An online and offline integrated retail inventory real-time synchronization method

Abstract

This invention discloses a method for real-time synchronization of online and offline retail inventory, relating to the field of retail information technology. The method includes: constructing a behavioral conversion probability model based on historical store transaction logs and electronic price tag touch event logs, according to product SKUs and business hours, to obtain initial probability parameters and decay rate parameters; real-time detection of picking signals in store shelf areas, combined with real-time environmental correction factors to generate pre-reservation records including a pre-reservation validity period; sending the pre-reservation records to the cloud inventory center to deduct online sellable inventory and store them in a pre-reservation record pool; monitoring the transaction logs generated by the POS system, and confirming the conversion of the pre-reservation record to a formal sales record when preset spatiotemporal matching conditions are met; if no confirmation is made within the time limit, canceling the pre-reservation record and restoring the inventory. This invention, through inventory pre-reservation triggered by picking signals, helps improve the real-time performance of online and offline inventory synchronization.

Description

Technical Field

[0001] This invention relates to the field of retail information technology, and more specifically, to a method for real-time synchronization of online and offline retail inventory. Background Technology

[0002] In a retail environment that integrates online and offline channels, the accuracy and timeliness of in-store inventory data impact both customer experience and business operations. Currently, common inventory synchronization methods rely on deterministic events such as POS transaction records, periodic manual inventory checks, or RFID exit detection to trigger inventory updates. This means that the online store only deducts the corresponding inventory after the item is checked out or confirmed as left. However, there is an inherent time lag between a customer picking up an item from the shelf and completing the checkout. During this time, the online system may still mark the item as available for sale, potentially leading to online orders selling items that have been picked up but not yet checked out, resulting in overselling.

[0003] Some solutions shorten information transmission time by increasing data transmission frequency and using message queues or distributed database transaction mechanisms. However, these solutions do not change the fundamental logic that the inventory update trigger point is after checkout, and the perceived latency problem still exists. Other solutions attempt to deploy RFID reader arrays in stores to track product movement in real time. However, this method requires significant hardware investment and suffers from problems such as read conflicts and difficulties in concurrent processing in tag-dense environments. Therefore, a real-time online-offline integrated retail inventory synchronization method is proposed to address the above problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a real-time online-offline integrated retail inventory synchronization method, which aims to solve the problem in the prior art where the online inventory update time lags behind the offline pickup time due to reliance on deterministic events to trigger inventory synchronization.

[0005] To achieve the above objectives, the present invention provides a method for real-time synchronization of online and offline retail inventory, comprising the following steps: Step S1: Based on the store's historical cashier transaction records and electronic price tag touch event logs, construct a behavioral conversion probability model by product SKU and business hours. This behavioral conversion probability model includes initial probability parameters. and decay rate parameter The initial probability parameter The decay rate parameter represents the probability that an item will be purchased after an electronic price tag is touched. This model represents the rate at which the purchase probability decays over time. By analyzing the temporal correlation between historical transactions and touch events, it extracts the time patterns of different goods from being picked up to the completion of the purchase during different business hours, providing a probabilistic basis for subsequent calculations of the pre-reservation validity period.

[0006] Furthermore, the construction of the behavior conversion probability model includes: for each cashier transaction, tracing back to the last electronic price tag touch event of the product SKU in the corresponding shelf area before the transaction, and recording the time interval between the touch time and the transaction completion time; For the sample set of time intervals recorded for each product SKU during each business period, an exponential decay function is used. The initial probability parameters were obtained by fitting. and decay rate parameter By establishing a sample of time intervals between pickup and transaction through backtracking, and using an exponential decay function to fit the probability decay curve, the probability model can quantitatively describe the decay of the transaction probability over time after pickup.

[0007] Step S2: Real-time detection of picking events in the store's shelf area. These picking events are determined based on continuous touch signals of electronic price tags within a preset short window, or by a combination of the dwell signal from a customer flow counting device in front of the shelf and the weight change signal from the shelf weight sensor. The physical behavior of consumers picking up goods is identified through the combined characteristics of continuous touches of electronic price tags or customer flow and weight changes, serving as the trigger condition for inventory pre-positioning.

[0008] Furthermore, the detection of the picking-up sign event includes: when the electronic price tag is touched twice in succession within 2 seconds, it is determined as a picking-up sign event, which corresponds to the typical behavior of a consumer picking up and viewing the goods; Alternatively, when the customer flow counting device detects that a person lingers in front of the shelf for more than 5 seconds and the shelf weight sensor detects a weight decrease exceeding a preset threshold, it is determined as a picking event. This combination of features corresponds to the typical behavior of a consumer taking goods from the shelf. The two determination methods complement each other, covering the identification of picking behavior under different sensor configurations.

[0009] Step S3: When a pickup indication event is detected, the initial probability parameters of the corresponding product SKU during the current business hours are used. and decay rate parameter Pre-occupancy records are generated by combining real-time environmental correction factors.

[0010] The pre-reservation record includes the pre-reservation validity period. The pre-occupancy validity period Take probability decay time With inventory constraint duration The smaller value in, where , To preset the release probability threshold, , To preset the maximum validity period, This refers to the current available inventory quantity for this product SKU. Set the preset safety stock quantity.

