A smart warehouse inventory management method based on big data analysis

Abstract

This invention discloses a smart warehouse inventory management method based on big data analysis, comprising the following steps: acquiring and cleaning multi-source warehouse data to form a standardized set of inventory event fragments; identifying valid demand and updating the quiescent state, generating a demand dormancy age sequence and an activity trigger propagation graph; estimating the wake-up risk based on the age sequence and propagation graph, generating a wake-up risk spectrum result; determining the released inventory objects and decoding demand impulses to generate demand impulse release packets; combining the release packets with inventory events to execute inventory envelope adjudication, generating inventory level results and inventory action results; updating the propagation graph and age sequence based on the level results, action results, and execution feedback to form a closed loop. This invention employs big data analysis and inventory event evolution modeling methods to achieve dynamic adjudication and closed-loop updating of smart warehouse inventory, possessing the advantages of accurate demand identification, fast decision response, and strong inventory risk control.

Description

Technical Field

[0001] This invention relates to the field of smart warehousing technology, and in particular to a smart warehousing inventory management method based on big data analysis. Background Technology

[0002] Existing smart warehousing inventory management technologies typically rely on warehouse management systems, order management systems, and procurement management systems to collect inventory flow data, inbound and outbound records, and replenishment records. They then generate replenishment suggestions and inventory warnings through rule thresholds, statistical analysis, or general demand forecasting models. These solutions can, to a certain extent, perform inventory monitoring, inventory counting, and replenishment planning, and are widely used in retail warehouses, manufacturing warehouses, and regional distribution warehouses. However, most existing technologies treat inventory objects as continuously changing, ordinary time-series sequences, usually making predictions directly based on historical outbound volumes, inventory levels, or sales volumes. They lack the separate identification of effective demand events and the specific characterization of demand intervals, quiet periods, and abnormal triggering periods.

[0003] When faced with low-frequency materials, substitute materials, project-driven materials, and regionally linked demands, existing technologies struggle to distinguish between inventory adjustments, allocation changes, and actual demand reductions, easily leading to distorted demand identification. Furthermore, existing technologies typically employ static association rules or single forecast results for inventory decisions, failing to express the timely changes in relationships such as demand co-occurrence, substitution migration, project release, regional linkage, and supply suppression. They also struggle to generate structured intermediate results based on the evolution of demand from dormant to awakened and then released, further resulting in inaccurate assessments of stockout, overstocking, and supply disruption risks. Additionally, there is a lack of a closed-loop linkage mechanism between inventory action generation and subsequent feedback updates.

[0004] Therefore, how to provide a smart warehousing and inventory management method based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a smart warehouse inventory management method based on big data analysis. This invention uses big data analysis and inventory event evolution modeling to achieve dynamic adjudication and closed-loop updating of smart warehouse inventory, which has the advantages of accurate demand identification, fast decision response and strong inventory risk control.

[0006] A smart warehouse inventory management method based on big data analysis according to an embodiment of the present invention includes the following steps: Acquire multi-source warehousing business data, align and clean it according to a unified time index, warehouse node identifier, and inventory object identifier, and generate a standardized set of inventory event fragments; Based on a standardized set of inventory event fragments, valid demand events are identified. Each inventory object is updated according to its quiescent state since the last valid demand event, generating a demand dormancy age sequence. Demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply inhibition relationships are extracted to generate an activity trigger propagation graph. Based on the demand dormancy age sequence and the activity trigger propagation map, the marginal risk estimation from dormancy to wakefulness is performed, and the wakefulness risk spectrum results are generated. Based on the wake-up danger spectrum results, the released inventory objects are identified, and demand pulse release decoding is performed on the released inventory objects to generate demand pulse release packets. Based on the demand impulse release packet and the standardized set of inventory event fragments, a joint decision on the inventory envelope is executed to generate inventory level results and inventory action results; Based on inventory level results, inventory action results, and execution feedback, an active trigger propagation graph and a closed loop for updating the demand dormancy age sequence are constructed.

[0007] Optionally, the generation of the standardized inventory event fragment set specifically includes: It collects original business records from warehouse management systems, order management systems, procurement management systems, transportation management systems, inventory management systems, master data management systems, and project planning management systems, and encapsulates them according to record source and business type to form a set of original warehouse business records; Perform unified field mapping and object primary key parsing on the original warehousing business record set to form a unified field business record set; By performing time merging, node merging, and object merging on a unified set of business records, an indexed set of business records is obtained. Alignment and cleaning are performed on the indexed business record set, including duplicate record elimination, record cancellation pairing, missing field completion, conflict status adjudication, and abnormal number correction, to form a clean business record set; The cleanup business record set is aggregated into fragments according to a unified time index, warehouse node identifier, and inventory object identifier to form a standardized inventory event fragment set.

[0008] Optionally, the generation of the required dormancy age sequence and the activity trigger propagation map specifically includes: Perform valid demand event identification on each standardized inventory event fragment in the standardized inventory event fragment set to form a valid demand event set; The set of effective demand events is expanded into a time-series record based on warehouse node identifier and inventory object identifier, forming an effective demand time series index table. The quiescent state of each inventory object is updated based on the effective demand time series index table and the standardized inventory event fragment set to form a demand dormancy age sequence. Construct a candidate set of trigger relationships based on a standardized set of inventory event fragments, a set of effective demand events, and a sequence of demand dormancy periods; Perform liveness determination and propagation encapsulation on the candidate set of trigger relationships to generate an live trigger propagation graph.

[0009] Optionally, the candidate set of triggering relationships includes demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply suppression relationships.

[0010] Optionally, the generation of the awakening hazard spectrum results specifically includes: Perform object alignment, time unrolling, and edge type splitting on the demand dormancy age sequence and activity trigger propagation graph to form a hazard estimation sample set; Based on the risk estimation sample set, calculate the dormant base risk value of each inventory object in the future review cycle sequence to obtain the dormant base risk sequence; The positive trigger gain values ​​of the positive active trigger edge set are aggregated in the future review cycle sequence to form a positive trigger gain sequence; The suppression deduction values ​​of the active suppression trigger edge set and the inactivation silencing marker on the future review cycle sequence are aggregated to form the suppression deduction sequence. The wake-up danger candidate sequence is calculated by combining the dormant base danger sequence and the positive trigger gain sequence. Lifetime capping and temporal smoothing are performed on wake-up hazard candidate sequences to generate wake-up hazard spectrum results.

[0011] Optionally, the suppression deduction sequence reads the relationship strength, effective start point, and effective end point of each active trigger edge in the set of suppression active trigger edges for each unified time index in the future review cycle sequence. When the unified time index is located between the effective start point and effective end point of the active trigger edge, the suppression attenuation value is calculated according to the time distance between the unified time index and the unified time index formed by the relationship. The relationship strength of all suppression active trigger edges within the effective interval is multiplied by the corresponding suppression attenuation value and then accumulated to obtain the suppression deduction value on the unified time index. When the inactivation silent mark corresponding to the current unified time index is valid, the preset inactivation deduction value is superimposed on the suppression deduction value on the unified time index. All suppression edge deduction values ​​in the future review cycle sequence are arranged in the order of the unified time index to form the suppression deduction sequence. For each unified time index in the future review cycle sequence, the wake-up danger candidate sequence is formed by adding the corresponding dormant base danger value and the positive trigger gain value, subtracting the corresponding suppression deduction value, comparing the calculation result with zero and taking the larger value. The wake-up danger candidate value is then arranged in the order of the unified time index to form the wake-up danger candidate sequence.

