Pet product intelligent replenishment system based on sales rate prediction

Abstract

The application discloses a pet product intelligent replenishment system based on sales rate prediction, and particularly relates to the technical field of supply chain intelligent replenishment, and comprises the following modules: a data acquisition module which outputs a real steady-state consumption rate of pet products by stripping off promotion interference; a dynamic performance time limit evaluation module which dynamically corrects a performance lead time and reconstructs a reordering point threshold value in combination with an operating capacity track and delivery data; a multi-target collaborative decision optimization module which introduces an exponential default penalty mechanism for multi-constraint solution with the minimum comprehensive cost as the target; a self-adaptive routing hard switching module which adaptively hard switches among mixed loading, carpooling, cross-warehouse direct delivery and a conventional replenishment mode according to a urgency difference value and a three-dimensional time slot matrix and issues end-to-end instructions; and an instruction execution and feedback interface module which issues instructions and receives execution feedback to update a system feature matrix in a closed loop.

Description

Technical Field

[0001] This invention relates to the field of intelligent replenishment technology in the supply chain, and more specifically, to an intelligent replenishment system for pet supplies based on sales rate prediction. Background Technology

[0002] With the continued boom in the pet economy, the pet supplies market is rapidly expanding, characterized by a wide variety of products, high purchase frequency, and significant demand volatility, placing higher demands on supply chain management. Intelligent replenishment systems, as a key tool connecting demand forecasting and inventory decisions, have become an important means for pet supplies companies to improve operational efficiency. Existing technologies include automated replenishment solutions based on sales forecasting, such as predicting future sales through time series analysis or machine learning models and generating replenishment suggestions accordingly. Some solutions also introduce multi-store collaborative management mechanisms, achieving differentiated replenishment through customer profiling analysis, or utilizing deep time series models to improve forecast accuracy in multi-distribution center scenarios.

[0003] However, existing replenishment systems generally suffer from the following technical defects: First, the accuracy of prediction is easily affected by external disturbances and the disconnect between product category attributes; sales pulses caused by promotional activities are often misjudged as normal consumption, resulting in replenishment quantities deviating from actual demand; at the same time, traditional systems lack a mechanism to fit and cross-validate the inherent attributes of pet products with the actual sales rate, and cannot effectively isolate false demand.

[0004] Second, replenishment decisions are disconnected from the dynamic execution of the supply chain; most systems take the static "how much to replenish" as the core output, treat the supplier's delivery cycle as a fixed constant, ignore the dynamic fluctuations in trunk logistics timeliness, and cannot make real-time corrections to reorder points and dynamic safety stock based on real-time transportation capacity trajectories, resulting in a misalignment between safety stock settings and actual arrival risks.

[0005] Third, there is a lack of end-to-end scheduling solutions for integrated procurement and logistics service scenarios. For third-party service providers that undertake the functions of coordinating multiple suppliers, warehouses, and channels, multiple hard constraints such as procurement costs, tiered discounts, vehicle loading rates, storage capacity, and minimum order quantities from suppliers need to be incorporated into the decision-making process simultaneously. Furthermore, there is a lack of a closed-loop mechanism that allows for adaptive hard switching between mixed loading and cross-warehouse direct delivery based on the urgency of replenishment, making it difficult to adapt to complex end-to-end integrated business scenarios. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a smart replenishment system for pet supplies based on sales rate prediction, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent replenishment system for pet supplies based on sales rate prediction, comprising:

[0008] The data acquisition module is used to collect order data from various sales channels, pet product attribute data, inventory level data from various warehousing nodes, and real-time transportation capacity trajectory data from the logistics system in real time. Based on the theoretical consumption cycle and coefficient of variation of pet products, it performs promotion interference removal and sales rate weight reduction correction on the original sales data and outputs the true consumption rate.

[0009] The dynamic fulfillment timeliness assessment module is used to dynamically fit the supplier's fulfillment delay probability based on the supplier's historical delivery data and the real-time trajectory of trunk logistics, correct the static lead time to the dynamic fulfillment lead time, and reconstruct the reorder point threshold of each warehousing node in real time.

[0010] The multi-objective collaborative decision optimization module is used to minimize the overall supply chain cost as the objective function. It sets the supplier's minimum order quantity, procurement tiered discount, vehicle volume loading rate and available warehouse capacity of the destination warehouse as hard constraints, and solves the problem by combining the actual consumption rate and the reorder point threshold. It also introduces an exponential time-sensitive default penalty mechanism to shrink the feasible region.

[0011] The adaptive routing hard switching module is used to automatically generate end-to-end scheduling instructions containing replenishment quantity, replenishment timing and transportation route based on the replenishment urgency difference and waiting tolerance threshold of SKUs on the same link, as well as the matching results of the procurement demand time window and the three-dimensional time slot feature matrix of the supply network, and to perform automatic hard switching between mixed loading and carpooling, cross-warehouse direct delivery and regular replenishment modes.

[0012] The instruction execution and feedback interface module is used to output the end-to-end scheduling instructions to the ERP system, WMS warehouse management system and TMS logistics management system, and to receive execution feedback data to dynamically update the three-dimensional time slot feature matrix.

[0013] Preferably, as a preferred embodiment of the intelligent pet supplies replenishment system based on sales rate prediction described in this invention, it includes the data acquisition module, specifically comprising the following:

[0014] A data stream of orders and purchases containing supplier historical fulfillment and delivery records is constructed using standardized API interfaces. Pet supplies attribute data Includes real-time available inventory. Inventory water level data And capacity tracking data including the supplier's historical fulfillment and delivery records. The unified four-dimensional dataset ;

[0015] The collected pet product attribute data is transmitted to the real consumption rate assessment logic. Combined with preset pet physiological and consumption benchmarks, the theoretical consumption cycle for each product category is calculated, specifically including:

[0016] For pet staple food: Access a pre-set pet physiology database to obtain the net content per package. Feed conversion rate Average weight of pets within the service area and the daily recommended intake coefficient The theoretical number of days a single bag of pet food can last under ideal feeding conditions is calculated as follows: ,in, This refers to the theoretical number of days a single pack of pet supplies can last.

[0017] For non-staple pet food products, extract the adjacent purchase timestamps of the same SKU from the same end customer within the statistical period. and The theoretical consumption period is calculated using a median function, and the specific formula is as follows:

[0018] ;in, This refers to the theoretical consumption cycle of non-staple pet food products. The median function is used to calculate the median number of purchases within the statistical period; m represents the total number of purchases made by a customer within the statistical period. At that time, the preset industry benchmark consumption cycle is directly retrieved as... reveal all the details;

[0019] Extract the raw sales figures for each day within the sliding time window, using the current date as the endpoint. Based on the standard deviation of daily sales with the mean Calculate the coefficient of variation And calculate the actual number of days consumed within the sliding window. When the actual number of days consumed is met And coefficient of variation At that time, it is determined that a promotional pulse is triggered; among which, This is the preset deviation sensitivity coefficient; The threshold for the coefficient of variation; Net content per single package. This represents the average daily sales volume within the window.

[0020] Iterate through each day within the window, and based on the promotional markers and promotional activity identifiers in the order and purchase data streams, record whether the day is a promotional day. For day t within the sliding time window:

[0021] When a day is determined to be a non-promotion day, the promotion intensity coefficient is zero, and the adjusted equivalent sales volume is... equal to original sales ;

[0022] When a day is determined to be a promotional day, an exponential decay factor is introduced. The original sales figures for the day will be adjusted for weight reduction as follows: The corrected equivalent daily sales are: ;in, Set the preset attenuation adjustment coefficient; set the lower limit protection value of the attenuation factor. , when the formula At that time, a mandatory order ;

[0023] The corrected equivalent sales for each day within the window are weighted and averaged to calculate and output the true steady-state consumption rate. ;in, To determine the true steady-state consumption rate, when the sliding time window consists entirely of promotional days, the average consumption rate from the previous non-promotional window is automatically extracted as the base rate and compared with the currently calculated rate. Smooth the surface using a preset weighted ratio.