[0011] Pre-occupancy validity period By combining the decay duration based on historical behavior probability with the constraint duration based on real-time inventory pressure, the pre-occupancy duration reflects both the time distribution characteristics of consumer transaction probability and is adjusted by the current available inventory level.

[0012] Furthermore, the real-time environmental correction factor includes a passenger flow correction factor. When the real-time customer flow in the shelf area exceeds the historical average customer flow for that business period, the decay rate parameter will be adjusted. Multiplied by the aforementioned passenger flow correction factor The passenger flow correction factor The value of is greater than 1.

[0013] Passenger flow correction factor The decay rate parameter is adjusted based on real-time passenger flow density. When passenger flow is dense, the probability decay rate is accelerated to shorten the pre-occupancy validity period, reflecting the behavioral characteristic that consumers tend to make decisions in crowded environments.

[0014] Based on this, the pre-occupancy validity period The method for determining it is as follows: , , , in, To preset the release probability threshold, To preset the maximum validity period, This refers to the current available inventory quantity for this product SKU. Preset safety stock quantity. Probability decay time. Release probability threshold Initial probability parameters Together with the decay rate adjusted for passenger flow factors, it represents the time required for the predicted transaction probability to drop to the release threshold; inventory constraint duration. Constrained by the ratio of maximum shelf life, current inventory, and safety stock, the pre-occupancy period is shortened proportionally when the inventory falls below the safety line.

[0015] Furthermore, the real-time environment correction factor also includes an online popularity correction factor. When the real-time pageviews of the product SKU on the online store exceed the historical average pageviews for that business period, the initial probability parameter is adjusted. Multiply by a preset factor greater than 1 to obtain the corrected initial probability parameters. .

[0016] The online popularity correction factor adjusts the initial probability parameter upwards based on changes in online browsing popularity, so that the calculation of the pre-occupancy validity period can reflect the potential impact of increased online traffic on the probability of offline transactions.

[0017] Based on this, the pre-occupancy validity period The method for determining it is as follows: , , , in, These are the corrected initial probability parameters. To preset the release probability threshold, To preset the maximum validity period, This refers to the current available inventory quantity for this product SKU. This is the preset safety stock quantity. The initial probability parameters are adjusted for online popularity. By incorporating the probability decay time calculation, the pre-occupancy validity period is adjusted according to changes in online attention, while the inventory constraint time is also considered. Continue to impose inventory-level-based restrictions on pre-occupancy periods.

[0018] Step S4: The pre-reservation record is sent to the cloud inventory center, which then deducts the online available inventory quantity of the product SKU based on the pre-reservation record and stores the pre-reservation record in the pre-reservation record pool. Through the pre-deduction triggered by the pickup indication, the online available inventory is updated as soon as the consumer picks up the product, shortening the time difference between physical pickup and changes in online inventory status.

[0019] Furthermore, the operation of the cloud inventory center deducting the online salable inventory quantity in advance includes: Current available inventory quantity for this product SKU If the value is greater than 0, decrease the available inventory by 1 and increase the reserved locked inventory by 1; Current available inventory quantity for this product SKU If the value is 0, the reserved record is rejected.

[0020] By linking the available inventory quantity with the pre-reserved locked inventory quantity, the online store displays the remaining quantity after deducting the pre-reserved quantity when querying available inventory, making the inventory seen by consumers closer to the physical reality; rejecting pre-reservation when the inventory is zero avoids the generation of invalid pre-reservations.

[0021] Step S5: Monitor the transaction logs generated by the store's POS system. When the product SKU and transaction completion time in the transaction log meet the preset spatiotemporal matching conditions with the pre-reserved record, send a confirmation command to the cloud inventory center to convert the pre-reserved record from the pre-reserved record pool into a formal sales record. Through multi-dimensional matching of POS transactions and pre-reserved records in terms of products, time, and store area, the previously deducted pre-reserved items are confirmed as actual sales, completing the closed loop of inventory synchronization.

[0022] Furthermore, the preset spatiotemporal matching conditions include: the product SKU in the cashier transaction log is the same as the product SKU in the pre-reservation record; the transaction completion time of the cashier transaction log is between the trigger time of the pre-reservation record and the expiration time of the pre-reservation validity period. The area of ​​the POS terminal corresponding to the cashier transaction and the shelf area associated with the pre-reservation record belong to the same or adjacent preset store area. Through the joint constraints of three dimensions—product, time window, and store area—the cashier transaction and the pre-reservation record are accurately associated, avoiding mismatches between pre-reservation records in different shelf areas or time periods.

[0023] Furthermore, when multiple pre-occupied records simultaneously meet the preset spatiotemporal matching conditions, the first sorting condition is based on the shortest spatial distance between the shelf area and the area where the checkout terminal is located, and the second sorting condition is based on the smallest absolute value of the difference between the time interval between the trigger time and the transaction completion time and the historical average transaction duration of the product SKU. This two-tiered sorting rule, prioritizing spatial distance and then temporal matching, helps to accurately identify the pre-occupied record corresponding to the actual transaction in scenarios where multiple customers are simultaneously picking up goods.