[0012] Optionally, the generation of the demand pulse release packet specifically includes: Based on the warehouse node identifier and the inventory object identifier, release candidate screening is performed on each object according to the wake-up hazard spectrum results to form a set of release candidate intervals; The inventory objects in the release status are determined according to the set of release candidate intervals, and a current release feature table is formed. By tracing back historical release segments based on the current release feature table, a historical release sample set is formed; Calculate the sample similarity for each historical release sample in the historical release sample set, perform sample filtering and pulse parameter decoding, and form a release parameter table; The current release feature table and release parameter table are encapsulated into a demand pulse release package.

[0013] Optionally, the generation of the inventory level results and inventory action results specifically includes: Extract records of the same object from the demand pulse release package and the standardized inventory event fragment set to form an inventory decision base table; For each record in the inventory decision base table, perform a shortage envelope calculation. When the shortage envelope value is greater than zero, mark the corresponding record as having a valid shortage envelope. When the shortage envelope value is equal to zero, mark the corresponding record as having an invalid shortage envelope, thus forming the shortage envelope result. Perform backlog envelope calculation around the inventory decision base table to generate backlog envelope results; The supply disruption envelope is calculated by combining the shortage envelope result, the inventory adjudication base table, and the abnormal delivery status to form the supply disruption envelope result; The results of stockout envelope, overstock envelope, and supply disruption envelope are subject to priority joint adjudication to generate inventory level results; Based on the inventory level results and the standardized set of inventory event fragments, execute action mapping to generate inventory action results.

[0014] Optionally, the generation of the update closed loop specifically includes: Collect subsequent business feedback on inventory actions within the feedback observation window to form an execution feedback record set. Align the execution feedback record set, inventory level results, inventory action results, activity trigger propagation diagram, and demand dormancy age sequence with the execution feedback to form an updated baseline table; The relationship strength, effective range, and failure condition re-evaluation of the activity trigger propagation graph are performed on the updated baseline table to form the updated activity trigger propagation graph. Based on the updated baseline table and the execution feedback record set, the demand dormancy age sequence is reset, accumulated, and marked to form the updated demand dormancy age sequence. Write back the updated activity trigger propagation graph and the updated demand dormancy age sequence to form a closed loop for marginal risk estimation update.

[0015] The beneficial effects of this invention are: Compared with existing technologies, this invention first performs unified alignment and cleaning of multi-source warehousing business data, and identifies valid demand events based on a standardized set of inventory event fragments. It separates actual demand deductions from non-demand inventory changes such as inventory count corrections, transfer changes, and return write-offs. Then, it constructs a demand dormancy age sequence based on the quiescent state since the last valid demand event. This approach allows for fine-grained characterization of the quiescent length, quiescent extension, quiescent deviation, and inactive quiescent state of inventory objects during continuous periods of no demand, thereby improving the accuracy of demand identification for low-frequency materials, slow-moving items, substitute materials, and project-driven materials. It avoids the demand distortion problems caused by directly modeling based on total outbound volume or ordinary time series data in existing technologies.

[0016] Furthermore, this invention extracts demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply suppression relationships to construct an active trigger propagation diagram with an effective start point, effective end point, and failure condition. It then combines this with a demand dormancy age sequence to generate a wake-up hazard spectrum result, and performs demand pulse release decoding on released inventory objects to form a demand pulse release package. Based on this, it generates inventory level results and inventory action results through joint adjudication of shortage envelopes, overstock envelopes, and supply disruption envelopes. This allows inventory actions to no longer rely on a single forecast value but to simultaneously consider demand release intensity, supply coverage capacity, lifecycle status, and abnormal delivery status. Therefore, it can more accurately complete decisions on replenishment, allocation, frozen procurement, clearance, and maintenance actions. Simultaneously, by using execution feedback to perform closed-loop updates on the active trigger propagation diagram and demand dormancy age sequence, it can continuously correct relationship strength, active range, and dormant state, improving the adaptability and stability of subsequent marginal hazard estimation and inventory management results. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a smart warehouse inventory management method based on big data analysis proposed in this invention; Figure 2 This invention presents a flowchart illustrating the generation process of a demand dormancy period sequence and an activity trigger propagation graph for a smart warehousing inventory management method based on big data analysis. Figure 3 This invention presents a flowchart illustrating the generation process of a wake-up hazard spectrum and demand pulse release package for a smart warehouse inventory management method based on big data analysis. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A smart warehouse inventory management method based on big data analytics includes the following steps: Acquire multi-source warehousing business data, align and clean it according to a unified time index, warehouse node identifier, and inventory object identifier, and generate a standardized set of inventory event fragments; Based on a standardized set of inventory event fragments, valid demand events are identified. Each inventory object is updated according to its quiescent state since the last valid demand event, generating a demand dormancy age sequence. Demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply inhibition relationships are extracted to generate an activity trigger propagation graph. Based on the demand dormancy age sequence and the activity trigger propagation map, the marginal risk estimation from dormancy to wakefulness is performed, and the wakefulness risk spectrum results are generated. Based on the wake-up danger spectrum results, the released inventory objects are identified, and demand pulse release decoding is performed on the released inventory objects to generate demand pulse release packets. Based on the demand impulse release packet and the standardized set of inventory event fragments, a joint decision on the inventory envelope is executed to generate inventory level results and inventory action results; Based on inventory level results, inventory action results, and execution feedback, an active trigger propagation graph and a closed loop for updating the demand dormancy age sequence are constructed.

[0020] In this embodiment, the generation of the standardized inventory event fragment set specifically includes: It collects original business records from warehouse management systems, order management systems, procurement management systems, transportation management systems, inventory management systems, master data management systems, and project planning management systems, and encapsulates them according to record source and business type to form a set of original warehouse business records; The original warehousing business record set includes inbound records, outbound records, transfer records, return records, reservation records, in-transit records, inventory records, purchase arrival records, delivery date records, substitution relationship records, project release records, regional order records, and lifecycle records. Each original business record is written with a source identifier, business type identifier, and original occurrence time. Perform unified field mapping and object primary key parsing on the original warehousing business record set to form a unified field business record set; The unified field mapping and object primary key parsing include mapping product codes from different sources to inventory object identifiers, mapping warehouse codes, storage area codes, and operation node codes from different sources to warehouse node identifiers, mapping the original occurrence time to the base time value corresponding to the unified time index, and merging quantity fields, status fields, relationship fields, and source fields into a unified field business record set. The quantity fields include inbound quantity, outbound quantity, transfer inbound quantity, transfer outbound quantity, returned quantity, reserved quantity, and in-transit quantity. The status fields include delivery date abnormal status and lifecycle status. The relationship fields include substitution relationship identifier, project release identifier, and regional order identifier. By performing time merging, node merging, and object merging on a unified set of business records, an indexed set of business records is obtained. The node merging process follows the warehouse node hierarchy mapping rules to merge warehouses, storage areas, and operation nodes into warehouse node identifiers. The object merging process follows the inventory object master data and substitution relationship records to merge master goods, substitute goods, and related goods into inventory object identifiers. Alignment and cleaning are performed on the indexed business record set, including duplicate record elimination, record cancellation pairing, missing field completion, conflict status adjudication, and abnormal number correction, to form a clean business record set; Among them, duplicate record elimination retains uniqueness based on source identifier, business type identifier, unified time index, warehouse node identifier, inventory object identifier, and original business document identifier; offset record pairing is performed to offset the opposite direction of inventory change and the same business document identifier; missing field completion is performed to fill in the missing field by filling in the missing field with the same inventory object identifier in consecutive records under adjacent unified time indexes; conflict status adjudication is performed to merge and adjudicate delivery abnormal status and life cycle status according to business type priority; and abnormal quantity correction is performed to correct the quantity field according to quantity non-negative constraint and inventory continuity constraint. The cleanup business record set is aggregated into fragments according to a unified time index, warehouse node identifier, and inventory object identifier to form a standardized inventory event fragment set.

