Clothing inventory intelligent management system based on big data

Abstract

The invention relates to the field of electric digital data processing, and discloses an intelligent clothing inventory management system based on big data, which comprises the steps of deploying an edge computing unit at each inventory node, and constructing a distributed network; each edge calculation unit calculates and broadcasts an inventory potential energy value in real time based on the current inventory and a recent sales rate subjected to multi-source data cross validation; any node autonomously compares the potential energy gradient with the potential energy gradient of an adjacent node, and when the gradient meets a triggering condition, an inventory pulling request is actively generated. The inherent information delay and calculation bottleneck of centralized decision making are eradicated, it is ensured that each allocation is driven by a real market signal subjected to multi-dimensional verification, and then the whole inventory network shows high agility and ordered self-adaption characteristics.

Description

Technical Field

[0001] This invention relates to an intelligent management system for apparel inventory based on big data, belonging to the field of electronic digital data technology. Background Technology

[0002] Current mainstream intelligent inventory management systems are generally built on a centralized data processing and decision-making architecture. This architecture typically relies on a powerful central server to aggregate massive amounts of historical sales data generated by all sales terminals or inventory nodes within a given period. It then uses complex macro-forecasting models to uniformly formulate inventory allocation and transfer plans, and finally issues instructions to each node for execution. However, this centralized processing paradigm, which has been adopted and continuously optimized by the industry for a long time, reveals its fundamental limitations in design philosophy when applied to application scenarios such as fashion apparel, where product lifecycles are short, market demand fluctuates wildly, and is highly fragmented. The reason is that the effectiveness of the system's decisions heavily depends on mining statistical patterns from historical data. However, the cutting-edge trends in fashion consumption are often sudden and have no historical patterns to follow. This results in a structural delay and inherent accuracy bottlenecks in the response of prediction models based on historical data when faced with rapidly changing market demands.

[0003] At a deeper level, this centralized architecture inherently dictates that the information flow path will inevitably be lengthy. The complete closed loop from terminal perception to central decision-making and then to terminal execution will inevitably produce time delays, which are often fatal when every second counts in the sales opportunity window. At the same time, the calculations and modeling performed by the central server to process global data are essentially an averaging or blurring of local dynamic information. This makes it easy for the system to ignore or filter out key early signals that occur at individual nodes but may indicate the outbreak of new demand.

[0004] Specifically, this model fundamentally suffers from the following inherent contradictions that are difficult to reconcile: 1. Lag in decision-making response: the centralized calculation and command issuance mode of the system cannot physically match the near-real-time dynamic change rate of end-market demand, resulting in inventory adjustments always being a step behind; 2. Limitations in decision-making basis: the system relies excessively on relatively singular historical sales data as the basis for decision-making, lacking the ability to effectively capture and integrate multi-dimensional, real-time on-site data that can more proactively represent the intensity of immediate demand. Therefore, how to break through the fundamental constraints of the existing centralized architecture and construct a new distributed intelligent paradigm that allows inventory decision-making power to be decentralized to the network edge, so that it no longer relies on macro-prediction of historical data, but achieves adaptive and predictive optimization of inventory resources through proximity interaction and autonomous negotiation of real-time, high-dimensional market signals among nodes, becomes the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a big data-based intelligent management system for apparel inventory. Its main purpose is to solve the problem that existing centralized inventory management systems, due to their inherent decision-making delays and reliance on single data sources, are unable to respond sensitively and efficiently to the ever-changing dynamics of the end market.

[0006] To achieve the above objectives, the present invention provides a big data-based intelligent management system for apparel inventory, comprising:

[0007] A distributed network consisting of at least two inventory nodes, each of which has an edge computing unit; in each inventory node, the edge computing unit is configured to calculate an inventory potential value in real time based on the current inventory quantity and recent sales rate of the inventory unit of that inventory node.

[0008] The edge computing unit is also configured to broadcast its calculated inventory potential value to other inventory nodes that have a predetermined network topology relationship with the inventory node through a distributed network.

[0009] At any first inventory node, its edge computing unit is also configured to receive and compare its own first inventory potential value with the second inventory potential value of at least one second inventory node to determine the potential gradient between the two.

[0010] Furthermore, the edge computing unit is configured to automatically generate and send an inventory transfer request when the potential gradient meets a triggering condition.

[0011] Preferably, the edge computing unit is configured to calculate the inventory potential value using the following relationship. : ,in, This represents the inventory potential value. This represents the current inventory quantity. Based on recent sales rate, It is a positive number set to prevent the denominator from being zero.

[0012] Preferably, the triggering condition is: the ratio of the second inventory potential value to the first inventory potential value exceeds a first threshold.

[0013] Preferably, the inventory transfer request is an inventory pull request actively sent by the first inventory node, whose first inventory potential value is lower than the second inventory potential value, to the second inventory node.