[0024] Preferably, as a preferred embodiment of the intelligent replenishment system for pet supplies based on sales rate prediction described in this invention, it includes the dynamic fulfillment timeliness assessment module, which retrieves capacity trajectory data and historical fulfillment delivery records from the four-dimensional dataset, and reconstructs the reorder point threshold (ROP) of each warehousing node in real time, specifically including:

[0025] Extract the location coordinates and traffic status of trunk transport vehicles, and calculate the real-time deviation duration of currently en route trunk logistics. Extract a sample dataset of the time difference between the actual production time and the agreed delivery time of the target supplier's past N batches of orders, and fit the supplier's expected production delay value. With variance ;

[0026] Retrieve preset static lead time of supplier contracts Combined with the real-time deviation time of trunk logistics Expected value of factory delay Calculate and generate dynamic lead time for contract fulfillment. ;in, For dynamic lead time for contract fulfillment;

[0027] The dynamic performance evaluation module uses a normal distribution to fit the supplier's factory delay probability distribution to obtain its expected factory delay value. When the fitted At that time, a mandatory order When calculated When this occurs, the lower limit protection mechanism is triggered, causing... When the transport capacity trajectory data is interrupted or abnormal, the historical average transit time of the past 30 days for that transport route is automatically retrieved as the real-time deviation time for the current trunk line logistics. The fallback alternative value;

[0028] Based on dynamic performance lead time Based on variance and demand fluctuations, reconstruct dynamic safety stock levels. ,in, This is a dynamic safety stock level, measured in units of pieces. The actual steady-state consumption rate output by the data acquisition module; This represents the variance of daily sales within the sliding time window. The variance of dynamic lead time is calculated based on the supplier's factory delay variance. With the variance of trunk logistics The result is obtained by superposition; The safety factor corresponding to the service level;

[0029] The true steady-state consumption rate output by the data acquisition module Calculated dynamic lead time and the calculated dynamic safety stock. Combine and reconstruct the reorder point thresholds for each warehousing node in real time: ;in, The reorder point threshold after reconstruction, in units of pieces;

[0030] Continuously monitor the current available inventory at each warehouse node. ,when At that time, a replenishment trigger signal will be automatically generated, which will include the target SKU and the current available inventory. True steady-state consumption rate and dynamic lead time for performance The feature data package is sent to the multi-objective collaborative decision optimization module for replenishment scheme calculation.

[0031] Preferably, as a preferred embodiment of the intelligent pet supplies replenishment system based on sales rate prediction described in this invention, it includes the multi-objective collaborative decision optimization module, specifically comprising the following:

[0032] When a replenishment trigger signal is received, the target SKU identifier and current available inventory are retrieved in real time from the feature data packet. True steady-state consumption rate and dynamic lead time for performance Set the optimization decision variable matrix as follows: ,in, This represents the replenishment quantity for purchasing the k-th SKU from the supplier, where k is the total number of candidate SKU types destined for the same warehouse.

[0033] An optimization model is constructed with the objective of minimizing the total supply chain cost (TC). The specific formula is as follows:

[0034] ;

[0035] Wherein, TC represents the total cost of the integrated supply chain; the specific derivation of each sub-cost item is as follows:

[0036] Procurement costs: ,in, For procurement costs, For the k-th SKU, the purchase quantity is The dynamic unit price at any given time is determined by the supplier's tiered discount function:

[0037] when hour, ,in, The minimum order quantity for the nth tier of the k-th SKU. This corresponds to the tiered discount unit price;

[0038] Transportation costs: Calculated using a tiered freight rate model based on vehicle volume and load capacity. ,in, For transportation costs, The number of vehicles to be called. For the freight of a single vehicle, For LTL (Less-than-Truckload) freight, The remaining volume is less than the entire vehicle's volume;

[0039] Warehouse holding costs: ,in, For storage and holding costs, Let k be the dynamic safety stock level for the k-th SKU. Let $k$ be the holding cost per unit of product for the $k$-th SKU per unit of time. This represents the actual steady-state consumption rate corresponding to the k-th SKU;

[0040] Cost of stockout losses: ,in, To avoid losses due to stockouts, Let k be the current available inventory of the k-th SKU. Let $k$ be the unit stockout penalty cost for the $k$-th SKU. This is a function to find the maximum value.

[0041] This is a non-linear, exponential, time-limited penalty term, with the specific formula as follows: ;

[0042] in, For the cost of penalties for breach of contract; The default coefficient is the preset base default coefficient; It is an exponential penalty factor; Set the maximum expected arrival time window for the destination warehouse; set a maximum protection value for penalty items. , For transportation costs, when the calculation result exceeds... At that time, a mandatory order ;

[0043] Define hard boundary constraints, including:

[0044] Supplier minimum order quantity constraint: For any selected k-th SKU, the following condition must be met. or ;in, Minimum order quantity;

[0045] Hard constraint on vehicle loading rate: The total loading volume must meet the following requirements. ,in, Let K be the volume of a single item of the k-th SKU. To meet the minimum load requirement, This refers to the vehicle's rated maximum volume.

[0046] Available storage capacity constraints for the destination warehouse: ,in, This refers to the current real-time remaining available storage capacity of the destination warehouse.

[0047] The hierarchical branch and bound algorithm is used to solve the optimization model; a 500ms timeout threshold is set. If the timeout occurs and no feasible solution is found, the reorder point shortfall is calculated. Minimum order quantity with supplier The maximum value is used to generate an emergency replenishment vector and issue the optimal replenishment decision vector. The optimal replenishment decision vector obtained by solving the problem. The instructions are sent to the adaptive routing hard handover module to generate capacity routes and scheduling commands. This represents the final optimal replenishment quantity calculated for the k-th SKU.

[0048] Preferably, as a preferred embodiment of the intelligent pet supplies replenishment system based on sales rate prediction described in this invention, it includes the adaptive routing hard switching module, specifically comprising the following:

[0049] Receive the optimal replenishment decision vector Based on the real-time inventory and consumption rate of each SKU, the replenishment urgency of each SKU in the same supply chain is calculated. ;in, The urgency of replenishing the k-th SKU; This is the reorder point threshold; This represents the current available inventory of the k-th SKU; This represents the true steady-state consumption rate.

[0050] Calculate the difference in replenishment urgency between the j-th SKU and the k-th SKU in the same supply chain. Extract the dynamic waiting tolerance threshold for the j-th SKU in the target warehouse. ;in, This is a function to find the maximum value.

[0051] Retrieve supply network time slot data from the TMS logistics management system and construct a three-dimensional time slot feature matrix containing node coordinate dimension X, time slot dimension Y, and transportation capacity status dimension Z. ; and calculate the procurement demand time window and the aforementioned Time slot overlap ;

[0052] Based on the aforementioned urgency difference Waiting tolerance threshold Vehicle loading rate and time slot overlap Adaptive hard switching between the three transportation modes;

[0053] The adaptive hard switching between the three transportation modes specifically includes:

[0054] When the following conditions are met: and and At that time, a hard switch to mixed-load carpooling mode is implemented; merging orders for carpooling improves loading efficiency, among which, This is the waiting tolerance threshold;

[0055] When the mixed-carpooling mode is not met, but the following conditions are met: Time slot overlap occurs in the destination warehouse Furthermore, there are matching time slots at the transit cross-database nodes. When necessary, a hard switch to cross-warehouse direct delivery mode is performed; a cross-warehouse instruction is generated to bypass the central warehouse and directly deliver to the destination warehouse on the transit platform;

[0056] If none of the above conditions are met, the system will default to the regular replenishment mode; otherwise, it will proceed as originally planned. Issue standard point-to-point trunk line transportation instructions in batches;

[0057] After completing the hard mode switch decision, the adaptive routing hard switch module automatically aggregates and generates an end-to-end integrated scheduling instruction data packet, including: the replenishment quantity of specific SKU combinations. Loading and shipping sequence The system adapts to transportation routes and target vehicle / container IDs; and pushes scheduling instructions to the WMS warehouse management system and capacity locking and electronic lock opening / closing instructions to the TMS logistics management system.