[0021] In this embodiment, the generation of the required dormancy age sequence and the activity trigger propagation map specifically includes: Perform valid demand event identification on each standardized inventory event fragment in the standardized inventory event fragment set to form a valid demand event set; The identification of effective demand events calculates the net deduction value of available inventory segment by segment according to the unified time index, warehouse node identifier, and inventory object identifier. The quantity of outbound goods is subtracted from the quantity of transfer outbound goods, and then the quantity of returned goods is subtracted. When the net deduction value of available inventory is greater than zero, the source identifier sequence contains the demand source identifier corresponding to the order management system or project planning management system, the business type sequence does not contain inventory count records and inventory correction records, and the ending inventory is less than the beginning inventory, the standardized inventory event segment is marked as an effective demand event; otherwise, it is marked as an ineffective demand event. The set of effective demand events is expanded into a time-series record based on warehouse node identifier and inventory object identifier, forming an effective demand time series index table. The time-series records all valid demand events under the same warehouse node identifier and the same inventory object identifier in ascending order according to a unified time index. The first valid demand event after sorting is written with a first event flag. The remaining valid demand events after sorting are respectively written with the current unified time index, the unified time index corresponding to the previous valid demand event, the available inventory net deduction value corresponding to the current valid demand event, the project release flag corresponding to the current valid demand event, and the regional order flag corresponding to the current valid demand event. The valid demand time-series index table contains an event order field, a previous event field, and a current event field. The quiescent state of each inventory object is updated based on the effective demand time series index table and the standardized inventory event fragment set to form a demand dormancy age sequence. The silent status update uses a unified time index as the update step size, and updates each warehouse node identifier and each inventory object identifier hourly. When there is a valid demand event in the current unified time index, the silent length corresponding to the unified time index is set to zero and a reset flag is written. When there is no valid demand event in the current unified time index, the silent length corresponding to the previous unified time index is incremented by one and written to the current unified time index. The reserved quantity, in-transit quantity, delivery abnormal status and life cycle status under the current unified time index are compared. If the reserved quantity increases continuously and the in-transit quantity does not increase, a silent extension flag is written. If the delivery abnormal status persists and the silent length is greater than the preset silent threshold, a silent deviation flag is written. If the life cycle status corresponds to the end stock, discontinued or replaced, an inactive silent flag is written. Construct a candidate set of trigger relationships based on a standardized set of inventory event fragments, a set of effective demand events, and a sequence of demand dormancy periods; Perform liveness determination and propagation encapsulation on the candidate set of trigger relationships to generate an live trigger propagation graph; The activity determination and propagation encapsulation process involves determining the effective start point for each relationship record in the candidate set of trigger relationships according to the unified time index of the relationship formation, determining the effective end point according to the preset activity window length corresponding to the relationship type, and determining the failure condition according to any one of the following conditions in the demand dormancy age sequence: the appearance of the inactive silent marker, the removal of the delivery abnormality state, the end of the project release window, or the life cycle state entering the deactivated or replacement completed state. If the current unified time index is between the effective start point and the effective end point and the failure condition is not met, the relationship record is retained as an active trigger edge; otherwise, it is discarded. All active trigger edges, along with the corresponding source inventory object node, target inventory object node, warehouse node constraints, relationship type, relationship strength, effective start point, effective end point, and failure condition, are encapsulated into an active trigger propagation graph.

[0022] In this embodiment, the candidate set of triggering relationships includes demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply suppression relationships; The demand co-occurrence relationship statistically analyzes the situation where two different inventory object identifiers under the same warehouse node identifier simultaneously have valid demand events. The number of times the two inventory object identifiers simultaneously have valid demand events in the same unified time index or adjacent unified time indexes is divided by the maximum value of the number of valid demand events of each of the two inventory object identifiers and the first value to obtain the demand co-occurrence intensity of the two inventory object identifiers. If it is not less than the preset co-occurrence threshold, it is written into the demand co-occurrence relationship. The substitution migration relationship is defined as follows: when there is a substitution relationship between two inventory object identifiers, and the net available inventory deduction value of the previous inventory object identifier in the current unified time index is less than that of the previous unified time index, and the net available inventory deduction value of the subsequent inventory object identifier in the current unified time index is greater than that of the previous unified time index, the difference obtained by subtracting the net available inventory deduction value of the previous unified time index from the net available inventory deduction value of the subsequent inventory object identifier in the current unified time index is compared with zero and the larger value is taken. Then, the difference obtained by adding the net available inventory deduction value of the previous inventory object identifier in the previous unified time index to the net available inventory deduction value of the current unified time index is compared with zero and the larger value is taken. The substitution migration strength is calculated, and if it is not less than the preset migration threshold, the substitution migration relationship is written. The release relationship is calculated when the project release identifier first appears in the unified time index and the corresponding inventory object identifier first appears in the preset release window. If the project release time delay value is not greater than the preset release time delay threshold, the project release relationship is written. When two different warehouse nodes corresponding to the same regional order identifier have valid demand events in the same unified time index or adjacent unified time index, the number of valid demand events that occur simultaneously will be divided by the maximum value of the number of times both nodes jointly participate in the same regional order identifier and the value in the first part to obtain the regional linkage strength. If the strength is not less than the preset linkage threshold, it will be written into the regional linkage relationship. When the supply suppression relationship is established, the following factors are considered: the number of unified time indices for continuous delivery abnormality, continuous decrease in the number of items in transit, continuous increase in the number of reserved items, and continuous increase in the silent length. The supply suppression strength is calculated by adding the number of unified time indices for continuous delivery abnormality, continuous increase in the number of reserved items, and continuous increase in the silent length, and then subtracting the number of unified time indices for continuous increase in the number of items in transit. If the supply suppression strength is not less than the preset suppression threshold, it is written into the supply suppression relationship. Each relationship record in the candidate set of triggering relationships includes the relationship type, source inventory object identifier, target inventory object identifier, relationship strength, unified time index of relationship formation, and relationship source field.