[0014] Preferably, the system also includes an application programming interface (API); the API is configured to automatically trigger an external logistics service to perform the physical transportation of the inventory units after the inventory transfer request is confirmed.

[0015] Preferably, the edge computing unit is also configured to switch to a high-load operation mode when it detects that the total sales fluctuation of the system exceeds a second threshold; in this high-load operation mode, the edge computing unit generates an inventory transfer request based on a fixed static rule table, which maps the combination relationship between the sales rate of inventory units and the inventory level to a determined transfer instruction.

[0016] Preferably, the edge computing unit is further configured to determine the recent sales rate through a three-source cross-arbitration logic; the three-source cross-arbitration logic specifically involves: acquiring the customer flow rate from camera customer flow statistics, the transaction rate from the sales terminal equipment, and the product exit rate from the RFID sensor door as three independent data sources; if the difference between the rate estimates of any two data sources is less than a third threshold, then the average of the rate estimates of the two data sources is taken as the recent sales rate; if the difference between the rate estimates of the three data sources is not less than the third threshold, then the transaction rate is adopted as the recent sales rate.

[0017] Preferably, the edge computing unit is also configured to execute an output decision arbitration mechanism; the edge computing unit is configured to, in addition to broadcasting the inventory potential value, periodically send a heartbeat packet to other inventory nodes through an independent asynchronous communication channel; before generating an inventory transfer request, the edge computing unit is also configured to: check the timestamp of the last heartbeat packet received from the target inventory node, and if the timestamp exceeds a predetermined time period, then suspend the generation and transmission of the inventory transfer request.

[0018] Preferably, when the inventory transfer request is a replenishment request sent to an external supplier system, the edge computing unit is further configured to: first send a probe order to the external supplier system before sending the formal replenishment request; extract the response delay time of the external supplier system to the probe order; generate a dynamic calibration coefficient based on the response delay time to characterize the current fulfillment capability of the external supplier system; and use the dynamic calibration coefficient to adjust the replenishment decision for the external supplier.

[0019] Preferably, when inventory transfer is performed by automated guided vehicles (AGVs) within the warehouse, the system also includes a dynamic immunity module for in-warehouse transport paths. The dynamic immunity module is configured to: extract the variance of the measurement residuals generated by the positioning system of the AGV in real time during operation; dynamically assess the signal interference intensity of the physical environment in which the AGV is currently located based on the variance of the measurement residuals; and automatically switch between multiple navigation strategy modes preset by the AGV according to the signal interference intensity.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. This invention constructs a distributed, self-regulating inventory circulation system by setting up edge computing units at each inventory node and calculating an inventory potential value in real time based on the current quantity of inventory units and recent sales rate. Under this system, inventory allocation no longer relies on a central unit that processes massive amounts of historical data, makes complex predictions, and issues instructions. Instead, each node autonomously judges the redundancy and scarcity of inventory by comparing their potential gradients, and the node with the more urgent need actively initiates allocation requests. This method of data generation locally, interaction nearby, and direct triggering of execution allows inventory adjustments to closely follow every subtle and real dynamic change in the market terminal, avoiding the information delay and computational bottleneck inherent in centralized decision-making, and giving the entire inventory network a more agile and orderly adaptive characteristic.

[0022] 2. This invention cross-arbitrates three independent physical world data sources: camera-based customer flow, sales terminal transactions, and RFID sensor gate product exit rates. This cross-arbitrates these data to determine a more reliable recent sales rate. This mechanism not only provides a solid and reliable data foundation for calculating inventory potential, but also ensures that inventory pull requests initiated by nodes with low inventory potential are no longer based on a single-dimensional sales record, but rather on a precise perception of immediate demand verified by multi-source information. In this way, the system tightly couples the front-end multi-dimensional perception capabilities with the back-end inventory allocation behavior, ensuring that every inventory movement is driven by real and verified market signals, thereby improving the accuracy and rationality of inventory resource allocation.

[0023] 3. By introducing an independent asynchronous heartbeat communication channel and an output decision arbitration mechanism, this invention ensures that the real-time validity of the target node's communication link is confirmed before any inventory transfer request is issued. This adds a crucial layer of security to the automated inventory flow at the execution level, effectively preventing erroneous transfers caused by network latency or interruptions. Simultaneously, the system has the ability to switch to a fixed static rule table for decision-making under high load. This design allows the system to switch from dynamic, fine-grained calculation to highly efficient rule matching when facing sudden large-scale sales fluctuations. While ensuring the continuous effectiveness of the core transfer function, it avoids resource overload of the edge computing unit, thus demonstrating strong environmental adaptability and system stability under complex, volatile, and extreme operating conditions. Attached Figure Description

[0024] Figure 1 This is a system architecture diagram of a big data-based intelligent management system for apparel inventory according to the present invention.