[0024] Step S6: For those exceeding the pre-occupancy validity period For unconfirmed pre-reservation records, a cancellation command is sent to the cloud inventory center to restore the online available inventory quantity for that product SKU. In scenarios where no purchase occurs after pickup, the pre-reservation is automatically cancelled and the inventory is replenished after a timeout, ensuring that the online inventory status returns to consistency with the physical reality.

[0025] Furthermore, the edge computing nodes periodically aggregate confirmed pre-sale records that have been converted into formal sales records and cancelled pre-sale records, along with their corresponding pickup indicator events, as feedback samples. Based on the feedback samples within the most recent preset number of days, the initial probability parameters for each product SKU in each business period are recalculated. and decay rate parameter The updated parameters are then distributed to the store's edge computing nodes. This self-updating mechanism allows the behavioral conversion probability model to continuously evolve as the store's actual operating data changes, helping to improve the accuracy of the pre-occupancy validity period.

[0026] The technical effects and advantages of this invention are as follows: (1) This invention quantifies pickup symptom events into pre-reservation records with a pre-reservation validity period by constructing a behavioral conversion probability model based on the store's historical cashier transaction flow and electronic shelf tag touch event logs. When an electronic shelf tag is detected to be touched continuously within a short window, or when a customer flow counting device detects that a person is lingering in front of the shelf and the shelf weight decreases, the edge computing node calculates the pre-reservation validity period based on the initial probability parameter and decay rate parameter of the product corresponding to the current business period, combined with the customer flow correction factor or online popularity correction factor, generates a pre-reservation record, and sends it to the cloud. The cloud inventory center then deducts the online available inventory quantity accordingly. This process moves the trigger point for inventory synchronization from the checkout completion time to the pickup action time, which helps to shorten the time difference between online inventory status and offline physical inventory changes.

[0027] (2) The calculation of the pre-occupancy validity period integrates two factors: probability decay duration and inventory constraint duration. The probability decay duration is determined by the preset release probability threshold, initial probability parameter, and decay rate parameter, reflecting the pattern of the probability of the product being eventually sold after being picked up during the business hours in historical data. The inventory constraint duration is jointly constrained by the preset maximum validity period, the ratio of the current available inventory quantity to the preset safety stock quantity. This two-factor dynamic determination of the validity period makes the pre-occupancy duration both related to consumer behavior patterns and adjusted by real-time inventory levels.

[0028] (3) After the pre-reservation record is generated, the edge computing node continuously monitors the cashier transaction flow generated by the POS system. When the product SKU in the transaction flow is the same as the pre-reservation record, the transaction completion time is within the pre-reservation validity period, and the area where the POS terminal is located and the shelf area belong to the same or adjacent store area, the pre-reservation record is confirmed and converted into a formal sales record. Pre-reservation records that are not confirmed after the pre-reservation validity period are automatically canceled by the cloud, and the corresponding online available inventory quantity is restored accordingly. This confirmation and cancellation closed loop helps to ensure the final consistency of inventory accounts.

[0029] (4) Edge computing nodes periodically summarize the confirmation and cancellation results of pre-occupancy records as feedback samples and upload them to the cloud analysis module. The cloud analysis module recalculates the initial probability parameters and decay rate parameters according to the product SKU and business hours and issues updates. The self-updating mechanism of these parameters enables the behavior conversion probability model to continuously evolve with changes in the actual operating data of the store, which helps to improve the accuracy of the pre-occupancy validity period.

[0030] (5) This invention can be implemented by relying on the existing electronic price tags, customer flow counting equipment and cash register system in the store. There is no need to deploy high-density RFID or visual recognition devices. It has low transformation cost and good engineering feasibility. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the overall implementation of the method of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example 1 As attached Figure 1 The method shown is a real-time online-offline integrated retail inventory synchronization method applied to a retail system with offline physical stores and an online mall. The system includes electronic shelf labels, customer flow counting devices, shelf weight sensors, POS systems, and edge computing nodes deployed at the store level, as well as a cloud-based inventory center and cloud-based analytics module deployed in the cloud.

[0034] Electronic shelf labels are deployed on store shelves. Each electronic shelf label is associated with a single product SKU and is used to sense consumer touch operations and generate touch signals.

[0035] Customer flow counting devices are deployed in various shelf areas of the store to count the number of people in the area in real time.

[0036] Shelf weight sensors are deployed under the load-bearing shelves to detect changes in the total weight of goods on the shelves in real time.

[0037] The POS system is the store's existing POS system, used to generate real-time transaction records.

[0038] An edge computing node is a local server or industrial control computer with computing and storage capabilities. It communicates with the above-mentioned devices via a local area network and with the cloud inventory center and cloud analysis module via the Internet.

[0039] The cloud-based inventory center is used to maintain a unified inventory ledger for all stores and to provide inventory query and deduction services to the online mall.

[0040] The cloud analytics module is used to collect feedback samples uploaded by edge computing nodes in various stores and perform update calculations for model parameters.

[0041] The steps of the method of the present invention will be described in detail below.