[0023] In this embodiment, the generation of the awakening hazard spectrum results specifically includes: Perform object alignment, time unrolling, and edge type splitting on the demand dormancy age sequence and activity trigger propagation graph to form a hazard estimation sample set; Object alignment and time expansion are performed object-by-object according to warehouse node identifier and inventory object identifier. For each warehouse node identifier and each inventory object identifier, the silent length, reset flag, silent extension flag, silent deviation flag, and inactive silent flag under the current unified time index are read. All active triggering edges with the target inventory object identifier are read. The edge type is divided into active triggering edges with relationship types of demand co-occurrence relationship, substitution migration relationship, project release relationship, and regional linkage relationship into a positive active triggering edge set, and active triggering edges with relationship type of supply suppression relationship into a suppression active triggering edge set. The risk estimation sample set is constructed by using several consecutive unified time indices after the current unified time index to construct a future review cycle sequence. For each unified time index in the future review cycle sequence, the warehouse node identifier, inventory object identifier, current silent state, positive active triggering edge set, and suppression active triggering edge set are written. Based on the risk estimation sample set, calculate the dormant base risk value of each inventory object in the future review cycle sequence to obtain the dormant base risk sequence; The dormant base risk value is obtained by extracting the unified time index corresponding to all reset marks under the same warehouse node identifier and the same inventory object identifier in the demand dormant age sequence. The difference between the unified time indexes corresponding to two adjacent reset marks is taken as a historical trigger interval. The average value of all historical trigger intervals is taken as the historical trigger interval benchmark value. For each unified time index in the future review cycle sequence, the interval step size of the unified time index relative to the current unified time index is calculated. The current silence length, interval step size, the extension value corresponding to the silence extension mark and the deviation value corresponding to the silence deviation mark are added together and divided by the larger value between the historical trigger interval benchmark value and one to obtain the dormant base risk value on the unified time index. When the inactivation silence mark corresponding to the current unified time index is valid, the dormant base risk value is multiplied by a preset inactivation inhibition coefficient less than one. The positive trigger gain values ​​of the positive active trigger edge set are aggregated in the future review cycle sequence to form a positive trigger gain sequence; The positive trigger gain value is calculated edge-by-edge for each unified time index in the future review cycle sequence. For each active trigger edge in the set of positive active trigger edges, the relationship strength, effective start point, and effective end point are read. When the unified time index is located between the effective start point and effective end point of the active trigger edge, the propagation attenuation value of the active trigger edge on the unified time index is calculated. The attenuation is reversed according to the time distance between the unified time index and the relationship forming the unified time index. The relationship strength of all active trigger edges within the effective interval is multiplied by the corresponding propagation attenuation value and then summed to obtain the positive trigger gain value on the unified time index. The suppression deduction values ​​of the active suppression trigger edge set and the inactivation silencing marker on the future review cycle sequence are aggregated to form the suppression deduction sequence. The wake-up danger candidate sequence is calculated by combining the dormant base danger sequence and the positive trigger gain sequence. Perform lifetime capping and temporal smoothing on wake-up hazard candidate sequences to generate wake-up hazard spectrum results; The lifecycle capping and time smoothing are performed one by one for each unified time index in the future review cycle sequence. When the lifecycle status of the corresponding inventory object identifier in the standardized inventory event fragment set is normal, end-of-life, discontinued, or replacement completed, the corresponding upper limit coefficient is used as the upper limit value of the danger. The wake-up danger candidate value on the corresponding unified time index is compared with the upper limit value of the danger and the smaller value is taken to obtain the capped danger value. The capped danger value on the current unified time index is weighted and averaged with the capped danger value on the previous unified time index to obtain the smoothed danger value. All smoothed danger values ​​in the future review cycle sequence are arranged in the order of unified time index and encapsulated together with the warehouse node identifier, inventory object identifier, unified time index, and upper limit value of danger to form the wake-up danger spectrum result.

[0024] In this embodiment, the suppression deduction sequence reads the relationship strength, effective start point, and effective end point of each active trigger edge in the suppression active trigger edge set for each unified time index in the future review cycle sequence. When the unified time index is located between the effective start point and effective end point of the active trigger edge, the suppression attenuation value is calculated according to the time distance between the unified time index and the unified time index formed by the relationship. The relationship strength of all suppression active trigger edges within the effective interval is multiplied by the corresponding suppression attenuation value and then accumulated to obtain the suppression deduction value on the unified time index. When the inactivation silent mark corresponding to the current unified time index is valid, the preset inactivation deduction value is superimposed on the suppression deduction value on the unified time index. All suppression edge deduction values ​​in the future review cycle sequence are arranged in the order of the unified time index to form the suppression deduction sequence. The wake-up hazard candidate sequence is formed by adding the corresponding dormant base hazard value and the positive trigger gain value to each unified time index in the future review cycle sequence, subtracting the corresponding suppression deduction value, comparing the calculation result with zero and taking the larger value to obtain the wake-up hazard candidate value, and arranging them in the order of unified time index to form the wake-up hazard candidate sequence.

[0025] In this embodiment, the generation of the demand pulse release packet specifically includes: Based on the warehouse node identifier and the inventory object identifier, release candidate screening is performed on each object according to the wake-up hazard spectrum results to form a set of release candidate intervals; The release candidate screening process involves sequentially reading the smoothed hazard value, the upper limit of the hazard value, and the set of positive active triggering edges on the corresponding unified time index in each unified time index of the future review cycle sequence. When the smoothed hazard value is not less than the preset release threshold, there is a positive active triggering edge on the corresponding unified time index, and the life cycle status is not in the disabled state or the replacement completed state, the unified time index is marked as a release candidate unified time index. All release candidate unified time indices that are consecutively adjacent in time are merged into a release candidate interval. For each release candidate interval, the interval start unified time index, the interval end unified time index, the interval peak hazard, and the interval smoothed hazard value sequence are written to obtain the release candidate interval set. The inventory objects in the release status are determined according to the set of release candidate intervals, and a current release feature table is formed. The process involves retrieving the earliest available release candidate interval from the release candidate interval set for each warehouse node identifier and each inventory object identifier. The unified time index of the interval's start is determined as the release start unified time index, the unified time index of the interval's end is determined as the candidate end unified time index, and the interval peak risk within the release candidate interval is determined as the candidate peak risk. The silent length, lifecycle status, and all positive active trigger edges covering the release start unified time index are read and sorted by relationship strength. The relationship type with the strongest relationship strength is determined as the primary trigger relationship type, and the corresponding relationship strength is determined as the primary trigger strength. The warehouse node identifier and inventory object identifier corresponding to the release candidate interval are marked as release status inventory objects. The release start unified time index, candidate end unified time index, candidate peak risk, silent length, lifecycle status, primary trigger relationship type, and primary trigger strength are all written into the current release feature table. By tracing back historical release segments based on the current release feature table, a historical release sample set is formed; The historical release segment backtracking involves retrieving the effective demand time sequence index table corresponding to the same warehouse node identifier and the same inventory object identifier within the historical statistics window before the current unified time index for each released state inventory object in the current release feature table. If the absolute value of the difference between the preceding silent length corresponding to a historical effective demand event and the silent length in the current release feature table is not greater than the preset silent tolerance, the lifecycle state corresponding to the historical effective demand event is consistent with the lifecycle state in the current release feature table, and there is an active trigger edge on the unified time index corresponding to the historical effective demand event that matches the main trigger relationship type, then the unified time index corresponding to the historical effective demand event is determined as the historical release starting unified time index. Starting from the historical release starting unified time index, subsequent effective demand events are continuously read along the unified time index sequence. When the unified time index interval between two adjacent effective demand events is not greater than the preset pulse interval threshold, the corresponding event will be... Effective demand events are merged into the same historical release segment. When the unified time index interval between two adjacent effective demand events is greater than the preset pulse interval threshold, the unified time index corresponding to the previous effective demand event is determined as the unified time index for the termination of historical release. The net deduction value of available inventory corresponding to all effective demand events in each historical release segment is accumulated to obtain the total historical release. The number of unified time indices contained between the unified time index for the start of historical release and the unified time index for the end of historical release is counted to obtain the duration of historical release. The unified time index interval between the current unified time index for the end of historical release and the unified time index for the start of the next historical release is counted to obtain the historical re-release interval value. The unified time index for the start of historical release, the unified time index for the end of historical release, the total historical release, the duration of historical release, the historical re-release interval value, the preceding silent length, the life cycle status, and the main trigger relationship type are all written into the historical release sample set. Calculate the sample similarity for each historical release sample in the historical release sample set, perform sample filtering and pulse parameter decoding, and form a release parameter table; The sample selection and pulse parameter decoding processes involve accumulating sample similarity scores after assigning hierarchical values ​​based on the degree of matching of preceding silent lengths, consistency of lifecycle states, and consistency of main trigger relationship types. The sample selection and pulse parameter decoding processes sort historical release samples from highest to lowest similarity. When the number of samples in the historical release sample set is not less than a preset sample number threshold, a preset number of historical release samples are selected as the decoding sample subset. The total historical release amount within the decoding sample subset is sorted from smallest to largest, and the median value is taken as the median value of the release magnitude. A preset upper quantile value is taken as the upper edge value of the release magnitude. The historical release duration is sorted from smallest to largest, and the median value is taken as the release duration. The historical re-release interval is also sorted from smallest to largest. The middle value is taken as the re-release interval value. When the number of samples in the historical release sample set is less than the preset sample number threshold, the available inventory net deduction value corresponding to the most recent valid demand event is read as the benchmark release amount. The benchmark release amount is determined as the median value of the release amount. The value after multiplying the benchmark release amount by the preset upper edge amplification factor is determined as the upper edge value of the release amount. One is determined as the release duration length. The silent length in the current release feature table is determined as the re-release interval value. When the main trigger relationship type is the substitution migration relationship and the main trigger strength is not less than the preset migration flag threshold, the substitution migration flag is determined as valid. Otherwise, the substitution migration flag is determined as invalid. The median value of the release amount, the upper edge value of the release amount, the release duration length, the re-release interval value, and the substitution migration flag are written together into the release parameter table. The current release feature table and release parameter table are encapsulated into a demand pulse release package.

