample library where the sample status is normal, with a value greater than 0 and less than or equal to 0.5; the smaller the baseline tail radix parameter, the wider the sample distribution covered by the original indicator baseline range. The fluctuation expansion coefficient is offline calibrated based on the proportion of normal samples falling outside the original indicator baseline range within the same operational lifecycle stage, with a value greater than or equal to 0 and less than or equal to the preset expansion upper limit; the higher the proportion of normal samples falling outside the original indicator baseline range, the larger the fluctuation expansion coefficient, but it stops increasing after reaching the preset expansion upper limit. The minimum baseline width is determined based on the smallest resolvable recording unit of the operational indicator identifier, ensuring a discernible interval between the baseline upper limit and the baseline lower limit. The baseline tail radix parameter, fluctuation expansion coefficient, and minimum baseline width are frozen within the same stage baseline version. After the stage baseline version is updated, the generated store indicator warning information continues to be associated with the stage baseline version used during generation.

[0079] To ensure that the baseline range of indicators between adjacent operating age positions matches the process of phased change, the indicators are arranged according to operating age position from... The original index baseline range is processed continuously in an increasing order; when hour, The baseline range for indicators used for anomaly comparison is obtained by considering the previous operational age position within the current operational lifecycle stage where continuous processing has been completed. , , in, Indicates the operating age position and operating indicators The corresponding lower limit of the baseline of the continuous processing indicator; Indicates the starting age position within the current business lifecycle stage; Indicates the stage of the business life cycle. and operating indicators The corresponding continuous smoothing coefficient takes a value greater than 0 and less than or equal to 1; Indicates the time granularity interval between adjacent operating age positions; Indicates the operating age position and operating indicators The corresponding lower limit of the baseline of the continuous processing indicator; Indicates the operating age position and operating indicators The corresponding upper limit of the baseline for the continuous processing indicator; Indicates the operating age position and operating indicators The corresponding upper limit of the baseline for the continuous processing indicator; , , , and The aforementioned definition will be used.

[0080] Continuity smoothing coefficient Offline calibration is performed based on the baseline jump amplitude of normal samples at adjacent operating age positions within the same operating life cycle stage; The larger the value, the closer the baseline range of the indicator after continuous processing is to the original baseline range of the indicator at the current operating age. The smaller the value, the closer the baseline range of the continuous-processed indicator is to the baseline range of the continuous-processed indicator at the previous operating age position. The continuous smoothing coefficient is frozen within the same stage baseline version and is not updated with real-time anomaly results for individual target stores.

[0081] When generating the indicator baseline range, the direction of indicator change is also generated based on the changes in the central tendency value of adjacent operating age positions. A continuous increase, a continuous decrease, or fluctuation within the smallest resolvable recording unit of the central tendency value are marked as an upward direction, a downward direction, or a stable direction, respectively. If the direction of indicator change for multiple consecutive operating age positions is consistent with the stage change baseline of the same historical segment, the indicator baseline range after continuity processing is retained. If the original indicator baseline range for a certain operating age position jumps beyond the allowable range of the stage change baseline relative to the previous operating age position, a low-confidence baseline is written to the indicator baseline range for that operating age position. When generating abnormal scores in subsequent stages, the single-point deviation state corresponding to the indicator baseline range marked with a low confidence baseline is only involved in the early warning triggering judgment when there is a same-direction deviation at least one adjacent existing operating age position with valid operating indicator records. Same-direction deviation includes continuously falling below the lower limit of the indicator baseline or continuously rising above the upper limit of the indicator baseline. If the indicator baseline range marked with a low confidence baseline is located at the starting or ending operating age position of the current operating life cycle stage, and there is no valid adjacent operating age position that can be used to confirm the same-direction deviation, then the single-point deviation state corresponding to the indicator baseline range marked with a low confidence baseline is not involved in the early warning triggering judgment on its own.

