A method for managing retail inventory of devices

Abstract

The application discloses a kind of equipment retail inventory management methods, it is related to equipment retail inventory control technical field, method includes: obtaining the physical operating parameter and quality inspection score data of each entity equipment under the same inventory unit, calculate health state value and construct the health state distribution matrix mapped with equipment serial number, the matrix and multi-channel demand data are as joint state input continuous time Markov decision process model, output joint action by replenishment, allocation, discount agent, based on the joint reward function containing health state salvage value training, according to the decay critical threshold screening device to be processed, generate serial number level promotion and picking instruction issue execution, and feedback update matrix forms closed loop, the present application can realize equipment individual differentiation management, reduce asset depreciation loss, improve decision and execution synergy efficiency.

Description

Technical Field

[0001] This invention relates to the field of equipment retail inventory management technology, specifically to a method for equipment retail inventory management. Background Technology

[0002] In high-value equipment retail scenarios, store display equipment and warehousing equipment typically incorporate IoT sensors and battery management systems to record physical operating parameters such as the number of charge-discharge cycles. Automated quality inspection equipment is used in the reverse logistics process to score the performance of returned equipment, providing a technological foundation for the quantitative assessment of the individual physical condition of each piece of equipment. Existing equipment retail inventory management systems, when managing multiple physical devices within the same inventory unit, typically treat them as completely homogeneous units, failing to perceive and utilize the differences in individual health status caused by physical degradation during storage and reverse logistics such as trade-ins and returns. This makes it difficult to implement differentiated pricing and turnover strategies for equipment in different health states, resulting in hidden asset depreciation losses that are difficult to quantify and control. The closest existing technical solution to this invention is an intelligent replenishment system based on time series forecasting algorithms. This solution periodically collects historical sales data from each inventory unit to form a time series. The existing technology uses historical sales data as input, which is inherently unable to perceive the differences in the physical health status of individual devices. Furthermore, when reverse logistics devices with randomly distributed health status are introduced, the model parameters are easily shifted, leading to a decrease in prediction confidence. In addition, the output of this solution is only an aggregated replenishment suggestion, with decision granularity remaining at the inventory unit level. It lacks a precise instruction mapping mechanism down to the serial number level between the inventory and physical execution systems, making it difficult to translate algorithmic decisions into executable physical transfer and picking actions. Therefore, we propose a new method for managing retail inventory. Summary of the Invention

[0003] The purpose of this invention is to provide a method for managing retail inventory of equipment.

[0004] Existing retail inventory management systems treat all physical equipment within the same inventory unit as completely equivalent units, failing to recognize the individual health differences caused by physical decay and reverse logistics during the inventory period. This results in physically decayed equipment being treated the same as brand-new equipment, randomly shipped according to the order of entry into the warehouse. Some highly decayed equipment continues to depreciate deep within the inventory, ultimately requiring clearance at prices far below normal, leading to hidden asset depreciation losses. Furthermore, existing predictive models based on historical sales sequences assume relatively stable statistical patterns, while trade-ins and returns create a significant risk. The health status of reverse logistics inbound equipment exhibits a random distribution. Once this randomness is mixed with historical sales sequences, it can cause the model parameter estimation to deviate, leading to a decrease in the confidence of the prediction results and thus triggering the risk of inventory backlog or supply disruption. In addition, the decision output of existing intelligent inventory optimization solutions is only a replenishment suggestion at the aggregation level, with the decision granularity remaining at the inventory unit level. It lacks an instruction mapping mechanism between the warehouse physical execution system and the equipment serial number level, which means that even if the system has identified certain equipment with depreciation risks, it cannot translate the identification results into picking or promotional actions that the warehouse execution system can directly execute, down to the individual equipment level.