[0025] Figure 2This is a comparison chart of inventory level changes between the proposed solution and the traditional centralized model when dealing with asymmetric demand shocks.

[0026] Figure 3 This is a flowchart illustrating the core operating logic of the edge computing unit of this invention.

[0027] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0029] A big data-based intelligent management system for apparel inventory comprises a distributed network of edge computing units located at various inventory nodes. Unlike centralized decision-making models, each inventory node in this network is endowed with independent sensing, computing, and decision-making capabilities. The system operates by having each edge computing unit first determine a recent sales rate representing the local real-time market situation through a multi-source data cross-arbitration process. This rate is then combined with the current local inventory quantity, and a standardized potential energy calculation model is used to generate a real-time inventory potential energy value. Each node broadcasts its own inventory potential energy value via the network and continuously monitors the potential energy status of neighboring nodes. When any node detects that the potential energy gradient between itself and its neighboring nodes meets a preset trigger condition, the node with lower potential energy autonomously generates and sends an inventory retrieval request to the node with higher potential energy. This achieves adaptive and optimized allocation of inventory resources within the network without a central unified scheduling command. In a specific application scenario… For example, a clothing brand operating numerous chain stores faces the common challenge that localized consumption hotspots may suddenly occur in a few stores in a specific region, leading to a surge in demand for certain items in a short period of time. Traditional inventory management models that rely on periodic data aggregation and analysis are unable to keep up with such demand changes, often resulting in rapid stockouts in popular stores while the same items remain idle in other stores, thus affecting sales and increasing the risk of inventory backlog. To address this challenge, the system of this invention deploys an edge computing unit on the local computing device of each store, such as an in-store server or an industrial computer. These units form a distributed network via the Internet, and each unit is configured to independently execute the following procedures.

[0030] To ensure that inventory decisions more comprehensively reflect the true demand in the end market, each edge computing unit is first configured to determine the near-term sales rate through a three-source cross-arbitration logic before calculating the inventory potential value. This logic aims to address the potential limitations of a single data source. For example, relying solely on transaction rates at sales terminals is insufficient to capture potential customer demand signals indicating high attention but no immediate purchase, while isolated customer flow statistics cannot be directly linked to specific products. Therefore, the edge computing unit is configured to simultaneously acquire three independent data streams: first, customer flow rates within specific product display areas, statistically analyzed using computer vision algorithms via in-store cameras; second, transaction rates for specific inventory units, directly from sales terminal devices; and third, exit rates for specific products, collected in real-time by RFID-sensor gates deployed at store exits. After acquiring these three rate estimates, the edge computing unit executes a deterministic arbitration procedure: setting a third threshold for evaluating data consistency, denoted as... For example, set to If the relative difference between the rate estimates of any two data sources is less than this threshold. If the two data sources with high consistency are used, their arithmetic mean will be taken as the final recent sales rate; if the rate estimates from the three data sources differ significantly, i.e., the relative difference between any two is not less than [a certain value], then [the remaining data sources will be used]. If the system adopts the transaction rate with final commercial confirmation as the recent sales rate, the basis for inventory allocation decisions will be transformed from single-dimensional transaction data into a quantitative representation of immediate demand that has been verified by multi-source information.

[0031] In order to obtain a reliable near-term sales rate Then, the edge computing unit calculates the inventory potential value. This value aims to quantify the adequacy of the current inventory level relative to the immediate consumption rate, and its physical dimension is time. For example, if the inventory quantity... The unit is pieces, and the sales rate is... If the unit is pieces / hour, then The unit is hours. To ensure the numerical stability of the calculation process, the inventory potential energy value is... It is calculated using the following relation: In this formula, This refers to the current inventory quantity of a specific inventory unit within the inventory node where the edge computing unit is located. This refers to the recent sales rate determined by the aforementioned three-source cross-arbitration logic; Then it is a with The smallest positive number with the same dimensions is defined to prevent the denominator in the expression from being zero. For example, its value can be specified as... This value exists only when The calculation remains effective when the value approaches zero, while during normal sales periods... The impact of the calculation results is negligible. The calculation of this inventory potential value generates a unified dynamic quantitative indicator for the inventory status of each node, which can be transmitted and compared within the network.