[0058] Preferably, as a preferred embodiment of the intelligent pet supplies replenishment system based on sales rate prediction described in this invention, it includes the instruction execution and feedback interface module, establishing a bidirectional long-connection channel to achieve positive instruction issuance and reverse data closed-loop feedback, specifically including the following:

[0059] The system connects to the ERP system, WMS warehouse management system, and TMS logistics management system through a pre-defined standardized open interface. It receives end-to-end integrated scheduling instruction data packets and pushes them to the back-end ERP system to lock purchase accounts and discount unit prices, to the WMS warehouse management system for automatic scheduling of picking and reservation of warehouse capacity, and to the TMS logistics management system to lock time slots and draw tracks.

[0060] Establish a two-way long connection channel to capture the physical node feedback data stream containing WMS actual picking data, TMS vehicle real-time trajectory and ERP warehouse confirmation status; after normalization and cleaning, when picking shortage or TMS dispatch timeout is detected, send a local recalculation request to the multi-objective collaborative decision optimization module.

[0061] The cleaned execution feedback data is then injected back into the upstream module's dataset to achieve incremental closed-loop updates of parameters across the entire process. Specifically, this includes:

[0062] The time difference between the actual factory departure time and the agreed delivery time of this order is used as a new sample and pushed into the supplier's fulfillment and delivery history dataset N. The oldest batch of data is automatically discarded, keeping window N constant. The expected value of the supplier's factory departure delay is dynamically updated. ,in, For the updated supplier delivery delay expectations, The previous supplier's expected delivery delay was used. This is the time difference between the actual manufacturing time and the agreed delivery time for this order. If delivery is delayed, When delivered on time, ; To smooth the update factor;

[0063] Based on real-time capacity utilization and delivery time feedback from the TMS logistics management system, the three-dimensional time slot feature matrix is ​​analyzed. Perform real-time dimensionality reduction and coverage updates, including:

[0064] Capacity status dimension Z: based on actual loading volume The remaining available volume of the corresponding road segment and time slot will be deducted. With remaining available load capacity ;

[0065] Time window dimension Y: Dynamically adjust the baseline of travel time for future time slots on the transportation route based on the actual travel time of vehicles.

[0066] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0067] By using a data acquisition module to filter out promotional interference and outputting a steady-state rate that reflects the true consumption rhythm of pets, false demand impulses are eliminated at the source. Combined with dynamic fulfillment lead time to reconstruct reorder points in real time, the lag in dealing with supply chain delays caused by traditional static lead time is overcome. The multi-objective optimization module introduces an exponential timeliness default penalty mechanism, achieving efficient optimization under hard constraints of procurement discounts, transportation capacity loading rate, and warehouse capacity. The adaptive routing module performs millisecond-level hard switching between mixed loading and carpooling, cross-warehouse direct delivery, and conventional modes, and realizes parameter adaptive iteration closed loop based on bidirectional interface, which significantly reduces the overall supply chain cost and warehouse backlog risk while ensuring order fulfillment rate. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0069] Figure 1 This is a system architecture block diagram according to an embodiment of the present invention;

[0070] Figure 2 This is a flowchart illustrating the data acquisition and actual consumption rate calculation process in an embodiment of the present invention.

[0071] Figure 3 This is a flowchart of the adaptive routing hard handover decision logic in an embodiment of the present invention;

[0072] Figure 4 This is a schematic diagram of the instruction execution and feedback closed-loop architecture in an embodiment of the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0074] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0075] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0076] Example 1

[0077] This embodiment provides, for example Figure 1 The illustrated intelligent replenishment system for pet supplies based on sales rate prediction specifically includes:

[0078] The data acquisition module is used to collect order data from various sales channels, pet product attribute data, inventory level data from various warehousing nodes, and real-time transportation capacity trajectory data from the logistics system in real time. Based on the theoretical consumption cycle and coefficient of variation of pet products, it performs promotion interference removal and sales rate weight reduction correction on the original sales data and outputs the true consumption rate.

[0079] The dynamic fulfillment timeliness assessment module is used to dynamically fit the supplier's fulfillment delay probability based on the supplier's historical delivery data and the real-time trajectory of trunk logistics, correct the static lead time to the dynamic fulfillment lead time, and reconstruct the reorder point threshold of each warehousing node in real time.

[0080] The multi-objective collaborative decision optimization module is used to minimize the overall supply chain cost as the objective function. It sets the supplier's minimum order quantity, procurement tiered discount, vehicle volume loading rate and available warehouse capacity of the destination warehouse as hard constraints, and solves the problem by combining the actual consumption rate and the reorder point threshold. It also introduces an exponential time-sensitive default penalty mechanism to shrink the feasible region.

[0081] The adaptive routing hard switching module is used to automatically generate end-to-end scheduling instructions containing replenishment quantity, replenishment timing and transportation route based on the replenishment urgency difference and waiting tolerance threshold of SKUs on the same link, as well as the matching results of the procurement demand time window and the three-dimensional time slot feature matrix of the supply network, and to perform automatic hard switching between mixed loading and carpooling, cross-warehouse direct delivery and regular replenishment modes.

[0082] The instruction execution and feedback interface module is used to output the end-to-end scheduling instructions to the ERP system, WMS warehouse management system and TMS logistics management system, and to receive execution feedback data to dynamically update the three-dimensional time slot feature matrix.

[0083] In this embodiment, the data acquisition module is specifically described. This module is used to collect real-time order data from various sales channels, pet product attribute data, inventory level data from various warehousing nodes, and real-time transportation capacity trajectory data from the logistics system. Based on the theoretical consumption cycle and coefficient of variation of pet products, it performs promotional interference removal and sales rate weighting correction on the original sales data, outputting the true consumption rate. Specifically, this includes the following:

[0084] By establishing bidirectional data pathways with various peripheral systems through standardized API interfaces, millisecond-level acquisition and structured parsing of multi-source heterogeneous data can be achieved, specifically including:

[0085] By connecting with front-end e-commerce platforms and offline POS systems, order data streams are collected, including order generation timestamps, product SKU identifiers, transaction quantities, and promotional activity identifiers. The promotional activity identifier includes a promotion type code and a promotion start and end timestamp. The promotion type code includes at least five types: discounts, buy-one-get-one-free, flash sales, and platform promotions. Different types of promotional activities correspond to different promotion intensity coefficient assignment ranges. Based on the promotion type code and promotion start and end timestamps, it automatically determines whether each order falls within the promotion time window. When the order timestamp is within the start and end time range of any promotional activity, the order is marked as a promotional order and the corresponding promotion type code is recorded. Furthermore, by connecting with the back-end ERP system, historical supplier fulfillment and delivery records, including purchase order issuance timestamps, agreed delivery timestamps, and actual factory delivery timestamps, are collected to jointly construct order and purchase data streams.

[0086] Pet product attribute data, including product packaging specifications, weight, and feeding conversion rate, is obtained by connecting to the back-end ERP system.

[0087] By connecting to the WMS warehouse management system of each warehouse node, real-time inventory level data of each warehouse node can be obtained. This includes current available inventory. Lock in inventory levels and pre-inspection records;

[0088] By connecting to a third-party TMS logistics management system, the system acquires transportation trajectory data and extracts real-time location information of trunk line transport vehicles, historical average travel time for each road segment, and node receipt timestamps. The historical average travel time for each road segment is dynamically calculated by time period, which is divided into three categories: peak season, off-season, and low season. Under each category, the historical average travel time and fluctuation variance of each road segment are stored. The location information of trunk line transport vehicles is matched with road segments using GPS coordinates and electronic map road networks to automatically identify the current road segment number and estimated remaining travel time of the vehicle. The node receipt timestamps are used to calculate the actual time consumed from the departure node to the current node and compare it with the historical average travel time for each road segment to generate a timeliness deviation index.