[0042] S1. Based on the store's historical cashier transaction records and electronic price tag touch event logs, construct a behavioral conversion probability model by product SKU and business hours. This behavioral conversion probability model includes initial probability parameters. and decay rate parameter The initial probability parameter The decay rate parameter represents the probability that an item will be purchased after an electronic price tag is touched. It represents the rate at which the probability of purchase decays over time.

[0043] The edge computing node retrieves all historical transaction records from the POS system within a preset number of historical days, and retrieves all electronic shelf label touch event logs within the same time period from the electronic shelf label management system. The preset number of historical days is a pre-configured value, such as 90 days.

[0044] Each checkout transaction record includes the product SKU, transaction completion time, checkout terminal identifier, and transaction quantity. Each electronic shelf tag touch event log record includes the product SKU, touch time, and shelf area identifier where the shelf tag was located.

[0045] Edge computing nodes perform backtracking and correlation operations on each checkout transaction. Specifically, for each checkout transaction, the product SKU and checkout terminal identifier are extracted. The edge computing nodes pre-store a mapping table between checkout terminals and store area zones, and retrieve the store area zone where the checkout terminal is located based on the checkout terminal identifier.

[0046] A store zone is defined as a physical area that is divided into several non-overlapping zones. Each zone contains at least one shelf area and at least one POS terminal, and each shelf area belongs to a unique zone. By querying the relationship table between shelf areas and zones, the set of all shelf area identifiers contained within a given zone can be determined.

[0047] From the electronic shelf tag touch event log, select touch event records that simultaneously meet the following conditions: same product SKU, touch time earlier than transaction completion time, and shelf area identifier of the shelf tag belonging to the aforementioned shelf area identifier set. Sort the selected touch event records in descending order of touch time, and take the record with the latest touch time as the last picking indication event corresponding to that transaction.

[0048] If a matching touch event record is found, the time interval between the touch time of the event and the transaction completion time is calculated, and this time interval is saved as a sample record. If no matching event record is found, the transaction is discarded and no sample record is generated.

[0049] After completing the backtracking and association of all cashier transactions, the edge computing node obtains a time interval sample set. The edge computing node groups this sample set by product SKU and business hours. The business hours are divided as follows: the entire day's business hours are divided into consecutive time periods of fixed duration, with the fixed duration being a pre-configured value, such as 30 minutes.

[0050] For each product SKU in each business period, extract all time interval samples belonging to that group. If a product SKU has no time interval samples in a certain business period, then use the overall parameters of that product SKU in all business periods; if there are still no overall parameters, then use the preset default parameters.

[0051] For each product SKU, based on the time interval sample set for each business period, the edge computing nodes use an exponential decay function for curve fitting. The form of the exponential decay function is as follows: ,in Indicates a time interval variable. Indicates the initial probability parameter. This represents the decay rate parameter.

[0052] The fitting process uses the least squares method, specifically: the time interval values ​​in the sample set are arranged in ascending order, and the empirical cumulative transaction probability corresponding to each time interval value is calculated. The empirical cumulative transaction probability is the proportion of the number of samples in the sample set that are less than or equal to the time interval value to the total number of samples. Using the time interval as the independent variable and the empirical cumulative transaction probability as the dependent variable, the least squares method is used to estimate the parameters of the exponential decay function, and the initial probability parameters of the product SKU corresponding to this business period are obtained. and decay rate parameter The value.

[0053] Simultaneously, the edge computing nodes calculate the arithmetic mean of the time interval values ​​in the sample set, which is taken as the historical average transaction duration of the product SKU during that business period, denoted as... .

[0054] Edge computing nodes will correspond to all business hours of all product SKUs. , as well as Stored in a local parameter table. Each record in the parameter table contains the following fields: Product SKU, Business Hours, and Initial Probability Parameters. Attenuation rate parameter Historical average transaction duration .

[0055] Edge computing nodes also calculate and store the historical average number of customers in each shelf area during each business period, as well as the historical average number of views for each product SKU during each business period. The specific calculation method is to calculate the arithmetic mean of the customer flow data or view data for the same period within the past preset statistical days.

[0056] S2. Real-time detection of picking events in the store shelf area. The picking events are determined based on continuous touch signals of electronic price tags within a preset short window, or based on the dwell signal of the customer flow counting device in front of the shelf and the weight change signal of the shelf weight sensor.

[0057] The detection of pickup warning events can be performed using one of the following two methods or in parallel.

[0058] In the implementation method that uses continuous touch signals from electronic shelf labels, each electronic shelf label monitors its own touch sensor status in real time during store operations. When a consumer touches an electronic shelf label, the label generates a touch signal and sends it to the edge computing node via the local area network. The touch signal includes the electronic shelf label identifier, the product SKU, and a timestamp of the touch time.

[0059] The edge computing node maintains a temporary record of touch signals for each electronic shelf label in its memory cache. The temporary record contains the touch time of the most recent touch signal and the product SKU, and is set to have a validity period of 2 seconds. If no subsequent touch signal matches within 2 seconds, it will be automatically cleared.