[0026] In this embodiment, the generation of inventory level results and inventory action results specifically includes: Extract records of the same object from the demand pulse release package and the standardized inventory event fragment set to form an inventory decision base table; Specifically, for each released state inventory object in the demand pulse release package, the standardized inventory event fragment with the unified time index no later than the unified time index of the release start and the closest time position in the standardized inventory event fragment set is retrieved. The reserved quantity is subtracted from the ending inventory and compared with zero to obtain the larger value to obtain the current available inventory quantity. When the delivery abnormality status is abnormal, the in-situ coverage quantity is set to zero. When the delivery abnormality status is not abnormal, the in-situ coverage quantity is determined as the in-transit quantity. The current available inventory quantity, in-situ coverage quantity, median value of release level, upper edge value of release level, release duration length, re-release interval value, substitution migration identifier, delivery abnormality status and life cycle status are written together into the inventory adjudication base table. For each record in the inventory decision base table, perform a shortage envelope calculation. When the shortage envelope value is greater than zero, mark the corresponding record as having a valid shortage envelope. When the shortage envelope value is equal to zero, mark the corresponding record as having an invalid shortage envelope, thus forming the shortage envelope result. The shortage envelope calculation adds the current available inventory to the supply coverage to obtain the supply coverage, and then subtracts the supply coverage from the upper limit of the release level and compares it with zero to obtain the larger value to obtain the shortage envelope value. Perform backlog envelope calculation around the inventory decision base table to generate backlog envelope results; The backlog envelope calculation involves adding the current available inventory to the in-situ coverage and subtracting the median value of the release level to obtain the basic backlog difference. When the basic backlog difference is less than zero, the backlog envelope value is set to zero. When the basic backlog difference is not less than zero and the lifecycle status is normal, the basic backlog difference is determined as the backlog envelope value. When the basic backlog difference is not less than zero and the lifecycle status is end-of-life, the basic backlog difference is multiplied by a preset end-of-life amplification factor to determine the backlog envelope value. When the basic backlog difference is not less than zero and the lifecycle status is discontinued or replacement completed, the basic backlog difference is multiplied by a preset discontinued amplification factor to determine the backlog envelope value. When the backlog envelope value is greater than zero, the corresponding record is marked as having a valid backlog envelope. When the backlog envelope value is equal to zero, the corresponding record is marked as having an invalid backlog envelope. The supply disruption envelope is calculated by combining the shortage envelope result, the inventory adjudication base table, and the abnormal delivery status to form the supply disruption envelope result; The supply disruption envelope calculation is performed on each record in the inventory adjudication base table. When the shortage envelope value is greater than zero, the delivery abnormality status is abnormal, the arrival coverage is less than the median value of the release level, and the re-release interval value is not greater than the release duration, the upper edge value of the release level minus the arrival coverage is compared with zero and the larger value is taken. Then, it is added to the shortage envelope value to obtain the supply disruption envelope value. When any condition is not met, the supply disruption envelope value is set to zero. When the supply disruption envelope value is greater than zero, the corresponding record is marked as a valid supply disruption envelope. When the supply disruption envelope value is equal to zero, the corresponding record is marked as an invalid supply disruption envelope. The results of stockout envelope, overstock envelope, and supply disruption envelope are subject to priority joint adjudication to generate inventory level results; The priority joint decision-making process involves the following steps: when the supply disruption envelope value is not less than the preset supply disruption threshold, the corresponding inventory object is marked as a supply disruption risk type; when the supply disruption envelope value is less than the preset supply disruption threshold and the stockout envelope value is not less than the preset stockout threshold, the corresponding inventory object is marked as a shortage type; when the supply disruption envelope value is less than the preset supply disruption threshold, the stockout envelope value is less than the preset stockout threshold, and the backlog envelope value is not less than the preset backlog threshold, the corresponding inventory object is marked as a stagnant type; and when the supply disruption envelope value is less than the preset supply disruption threshold, the stockout envelope value is less than the preset stockout threshold, and the backlog envelope value is less than the preset backlog threshold, the corresponding inventory object is marked as a balanced type. The inventory level results include the inventory level, stockout envelope value, backlog envelope value, supply disruption envelope value, release start unified time index, and release end unified time index. Based on the inventory level results and the standardized set of inventory event fragments, execute action mapping to generate inventory action results; In the action mapping, when the inventory level is of supply disruption risk type, firstly, candidate supply nodes with the same identifier as the current inventory object but different warehouse node identifiers are retrieved from the standardized inventory event fragment set. The ending inventory and reserved quantity corresponding to each candidate supply node are read. The ending inventory minus the reserved quantity is compared with the preset transfer retention quantity to filter out transferable candidate supply nodes. The warehouse node identifier with the largest current available inventory is determined as the source warehouse node identifier. When a source warehouse node identifier exists, the action type is determined as transfer replenishment. The smaller value between the shortage envelope value and the difference between the current available inventory corresponding to the source warehouse node identifier and the preset transfer retention quantity is determined as the action quantity. When no source warehouse node identifier exists, ... When the action type is determined as emergency replenishment, the shortage envelope value is determined as the action quantity. When the inventory level is scarce, the action type is determined as replenishment, and the shortage envelope value is determined as the action quantity. When the inventory level is stagnant and the life cycle status is end-of-line stock, discontinued, or replacement completed, the action type is determined as clearance and the backlog envelope value is determined as the action quantity. When the inventory level is stagnant and the life cycle status is normal, the action type is determined as frozen procurement, and the backlog envelope value is determined as the action quantity. When the inventory level is balanced, the action type is determined as maintenance, and the action quantity is zero. When the replacement migration flag is valid and the inventory level is scarce or at risk of supply disruption, the action priority is increased by one level.