[0032] Once the edge computing units of each inventory node in the network have completed the calculation of their own inventory potential energy value, the system enters the autonomous allocation phase driven by the potential energy gradient. Each edge computing unit is configured to allocate resources at a predetermined period, for example, every... Every second, the system broadcasts its calculated inventory potential value for key inventory units to its pre-defined neighboring nodes within the distributed network topology. At any first inventory node, its edge computing unit receives a second inventory potential value from any second inventory node. Then, the potential energy gradient between the two is determined. To avoid unnecessary transfer caused by small potential energy differences, the system sets a clear trigger condition, which is a preset value serving as the first threshold. By definition, in practice, this threshold The reasonable range is usually set at to Between, therefore, if and only if the node Determine its own inventory potential value Below the node Inventory potential value And satisfy the inequality: As the first inventory node with more urgent demand Its edge computing unit is then configured to automatically generate inventory pull requests and send them to the second inventory node. The sending mechanism, which involves resources being actively pulled from a low-potential-energy region to a high-potential-energy region, aims to avoid the information delays and decision lags inherent in centralized command issuance. To achieve closed-loop execution of inventory transfers, the system also includes an application programming interface (API) to connect the system with external logistics execution services. When an inventory transfer request, such as a pull request from the first inventory node to the second inventory node, is confirmed and responded to by the edge computing unit of the second inventory node, the edge computing unit of the first inventory node is configured to automatically call the interface of the external logistics service provider through its API and send out instruction data containing transfer order information such as inventory unit code, quantity, pickup and delivery address, thereby triggering a physical transportation task.

[0033] To improve the system's operational stability under complex or extreme conditions, two collaborative mechanisms are designed. First, there is the ability to switch between high-load operating modes. The edge computing unit continuously monitors the total sales fluctuations aggregated from data reported by each node. If the calculated total sales fluctuation rate exceeds a preset value serving as a second threshold within a predetermined time window... For example If the system determines that a global event may occur, the edge computing unit will automatically switch to a high-load operation mode. In this mode, the system suspends the dynamic calculation of the potential gradient and instead generates inventory transfer requests based on a pre-installed local static rule table. This rule table directly maps the combination of sales rate and inventory level of inventory units to a specific transfer instruction, thereby reducing the computational load of edge nodes while maintaining core transfer functions. Secondly, there is a communication validity arbitration mechanism for output decisions. In addition to the main communication channel broadcasting inventory potential values, the edge computing unit also sends heartbeat packets to neighboring nodes at a higher frequency through an independent asynchronous communication channel. Before generating any inventory transfer request, the edge computing unit is forced to perform a pre-check: verifying the timestamp of the last heartbeat packet received from the target inventory node. If the timestamp exceeds a predetermined time period, the system determines that the target node's communication link may be invalid and suspends the generation and transmission of this transfer request, providing a communication validity verification mechanism for distributed automated decision-making. The system design can also be extended to deep integration with the upstream of the supply chain and warehouse automation. When an inventory transfer request is a replenishment request sent to an external supplier system, the edge computing unit is configured to send a very small probe order to the supplier system before sending the formal replenishment request, and record the delay time from order sending to receiving confirmation. Based on this delay time, the unit generates a dynamic calibration coefficient characterizing the supplier's current fulfillment capability and uses this coefficient to adjust subsequent replenishment decisions for that supplier. When inventory transfer is performed by automated guided vehicles (AGVs) within the warehouse, the system also includes a dynamic immune module for in-warehouse transportation paths. This module extracts the variance of the measurement residuals generated by the AGV positioning system in real time, dynamically assesses the signal interference strength of the vehicle's current physical environment based on this variance, and automatically switches between various preset navigation strategy modes of the AGV, such as LiDAR navigation or QR code navigation, according to the interference strength, thereby ensuring the reliability of in-warehouse logistics execution.

[0034] Example 1: During a major holiday promotion, a clothing brand simultaneously launched a flagship outerwear item in two core cities located in different climate zones. Based on macro sales forecasts from a central system, the brand allocated the same initial inventory to stores in both locations. After the promotion began, the inventory in both cities... Due to a sudden and sharp drop in temperature, consumer demand for this jacket surged within hours, while demand in urban areas also increased during the same period. The weather remained above historical averages, leading to a lack of demand for the jacket in the area. This created an extremely unbalanced demand scenario between the two inventory points, driven by external environmental factors and unpredictable by traditional forecasting models; in the city Within the store, the three-source cross-arbitration logic executed by its edge computing unit immediately responded to this change. Although the initial transaction rate did not fully reflect the surge in demand, both the customer flow statistics from cameras and the product exit rate from RFID sensor gates showed significant increases. The cross-arbitration of these two data points with the transaction rate outputs a revised recent sales rate. ,Should The value, as a high-weight denominator, is directly substituted into the formula for calculating the inventory potential energy value. Even in the initial inventory quantity Even with sufficient supplies, it still led to urban [problems]. Inventory potential value The sharp decline, and the rapid decay of this value, quantifies local demand pressure as a clear and communicable signal.