[0060] When an edge computing node receives a touch signal, it checks whether there is a valid temporary record of the touch signal in the memory cache for the electronic shelf label. If there is, and the time interval between the current touch moment and the touch moment in the temporary record does not exceed 2 seconds, the edge computing node determines that a picking event has occurred, records the product SKU corresponding to the picking event, records the touch moment in the temporary record as the trigger time, and records the shelf area where the electronic shelf label is located as the event occurrence area.

[0061] If no valid temporary record exists, the edge computing node will write the touch time and product SKU of this touch signal into the temporary record corresponding to the electronic price tag, and wait for subsequent touch signals to complete the judgment.

[0062] In the implementation method that uses both customer flow counting equipment and shelf weight sensors for determination, during store operation, the customer flow counting equipment reports the real-time number of people in each shelf area to the edge computing node at a fixed frequency, and the shelf weight sensors report the current total weight value of each shelf to the edge computing node at a fixed frequency.

[0063] Edge computing nodes maintain a status monitoring window for each shelf area. When the number of people in a certain shelf area reported by the customer flow counting device changes from zero to greater than zero, the edge computing node records this moment as the time when the person entered and begins to continuously monitor the number of people and changes in shelf weight in that area.

[0064] The edge computing node continuously records the weight readings of the corresponding weight sensor for the shelf, and calculates the weight baseline value using the moving average method. The sliding window size is set to 10 recent weight readings, and the arithmetic mean of these 10 readings is used as the moving average weight value.

[0065] When a person is detected to have lingered in front of the shelf for more than 5 seconds, the edge computing node calculates the difference between the current moving average weight value and the moving average weight value at the time the person entered. If the difference is negative and its absolute value is greater than a preset weight change threshold, a picking indicator event is determined to have occurred.

[0066] The preset weight change threshold is set as the average weight of each item displayed on the shelf multiplied by a preset ratio coefficient, which is pre-configured in the edge computing node, for example, 0.5. The average weight of each item is obtained by querying the basic product information table pre-maintained in the edge computing node. The edge computing node records the product SKU corresponding to the picking event, records the moment when the weight begins to change as the trigger time, and records the area where the shelf is located as the event occurrence area.

[0067] When the two implementation methods described above are executed in parallel and a pickup indication event for the same product SKU is detected simultaneously within 2 seconds, the edge computing node will use the earlier trigger time as the standard and generate only one pickup indication event record; if the trigger times are the same, one of the records will be randomly selected.

[0068] When any implementation determines that a pickup indication event has occurred, the edge computing node passes the product SKU, trigger time, and shelf area corresponding to the pickup indication event as input parameters to the processing flow of step S3.

[0069] S3. When a pickup signal event is detected, the initial probability parameters of the corresponding product SKU during the current business hours are used. and decay rate parameter A pre-occupancy record is generated by combining real-time environmental correction factors, and the pre-occupancy record includes the pre-occupancy validity period. The pre-occupancy validity period Take probability decay time With inventory constraint duration The smaller value in, where , To preset the release probability threshold, , To preset the maximum validity period, This refers to the current available inventory quantity for this product SKU. Set the preset safety stock quantity.

[0070] After receiving a pickup indication event, the edge computing node determines the business hours to which the current time belongs and reads the initial probability parameters of the product SKU corresponding to that business hours from the local parameter table. and decay rate parameter .

[0071] Preset release probability threshold Preset maximum validity period Preset safety stock quantity These are all fixed parameters pre-configured on the edge computing nodes according to business needs.

[0072] Real-time environmental correction factors include passenger flow correction factors. And at least one of the online popularity correction factors. The processing methods under different combinations of correction factors are described below.

[0073] When applying passenger flow correction factor The processing steps are as follows: The edge computing node obtains the real-time customer flow of the shelf area from the customer flow counting device at the current moment, and reads the historical average customer flow of the shelf area during the business hours from the local parameter table. The real-time customer flow is compared with the historical average customer flow; if the real-time customer flow is greater than the historical average customer flow, a customer flow correction factor is applied. Set to a pre-configured fixed value, such as 1.2; otherwise, use the passenger flow correction factor. Set to 1.

[0074] Edge computing nodes query the current available inventory quantity of the product SKU from the cloud inventory center. Read the preset safety stock quantity Preset release probability threshold and preset maximum validity period .

[0075] Pre-occupancy validity period The calculation process is as follows: calculate the probability decay time. ;Calculate inventory constraint duration ;Pick .like If the value is greater than or equal to 1, it is determined that no pre-reservation record will be generated, and the process will terminate. At that time, the formula degenerates into its uncorrected form.

[0076] When applying the online popularity correction factor, the processing procedure is as follows: The edge computing node obtains the real-time pageviews of the product SKU during the current business hours from the online store system, and reads the historical average pageviews of the product SKU during the business hours from the local parameter table.

[0077] The real-time pageviews are compared with the historical average pageviews. When the real-time pageviews are greater than the historical average pageviews, the initial probability parameter is adjusted. Updated to the original The value after multiplying by a pre-configured fixed factor, such as 1.3; otherwise, it remains unchanged. constant.

[0078] Edge computing nodes query the current available inventory quantity of the product SKU from the cloud inventory center. Read the preset safety stock quantity Preset release probability threshold and preset maximum validity period .