[0005] The technical problem to be solved by the present invention is to provide a device retail inventory management method that can continuously and quantitatively perceive the individual physical health status of each physical device under the same inventory unit, and incorporate the health status distribution information, multi-channel demand, and reverse logistics randomness into the reinforcement learning decision space, while directly mapping the optimization decision into physical execution instructions accurate to the device serial number level.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by this invention is as follows: Obtain the physical operating parameters and quality inspection score data of each physical device under the same inventory unit; calculate the health status value of each physical device based on the physical operating parameters and the quality inspection score data; construct a health status distribution matrix corresponding to the inventory unit based on the health status values ​​of all physical devices under the same inventory unit; establish a mapping relationship between each element in the health status distribution matrix and the corresponding device serial number; input the health status distribution matrix and the front-end multi-channel demand data as a joint state into a continuous-time Markov decision process environment model; and output the replenishment quantity, transfer path, and dynamic discount rate respectively through three types of strategy networks: replenishment agent, transfer agent, and discount agent, constituting a joint action based on sales profit reward, inventory holding cost reward, stockout penalty reward, and health status. A joint reward function for status recovery value rewards iteratively updates the parameters of the three types of strategy networks. Based on a preset health decay threshold, it filters out device serial numbers with health status values ​​lower than the threshold from the health status distribution matrix of the inventory units corresponding to at least one of the dynamic discount rate and the allocation path, forming a set of devices to be processed. For devices in the set of devices to be processed that trigger dynamic discount actions, it generates a promotional instruction containing the device serial number and discount amount and sends it to the store sales terminal system. For devices that trigger priority delivery actions, it generates a picking instruction containing the device serial number and target location information and sends it to the warehouse management system. It receives the execution feedback results of the promotional instruction and the picking instruction, and updates the status of the corresponding device serial number in the health status distribution matrix of the corresponding inventory unit according to the execution feedback results.

[0007] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0008] 1. This invention constructs a continuous health status distribution matrix to achieve quantitative identification of individual differences of equipment under the same inventory unit, providing a basis for decision-making on differentiated pricing and priority delivery, thereby reducing hidden asset depreciation losses. This effect is directly supported by the continuous calculation method of health status values ​​and the sequence number mapping relationship of the distribution matrix.

[0009] 2. This invention, through the design of a joint reward function based on a continuous-time Markov decision process, enables reinforcement learning models to maintain stable convergence of decisions under reverse logistics conditions with random state distributions, avoiding the parameter offset problem of traditional prediction models. This effect is directly supported by the specific calculation methods of the health state recovery value reward item and the inventory holding cost reward item in the joint reward function.

[0010] 3. This invention constructs an instruction mapping mechanism from algorithm output to the physical execution layer, enabling decision results to be implemented in the form of specific picking, allocation, and promotion instructions at the device serial number level, forming a closed-loop control between algorithm decision-making and physical execution. This effect is directly supported by the screening method of the set of devices to be processed and the mechanism of execution feedback and update matrix. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the overall four-layer closed-loop process in an embodiment of the present invention;

[0012] Figure 2 This is a schematic diagram of the health status matrix construction process in an embodiment of the present invention;

[0013] Figure 3 This is a schematic diagram of the closed-loop process of reinforcement learning decision-making and instruction implementation in an embodiment of the present invention. Detailed Implementation

[0014] The following is in conjunction with the appendix Figure 1-3 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0015] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0016] The technical solution of the present invention will be further described below with reference to the accompanying drawings. The overall technical solution of the present invention includes four interconnected functional layers: a data perception layer, a health status distribution matrix construction layer, a core algorithm processing layer, and a physical execution command issuing layer. These layers form a closed-loop control system through data and command flows. The data perception layer is deployed in store display equipment and warehousing equipment, and includes an IoT sensor module, a battery management system log collection module, and an automated quality inspection station module for reverse logistics warehousing. The outputs of these three modules converge at the health status data collection interface. The health status distribution matrix construction layer receives the collected data and classifies it according to inventory units, dynamically maintaining a health status distribution matrix for each inventory unit. The matrix is ​​stored in a matrix database and refreshed periodically. The core algorithm processing layer is deployed on edge computing nodes or cloud servers. It includes a state observation module, a multi-agent policy network module, a continuous-time Markov decision process environment interaction module, and a policy update module. This layer takes health status distribution matrix data and front-end multi-channel real-time demand data as inputs and outputs three types of actions: replenishment quantity, allocation path, and dynamic discount rate. The physical execution instruction distribution layer includes an instruction generation module, which parses the above abstract actions into operation instructions at the specific device serial number level and distributes them to the warehouse management system, store sales terminal system, and procurement module, respectively. The execution results are returned to the data perception layer through the feedback channel, forming a complete closed loop.