[0035] In the city Inside the store, sales of this jacket have almost stagnated recently, resulting in a slowdown in sales. Its inventory potential value approaches zero. It remains at an extremely high level. The existence of distributed networks enables cities to... Edge computing units can receive data from the city in real time. Broadcast The value is determined immediately, and the potential gradient between the two is calculated. Much larger The ratio of the inventory potential values ​​of the two nodes quickly met the preset triggering condition. The fulfillment of this condition allows the system's decision-making to no longer rely on causal analysis of why this demand difference occurs, but instead reacts directly based on the fact that an inventory potential gradient already exists. The edge computing unit then automatically generates an inventory retrieval request and sends it directly to the city via the application programming interface. The inventory nodes and associated logistics systems issue allocation instructions; the generation and execution of this inventory retrieval request are not subject to any central server intervention, but are completed autonomously by two nodes at the network edge. This avoids the inherent constraint of sacrificing local response speed in pursuit of global data consistency in traditional architectures. A process that would take several days for data reporting, analysis, decision-making, and instruction issuance in the traditional model is replaced by a local self-regulating behavior driven directly by data potential gradients, measured in minutes. Ultimately, some of the inventory retrieval requests in the city are processed automatically. The idle inventory in the city Before the window of potential sales opportunities closed, they were physically transferred to the city. Within the store, it effectively replenished the latter's shelves, in the city The explosive demand caused by sudden weather changes was met, and the city It also frees up inventory space occupied by ineffective items. The entire inventory network does not attempt to predict an unpredictable local weather, but instead builds an internal mechanism that can sense and respond to changes in demand caused by such events in real time. This transforms the allocation of inventory resources from a pre-emptive, macro-forecast-based push allocation to a real-time, real-time, potential-based pull adjustment.

[0036] Example 2: To objectively verify the response speed and effectiveness of the distributed self-regulating inventory circulation system driven by inventory potential energy gradient of the present invention in dealing with local sudden demand, a system was built... A software simulation test platform consisting of several inventory nodes; the nodes in this platform ,node With nodes The experiment utilizes a simulated distributed network interconnection, with each node running the edge computing unit described in the aforementioned specific implementation. The objective is to quantitatively compare the differences in key performance indicators between the proposed solution and a centralized decision-making model based on a fixed period, when facing the same asymmetric demand shocks. A key parameter of the experimental platform is the inventory potential value. The broadcast cycle needs to be set to balance the real-time nature of data interaction with the network and computing load. The value of this cycle is directly related to the time scale of the demand fluctuation characteristics of the monitored product. An excessively long cycle will fail to capture short-term demand pulses, while an excessively short cycle will generate redundant communication overhead. Therefore, the decision rule for this cycle is set as follows: its duration should be significantly shorter than the time scale of a significant change in inventory level in the target scenario, such as consumption. In this experiment, the typical timeframe for demand peaks in fast-moving consumer goods (FMCG) apparel items during promotional events was selected, which may last for several hours. Seconds as The broadcast cycle is set to allow for multiple dynamic adjustments during the evolution of demand events. As a control group, the centralized decision-making model is configured to broadcast every... Every hour, a central server aggregates data from all nodes and then uniformly calculates and issues allocation instructions after this point in time. This cycle is used to simulate the batch processing decision delays that are common in existing technologies.

[0037] When the test starts, the node ,node With nodes The quantity of inventory initialized to hold the same specific inventory unit, i.e. Item. In At the moment, at the node Apply a simulated demand shock to its recent sales rate Instantly boost and maintain at Items / hour, and nodes With nodes Sales rate and Then remain at Baseline level of units / hour, total number of test runs Hours were recorded, during which the inventory level and cumulative sales at each node were continuously recorded; in the experimental group of the present invention, as the nodes... Sales activities are being conducted, and the inventory potential value is... It began to drop rapidly within minutes, and the node Inventory potential value Because its sales rate is extremely low and has remained basically stable, Hour, node Edge computing units have been detected The ratio exceeded the set first threshold. and autonomously to the node The first batch of inventory pull requests was issued, while in the control group, until... The system only first identifies the node at the first decision cycle point of the hour. The inventory was depleted abnormally. Table 1 compares the key performance data of the two experimental groups at different time points.

[0038] Table 1: Performance data comparison of the two models at different time points.

[0039]

[0040] See table Experimental data show that, due to its real-time sensing mechanism of inventory potential gradient, the solution of this invention... Hour, node The inventory has been obtained from the node Supplement, maintain at The level of the component, while the node of the control group. Inventory has decreased Item. In Hours, the node of the present invention Inventory still exists The nodes of the control group The inventory depletion at this moment led to a disruption in subsequent sales. The underlying mechanism of this phenomenon lies in the fact that the solution of this invention decentralizes the decision-making power of inventory allocation to the network edge, allowing allocation behavior to be directly driven by real-time market signals. This avoids the inherent information delay and batch processing cycle of the central decision-making unit in the control group model. The data from this experiment shows that when dealing with asymmetric local demand shocks, the system that autonomously adjusts based on inventory potential gradients has significantly better response time for triggering inventory allocation and its ability to maintain the health of inventory at the front-end sales nodes than the centralized decision-making model with a fixed cycle. The experimental results confirm that the solution of this invention can effectively avoid the inherent information delay of centralized decision-making, ensure that inventory allocation is driven by real market signals, and thus improve the agility and adaptability of the entire inventory network.