[0079] Pre-occupancy validity period The calculation process is as follows: calculate the probability decay time. ;Calculate inventory constraint duration ;Pick .like If the value is greater than or equal to 1, it is determined that no pre-reservation record will be generated, and the process will terminate.

[0080] When passenger flow correction factor is applied simultaneously When adjusting for online popularity, edge computing nodes determine the passenger flow adjustment factor according to the aforementioned rules. and the updated initial probability parameters .

[0081] Pre-occupancy validity period The calculation process is as follows: calculate the probability decay time. ;Calculate inventory constraint duration ;Pick .like If the value is greater than or equal to 1, it is determined that no pre-reservation record will be generated, and the process will terminate.

[0082] Edge computing nodes calculate the pre-occupancy validity period. Then, a pre-reservation record is generated. The pre-reservation record contains a unique identifier for the pre-reservation record, the product SKU, the trigger time, and the pre-reservation validity period. Expiry date, shelf area, initial probability parameters numerical values, decay rate parameters The numerical value, the available inventory quantity at the time of generation The expiration date is the trigger time plus the pre-occupied validity period. .

[0083] The edge computing node saves the pre-occupancy record in its local cache and sends the complete content of the pre-occupancy record to the cloud inventory center.

[0084] S4. Send the pre-reservation record to the cloud inventory center, and the cloud inventory center will deduct the online available inventory quantity of the product SKU based on the pre-reservation record and store the pre-reservation record in the pre-reservation record pool.

[0085] After receiving the pre-reservation record from the edge computing node, the cloud inventory center parses out the product SKU and the unique identifier of the pre-reservation record. The cloud inventory center then queries the inventory database to determine the current available inventory quantity of that product SKU.

[0086] If the available inventory quantity is greater than 0, the cloud inventory center will execute an atomic transaction using the database transaction mechanism. This atomic transaction includes the following operations: decrementing the value of the available inventory quantity field by 1; incrementing the value of the reserved locked inventory quantity field by 1; and inserting a record into the reserved record pool data table. This record contains all the fields of the reserved record and a status field, with the initial value of the status field set to valid.

[0087] If the available inventory quantity is 0, the cloud inventory center rejects the pre-reserved record and returns a rejection response to the edge computing node, containing the unique identifier of the rejected pre-reserved record. Upon receiving the rejection response, the edge computing node deletes the corresponding pre-reserved record from its local cache based on the unique identifier.

[0088] When an online store receives a request from a consumer to inquire about the available inventory of a particular SKU, it initiates a query with the cloud-based inventory center. The available inventory quantity returned by the cloud-based inventory center is the value of the "Currently Available Inventory Quantity" field minus the value of the "Pre-reserved Locked Inventory Quantity" field.

[0089] S5. Monitor the cash register transaction flow generated by the store's cash register system. When the product SKU and transaction completion time in the cash register transaction flow meet the preset spatiotemporal matching conditions with the pre-reserved record, send a confirmation instruction to the cloud inventory center to convert the pre-reserved record from the pre-reserved record pool into a formal sales record.

[0090] The edge computing nodes and the POS system establish a data channel via message queues or application programming interfaces (APIs). Each time the POS system completes a transaction, it generates a transaction log and pushes it to the edge computing nodes in real time. Each transaction log includes the product SKU, transaction completion time, POS terminal identifier, and transaction quantity.

[0091] After receiving the transaction log, the edge computing node performs a matching search among all valid pre-registered records in its local cache. The matching search is based on the following condition: the product SKU in the transaction log is the same as the product SKU in the pre-registered record. The transaction completion time of the cash register transaction is between the trigger time and the expiration time of the pre-reservation record; the area where the cash register terminal corresponding to the cash register transaction is located and the shelf area associated with the pre-reservation record belong to the same preset store area, or belong to two store areas that are pre-configured as adjacent.

[0092] The store's physical space is divided into several non-overlapping zones. Each zone has a unique identifier, and each shelf area and each POS terminal belongs to a single zone. These affiliations are stored in the configuration table of the edge computing nodes. The correspondence between adjacent zones is pre-configured in the edge computing nodes.

[0093] When multiple pre-occupied records simultaneously meet the above matching conditions, the edge computing node selects one pre-occupied record as the matching object according to the following sorting rules: First, compare the spatial distance between the shelf area and the area where the POS terminal is located, with the smaller spatial distance value taking priority; when the spatial distance values ​​are the same, calculate the time interval between the trigger time and the transaction completion time and the historical average transaction duration of the product SKU. The absolute value of the difference is used, and the one with the smaller absolute value of the difference takes priority.

[0094] Spatial distance values ​​are obtained by querying a distance matrix table pre-stored in the edge computing nodes, representing the distance between shelf areas and POS terminal areas. This distance matrix table is pre-measured and configured based on the store's floor plan, and stores the spatial distance values ​​between any shelf area and the POS terminal area. Historical average transaction time. It is obtained by querying the parameter table stored in step S1.