[0017] In terms of data perception and matrix construction, the system periodically collects physical operating parameters such as the number of charge and discharge cycles, power-on activation time, and environmental temperature and humidity exposure of individual devices through IoT sensors deployed on store display equipment and warehouse equipment and the battery management system log interface. For devices that return through trade-in or return channels, the system collects quality inspection scoring data such as appearance damage and functional performance test through an automated quality inspection station when they enter the warehouse.

[0018] The health status value of a single device is calculated in the following manner:

[0019] ;

[0020] in, Indicates the first The health status value of the equipment ranges from 0 to 1. The lower the value, the more severe the degradation of the equipment's physical performance. , , These are the weighting coefficients corresponding to the charging / discharging loss factor, the environmental exposure loss factor, and the quality inspection scoring factor, respectively, and the sum of the three is 1. Indicates the first The cumulative number of charge-discharge cycles for the device This indicates the rated cycle count threshold for this model of equipment. Indicates the first The cumulative duration of adverse environmental exposure for the equipment. This indicates the permissible exposure duration threshold for this device model. Indicates the first The standardized quality inspection scoring function obtained by the equipment in the automated quality inspection station has its value range normalized to 0 to 1.

[0021] For a specific inventory unit (denoted as The system will store all of them in its database. The health status values ​​of the devices are collected and form the following dynamic distribution matrix:

[0022] ;

[0023] in, Indicates the first The health status distribution matrix corresponding to each inventory unit has a number of elements that dynamically changes with the increase or decrease of the number of devices in the inventory unit. Each element in the matrix is ​​uniquely mapped to its corresponding device serial number so that the subsequent core algorithm processing layer and physical execution instruction issuing layer can accurately locate the specific physical device.

[0024] The specific construction process of this matrix is ​​as follows: The system periodically reads the original sensor data of each device in the warehouse through the Internet of Things interface, including charge and discharge cycle count values ​​and cumulative exposure values ​​of ambient temperature and humidity. For newly warehoused reverse logistics equipment, the system reads the quality inspection scoring function after the equipment completes the inspection through the automated quality inspection station. The output results are used to substitute the raw data obtained above into the aforementioned health status numerical calculation method to obtain the current status of each device. The system calculates the numerical value based on the correspondence between the device serial number and the inventory unit. The distribution matrix of the updated values ​​to the corresponding inventory units In the process, for existing device serial numbers, numerical overwrite updates are performed. For newly added or already outdated device serial numbers, matrix elements are added or deleted respectively. The system stores the updated distribution matrix in the matrix database and uses it as one of the state inputs of the core algorithm processing layer for subsequent calls.

[0025] In terms of core algorithm processing, this invention constructs a multi-agent deep reinforcement learning model based on a continuous-time Markov decision process, which integrates the health status distribution matrix and real-time demand information from multiple channels into a unified decision space, thereby achieving joint optimization of three types of actions: replenishment, allocation, and discount.