[0041] Example 3: This example combines Figures 1 to 3 This section describes an intelligent management system for apparel inventory based on big data, such as... Figure 1 As shown, the diagram is logically divided into four collaborative layers. The top layer is the distributed network layer, which consists of multiple physically dispersed inventory nodes A, B, and C, with each node containing a core edge computing unit. Next, the system's perception foundation is the multi-source data acquisition layer, responsible for acquiring data in parallel from three independent physical sources: customer flow rate monitoring via camera statistics, transaction rate collection via point-of-sale (POS) devices, and product exit rate sensing via RFID-enabled gates. The collected multi-dimensional data is then sent to the core processing layer for in-depth processing. This layer first verifies and fuses the input data using a three-source cross-arbitration logic to determine a more reliable recent sales rate. Subsequently, this rate, along with the current local inventory level, is used in the inventory potential value calculation model. A standardized inventory potential value is generated in real time. This potential value is used for potential gradient judgment trigger condition detection. On the other hand, before the decision output, it also needs to be validated by the heartbeat packet monitoring of the target node through the output decision arbitration mechanism. Finally, all decisions are materialized at the execution layer. A valid decision after arbitration will trigger the automatic generation and sending of inventory transfer requests and trigger external logistics services through the application programming interface. In addition, this layer also includes a backup high-load operation mode, which can switch to the decision logic based on static rule tables when the total sales of the system fluctuate abnormally.

[0042] like Figure 2 As shown in the figure, the curve of the present invention, marked with a solid line, shows that the inventory quantity gradually decreases from the initial 200 units over time, and there is still inventory at the 8th hour. This indicates that the system effectively replenishes the consumption through timely cross-node allocation. In contrast, the curve of the traditional centralized model, marked with a dashed line, shows a sharp depletion of inventory, which is completely exhausted at the 4th hour, with subsequent inventory at zero. This intuitively demonstrates the significant advantages of the distributed self-regulating system of the present invention in terms of response speed and maintaining inventory health compared to the centralized model with inherent decision-making delay.

[0043] like Figure 3 As shown in the diagram, this further reveals the core operating logic and decision-making loop within an edge computing unit deployed at any inventory node. The process begins with data acquisition and rate calculation, its core task being to execute a three-source cross-arbitration logic to determine the recent sales rate. After obtaining a valid data source, the process enters the inventory potential energy calculation and broadcasting stage. The unit combines inventory and rate to calculate the potential energy value and broadcasts this value to neighboring nodes. Simultaneously, the unit continuously listens to and compares the potential energy gradient, i.e., receiving the potential energy values ​​from neighboring nodes and continuously monitoring potential energy changes. When the potential energy gradient is detected to exceed a trigger threshold, the process enters the decision-making arbitration stage. In the arbitration phase, the output decision arbitration mechanism is executed and the validity of the target node's heartbeat packet is checked. If the target node's communication is valid, the process ultimately triggers the generation of a transfer request phase to generate an inventory retrieval request and send the request via API. The flowchart also clearly shows two key feedback loops: First, if the target node's communication is invalid, the transfer is aborted and the system returns to the listening state. Second, if the total sales fluctuation is detected to exceed the threshold, the system switches to a high-load operation mode. In this mode, the dynamic calculation of potential energy is suspended and decisions are made based on a fixed static rule table until the system load returns to normal before returning to the active listening state.

[0044] Example 4: When the system of the present invention is first deployed in a new network environment consisting of a plurality of inventory nodes, in order to initially calibrate the key operating parameters and decision model within the system, the system is configured to execute the following procedure; this procedure aims to set the first threshold in the inventory potential gradient triggering condition. Find a setting value that matches the current network cost structure and sales model, and set a threshold for switching to high-load operation mode. A generation method is established for the static rule table it depends on; for the first threshold The system performs an offline, historical data-based simulation optimization process for calibration. The inputs to this process are the hourly sales records of the inventory network over a complete sales cycle, the logistics cost matrix between nodes, and the unit profit of the goods. The process first defines a comprehensive cost function, whose value consists of two parts: one part is the estimated sales loss caused by inventory depletion during the entire simulation cycle, calculated as the product of the out-of-stock duration and the average sales per unit time before the out-of-stock, multiplied by the unit profit; the other part is the sum of the total logistics costs incurred from performing inventory transfers. Subsequently, the system runs the simulation iteratively, where the first threshold... The value can be taken from a set lower limit. Start with a fixed step size. Increment to a set upper limit In each iteration, the entire system is based entirely on the current... The value is calculated by performing a complete simulation based on historical sales data. The system sets the resulting comprehensive cost function value; after all iterations are completed, the system generates a... The correspondence between the value and the comprehensive cost, and the point corresponding to the lowest comprehensive cost in this correspondence. The value is the setting that is determined to be applicable to the current network.