[0089] The collected data was standardized and transformed to construct a unified four-dimensional dataset. ,in, For order and procurement data flow, For pet supplies attribute data, This provides real-time inventory level data for each storage node. This data represents the transportation trajectory; it is stored in an in-memory database for high-speed retrieval by subsequent modules.

[0090] The collected pet product attribute data is transmitted to the real consumption rate assessment logic. Combined with preset pet physiological and consumption benchmarks, the theoretical consumption cycle for each product category is calculated, specifically including:

[0091] For pet staple food: Access a pre-set pet physiology database to obtain the net content per package. Feed conversion rate Average weight of pets within the service area and the daily recommended intake coefficient The pet physiology database is a built-in knowledge base of the system, storing corresponding physiological parameter standard values ​​according to pet category, breed, and age stage; it calculates the theoretical number of days a single bag of pet food can be consumed under ideal feeding conditions. ,in, This refers to the theoretical number of days a single pack of pet supplies can last, expressed in days per pack. This is the net content per single package, in kilograms, directly read from the pet supplies attribute data; Feeding conversion rate is the ratio of the actual amount consumed by the pet to the amount labeled on the packaging. It is derived by the system based on historical actual consumption data of the SKU. The default value is 0.95, and the value range is from 0.85 to 1.15. The average weight of pets within the service area, in kilograms, is obtained by statistically analyzing pet information in user registration profiles and is updated monthly. The recommended daily food intake coefficient for pets, in kilograms per kilogram of body weight per day, is obtained by looking up tables in the pet physiology database based on pet type, breed, and age group.

[0092] For non-staple pet products, including but not limited to cat litter, pet pee pads, and grooming products, the theoretical consumption cycle is determined based on the median purchase interval of historical sales data:

[0093] ;in, This refers to the theoretical consumption cycle of non-staple pet food products. and These are the timestamps of the i-th and (i-1)-th purchases of the same SKU by the same end customer. The median function is used, where m represents the total number of purchases within the statistical period; the statistical period defaults to the past 180 days, and the number of customer purchases within that period is used as the median function. In this case, the cold start fallback logic is automatically activated, directly retrieving the preset industry benchmark consumption cycle for non-staple grain categories from the ERP master database as... This is to ensure the continuity of evaluation during cold starts;

[0094] Starting from the current date, a sliding time window of a preset length is drawn backward, and the raw sales figures for each day within the window are extracted. Based on the standard deviation of daily sales with the mean Calculate the coefficient of variation And calculate the actual number of days consumed within the sliding window. When the actual number of days consumed is met And coefficient of variation At that time, it is determined that a promotional pulse is triggered; among which, The preset deviation sensitivity coefficient ranges from 0.3 to 0.6, with a default value of 0.5. This is the threshold for the coefficient of variation, set based on the historical sales stability of the product, with a value range of 0.6 to 1.0, and a default value of 0.8. Net content per single package. This represents the average daily sales volume within the window.

[0095] Iterate through each day within the window, and based on the promotional markers and promotional activity identifiers in the order and purchase data streams, record whether the day is a promotional day. For day t within the sliding time window:

[0096] When a day is determined to be a non-promotion day, the promotion intensity coefficient is zero, and the adjusted equivalent sales volume is... equal to original sales ;

[0097] When a day is determined to be a promotional day, an exponential decay factor is introduced. The original sales figures for the day will be adjusted for weight reduction as follows: The corrected equivalent daily sales are: ;in, This is a preset attenuation adjustment coefficient, ranging from 0.5 to 2.0, with a default value of 1.0; to prevent overfitting from causing excessive sales reduction, a lower limit protection value for the attenuation factor is set. , when the formula At that time, a mandatory order ;

[0098] The corrected equivalent sales for each day within the window are weighted and averaged to calculate and output the true steady-state consumption rate. ;in, To determine the true steady-state consumption rate, when the sliding time window consists entirely of promotional days, the average consumption rate from the previous non-promotional window is automatically extracted as the base rate and compared with the currently calculated rate. Weighted smoothing is performed according to a preset ratio to avoid distortion of the steady-state rate due to prolonged promotional periods.

[0099] like Figure 2 As shown, the theoretical consumption cycle of a product category is calculated by collecting order and SKU attribute data from all channels. Extract sales figures from the sliding window and calculate the coefficient of variation. When satisfied and When a promotion is triggered, an exponential decay factor is introduced. Sales figures are adjusted by reducing their weight; if the conditions are not met, the original sales figures are retained, and the final steady-state consumption rate is calculated and output using a weighted average. .

[0100] In this embodiment, the dynamic fulfillment timeliness assessment module needs to be specifically described. This module is used to dynamically fit the supplier's fulfillment delay probability based on the supplier's historical delivery data and the real-time trajectory of trunk logistics, correcting the static lead time to a dynamic lead time, and reconstructing the reorder point threshold for each warehousing node in real time. Specifically, it includes the following:

[0101] Retrieve 4D dataset Capacity trajectory data in And the target supplier's historical performance and delivery records in the ERP system, specifically including:

[0102] In-transit trunk logistics delay calculation: Real-time extraction of the location coordinates of trunk transport vehicles and the traffic status of road segments along the route, comparison with the historical average travel time of each road segment, and calculation of the real-time deviation time of the current in-transit trunk logistics. The unit is days; when the transport capacity trajectory data is interrupted or abnormal, the historical average travel time of the past 30 days for that transport route is automatically retrieved as... The fallback alternative value;

[0103] Supplier delivery delay probability distribution fitting: Extract a sample dataset of the time difference between the actual delivery time and the agreed delivery time of the target supplier's past N batches of orders, where N is preferably 20 to 50 batches; fit the supplier's delivery delay probability distribution using a normal distribution to obtain its expected delivery delay value. With variance Among them, the expected value of factory delay The unit is days; when the historical delivery sample is less than the preset batch N, the historical average expected delay value of suppliers of the same category in the ERP master database is automatically retrieved as a fallback data; when the fitted... This indicates that the supplier has the ability to deliver ahead of schedule, thus enforcing the order. To prevent premature deduction of lead time from causing replenishment delays;

[0104] Retrieve the preset static lead time of supplier contracts from the ERP master database. The unit is days, combined with the real-time deviation duration of the main line. Expected value of factory delay Calculate and generate dynamic lead time for contract fulfillment. ;in, The dynamic lead time is expressed in days; when the calculated lead time is... When abnormal fluctuations are detected in the trajectory data, the lower limit protection mechanism is forcibly triggered, causing... This is to avoid illogical compression of lead time due to data noise;

[0105] Based on dynamic performance lead time Based on variance and demand fluctuations, reconstruct dynamic safety stock levels. ,in, This is a dynamic safety stock level, measured in units of pieces. The actual steady-state consumption rate output by the data acquisition module; This represents the variance of daily sales within the sliding time window. The variance of dynamic lead time is calculated based on the supplier's factory delay variance. With the variance of trunk logistics The result is obtained by superposition; To correspond to the safety factor of the service level, The value ranges from 1.65 to 2.33, corresponding to a target order fulfillment rate of 95% to 99%; the default value is 1.96, corresponding to a fulfillment rate of 97.5%; the safety factor is determined based on the ABC classification of the SKU. Perform adaptive dynamic mapping: for Similar to core blockbuster products, making This corresponds to a high fulfillment rate of 99%; targeting Similar to regular goods, making This corresponds to a 97.5% fulfillment rate; Regarding Long-tail slow-moving products, making This corresponds to a 95% fulfillment rate, achieving an efficient balance between inventory capital occupation and stockout risk;

[0106] The true steady-state consumption rate output by the data acquisition module Calculated dynamic lead time and the calculated dynamic safety stock. Combine and reconstruct the reorder point thresholds for each warehousing node in real time: ;in, The reorder point threshold after reconstruction, in units of pieces;

[0107] Continuously monitor the current available inventory at each warehouse node. When the current available inventory of a certain warehouse node is monitored At that time, a replenishment trigger signal will be automatically generated, which will include the target SKU and the current available inventory. True steady-state consumption rate and dynamic lead time for performance The feature data package is sent to the multi-objective collaborative decision optimization module for replenishment scheme calculation.