[0026] The state of the inventory system may transition at any consecutive moment due to events such as sales, returns, physical decay, and replenishment, and the transition probabilities satisfy the following:

[0027] ;

[0028] in, Indicates the system at time 10:00 The joint state is composed of the current inventory quantity of each inventory unit and the corresponding health status distribution matrix. , These represent the values ​​of two specific states in the state space. Indicates the system state is determined by Leap to The instantaneous rate is determined by the sales rate, the reverse return rate, the physical decay rate, and the replenishment rate. This represents a very small time increment. Indicates when The error term of higher-order infinitesimals that approaches 0;

[0029] The multi-agent policy network corresponds to three roles: replenishment agent, allocation agent, and discount agent. Each agent's policy network can adopt a network structure with the health status distribution matrix and demand feature vector as input layers, several fully connected hidden layers for feature extraction, and the corresponding action dimension as the output layer. Its joint action space includes the replenishment quantity to suppliers, the specific path for inter-store inventory allocation, and the dynamic discount rate for devices within a specific health status range. The three types of agents share a unified joint reward function, specifically defined as follows:

[0030] ;

[0031] in, Indicates at time The joint rewards received This represents the profit bonus earned from sales in the current period. This represents the current inventory holding cost, which includes the physical depreciation loss of equipment in the calculation. This indicates the penalty for opportunity loss due to stockouts. This refers to the reward for asset value recovered due to priority delivery or dynamic discounts for degraded equipment. , , , These are the weighting coefficients for each reward component, used to balance the relative importance of different optimization objectives in the total reward;

[0032] The calculation method for inventory holding costs, which takes into account physical depreciation losses, is as follows:

[0033] ;

[0034] in, This indicates the total number of devices currently in stock. Indicates the first The original value of the equipment Indicates the first The current health status value of the device. The holding cost discount factor per unit time, numerical value. The larger the value, the higher the degree of physical degradation of the device, and the higher the proportion of its implicit depreciation loss in the total holding cost, thus guiding the policy network to prioritize devices with lower health.

[0035] The calculation method for the recovery value reward obtained for the set of devices that have triggered discounts or priority shipments is as follows:

[0036] ;

[0037] in, This refers to the set of devices that were determined to have triggered discount or priority shipping actions in the current period. This represents an estimated probability that the device will actually be sold after the action is triggered. This represents the preset critical threshold for health degradation. This indicates the current market valuation of the equipment;

[0038] The parameter updates of the policy network employ the near-end policy optimization method, with the objective function being:

[0039] ;

[0040] in, The parameters of the policy network are represented. Let represent the objective function for policy optimization with pruning mechanism. Indicates time step Take the expected value from the sample above. This represents the advantage function estimate, used to measure the relative benefit of the current action compared to the average level. This indicates the preset cropping range hyperparameter. This is a clipping function that restricts the values ​​within parentheses to a specified range. The ratio of the probabilities of the old and new strategies for the same state-action pair is expressed as follows:

[0041] ;

[0042] in, This indicates the policy network that is currently pending updates. This represents the old policy network before the update. Indicates time step The joint actions taken Indicates time step The observed joint state is composed of the aforementioned health state distribution matrix and multi-channel demand information;

[0043] The specific implementation process of this core algorithm is as follows: First, initialize the parameters of the three types of policy networks—replenishment agent, allocation agent, and discount agent—as well as the corresponding value network parameters. Simultaneously, initialize the initial joint state of the continuous-time Markov decision process environment model. At the beginning of each decision cycle, read the current health status distribution matrix of each inventory unit. Together with real-time demand data from multiple front-end channels, they form the current joint status. ,Will Inputting the policy networks of three types of agents respectively yields three types of actions: replenishment quantity, allocation path, and dynamic discount rate, which are then combined into joint actions. Joint action Input a continuous-time Markov decision process environment model, based on the aforementioned state transition rate The joint state at the next time step is calculated. Simultaneously, the current reward is calculated based on the joint reward function. The value network is used to evaluate the current state action, combined with the actual reward obtained. Calculate the estimate of the advantage function The objective function is optimized based on the aforementioned strategy with pruning mechanism. The policy network parameters are adjusted using the gradient ascent method. Perform iterative updates, determine whether the current iteration number or the stability of the policy network output meets the preset convergence condition. If not, return to the process of reading the joint state and continue iterative training. If it meets the condition, solidify the current policy network parameters into a decision model and output it to the physical execution instruction delivery layer for real-time calling.