[0095] Upon successful matching, the edge computing node sends a confirmation command to the cloud inventory center. This command contains the unique identifier of the successfully matched pre-reserved record. Upon receiving the confirmation command, the cloud inventory center searches for the corresponding record in the pre-reserved record pool based on the unique identifier. If the record exists and is valid, an atomic operation is performed: Update the status field of the reserved record to "confirmed"; Decrease the value of the pre-reserved locked inventory quantity field by 1; Insert a new sales record into the official sales record table.

[0096] S6. For those exceeding the pre-occupancy validity period For any unconfirmed pre-reservation records, a cancellation command is sent to the cloud inventory center to restore the online available inventory quantity for that product SKU.

[0097] When an edge computing node starts, it initiates a background timed scanning task that iterates through all pre-occupied records stored in its local cache every 10 seconds. For each pre-occupied record, the edge computing node obtains the current system time and compares it with the expiration time field in that record.

[0098] If the current system time is later than the expiration date, and the pre-reservation record is valid, the edge computing node generates a cancellation command containing the unique identifier of the pre-reservation record to be cancelled. The edge computing node sends the cancellation command to the cloud inventory center. If the transmission is successful, the corresponding pre-reservation record is deleted from the local cache; if the transmission fails, it will retry sending in the next scanning cycle.

[0099] After receiving a cancellation command, the cloud inventory center searches for the corresponding record in the pre-reserved record pool based on the unique identifier of the pre-reserved record. If the record exists and its status field is valid, the cloud inventory center executes an atomic transaction: Increment the value of the available inventory quantity field for the corresponding product SKU by 1; Decrease the value of the pre-reserved locked inventory quantity field by 1; Update the status field of the reserved record to "revoked", or delete the record directly from the reserved record pool.

[0100] The quantity of available inventory displayed on the online store will be restored in tandem with the increase in the quantity of available inventory mentioned above.

[0101] S7. Periodically summarize the pre-sale records that have been confirmed and converted into formal sales records, as well as the pre-sale records that have been cancelled, and their corresponding pickup indication events as feedback samples; recalculate the initial probability parameters of each product SKU in each business period based on the feedback samples within the most recent preset number of days. and decay rate parameter and update the initial probability parameters and decay rate parameter Distribute to the edge computing nodes of the store.

[0102] Edge computing nodes perform a summary task after the end of each business day. The edge computing nodes traverse the processing logs of all processed pre-occupied records in local storage for the day.

[0103] For a pre-occupancy record with a confirmed status, the edge computing node extracts the following information as a positive feedback sample: product SKU, the business period to which the pickup symptom event is triggered, and the actual time interval value corresponding to the pre-occupancy record. The actual time interval value is the difference between the trigger time and the transaction completion time when the matching confirmation is performed in step S5.

[0104] For a pre-booking record with a cancelled status, the edge computing node extracts the following information as a negative feedback sample: the product SKU, the business hours during which the pickup warning event was triggered, and the pre-booking validity period in the pre-booking record. Used as a reference value for time intervals.

[0105] Edge computing nodes package and upload all positive and negative feedback samples for the day to the cloud analytics module. The cloud analytics module collects and stores the feedback samples uploaded by each store, retaining the sample data for the most recent preset number of days, such as the last 30 days. Sample data exceeding the retention period is automatically deleted.

[0106] After completing daily sample collection, the cloud-based analytics module groups all feedback samples within the retention period by product SKU and business hours. For each product SKU grouped within a specific business hour, if there are no feedback samples in that group, the group's initial probability parameters are maintained. and decay rate parameter constant.

[0107] If there are feedback samples within this group, let the number of positive feedback samples within this group be . The number of negative feedback samples is .like If the value is 0, then the original initial probability parameters are retained. No change, no update. If the value is greater than 0, then the updated initial probability parameter .

[0108] Simultaneously, the actual time interval values ​​of all positive feedback samples within the group are extracted to form a new time interval sample set. If the number of positive feedback samples within the group is greater than or equal to the preset minimum fitted sample number, then the least squares method is used to fit the exponential decay function. The fitting process involves updating the initial probability parameters. As a known constant, only the decay rate parameter is fitted. The fitted decay rate parameter is used as ; If the number of positive feedback samples in the group is less than the preset minimum fitted sample number, the original decay rate parameter is retained. Unchanged, that is The minimum number of fitted samples is preset to a pre-configured value, such as 5.

[0109] The cloud-based analytics module will calculate all the results. and The parameters are packaged into a parameter update file and distributed to the edge computing nodes of each store. After receiving the parameter update file, the edge computing nodes update the initial probability parameters of the corresponding product SKU and business hours in the parameter table one by one. and decay rate parameter The updated parameters will take effect in the handling of subsequent pickup warning events.