[0027] In this embodiment, the generation of the updated closed loop specifically includes: Collect subsequent business feedback on inventory actions within the feedback observation window to form an execution feedback record set. The subsequent business feedback includes the following: When the action type is replenishment or emergency replenishment, the increase in the quantity in transit, the increase in the quantity received, and the increase in the ending inventory are counted to form replenishment execution feedback; when the action type is transfer replenishment, the decrease in the ending inventory of the inventory object corresponding to the source warehouse node identifier and the increase in the ending inventory of the inventory object corresponding to the target warehouse node identifier are counted to form transfer execution feedback; when the action type is frozen procurement, whether the procurement arrival record is added, whether the quantity in transit increases, and whether the ending inventory continues to rise are counted to form freeze execution feedback; when the action type is clearance and destocking, the increase in the quantity outbound, the decrease in the ending inventory, and the change in the life cycle status are counted to form clearance execution feedback; when the action type is maintenance, whether the effective demand event occurs, whether the silent length increases continuously, and whether the abnormal delivery status changes are counted to form maintenance execution feedback. Align the execution feedback record set, inventory level results, inventory action results, activity trigger propagation diagram, and demand dormancy age sequence with the execution feedback to form an updated baseline table; The relationship strength, effective range, and failure condition re-evaluation of the activity trigger propagation graph are performed on the updated baseline table to form the updated activity trigger propagation graph. The relationship strength update is performed on each record in the update baseline table separately. When the action type is replenishment, emergency replenishment, or transfer replenishment, and the abnormal delivery status is resolved, the quantity in transit increases, or the ending inventory rebounds within the feedback observation window, the relationship strength of the corresponding supply suppression relationship is reduced by a preset weakening step. When the action type is replenishment, emergency replenishment, or transfer replenishment, and the silent length within the feedback observation window continues to increase, the abnormal delivery status is not resolved, and no effective demand event occurs, the relationship strength of the corresponding supply suppression relationship is increased by a preset strengthening step. When the action type is clearing out inventory or freezing purchases, and no effective demand event occurs within the feedback observation window, and the silent length continues to increase, the relationship strength is adjusted according to the current inventory level. The strength of the demand co-occurrence relationship, project release relationship, and regional linkage relationship related to the inventory object identification is reduced by a preset weakening step. When the action type is maintenance, replenishment, emergency replenishment, or transfer replenishment and a valid demand event with the same type as the main trigger relationship appears in the feedback observation window, the strength of the corresponding demand co-occurrence relationship, substitution migration relationship, project release relationship, or regional linkage relationship is increased by a preset strengthening step. When the substitution migration identifier is valid and the number of valid demand events for the substitution inventory object increases while the number of valid demand events for the original inventory object decreases in the feedback observation window, the strength of the corresponding substitution migration relationship is increased by a preset strengthening step; otherwise, it is reduced by a preset weakening step. The effective interval update extends the effective endpoint of the relationship record that has been verified again in the feedback observation window by a preset active window length, and shrinks the effective endpoint of the relationship record that has not been verified in the feedback observation window and for which no effective demand event has occurred by a preset shrinkage length. The failure condition re-judgment is marked as a failure relationship and removed from the active trigger propagation graph when the inactive silent mark appears in the feedback observation window, the life cycle status enters the discontinued state or the replacement completion state, the project release window ends and no effective demand event occurs again. Based on the updated baseline table and the execution feedback record set, the demand dormancy age sequence is reset, accumulated, and marked to form the updated demand dormancy age sequence. The reset update process involves updating each record in the baseline table. When a valid demand event occurs within the feedback observation window, the silent length corresponding to the unified time index of the first valid demand event is set to zero and written to the reset flag. Then, starting from this unified time index, the silent length is re-accumulated hourly for subsequent unified time indices. When no valid demand event appears in the feedback observation window, the cumulative update starts from the silent length corresponding to the unified time index of the action execution and updates each unified time index in the feedback observation window by one hourly. The marking update is as follows: when the reserved quantity in the feedback observation window increases continuously and the quantity in transit does not increase, a silent extension mark is written; when the abnormal delivery status persists and the silent length is greater than the preset silent threshold, a silent deviation mark is written; when the life cycle status corresponds to the tail stock status, discontinued status, or replacement completed status and the silent length continues to increase, an inactive silent mark is written; when the abnormal delivery status is resolved, the ending inventory recovers, and the effective demand event reappears, the silent extension mark and the silent deviation mark are cleared. Write back the updated activity trigger propagation graph and the updated demand dormancy age sequence to form a closed loop for marginal risk estimation update.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to the inventory management scenario of a raw material warehouse, spare parts warehouse, and finished goods turnover warehouse of an electronics manufacturing company in East China. This company's warehouses contain a variety of materials, including regularly used materials that are continuously requisitioned, as well as low-frequency spare parts, substitute materials, and order-driven materials that are suddenly released after a long period of inactivity. Previously, management methods relied mainly on historical outbound records and manual experience to determine replenishment rhythm. This easily led to misinterpreting inventory changes such as inventory count corrections, transfers, and returns as actual demand, resulting in the wrong replenishment of slow-moving materials, delayed identification of substitute material switching, and inaccurate timing of transfers between regional warehouses, leading to shortages on one side and overstocking on the other. Especially when delivery fluctuations, abnormal incoming materials, and substitute switching occur simultaneously, the existing methods struggle to determine whether a particular inventory item is in a state of continuous inactivity, about to be activated, or already in the release phase. Inventory warnings are often delayed, and the handling actions lack specificity.

[0029] In practical applications, business records from warehouse management, order management, procurement management, transportation management, inventory management, master data management, and production scheduling are first integrated. Inbound, outbound, transfer, return, reservation, in transit, delivery date, substitution relationships, order attribution, and lifecycle status are then standardized into a set of inventory event fragments. Next, effective demand events that truly cause a decrease in available inventory are identified, excluding records that only reflect book corrections or in-warehouse movements. Then, the dormant state of each inventory object since the last effective demand event is updated, forming a demand dormancy age sequence. For frequently used matching materials, substitute materials switching from old to new materials, materials driven by concentrated delivery plans, and materials affected by cross-warehouse orders, demand co-occurrence relationships, substitution migration relationships, release relationships, regional linkage relationships, and supply inhibition relationships are extracted to construct an activity trigger propagation diagram. Finally, the awakening hazard spectrum is estimated by combining the dormant state and trigger relationships to determine which inventory objects have entered the release edge, and demand pulse release is decoded to obtain demand pulse release packages. Finally, the release package, along with the current inventory, reservations, in transit, delivery anomalies, and lifecycle status, is sent to the inventory envelope for joint adjudication. The inventory level result and inventory action result are output. For inventory objects that need to be replenished, transferred, frozen for purchase, cleared or maintained, the corresponding processing is performed respectively. The feedback after execution is continued to be written back to update the activity trigger propagation graph and the demand dormancy age sequence.