[0108] In this embodiment, the multi-objective collaborative decision optimization module needs to be specifically described. This module uses minimizing the overall supply chain cost as the objective function, setting supplier minimum order quantity, tiered purchase discounts, vehicle volume loading rate, and available warehouse capacity at the destination as hard constraints. It combines the actual consumption rate with the reorder point threshold to solve the problem, and introduces an exponential time-sensitive default penalty mechanism to narrow the feasible region. Specifically, it includes the following:

[0109] When a replenishment trigger signal is received, the target SKU identifier and current available inventory are retrieved in real time from the feature data packet. True steady-state consumption rate and dynamic lead time for performance Set the optimization decision variable matrix as follows: ,in, This represents the replenishment quantity for purchasing the k-th SKU from the supplier, in units of pieces, where k is the total number of candidate SKU types destined for the same warehouse.

[0110] An optimization model is constructed with the objective of minimizing the total supply chain cost (TC). The specific formula is as follows:

[0111] ;

[0112] Where TC represents the total supply chain cost, in yuan; the specific derivation of each sub-cost item is as follows:

[0113] Procurement costs: ,in, For procurement costs, For the k-th SKU, the purchase quantity is The dynamic unit price at any given time is determined by the supplier's tiered discount function:

[0114] when hour, ,in, The minimum order quantity for the nth tier of the k-th SKU. This corresponds to the tiered discount unit price;

[0115] Transportation costs: Calculated using a tiered freight rate model based on vehicle volume and load capacity. ,in, For transportation costs, The number of vehicles to be called. For the freight of a single vehicle, For LTL (Less-than-Truckload) freight, The remaining volume is less than the entire vehicle's volume;

[0116] Warehouse holding costs: ,in, For storage and holding costs, Let k be the dynamic safety stock level for the k-th SKU. Let $k$ be the holding cost per unit of product for the $k$-th SKU per unit of time. This represents the actual steady-state consumption rate corresponding to the k-th SKU;

[0117] Cost of stockout losses: ,in, To avoid losses due to stockouts, Let k be the current available inventory of the k-th SKU. Let $k$ be the unit stockout penalty cost for the $k$-th SKU. This is a function to find the maximum value.

[0118] This is a non-linear, exponential, time-limited penalty term, with the specific formula as follows: ;

[0119] in, The penalty cost for breach of contract is expressed in yuan. The default coefficient is the preset base default coefficient; This is the exponential penalty factor, with a value ranging from 0.5 to 2.0, and a default value of 1.2. The maximum expected delivery time window for the destination warehouse is set at the preset date; when the calculated dynamic fulfillment lead time is used... At that time, the cost of punishment The exponential increase causes the objective function TC to approach infinity. In the early stages of the algorithm's solution process, the feasible decision region is rapidly narrowed, automatically eliminating cheap but high-risk suppliers or transportation routes with delivery delays. To prevent extreme delays from causing floating-point overflow in the computer, a penalty term upper limit is set as a protection value. When the calculation result exceeds At that time, a mandatory order ;

[0120] Within the shrunk feasible region, the following physical hard constraints are used as boundary limits for the optimization model:

[0121] Supplier minimum order quantity constraint: For any selected k-th SKU, the following condition must be met. or ;in, This refers to the minimum order quantity limit set by the supplier for the k-th SKU;

[0122] Hard constraint on vehicle loading rate: The total loading volume must meet the following requirements. ,in, Let K be the volume of a single item of the k-th SKU. To meet the minimum load requirement, 80% of the vehicle's volume is preferred. This refers to the vehicle's rated maximum volume.

[0123] Available storage capacity constraints for the destination warehouse: ,in, The current real-time remaining available warehouse capacity of the destination warehouse is extracted from the WMS warehouse management system by the data acquisition module.

[0124] The hierarchical branch and bound algorithm is used to solve the mixed-integer nonlinear programming model; the termination condition is set as a timeout degradation protection logic.

[0125] When the algorithm satisfies the global relative deviation within a 500ms threshold When the time is right, directly output the obtained global optimal solution;

[0126] When the computation time reaches the 500ms threshold but is not met When the iteration is stopped, the best feasible solution found so far is directly extracted as the decision output.

[0127] When the computation time reaches 500ms and no feasible solution satisfying the hard constraints is found, an emergency degradation mechanism is automatically triggered, and the calculated reorder point gap is adjusted. Minimum order quantity with supplier The maximum value is used to generate an emergency replenishment vector to ensure the continuous issuance of replenishment instructions;

[0128] Through the above solution and degradation mechanism, the final output is the optimal replenishment decision vector that minimizes the total cost (TC). ;in, This represents the final optimal replenishment quantity calculated for the k-th SKU; the resulting optimal replenishment decision vector is... The instructions are sent to the adaptive routing hard switching module to generate capacity routes and scheduling commands.

[0129] In this embodiment, the adaptive routing hard switching module is specifically described. This module is used to automatically generate end-to-end scheduling instructions containing replenishment quantity, replenishment sequence, and transportation route based on the replenishment urgency difference and waiting tolerance threshold of SKUs on the same link, as well as the matching results of the procurement demand time window and the three-dimensional time slot feature matrix of the supply network. It then performs automatic hard switching between mixed loading / cargo sharing, cross-warehouse direct delivery, and regular replenishment modes. Specifically, it includes the following:

[0130] Receive the optimal replenishment decision vector it sends. Based on the real-time inventory and consumption rate of each SKU, the replenishment urgency of each SKU in the same supply chain is calculated. ;

[0131] in, The replenishment urgency for the kth SKU is expressed in days, representing the duration that the current available inventory can support. This is the reorder point threshold; Let k be the current available inventory of the k-th SKU; This represents the true steady-state consumption rate.

[0132] For any two SKUs in the same supply chain, calculate the difference in replenishment urgency between the j-th SKU and the k-th SKU. Extract the dynamic waiting tolerance threshold for the j-th SKU in the target warehouse. This is used to assess whether delayed shipping due to waiting for carpooling will lead to premature stockouts; among them, This is a function to find the maximum value.

[0133] Retrieve supply network time slot data from the TMS logistics management system and construct a three-dimensional time slot feature matrix. The node dimension X includes the spatial coordinates of the originating supplier's factory area, intermediate cross-border transshipment centers, and the destination warehouse; the time window dimension Y is divided into continuous loading, unloading, and passage time slots in 1-hour units. Capacity status dimension Z: includes the remaining available capacity of the corresponding trunk line vehicles within each time slot. With remaining available load capacity ;

[0134] For the optimal replenishment decision vector Procurement demand time window The adaptive routing hard handover module combines it with the three-dimensional time slot feature matrix. Perform cross-matching and calculate time slot overlap. It is used to assess the degree of fit between current transportation capacity resources and procurement needs in the spatial and temporal dimensions;

[0135] Based on the aforementioned urgency difference Waiting tolerance threshold Vehicle loading rate and time slot overlap Adaptive hard switching between the three transportation modes is achieved through the following non-overlapping hard branch conditions:

[0136] Switching condition A: Mixed carpooling mode, when the following conditions are met: and and At that time, a hard switch to mixed-carpooling mode was initiated;

[0137] Execution logic: Suspend individual shipments of SKUs with lower urgency, merge them with high-urgency SKUs in the same supply chain, and perform volume-based order consolidation to forcibly increase vehicle loading rates. Then, the shipping instructions were issued;

[0138] Switching condition B: Cross-database direct delivery mode, when the following conditions are met: Furthermore, there is time slot overlap in the destination warehouse. Furthermore, there are matching time slots at the transit cross-database nodes. At that time, a hard switch to cross-database direct send mode is performed;

[0139] Execution logic: Generate cross-warehouse dispatch instructions that skip the traditional central warehouse temporary storage. Goods are sent directly from the supplier to the cross-warehouse transfer center. After the batching and unpacking are completed on the transfer platform, the goods are directly unloaded and transferred to urban distribution vehicles and sent directly to the destination warehouse, reducing the transfer downtime to less than 2 hours.