[0110] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time synchronization of online and offline retail inventory, characterized in that, include: S1. Based on the store's historical cashier transaction records and electronic price tag touch event logs, construct a behavioral conversion probability model by product SKU and business hours. This behavioral conversion probability model includes initial probability parameters. and decay rate parameter The initial probability parameter The decay rate parameter represents the probability that an item will be purchased after an electronic price tag is touched. Characterizes the rate at which the probability of purchase decays over time; S2. Real-time detection of picking events in the store shelf area. The picking events are determined based on the continuous touch signals of electronic price tags within a preset short window, or based on the dwell signal of the customer flow counting device in front of the shelf and the weight change signal of the shelf weight sensor. S3. When a pickup signal event is detected, the initial probability parameters of the corresponding product SKU during the current business hours are used. and decay rate parameter A pre-occupancy record is generated by combining real-time environmental correction factors, and the pre-occupancy record includes the pre-occupancy validity period. The pre-occupancy validity period Take probability decay time With inventory constraint duration The smaller value in, where , To preset the release probability threshold, , To preset the maximum validity period, This refers to the current available inventory quantity for this product SKU. Preset safety stock quantity; S4. Send the pre-reservation record to the cloud inventory center, and the cloud inventory center will first deduct the online available inventory quantity of the product SKU based on the pre-reservation record and store the pre-reservation record in the pre-reservation record pool. S5. Monitor the cashier transaction flow generated by the store's cashier system. When the product SKU and transaction completion time in the cashier transaction flow meet the preset spatiotemporal matching conditions with the pre-reserved record, send a confirmation instruction to the cloud inventory center to convert the pre-reserved record from the pre-reserved record pool into a formal sales record. S6. For those exceeding the pre-occupancy validity period For any unconfirmed pre-reservation records, a cancellation command is sent to the cloud inventory center to restore the online available inventory quantity for that product SKU.

2. The method for real-time synchronization of online and offline retail inventory according to claim 1, characterized in that, The construction of the behavior conversion probability model in step S1 includes: For each cashier transaction, trace back to the last electronic price tag touch event of the product SKU in the corresponding shelf area before the transaction, and record the time interval between the touch time and the transaction completion time; For the sample set of time intervals recorded for each product SKU during each business period, an exponential decay function is used. The initial probability parameters were obtained by fitting. and decay rate parameter .

3. The method for real-time synchronization of online and offline retail inventory according to claim 2, characterized in that, The real-time environmental correction factor mentioned in step S3 includes the passenger flow correction factor. When the real-time customer flow in the shelf area exceeds the historical average customer flow for that business period, the decay rate parameter will be adjusted. Multiplied by the aforementioned passenger flow correction factor The passenger flow correction factor The value of is greater than 1.

4. The method for real-time synchronization of online and offline retail inventory according to claim 2, characterized in that, The real-time environment correction factor mentioned in step S3 includes an online popularity correction factor. When the real-time pageviews of the product SKU on the online store are greater than the historical average pageviews during the business hours, the initial probability parameter is adjusted. Multiply by a preset factor greater than 1 to obtain the corrected initial probability parameters. .

5. The method for real-time synchronization of online and offline retail inventory according to claim 3, characterized in that, The pre-occupancy validity period mentioned in step S3 The method for determining it is as follows: ， ， ， in, To preset the release probability threshold, To preset the maximum validity period, This refers to the current available inventory quantity for this product SKU. Set the preset safety stock quantity.

6. The method for real-time synchronization of online and offline retail inventory according to claim 4, characterized in that, The pre-occupancy validity period mentioned in step S3 The method for determining it is as follows: ， ， ， in, These are the corrected initial probability parameters. To preset the release probability threshold, To preset the maximum validity period, This refers to the current available inventory quantity for this product SKU. Set the preset safety stock quantity.

7. The method for real-time synchronization of online and offline retail inventory according to claim 1, characterized in that, The detection of the picking-up sign event in step S2 includes: when the electronic price tag is touched twice in a row within 2 seconds, it is determined to be a picking-up sign event; or when the customer flow counting device detects that a person stays in front of the shelf for more than 5 seconds and the shelf weight sensor detects that the weight reduction exceeds a preset threshold, it is determined to be a picking-up sign event.

8. The method for real-time synchronization of online and offline retail inventory according to claim 1, characterized in that, The preset spatiotemporal matching conditions mentioned in step S5 include: The SKUs of the products in the cashier transaction record are the same as the SKUs of the products in the pre-registration record; The transaction completion time of the cashier transaction record falls between the trigger time of the pre-reservation record and the expiration time of the pre-reservation validity period; The area where the POS terminal corresponding to the POS transaction flow is located and the shelf area associated with the pre-reserved record belong to the same preset store area or adjacent preset store areas.

9. The method for real-time synchronization of online and offline retail inventory according to claim 8, characterized in that, Step S5 further includes: when multiple pre-occupied records simultaneously meet the preset spatiotemporal matching conditions, the first sorting condition is based on the shortest spatial distance between the shelf area and the area where the cashier terminal is located, and the second sorting condition is based on the smallest absolute value of the difference between the time interval between the trigger time and the transaction completion time and the historical average transaction duration of the product SKU, and the matching pre-occupied record is selected.

10. The method for real-time synchronization of online and offline retail inventory according to claim 1, characterized in that, The operation of deducting the online available inventory quantity in the cloud inventory center in step S4 includes: Current available inventory quantity for this product SKU If the value is greater than 0, decrease the available inventory by 1 and increase the reserved locked inventory by 1; Current available inventory quantity for this product SKU If the value is 0, the reserved record is rejected.

Images