[0030] During implementation, a warehousing operation area located in Suzhou Industrial Park was selected as the site, covering daily business data during the continuous operation period from spring to summer. The data processed included multi-source inventory flow records, order attribution records, purchase arrival records, transportation in-transit records, in-warehouse transfer records, inventory correction records, substitution relationship records, lifecycle status records, and delivery arrangement change records. After application, it can be directly observed that this invention can separate previously mixed inventory changes into two categories: those driven by real demand and those not driven by demand. Furthermore, it can distinguish between continuously silent, abnormally silent, soon-to-be-activated, and already released inventory objects. For low-frequency spare parts and substitute materials, this invention no longer only looks at a single outbound shipment, but combines silent trajectories, trigger relationships, and subsequent feedback to determine whether action is needed. Therefore, it is easier to identify materials that appear "quiet on the surface but will actually be pulled," and materials that appear to fluctuate but are actually just being transferred or corrected. For multi-warehouse collaboration scenarios, this invention can identify the risks brought about by regional linkages and supply suppression before demand surges, shifting the original reactive approach to proactive adjudication. Based on the continuous feedback during the implementation period, it can be seen that the present invention has stable feasibility in terms of inventory object identification, risk assessment and action closure, and can improve the accuracy of demand identification, the targeting of inventory actions and the timeliness of internal and external warehouse collaborative processing.

[0031] Table 1. Comparison of Overall Inventory Management Performance for Intermittent Demand Scenarios in Smart Warehousing

[0032] As shown in Table 1, the method of this invention achieved optimal results in core indicators such as the accuracy of valid demand event identification, the F1 score of demand wake-up prediction, the recall rate of stockout warnings, and the accuracy of backlog identification. Specifically, the accuracy of valid demand event identification reached 91.8%, an improvement of 13.2 percentage points compared to Croston, 10.6 percentage points compared to TSB, and 4.3 percentage points compared to TemporalFusion Transformer. This indicates that identifying valid demand events first in warehousing business data, and then separating non-real demand changes such as inventory corrections, transfer offsets, and return reversals from the demand modeling entry point, can significantly reduce demand sample contamination and improve the reliability of subsequent demand status analysis. This is particularly crucial for spare parts, substitute materials, and low-frequency requisition materials, because the outflow of these types of inventory objects is inherently discontinuous; if the demand entry point is impure, subsequent prediction errors will be further amplified.

[0033] In demand wake-up prediction, the method of this invention achieves an F1 score of 88.7%, which is 9.4 percentage points higher than XGBoost, 7.3 percentage points higher than LSTM, and 5.1 percentage points higher than Temporal Fusion Transformer. This result demonstrates that organizing dormancy length, dormancy extension, dormancy deviation, and inactive dormancy states into a demand dormancy age sequence, and combining this with demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply inhibition relationships to construct an activity trigger propagation graph, can more accurately characterize the process of inventory objects transitioning from a dormant state to a wake-up state than simply relying on historical time series or general feature models. Especially in situations where order-driven, substitution switching, and cross-warehouse linkage coexist, traditional time series models are better at fitting continuous trends, while the method of this invention is more suitable for identifying sudden, intermittent, and relationship-triggered demands.

[0034] From the perspective of MAPE (Momentum Per Expectation) prediction, the method of this invention achieves 13.6%, significantly lower than Croston's 24.8%, TSB's 22.9%, LightGBM's 18.9%, and Temporal Fusion Transformer's 17.2%. This result reflects that simply predicting the demand at a future point in time is insufficient to accurately describe the actual requisition process in warehousing scenarios. Further decoding the released inventory objects into demand pulse release packets, providing the median value of the release magnitude, the upper limit value of the release magnitude, the release duration, and the re-release interval, better reflects the actual occurrence of intermittent and clustered demand. For the same inventory object, demand is not always evenly distributed along the timeline, but often manifests as a concentrated requisition after a long period of silence, followed by another round of silence. The method of this invention organizes the prediction chain around this pattern, thus providing a more accurate prediction of release volume compared to actual business performance.

[0035] In terms of risk warning and inventory management, the method of this invention achieves a stockout warning recall rate of 89.4% and a backlog identification accuracy rate of 89.1%, both superior to the comparison algorithms. In particular, compared to Temporal Fusion Transformer, the stockout warning recall rate is improved by 5.3 percentage points, and the backlog identification accuracy rate is improved by 4.5 percentage points. This is because the method of this invention does not directly make inventory judgments based on a single demand forecast value. Instead, it incorporates the demand impulse release package along with current available inventory, in-transit coverage, reserved occupancy, abnormal delivery status, and lifecycle status into a joint inventory envelope for adjudication, forming stockout envelopes, backlog envelopes, and supply disruption envelopes, respectively. In this way, inventory decisions no longer rely solely on a single forecast point but can distinguish between three fundamentally different risks: "imminent stockout," "already stockpiled," and "imminent supply disruption." This improves the accuracy of inventory level classification and makes the selection of replenishment, allocation, frozen procurement, and clearance actions more aligned with the actual situation.

[0036] From the perspective of operational effectiveness, the method of this invention controls the emergency replenishment ratio at 7.4%, significantly lower than Croston's 13.8%, TSB's 12.9%, LSTM's 10.5%, and Temporal Fusion Transformer's 9.9%; the average inventory turnover days are 31.8 days, also better than all the compared algorithms. This indicates that the method of this invention not only performs better in identifying and predicting indicators, but also truly translates into improved inventory management efficiency. The decrease in the emergency replenishment ratio means a stronger ability to proactively move inventory actions, completing replenishment or allocation before risks escalate; the decrease in the average inventory turnover days indicates that the method of this invention has better execution effects in suppressing backlogs, identifying inactive and stagnant states, and promoting the clearance of abnormal inventory. This improvement is not achieved by simply reducing inventory, but by optimizing the inventory structure through the combined effects of demand identification, relationship triggering, risk assessment, and action feedback loops.

[0037] In summary, the method of this invention organizes multi-source warehousing business data into a standardized set of inventory event fragments, uses effective demand events as the sole demand entry point, expresses the silent evolution state of inventory objects using a demand dormancy age sequence, expresses the timeliness propagation characteristics of demand relationships using an active trigger propagation graph, connects the prediction layer and the decision layer using the wake-up danger spectrum results and demand pulse release packets, and then generates inventory level results and inventory action results through joint adjudication of inventory envelopes. By continuously updating the active trigger propagation graph and demand dormancy age sequence using execution feedback, it can simultaneously improve the ability to identify low-frequency demand, the ability to judge demand wake-up, the ability to characterize release volume, and the ability to control inventory risk. As a result, it exhibits higher identification accuracy, faster action response, and better inventory turnover in complex warehousing scenarios.