[0045] To specify the switching logic for high-load operation mode, the system adopts a two-stage procedure; in the first stage, the system quantifies and defines the fluctuation of total sales and determines its switching threshold. The system first calculates the time series of the total sales rate of the entire network during non-major promotional periods within a baseline time period, and then calculates the coefficient of variation of this series, i.e., the ratio of the standard deviation to the mean, which is used as the baseline volatility; the second threshold... This is set as a predetermined multiple of the benchmark volatility, which is set to 3 in one implementation. In the second stage, the system generates a static rule table by analyzing historical data. This process discretizes all historical inventory levels and sales rates, dividing inventory levels into five intervals: too low / low / medium / high / too high, and sales rates into three intervals: low / medium / high. Furthermore, the system statistically analyzes the subsequent sales performance after each historical inventory-rate combination state occurs, whether or not a transfer is executed with different degrees of intensity. Finally, for each combination state, the system selects the transfer instruction that historically brings the highest sales conversion or the lowest stockout rate. The instruction is in the form of pulling a specific number of inventory units from neighboring nodes to fill the static rule table. By executing the above offline calibration and data analysis procedures, the core parameters and decision models within the system are assigned specific values ​​and rules corresponding to the historical operating characteristics of a specific business environment. This calibration result is then deployed in the edge computing units of all nodes in the network as the initial configuration for its online operation.

[0046] Example 5: When deploying the system of the present invention on multiple inventory nodes with different physical environment characteristics, in order to ensure the consistency and comparability of the data source used by the three-source cross-arbitration logic across nodes, a standardized on-site deployment pre-calibration procedure is executed before the edge computing unit of any new node is officially put into operation; after the procedure is started, the unit first enters a period of The baseline data acquisition cycle is 24 hours. During this period, it continuously records raw rate data from three independent data sources: camera customer flow statistics, sales terminal equipment, and RFID sensor gates. It also calculates the mean and standard deviation of the output rate of each data source during normal business hours and off-peak hours for the store. Based on this, the system generates a set of localized calibration coefficients for each data source of the node. In subsequent formal operation, all raw data collected from the sensors are normalized using this set of calibration coefficients before entering the three-source cross-arbitration logic. This is to correct for systematic deviations in data acquisition that may be caused by differences in store layout, lighting conditions, or local electromagnetic interference intensity among the various inventory nodes.

[0047] Furthermore, in scenarios where the system interacts with external supplier systems to perform automatic replenishment, to dynamically adapt to fluctuations in the fulfillment capabilities of different suppliers, the edge computing unit is configured to maintain a dynamic performance profile for each supplier. This profile includes an expected response latency time, which is not a fixed value but is calculated based on a configurable quantity from that supplier in the past. The system automatically triggers the sending of probe orders to suppliers under two conditions: firstly, if no transaction occurs with the supplier within a preset quiet period, probe orders are sent periodically to refresh the performance baseline; secondly, if the response time of a formal replenishment request deviates from the current expected response delay time by more than a preset ratio, probe orders are sent in an event-driven manner. Correspondingly, when calculating the dynamic calibration coefficient used to adjust replenishment decisions, the gain coefficient is included. It is not fixed, but rather inversely proportional to the standard deviation of the supplier's historical response delay time. In other words, for a supplier with stable performance capabilities, its... The value is relatively small, and for a supplier with unstable fulfillment capabilities, its The relatively large value increases the adjustment range of the calibration coefficient to changes in response time.

[0048] Example 6: To ensure the repeatability of the switching behavior of the dynamic immunity module for in-warehouse transportation routes under different electromagnetic environments, the system is configured to execute an offline calibration procedure for navigation strategy switching thresholds before the formal deployment of automated guided vehicles (AGVs). This procedure first pre-identifies areas with high potential signal interference risks on the warehouse map. Then, the AGV is instructed to repeatedly traverse all preset paths within the warehouse using its highest-precision but interference-sensitive navigation strategy mode. During this process, the dynamic immunity module continuously records the variance of the measurement residuals generated by its positioning system using high-frequency sampling and correlates the time-series data of this variance with the real-time position of the vehicle. After calibration, the system extracts all variance values ​​measured within the pre-identified interference areas and determines the switching threshold based on the statistical distribution of these values. In a specific setting, this threshold is set as the [missing value] of the set of variance values ​​for that interference area. Percentiles allow automated guided vehicles to switch to a backup navigation strategy that is insensitive to such interference before entering an area of ​​interference and experiencing significant positioning drift.