[0082] Low-confidence baseline markers are data markers used to record data where the indicator baseline range changes beyond the permissible range between adjacent operating age positions. Low-sample baseline location markers are data markers used to record data where the sample set of operating ages in the same stage does not meet the minimum sample size requirement. Both low-confidence baseline markers and low-sample baseline location markers are saved along with the indicator baseline range and are used as the basis for determining whether to participate in the single-point early warning determination when generating stage anomaly scores.

[0083] After the above processing, for operating age locations that meet the minimum sample size requirement after sample size verification, the indicator baseline range after continuous processing is used as the indicator baseline range corresponding to that operating age location and the operating indicator identifier. For operating age locations where the indicator baseline range has already been generated from adjacent operating age locations due to insufficient sample size, the conservative indicator baseline range is used as the indicator baseline range corresponding to that operating age location and the operating indicator identifier. The indicator baseline range is associated and saved with the similar store segment identifier, operating life cycle stage identifier, operating age location, sample size, quantile parameter, fluctuation expansion coefficient, low confidence baseline marker, and low sample baseline location marker used to generate the range, and is combined with the stage change baseline as part of the stage baseline interval.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0085] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for early warning of store indicators based on time-series anomaly detection, characterized in that, include: Obtain time-series data of store operation indicators, store attribute data, and store status event data for the target store; Based on the event source, event status, event coverage, event type, event occurrence time, and event effective time in the store status event data, valid status events are identified. From the valid status events, the phase start event is determined. When multiple valid status events exist within the same conflict resolution time unit, the phase start event is determined according to the event priority. The current business life cycle stage is determined based on the phase start event. Using the start time of the stage-starting event as the starting point of the operating age, the time-series data of the store's operating indicators in the current operating life cycle stage of the target store are converted into a series of operating age indicators. The stage start time is taken as the event effective time when the event effective time exists, and as the event occurrence time when the event effective time is missing. Based on the store attribute data and the current business life cycle stage, similar stores in the same stage are identified in the historical store sample library, and a stage baseline interval and a stage baseline source record are generated based on the similar stores in the same stage. The operating age indicator sequence is time-aligned with the stage baseline interval according to the operating age position, and a stage anomaly score is generated based on the deviation status of the aligned indicators. Determine whether the abnormal score of the stage meets the early warning triggering rules. If the early warning triggering rules are met, generate and output the store indicator early warning information.

2. The store indicator early warning method based on time-series anomaly detection according to claim 1, characterized in that, The time-series data of the store operation indicators include at least one of the following: sales revenue, order volume, customer traffic, new members, repeat purchases, platform exposure, platform visits, sales volume, positive reviews, and negative reviews. After obtaining the time-series data of the target store's operating indicators, the different operating indicator values ​​are time-aligned according to the preset time granularity, duplicate records are deduplicated, and missing markers or filler values ​​are generated based on the record status of adjacent operating age positions within the same operating life cycle stage.

3. The store indicator early warning method based on time-series anomaly detection according to claim 1, characterized in that, Convert the time-series data of the target store's operating indicators during its current operational lifecycle stage into a series of operational age indicators, including: Convert each point in time within the current business lifecycle stage into its business age position relative to the starting point of the business age; The operating indicator values ​​are arranged according to the operating age position described above; Write the business life cycle stage identifier, business age position, and time granularity identifier into each business indicator value; When a new valid status event occurs at the target store, a new sequence of operating age indicators is generated based on the new stage start event.

4. The store indicator early warning method based on time-series anomaly detection according to claim 1, characterized in that, Identify similar stores from the same period in the historical store sample database, including: Generate store matching tags based on at least two of the following: the target store's region, store type, business entity attributes, business district attributes, business status, and platform operation status; Candidate stores with corresponding store matching tags and corresponding business life cycle stages are selected from the historical store sample database; Extract time-series segments from the historical operating indicator time-series data of the candidate stores that correspond to the operating age range of the target stores; Time-series segments that meet the segment integrity criteria are identified as segments from the same stage in similar stores.