[0038] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart warehouse inventory management method based on big data analytics, characterized in that, Includes the following steps: Acquire multi-source warehousing business data, align and clean it according to a unified time index, warehouse node identifier, and inventory object identifier, and generate a standardized set of inventory event fragments; Based on a standardized set of inventory event fragments, valid demand events are identified. Each inventory object is updated according to its quiescent state since the last valid demand event, generating a demand dormancy age sequence. Demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply inhibition relationships are extracted to generate an activity trigger propagation graph. Based on the demand dormancy age sequence and the activity trigger propagation map, the marginal risk estimation from dormancy to wakefulness is performed, and the wakefulness risk spectrum results are generated. Based on the wake-up danger spectrum results, the released inventory objects are determined, and demand pulse release decoding is performed on the released inventory objects to generate demand pulse release packets. Based on the demand impulse release packet and the standardized set of inventory event fragments, a joint decision on the inventory envelope is executed to generate inventory level results and inventory action results; Based on inventory level results, inventory action results, and execution feedback, an active trigger propagation graph and a closed loop for updating the demand dormancy age sequence are constructed.

2. The intelligent warehousing and inventory management method based on big data analysis according to claim 1, characterized in that, The generation of the standardized inventory event fragment set specifically includes: It collects original business records from warehouse management systems, order management systems, procurement management systems, transportation management systems, inventory management systems, master data management systems, and project planning management systems, and encapsulates them according to record source and business type to form a set of original warehouse business records; Perform unified field mapping and object primary key parsing on the original warehousing business record set to form a unified field business record set; By performing time merging, node merging, and object merging on a unified set of business records, an indexed set of business records is obtained. Alignment and cleaning are performed on the indexed business record set, including duplicate record elimination, record cancellation pairing, missing field completion, conflict status adjudication, and abnormal number correction, to form a clean business record set; The cleanup business record set is aggregated into fragments according to a unified time index, warehouse node identifier, and inventory object identifier to form a standardized inventory event fragment set.

3. The intelligent warehouse inventory management method based on big data analysis according to claim 1, characterized in that, The generation of the required dormancy age sequence and the activity trigger propagation graph specifically includes: Perform valid demand event identification on each standardized inventory event fragment in the standardized inventory event fragment set to form a valid demand event set; The set of effective demand events is expanded into a time-series record based on warehouse node identifier and inventory object identifier, forming an effective demand time series index table; The quiescent state of each inventory object is updated based on the effective demand time series index table and the standardized inventory event fragment set to form a demand dormancy age sequence. Construct a candidate set of trigger relationships based on a standardized set of inventory event fragments, a set of effective demand events, and a sequence of demand dormancy ages; Perform liveness determination and propagation encapsulation on the candidate set of trigger relationships to generate an live trigger propagation graph.

4. The intelligent warehouse inventory management method based on big data analysis according to claim 3, characterized in that, The candidate set of triggering relationships includes demand co-occurrence relationships, substitution migration relationships, project release relationships, regional linkage relationships, and supply suppression relationships.

5. The intelligent warehouse inventory management method based on big data analysis according to claim 1, characterized in that, The generation of the awakening hazard spectrum results specifically includes: Perform object alignment, time unrolling, and edge type splitting on the demand dormancy age sequence and activity trigger propagation graph to form a hazard estimation sample set; Based on the risk estimation sample set, calculate the dormant base risk value of each inventory object in the future review cycle sequence to obtain the dormant base risk sequence; The positive trigger gain values ​​of the positive active trigger edge set are aggregated in the future review cycle sequence to form a positive trigger gain sequence; The suppression deduction values ​​of the active suppression trigger edge set and the inactivation silencing marker on the future review cycle sequence are aggregated to form the suppression deduction sequence. The wake-up danger candidate sequence is calculated by combining the dormant base danger sequence and the positive trigger gain sequence. Lifetime capping and temporal smoothing are performed on wake-up hazard candidate sequences to generate wake-up hazard spectrum results.

6. The intelligent warehouse inventory management method based on big data analysis according to claim 5, characterized in that, The suppression deduction sequence is for each unified time index in the future review cycle sequence. For each active triggering edge in the set of suppression active triggering edges, the relationship strength, effective start point, and effective end point are read. When the unified time index is located between the effective start point and effective end point of the active triggering edge, the suppression attenuation value is calculated according to the time distance between the unified time index and the unified time index. The relationship strength of all suppression active triggering edges within the effective interval is multiplied by the corresponding suppression attenuation value and then accumulated to obtain the suppression deduction value on the unified time index. When the inactivation silent mark corresponding to the current unified time index is valid, the preset inactivation deduction value is superimposed on the suppression deduction value on the unified time index. All suppression edge deduction values ​​in the future review cycle sequence are arranged in the order of the unified time index to form the suppression deduction sequence. For each unified time index in the future review cycle sequence, the wake-up danger candidate sequence is formed by adding the corresponding dormant base danger value and the positive trigger gain value, subtracting the corresponding suppression deduction value, comparing the calculation result with zero and taking the larger value. The wake-up danger candidate value is then arranged in the order of the unified time index to form the wake-up danger candidate sequence.

7. The intelligent warehouse inventory management method based on big data analysis according to claim 1, characterized in that, The generation of the demand pulse release packet specifically includes: Based on the warehouse node identifier and the inventory object identifier, release candidate screening is performed on each object according to the wake-up hazard spectrum results to form a set of release candidate intervals; The inventory objects in the release status are determined according to the set of release candidate intervals, and a current release feature table is formed. By tracing back historical release segments based on the current release feature table, a historical release sample set is formed; Calculate the sample similarity for each historical release sample in the historical release sample set, perform sample filtering and pulse parameter decoding, and form a release parameter table; The current release feature table and release parameter table are encapsulated into a demand pulse release package.

8. The intelligent warehouse inventory management method based on big data analysis according to claim 1, characterized in that, The generation of the inventory level results and inventory action results specifically includes: Extract records of the same object from the demand pulse release package and the standardized inventory event fragment set to form an inventory decision base table; For each record in the inventory decision base table, perform a shortage envelope calculation. When the shortage envelope value is greater than zero, mark the corresponding record as having a valid shortage envelope. When the shortage envelope value is equal to zero, mark the corresponding record as having an invalid shortage envelope, thus forming the shortage envelope result. Perform backlog envelope calculation around the inventory decision base table to generate backlog envelope results; The supply disruption envelope is calculated by combining the shortage envelope result, the inventory adjudication base table, and the abnormal delivery status to form the supply disruption envelope result; The results of stockout envelope, overstock envelope, and supply disruption envelope are subject to priority joint adjudication to generate inventory level results; Based on the inventory level results and the standardized set of inventory event fragments, execute action mapping to generate inventory action results.

9. The intelligent warehouse inventory management method based on big data analysis according to claim 1, characterized in that, The generation of the updated closed loop specifically includes: Collect subsequent business feedback on inventory actions within the feedback observation window to form an execution feedback record set. Align the execution feedback record set, inventory level results, inventory action results, activity trigger propagation diagram, and demand dormancy age sequence with the execution feedback to form an updated baseline table; The relationship strength, effective range, and failure condition re-evaluation of the activity trigger propagation graph are performed on the updated baseline table to form the updated activity trigger propagation graph. Based on the updated baseline table and the execution feedback record set, the demand dormancy age sequence is reset, accumulated, and marked to form the updated demand dormancy age sequence. Write back the updated activity trigger propagation graph and the updated demand dormancy age sequence to form a closed loop for marginal risk estimation update.

Images