[0140] Switching condition C: Normal replenishment mode. When conditions A and B above are not met, the system will switch to normal replenishment mode by default.

[0141] Execution logic: As originally planned Issue standard point-to-point trunk line transportation instructions in batches;

[0142] After completing the hard mode switch decision, the adaptive routing hard switch module automatically aggregates and generates an end-to-end integrated scheduling instruction data packet, including: the replenishment quantity of specific SKU combinations. Loading and shipping sequence Adaptive transportation routes and target vehicle / container IDs;

[0143] The end-to-end integrated scheduling instruction data packet is pushed to the WMS warehouse management system and TMS logistics management system in real time, including:

[0144] WMS (Web Management System) automatically triggers supplier-side AGV (Automated Picking, Packaging, and Outbound Loading) scheduling;

[0145] TMS Terminal: Automatically locks the 3D time slot feature matrix The target vehicle and time slot are used to generate electronic waybills and electronic lock opening / closing instructions for the transport capacity;

[0146] The adaptive routing hard handover module receives the optimal replenishment decision vector obtained by the multi-objective collaborative decision optimization module. Based on the difference in replenishment urgency of SKUs in the same supply chain Combined with the three-dimensional time slot feature matrix of the supply network Time slot overlap Automatic hard switching of transportation modes; the specific decision logic for hard switching of routing modes is as follows: Figure 3 As shown:

[0147] Determine if the conditions for mixed loading and carpooling are met: And vehicle loading rate and When the conditions are met, the system will automatically switch to the mixed-carpooling mode.

[0148] If the conditions are not met, continue to check whether the conditions for direct cross-database transmission are met: And time slot overlap Furthermore, there are matching time slots at the transit cross-database nodes. When the conditions are met, the system will automatically switch to cross-database direct delivery mode.

[0149] If none of the above conditions are met, the system will switch to the regular replenishment mode by default. Finally, an end-to-end integrated scheduling instruction data packet containing the specific quantity, delivery sequence, and transportation route will be generated and issued.

[0150] In this embodiment, the instruction execution and feedback interface module needs to be specifically described. This module is used to output the end-to-end scheduling instruction to the ERP system, WMS warehouse management system, and TMS logistics management system, and to receive execution feedback data to dynamically update the three-dimensional time slot feature matrix. Specifically, it includes the following:

[0151] It connects to ERP systems, WMS warehouse management systems, and TMS logistics management systems through pre-defined standardized open interfaces, and receives end-to-end integrated scheduling instruction data packets, which are then asynchronously pushed and parsed through the following distribution channels:

[0152] Pushed to the ERP system: Automatically generates purchase orders and accounts payable accrual vouchers, locking in the tiered discount unit price in the purchase contract. ;

[0153] Push to WMS system: Issue warehouse picking instructions and loading schedules to the supplier's WMS; and issue pre-arrival notifications and warehouse capacity reservation instructions to the destination warehouse's WMS.

[0154] Push to TMS system: Automatically lock the matching transport vehicle and time slot, and issue scheduling instructions containing electronic waybill, transport route and electronic fence / smart door lock opening and closing key;

[0155] Establish a bidirectional long-term connection channel to capture real-time data streams from physical nodes of downstream systems during execution. :

[0156] WMS Execution Feedback Includes actual picking completion time and actual number of items loaded onto trucks. And the loading volume deviation rate;

[0157] TMS Execution Feedback By using onboard GPS / BeiDou terminals and RFID at checkpoints along the route, the real-time location coordinates and actual departure time of transport vehicles can be extracted. Actual travel time along the route and check-in and delivery times at key points;

[0158] ERP Execution Feedback This includes the purchase order receipt confirmation status and the actual invoicing and reconciliation status.

[0159] The captured feedback data is formatted and noise filtered; when an actual picking shortage is detected in the WMS... Or TMS departure time is seriously overdue When an abnormal interruption event occurs, a local recalculation request is automatically sent to the multi-objective collaborative decision optimization module to prevent the erroneous instruction from continuing to spread.

[0160] The cleaned execution feedback data is then injected back into the upstream module's dataset to achieve incremental closed-loop updates of parameters across the entire process. Specifically, this includes:

[0161] The time difference between the actual factory departure time and the agreed delivery time of this order is used as a new sample and pushed into the supplier's fulfillment and delivery history dataset N. The oldest batch of data is automatically discarded, keeping window N constant. The expected value of the supplier's factory departure delay is dynamically updated. ,in, This is the updated expected supplier delivery delay, in days. The expected supplier delivery delay before the update is in days. This is the time difference between the actual production time and the agreed delivery time for this order, expressed in days. If delivery is delayed, When delivered on time, ; The smoothing update factor has a value range of 1. The preferred value is 0.2, which is used to reflect recent fluctuations in the supplier's ability to fulfill its obligations.

[0162] Based on real-time capacity utilization and delivery time feedback from the TMS logistics management system, the three-dimensional time slot feature matrix is ​​analyzed. Perform real-time dimensionality reduction and coverage updates:

[0163] Capacity status dimension Z: based on actual loading volume The remaining available volume of the corresponding road segment and time slot will be deducted. With remaining available load capacity ;

[0164] Time window dimension Y: Dynamically adjust the baseline of travel time for future time slots on the transportation route based on the actual travel time of vehicles;

[0165] The instruction execution and feedback interface module establishes a bidirectional long-lived connection channel for issuing scheduling instructions and receiving feedback data from physical nodes; such as Figure 4 As shown:

[0166] Forward command issuance: based on a three-dimensional time slot matrix Generate end-to-end scheduling instruction data packets and push them to the ERP system to lock in purchase accounts and discount unit prices; to the WMS system for automatic scheduling of picking and loading; and to the TMS system to lock in time slot train numbers and draw trajectory segments.

[0167] Reverse data closed-loop feedback: Feedback data generated by the ERP, WMS, and TMS systems during physical execution is uniformly aggregated and injected into the "Feedback Data Cleaning and Parameter Incremental Update" node via an inverted tree-like bus; after data cleaning, the expected supplier delivery delay is dynamically updated. Real-time subtraction of three-dimensional time slot matrix The remaining available volume is used to achieve closed-loop self-correction of the parameters of the top system decision-making layer model.

[0168] Example 2

[0169] The following is another embodiment of the present invention, which provides a smart replenishment system for pet supplies based on sales rate prediction. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0170] This experiment aims to verify the effectiveness of a smart replenishment system for pet supplies based on sales rate prediction. Through data collection and weight reduction correction, dynamic fulfillment timeliness assessment, multi-objective collaborative decision optimization, adaptive route hard switching, and instruction execution and feedback, the system enhances the dynamic replenishment and risk management capabilities across the entire pet supplies supply chain. The experiment uses simulated and actual collected operational data from e-commerce platforms, ERP, WMS, and TMS, including real consumption rates, dynamic fulfillment lead times, reorder point thresholds, comprehensive supply chain costs, and route hard switching modes under different sales scenarios. By analyzing the consistency between the system's replenishment decision results and actual warehousing and logistics fulfillment labels, the system's accuracy and robustness in eliminating promotional interference, reducing total supply chain costs, and ensuring fulfillment rates are verified.