[0044] Regarding the issuance of physical execution instructions, when the core algorithm processing layer outputs a discount action or transfer action for a specific inventory unit, the system further calculates the health status based on the preset health decay threshold in the health status distribution matrix corresponding to that inventory unit. Filter out all specific device serial numbers whose health status values ​​are below the threshold and add them to the set of devices to be processed. ;

[0045] For sets For each device in the system, the system generates a corresponding application programming interface (API) call instruction: if the device is determined to trigger a dynamic discount action, the system sends a limited-time promotion instruction along with the device's specific serial number to the store's sales terminal system through the interface, so that the specific physical device displays a differentiated discount price at the sales terminal; if the device is determined to trigger a priority delivery action, the system sends an instruction to the warehouse management system through the interface, requiring the automated warehouse's picking mechanism to prioritize locating and picking the specific storage location corresponding to the serial number to meet online order delivery, rather than randomly picking other devices in the same inventory unit according to the traditional first-in-first-out rule;

[0046] For replenishment and transfer operations, the system generates corresponding purchase orders and transfer work orders, which are sent to the enterprise resource planning procurement module and the store inventory transfer module through the application programming interface.

[0047] The system continuously monitors the execution feedback results of the above instructions, including whether the equipment has been actually sold and whether the picking and warehousing has been completed. The system then sends the feedback results back to the data perception and matrix construction layer, triggering the status update of the corresponding equipment serial number element in the health status distribution matrix, thus completing the closed loop between algorithm decision-making and physical execution.

[0048] To facilitate understanding and implementation of this invention by those skilled in the art, the technical solution of this invention is further described below with reference to a specific application example. Assume that the physical equipment corresponding to a certain inventory unit is a certain model of robotic vacuum cleaner. Currently, there are three devices in the inventory, with serial numbers SN001, SN002, and SN003. The system reads through the IoT interface that the cumulative charge / discharge cycle count of SN001 accounts for 0.10 of the rated cycle count threshold, and the environmental exposure time accounts for 0.08 of the exposure time threshold. This device is brand new and its quality inspection scoring function is set to 1. SN002 is a device returned through a trade-in program, with a cumulative charge / discharge cycle count ratio of 0.65 and an environmental exposure time ratio of 0.55, scoring 0.6 by the automated quality inspection station. SN003 is also a device returned through a trade-in program, with a cumulative charge / discharge cycle count ratio of 0.82 and an environmental exposure time ratio of 0.70, scoring 0.4 by the automated quality inspection station. The weighting coefficients are then used. It is 0.4. It is 0.3. The value is 0.3. Based on the aforementioned health status value calculation method, the health status value of SN001 is approximately 0.94, the health status value of SN002 is approximately 0.60, and the health status value of SN003 is approximately 0.44. The system aggregates the above three health status values ​​to form the health status distribution matrix corresponding to the inventory unit, and establishes a mapping relationship with the three device serial numbers respectively.

[0049] The core algorithm processing layer uses the health status distribution matrix and real-time demand data from current stores and online channels as joint state inputs to the policy network of three types of agents. In the joint actions output by the policy network, the discount agent triggers a discount action corresponding to the dynamic discount rate output for the inventory unit. The physical execution instruction issuing layer filters the health status distribution matrix of the inventory unit based on a preset health decay threshold (set to 0.5), identifying that the health status value of SN003 is below this threshold, while the health status values ​​of SN001 and SN002 are not below the threshold. Therefore, [they are] [selected as] [the appropriate action]. SN003 is included in the set of devices to be processed. The system then generates a promotional instruction containing the SN003 serial number and the corresponding discount, which is sent to the store sales terminal system through the application programming interface. This causes SN003 to be displayed at a differentiated discount price at the sales terminal. At the same time, other devices with higher health status values ​​in this inventory unit are still sold at the regular price. When SN003 is purchased by a customer after the promotion and the shipment is completed, the system receives the execution feedback result of the shipment event and deletes the matrix element corresponding to SN003 from the health status distribution matrix of this inventory unit, thus completing a complete closed-loop control process.