[0049] Furthermore, to enhance the system's fault tolerance and communication link stability during operation, the three-source cross-arbitration logic has been supplemented with a data source failure handling mechanism. If any data source, including camera-based customer flow statistics, sales terminal transaction rates, or RFID sensor gate product exit rates, experiences a data stream interruption or continuously returns invalid values ​​for more than a preset short period, that data source will be temporarily marked as failed. The arbitration logic will automatically compare and decide between the remaining two valid data sources. If only one data source remains valid, the system will accept that single data source. Simultaneously, the heartbeat mechanism used to determine the validity of inter-node communication links does not have a globally fixed timeout period. Instead, during the initialization of each edge computing unit, a series of probe packets are sent to all its neighboring nodes to measure the network round-trip latency. Based on the moving average of this latency, an adaptive timeout period is independently set for each communication link. This period is calculated as the average round-trip latency. It is twice the sum of a fixed buffer duration.

[0050] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart management system for apparel inventory based on big data, characterized in that, include: A distributed network consisting of at least two inventory nodes, with each inventory node containing an edge computing unit; At each inventory node, its edge computing unit is configured to calculate an inventory potential value in real time based on the current inventory quantity and recent sales rate of the inventory unit at that inventory node. The edge computing unit is also configured to broadcast its calculated inventory potential value to other inventory nodes that have a predetermined network topology relationship with the inventory node through a distributed network. At any first inventory node, its edge computing unit is also configured to receive and compare its own first inventory potential value with the second inventory potential value of at least one second inventory node to determine the potential gradient between the two. Furthermore, the edge computing unit is configured to automatically generate and send an inventory transfer request when the potential gradient meets a triggering condition.

2. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, The edge computing unit is configured to calculate the inventory potential value using the following relationship. : ,in, This represents the inventory potential value. This represents the current inventory quantity. Based on recent sales rate, It is a positive number set to prevent the denominator from being zero.

3. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, The trigger condition is: the ratio of the second inventory potential value to the first inventory potential value exceeds a first threshold.

4. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, An inventory transfer request is an inventory retrieval request actively sent by the first inventory node, whose first inventory potential value is lower than the second inventory potential value, to the second inventory node.

5. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, The system also includes an application programming interface (API); the API is configured to automatically trigger an external logistics service to perform the physical transportation of the inventory units after the inventory transfer request is confirmed.

6. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, The edge computing unit is also configured to switch to a high-load operation mode when it detects that the total sales fluctuation of the system exceeds a second threshold. In this high-load operation mode, the edge computing unit generates inventory transfer requests based on a fixed static rule table, which maps the combination of sales rate and inventory level of inventory units to a defined transfer instruction.

7. The intelligent management system for apparel inventory based on big data according to claim 2, characterized in that, The edge computing unit is also configured to determine the recent sales rate through a three-source cross-arbitration logic. Specifically, the three-source cross-arbitration logic is as follows: the customer flow rate from camera customer flow statistics, the transaction rate from the sales terminal equipment, and the product exit rate from the RFID sensor door are obtained as three independent data sources; if the difference between the rate estimates of any two data sources is less than a third threshold, the average of the rate estimates of the two data sources is taken as the recent sales rate; if the difference between the rate estimates of the three data sources is not less than the third threshold, the transaction rate is adopted as the recent sales rate.

8. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, The edge computing unit is also configured to execute an output decision arbitration mechanism; the edge computing unit is configured to, in addition to broadcasting the inventory potential value, periodically send a heartbeat packet to other inventory nodes through an independent asynchronous communication channel; before generating an inventory transfer request, the edge computing unit is also configured to: check the timestamp of the last heartbeat packet received from the target inventory node, and if the timestamp exceeds a predetermined time period, then suspend the generation and transmission of the inventory transfer request.

9. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, When the inventory transfer request is a replenishment request sent to an external supplier system, the edge computing unit is also configured to: first send a probe order to the external supplier system before sending the formal replenishment request; and extract the response delay time of the external supplier system to the probe order. Based on the response delay time, a dynamic calibration coefficient is generated to characterize the current fulfillment capability of the external supplier system, and the replenishment decision for the external supplier is adjusted using the dynamic calibration coefficient.

10. The intelligent management system for apparel inventory based on big data according to claim 1, characterized in that, When inventory transfers are performed by automated guided vehicles (AGVs) within the warehouse, the system also includes a dynamic immunity module for in-warehouse transport paths. The dynamic immunity module is configured to: extract the variance of the measurement residuals generated by the positioning system of the AGV in real time during operation; dynamically assess the signal interference strength of the physical environment in which the AGV is currently located based on the variance of the measurement residuals; and automatically switch between multiple navigation strategy modes preset by the AGV according to the signal interference strength.

Images