5. The store indicator early warning method based on time-series anomaly detection according to claim 4, characterized in that, The fragment integrity conditions include: There is a corresponding phase start event, there is a corresponding business life cycle phase identifier, the business indicator records cover the target store's operating age range, and the missing business indicator ratio meets the preset missing ratio condition. After identifying similar stores in the same stage, record the source store identifier, stage start event identifier, business life cycle stage identifier, business age range, and business indicator coverage status for each similar store in the same stage to form a stage baseline source record.

6. The store indicator early warning method based on time-series anomaly detection according to claim 1, characterized in that, Based on the similar stores and the same stage segments, a stage baseline interval is generated, including: Based on the operating age and location, the segments of the same stage of multiple similar stores are aligned, and the segments of the same stage of similar stores that meet the segment integrity condition are aggregated. Extract the distribution status, direction of change, and continuous change status of operating indicators at each operating age position; Generate an indicator baseline range based on the distribution status of the operating indicators, and generate a stage change baseline based on the direction of indicator change and continuous change status; The baseline range of the indicator and the baseline of the stage change are combined into a stage baseline interval, and the stage baseline interval is associated with and saved with the stage baseline source record.

7. The store indicator early warning method based on time-series anomaly detection according to claim 6, characterized in that, The stage baseline interval is generated by baseline generation rules or a time-series baseline model; The baseline generation rule is a processing rule that aggregates similar stores in the same stage according to their operating age and location, and outputs the indicator baseline range and the stage change baseline. The time-series baseline model takes similar stores in the same stage segment and the target store's operating age range as inputs, and outputs the indicator baseline range and stage change baseline. The training data sources for the time-series baseline model include historical time-series data of store operating indicators, store attribute data, and store status event data. The annotation methods include generating operating life cycle stage identifiers and operating age position identifiers based on the stage start event. The training objectives include fitting the indicator distribution status, indicator change direction, and continuous change status corresponding to the operating age position. After receiving the warning processing results and review markers associated with the store indicator warning information, the sample status in the historical store sample library is updated based on the warning processing results and review markers. The updated sample status will take effect in the next stage baseline version.

8. The store indicator early warning method based on time-series anomaly detection according to claim 1, characterized in that, Anomaly scores for each stage are generated based on the aligned metrics deviation, including: Based on the operating age position, the operating indicator values ​​of the target store are compared with the indicator baseline range in the stage baseline interval to obtain the stage deviation status of the single indicator. The continuous deviation status is obtained by determining the single-indicator stage deviation status at the continuous operating age position; The correlation deviation status is obtained based on the sequential changes of multiple operating indicators within the same operating age range; A stage anomaly score is generated based on the single-indicator stage deviation state, continuous deviation state, and associated deviation state.

9. The store indicator early warning method based on time-series anomaly detection according to claim 8, characterized in that, Determining whether the abnormal score of the stage meets the early warning triggering rules includes: When the abnormal score corresponding to the sales volume, order volume or customer traffic reaches the trigger condition of the result indicator, a result deviation warning mark is generated. When the abnormal scores corresponding to the platform exposure, platform visits, new members, core sales, positive reviews or negative reviews reach the trigger conditions of the preceding indicators, but the result indicators do not reach the trigger conditions of the result indicators, a potential transmission warning mark is generated. The warning type is determined based on the deviation warning marker or the potential transmission warning marker.

10. The store indicator early warning method based on time-series anomaly detection according to claim 9, characterized in that, Generate and output store performance indicator alerts, including: The recipients of the early warning are determined based on the organizational affiliation of the target stores; The warning level is determined based on the stage anomaly score, indicator deviation status, and warning type; Generate store indicator warning information that includes the target store identifier, current operating life cycle stage, stage anomaly score, warning type, warning level, and stage baseline source record identifier; After receiving the warning processing result, the warning processing result, processing time, and review mark are associated with and saved with the corresponding store indicator warning information.

Images