[0050] It should be understood that the above embodiments are merely preferred embodiments of the present invention, and the embodiments of the present invention are not limited thereto. In the core algorithm processing layer, in addition to using the proximal policy optimization method with pruning mechanism for policy network updates, other reinforcement learning algorithms suitable for continuous action spaces can also be selected according to actual needs. Similarly, policy training can be performed based on the joint reward function and continuous-time Markov decision process state transition model of the present invention to achieve joint optimization of replenishment, allocation, and discount actions, achieving the same or similar technical effects as the present invention. In the data perception and matrix construction layer, in addition to relying on the built-in IoT sensors of the device to collect physical operating parameters in real time, regular manual inspections can be set up in stores and warehouses in combination with handheld detection devices to periodically obtain the physical performance score of the device. The score data can be substituted into the health status numerical calculation method and distribution matrix construction method of the present invention to replace the data source of automated sensors. This can also support the normal operation of the subsequent core algorithm processing layer and physical execution command issuance layer. All equivalent substitutions or improvements made on the basis of the technical solution of the present invention should fall within the scope of protection claimed by the present invention.

[0051] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for managing retail inventory of equipment, characterized in that, include: Obtain the physical operating parameters and quality inspection score data of each physical device under the same inventory unit, and calculate the health status value of each physical device based on the physical operating parameters and the quality inspection score data; Based on the health status values ​​of all physical devices under the same inventory unit, a health status distribution matrix corresponding to the inventory unit is constructed, and each element in the health status distribution matrix is ​​mapped to the corresponding device serial number. The health status distribution matrix and the front-end multi-channel demand data are used as a joint state and input into the continuous-time Markov decision process environment model. The replenishment quantity, transfer path and dynamic discount rate are output by three types of policy networks: replenishment agent, transfer agent and discount agent, respectively, to form a joint action. The parameters of the three types of policy networks are iteratively updated based on the joint reward function, which includes sales profit reward, inventory holding cost reward, stockout penalty reward and health status recovery value reward. The inventory holding cost reward is calculated based on the health status value of each physical device, and the health status recovery value reward is calculated based on the health status value of the device that triggered the dynamic discount action and the priority delivery action. Based on the preset health decay threshold, the device serial numbers with health status values ​​lower than the threshold are selected from the health status distribution matrix of the inventory units corresponding to at least one of the dynamic discount rate and the allocation path, and a set of devices to be processed is formed. For the devices in the set of devices to be processed that trigger a dynamic discount action, a promotional instruction containing the device serial number and the discount amount is generated and sent to the store sales terminal system. For the devices in the set of devices to be processed that trigger a priority delivery action, a picking instruction containing the device serial number and the target storage location information is generated and sent to the warehouse management system. Receive the execution feedback results of the promotion instruction and the picking instruction, and update the status of the corresponding equipment serial number in the health status distribution matrix of the corresponding inventory unit according to the execution feedback results.

2. The equipment retail inventory management method according to claim 1, characterized in that: The health status values ​​of each physical device are obtained by weighted summation of the charge-discharge cycle loss item, the environmental exposure loss item, and the quality inspection score item. The charge-discharge cycle loss item is determined based on the ratio of the cumulative number of charge-discharge cycles of the physical device to the rated cycle number threshold. The environmental exposure loss item is determined based on the ratio of the cumulative environmental exposure duration of the physical device to the exposure duration threshold. The quality inspection score item is determined based on the standardized quality inspection score function obtained by the physical device on the automated quality inspection station. The sum of the weight coefficients corresponding to the three loss items is 1.

3. The equipment retail inventory management method according to claim 1, characterized in that: The state transition rate in the continuous-time Markov decision process environment model is jointly determined by the sales rate, return arrival rate, physical decay rate, and replenishment arrival rate. The joint state of the environment model is composed of the inventory quantity of each inventory unit and the corresponding health state distribution matrix.

4. The equipment retail inventory management method according to claim 1, characterized in that: The calculation method for the inventory holding cost incentive is as follows: for all physical equipment currently in stock, the product of the original value of each physical equipment and 1 minus the difference between the health status value of the equipment is accumulated, and then multiplied by the unit time holding cost discount factor.

5. The equipment retail inventory management method according to claim 1, characterized in that: The calculation method for the health status recovery value reward is as follows: For each device in the set of devices that triggers the dynamic discount action and the priority delivery action, the estimated probability of the device being sold after the action is triggered, the difference between 1 and the ratio of the device's health status value to the critical threshold of health decay, and the current market valuation of the device are multiplied together, and then the product results of all devices in the set of devices are accumulated.

6. The equipment retail inventory management method according to claim 1, characterized in that: The parameter iterative update of the three types of policy networks adopts a near-end policy optimization method with pruning mechanism. By limiting the probability ratio of the new and old policies on the same state action pair to a preset pruning range, the policy optimization objective function with pruning mechanism is calculated, and the policy network parameters are updated by gradient based on the objective function.

7. The equipment retail inventory management method according to claim 1, characterized in that: The construction of the health status distribution matrix also includes: periodically reading the charge / discharge cycle count and cumulative environmental exposure value of the equipment in the warehouse through the Internet of Things interface; for newly added reverse logistics equipment in the current period, reading the output result of the quality inspection scoring function after the equipment completes automated quality inspection; calculating the health status value of the corresponding physical equipment based on the read data; updating the values ​​of the existing equipment serial numbers in the health status distribution matrix according to the correspondence between equipment serial numbers and inventory units; adding matrix elements to the serial numbers of newly added equipment; and deleting matrix elements from the serial numbers of equipment that have been shipped out.

8. The equipment retail inventory management method according to claim 1, characterized in that, Also includes: For the replenishment quantity and the transfer route, corresponding purchase orders and transfer work orders are generated respectively, and sent to the enterprise resource planning procurement module and the store inventory transfer module respectively through the application programming interface.

9. A retail inventory management system for equipment, characterized in that, It includes a data perception and matrix construction module, a core algorithm processing module, and a physical execution instruction issuance module; The data perception and matrix construction module is used to obtain the physical operating parameters and quality inspection score data of each physical device under the same inventory unit, calculate the health status value of each physical device based on the physical operating parameters and the quality inspection score data, and construct the health status distribution matrix corresponding to the inventory unit based on the health status values ​​of all physical devices under the same inventory unit. Each element in the health status distribution matrix is ​​mapped to the corresponding device serial number. The core algorithm processing module is used to input the health status distribution matrix and the front-end multi-channel demand data as a joint state into the continuous-time Markov decision process environment model. Through three types of strategy networks—replenishment agent, allocation agent, and discount agent—the module outputs the replenishment quantity, allocation path, and dynamic discount rate, respectively, forming a joint action. The module iteratively updates the parameters of the three types of strategy networks based on a joint reward function, which includes sales profit reward, inventory holding cost reward, stockout penalty reward, and health status recovery value reward. The physical execution instruction issuing module is used to select device serial numbers with health status values ​​lower than the preset health decay threshold from the health status distribution matrix of the inventory unit corresponding to at least one of the dynamic discount rate and the transfer path to form a set of devices to be processed. For the devices in the set of devices to be processed that trigger dynamic discount actions, a promotional instruction containing the device serial number and discount amount is generated and sent to the store sales terminal system. For the devices that trigger priority delivery actions, a picking instruction containing the device serial number and target location information is generated and sent to the warehouse management system. The module also receives the execution feedback results of the promotional instruction and the picking instruction and sends the execution feedback results back to the data perception and matrix construction module to update the status of the corresponding device serial number in the health status distribution matrix of the corresponding inventory unit.

Images