[0171] The simulation experiment steps are implemented according to the content of the intelligent replenishment system for pet supplies based on sales rate prediction provided in Example 1, and the specific steps include:

[0172] Collect order data from various sales channels, pet product attribute data, inventory level data from various warehousing nodes, and real-time transportation capacity trajectory data from the logistics system, with the collection frequency set to millisecond level; based on the theoretical consumption cycle and coefficient of variation of pet products, remove promotional interference and correct sales rate by reducing weight on the original sales data, and output the true consumption rate.

[0173] Based on the historical delivery data of each supplier and the real-time trajectory of trunk logistics, the probability distribution of supplier factory delays is fitted, the static lead time is dynamically corrected into the dynamic fulfillment lead time, and the reorder point threshold of each warehousing node is reconstructed in real time by combining the adaptive safety factor.

[0174] A mixed-integer nonlinear programming model is constructed with the goal of minimizing the total cost of the integrated supply chain. The minimum order quantity of suppliers, tiered discounts for procurement, vehicle loading rate and available warehouse capacity of the destination warehouse are set as hard constraints. The model is solved by combining the actual consumption rate and the reorder point threshold, and an exponential time-delay penalty mechanism is introduced to quickly shrink the feasible region.

[0175] Based on the replenishment urgency difference and waiting tolerance threshold of SKUs in the same link, as well as the matching results of the procurement demand time window and the three-dimensional time slot feature matrix of the supply network, automatic hard switching is performed between mixed loading and carpooling, cross-warehouse direct delivery and regular replenishment modes; end-to-end scheduling instructions are generated.

[0176] The system outputs end-to-end scheduling instructions to the ERP system, WMS warehouse management system, and TMS logistics management system, and receives execution feedback data to dynamically update the three-dimensional time slot feature matrix and the supplier's historical performance parameters.

[0177] The specific data from the above simulation experiment are shown in Table 1.

[0178] Table 1 Data Recording Table of Simulation Experiment of the Invention

[0179] Time / day Sales channels and promotional labeling Original daily sales volume / pieces True steady-state consumption rate (units / day) Dynamic lead time / day Reorder Point (ROP) / piece Optimize the generation of replenishment quantity / piece Routing switching mode Overall cost reduction / % Performance status label Replenishment control measures 1-5 Regular e-commerce sales / no promotions 100 98 3.0 385 0 Standard replenishment model 12.5% Normal / Normal Continuous monitoring of inventory 5-10 Influencer live-streaming sales / promotions 850 105 3.2 410 0 Standard replenishment model 18.2% Pulse / Normal Strip away promotional distractions and maintain monitoring 10-15 Platform promotions / major sales 1200 110 4.5 580 600 Mixed carpooling mode 24.6% Trigger / Alert Issue carpooling instructions and combine shipments. 15-20 Main road congestion / No promotions 110 102 6.8 780 850 Cross-database direct delivery mode 21.0% Delay / Emergency Enable cross-database direct delivery, bypassing the central warehouse. 20-25 Normal operation resumed / No promotions 95 96 3.1 375 0 Standard replenishment model 15.3% Normal / Normal Resumption of routine inspections 25-30 Offline POS boom / promotion 500 108 3.5 450 500 Mixed carpooling mode 19.8% Triggered / Normal Decrease in weighting and adjustment, replenishment order issued.

[0180] Experimental Analysis:

[0181] By comparing the conformity between the model's decision results and the actual warehousing and logistics fulfillment labels, the accuracy and robustness of the present invention in stripping away false demand impulses, adaptively responding to capacity delays, and optimizing multi-objective supply chain costs are verified. Experiments show that a pet supplies intelligent replenishment system based on sales rate prediction has high accuracy and reliability, and can effectively improve risk management and efficient replenishment capabilities throughout the entire lifecycle of the pet supplies supply chain.

[0182] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A smart replenishment system for pet supplies based on sales rate prediction, characterized in that: include: The data acquisition module is used to collect order data from various sales channels, pet product attribute data, inventory level data from various warehousing nodes, and real-time transportation capacity trajectory data from the logistics system in real time. Based on the theoretical consumption cycle and coefficient of variation of pet products, it performs promotion interference removal and sales rate weight reduction correction on the original sales data and outputs the true consumption rate. The dynamic fulfillment timeliness assessment module is used to dynamically fit the supplier's fulfillment delay probability based on the supplier's historical delivery data and the real-time trajectory of trunk logistics, correct the static lead time to the dynamic fulfillment lead time, and reconstruct the reorder point threshold of each warehousing node in real time. The multi-objective collaborative decision optimization module is used to minimize the overall supply chain cost as the objective function. It sets the supplier's minimum order quantity, procurement tiered discount, vehicle volume loading rate and available warehouse capacity of the destination warehouse as hard constraints, and solves the problem by combining the actual consumption rate and the reorder point threshold. It also introduces an exponential time-sensitive default penalty mechanism to shrink the feasible region. The adaptive routing hard switching module is used to automatically generate end-to-end scheduling instructions containing replenishment quantity, replenishment timing and transportation route based on the replenishment urgency difference and waiting tolerance threshold of SKUs on the same link, as well as the matching results of the procurement demand time window and the three-dimensional time slot feature matrix of the supply network, and to perform automatic hard switching between mixed loading and carpooling, cross-warehouse direct delivery and regular replenishment modes. The instruction execution and feedback interface module is used to output the end-to-end scheduling instructions to the ERP system, WMS warehouse management system and TMS logistics management system, and to receive execution feedback data to dynamically update the three-dimensional time slot feature matrix.

2. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 1, characterized in that: The data acquisition module specifically includes the following: A data stream of orders and purchases containing supplier historical fulfillment and delivery records is constructed using standardized API interfaces. Pet supplies attribute data Includes real-time available inventory. Inventory water level data And capacity tracking data including the supplier's historical fulfillment and delivery records. The unified four-dimensional dataset ; The collected pet product attribute data is transmitted to the real consumption rate assessment logic. Combined with preset pet physiological and consumption benchmarks, the theoretical consumption cycle for each product category is calculated, specifically including: For pet staple food: Access a pre-set pet physiology database to obtain the net content per package. Feed conversion rate Average weight of pets within the service area and the daily recommended intake coefficient The theoretical number of days a single bag of pet food can last under ideal feeding conditions is calculated as follows: ; For non-staple pet food products, extract the adjacent purchase timestamps of the same SKU from the same end customer within the statistical period. and The theoretical consumption period is calculated using a median function, with the specific formula as follows: ;in, This refers to the theoretical consumption cycle of non-staple pet food products. The median function is used to calculate the median number of purchases within the statistical period; m represents the total number of purchases made by a customer within the statistical period. At that time, the preset industry benchmark consumption cycle is directly retrieved as... reveal all the details; Extract the raw sales figures for each day within the sliding time window, using the current date as the endpoint. Based on the standard deviation of daily sales with the mean Calculate the coefficient of variation And calculate the actual number of days consumed within the sliding window. When the actual number of days consumed is met And coefficient of variation At that time, it is determined that a promotional pulse is triggered; among which, This is the preset deviation sensitivity coefficient; The threshold for the coefficient of variation; Net content per single package. This represents the average daily sales volume within the window. Iterate through each day within the window, and based on the promotional markers and promotional activity identifiers in the order and purchase data streams, record whether the day is a promotional day. For day t within the sliding time window: When a day is determined to be a non-promotion day, the promotion intensity coefficient is zero, and the adjusted equivalent sales volume is... equal to original sales ; When a day is determined to be a promotional day, an exponential decay factor is introduced. The original sales figures for the day will be adjusted for weight reduction as follows: The corrected equivalent daily sales are: ;in, Set the preset attenuation adjustment coefficient; set the lower limit protection value of the attenuation factor. , when the formula At that time, a mandatory order ; The corrected equivalent sales for each day within the window are weighted and averaged to calculate and output the true steady-state consumption rate. ;in, To determine the true steady-state consumption rate, when the sliding time window consists entirely of promotional days, the average consumption rate from the previous non-promotional window is automatically extracted as the base rate and compared with the currently calculated rate. Smooth the surface using a preset weighted ratio.

3. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 1, characterized in that: The dynamic fulfillment timeliness assessment module retrieves capacity trajectory data and historical fulfillment and delivery records from the four-dimensional dataset to reconstruct the reorder point threshold (ROP) for each warehousing node in real time. Specifically, this includes: Extract the location coordinates and traffic status of trunk transport vehicles, and calculate the real-time deviation duration of currently en route trunk logistics. Extract a sample dataset of the time difference between the actual production time and the agreed delivery time of the target supplier's past N batches of orders, and fit the supplier's expected production delay value. With variance ; Retrieve preset static lead time of supplier contracts Combined with the real-time deviation time of trunk logistics Expected value of factory delay Calculate and generate dynamic lead time for contract fulfillment. ;in, For dynamic lead time for contract fulfillment; Based on dynamic lead time Based on variance and demand fluctuations, reconstruct dynamic safety stock levels. ,in, This is a dynamic safety stock level, measured in units of pieces. The actual steady-state consumption rate output by the data acquisition module; This represents the variance of daily sales within the sliding time window. The variance of dynamic lead time is calculated based on the supplier's factory delay variance. With the variance of trunk logistics The result is obtained by superposition; The safety factor corresponding to the service level; The true steady-state consumption rate output by the data acquisition module Calculated dynamic lead time and the calculated dynamic safety stock. Combine and reconstruct the reorder point thresholds for each warehousing node in real time: ;in, The reorder point threshold after reconstruction; Continuously monitor the current available inventory at each warehouse node. ,when At that time, a replenishment trigger signal will be automatically generated, which will include the target SKU and the current available inventory. True steady-state consumption rate and dynamic lead time for performance The feature data package is sent to the multi-objective collaborative decision optimization module for replenishment scheme calculation.

4. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 3, characterized in that: The dynamic performance evaluation module uses a normal distribution to fit the supplier's factory delay probability distribution to obtain its expected factory delay value. When the fitted At that time, a mandatory order When calculated When this occurs, the lower limit protection mechanism is triggered, causing... When the transport capacity trajectory data is interrupted or abnormal, the historical average transit time of the past 30 days for that transport route is automatically retrieved as the real-time deviation time for the current trunk line logistics. The fallback alternative value.

5. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 1, characterized in that: The multi-objective collaborative decision optimization module specifically includes the following: When a replenishment trigger signal is received, the target SKU identifier and current available inventory are retrieved in real time from the feature data packet. True steady-state consumption rate and dynamic lead time for performance ; Set the optimization decision variable matrix as follows: ,in, This represents the replenishment quantity for purchasing the k-th SKU from the supplier, where k is the total number of candidate SKU types destined for the same warehouse. Construct an optimization model with the objective of minimizing the total supply chain cost (TC): ； Wherein, TC represents the total cost of the integrated supply chain; the specific derivation of each sub-cost item is as follows: Procurement costs: ,in, For procurement costs, For the k-th SKU, the purchase quantity is The dynamic unit price at any given time is determined by the supplier's tiered discount function: when hour, ,in, The minimum order quantity for the nth tier of the k-th SKU. This corresponds to the tiered discount unit price; Transportation costs: Calculated using a tiered freight rate model based on vehicle volume and load capacity. ,in, For transportation costs, The number of vehicles to be called. For the freight of a single vehicle, For LTL (Less-than-Truckload) freight, The remaining volume is less than the entire vehicle's volume; Warehouse holding costs: ,in, For storage and holding costs, Let k be the dynamic safety stock level for the k-th SKU. Let $k$ be the holding cost per unit of product for the $k$-th SKU per unit of time. This represents the actual steady-state consumption rate corresponding to the k-th SKU; Cost of stockout losses: ,in, To avoid losses due to stockouts, Let k be the current available inventory of the k-th SKU. Let $k$ be the unit stockout penalty cost for the $k$-th SKU. This is a function to find the maximum value. This is a non-linear, exponential, time-limited penalty term, with the specific formula as follows: ; in, For the cost of penalties for breach of contract; The default coefficient is the preset base default coefficient; It is an exponential penalty factor; Set the maximum expected arrival time window for the destination warehouse; set a maximum protection value for penalty items. , For transportation costs, when the calculation result exceeds... At that time, a mandatory order .

6. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 5, characterized in that: The multi-objective collaborative decision optimization module sets the following hard boundary constraints during the solution process: Supplier minimum order quantity constraint: For any selected k-th SKU, the following condition must be met. or ;in, Minimum order quantity; Hard constraint on vehicle loading rate: The total loading volume must meet the following requirements. ,in, Let K be the volume of a single item of the k-th SKU. To meet the minimum load requirement, This refers to the vehicle's rated maximum volume. Available storage capacity constraints for the destination warehouse: ,in, This refers to the current real-time remaining available storage capacity of the destination warehouse. The hierarchical branch and bound algorithm is used to solve the optimization model; a 500ms timeout threshold is set. If the timeout occurs and no feasible solution is found, the reorder point shortfall is calculated. Minimum order quantity with supplier The maximum value is used to generate an emergency replenishment vector and issue the optimal replenishment decision vector. The optimal replenishment decision vector obtained by solving the problem. The instructions are sent to the adaptive routing hard handover module to generate capacity routes and scheduling commands. This represents the final optimal replenishment quantity calculated for the k-th SKU.

7. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 1, characterized in that: The adaptive routing hard handover module specifically includes the following: Receive the optimal replenishment decision vector Based on the real-time inventory and consumption rate of each SKU, the replenishment urgency of each SKU in the same supply chain is calculated. ;in, The urgency of replenishing the k-th SKU; This is the reorder point threshold; This represents the current available inventory of the k-th SKU; This represents the true steady-state consumption rate. Calculate the difference in replenishment urgency between the j-th SKU and the k-th SKU in the same supply chain. Extract the dynamic waiting tolerance threshold for the j-th SKU in the target warehouse. ;in, This is a function to find the maximum value. Retrieve supply network time slot data from the TMS logistics management system and construct a three-dimensional time slot feature matrix containing node coordinate dimension X, time slot dimension Y, and transportation capacity status dimension Z. ; and calculate the procurement demand time window and the aforementioned Time slot overlap ; Based on the aforementioned urgency difference Waiting tolerance threshold Vehicle loading rate and time slot overlap Adaptive hard switching is performed between the three transportation modes.

8. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 7, characterized in that: Adaptive hard switching between the three transportation modes includes: When the following conditions are met: and and At that time, a hard switch to mixed-load carpooling mode is implemented; merging orders for carpooling improves loading efficiency, among which, This is the waiting tolerance threshold; When the mixed-carpooling mode is not met, but the following conditions are met: Time slot overlap occurs in the destination warehouse Furthermore, there are matching time slots at the transit cross-database nodes. When necessary, a hard switch to cross-warehouse direct delivery mode is performed; a cross-warehouse instruction is generated to bypass the central warehouse and directly deliver to the destination warehouse on the transit platform; If none of the above conditions are met, the system will default to the regular replenishment mode; otherwise, it will proceed as originally planned. Issue standard point-to-point trunk line transportation instructions in batches; After completing the hard mode switch decision, the adaptive routing hard switch module automatically aggregates and generates an end-to-end integrated scheduling instruction data packet, including: the replenishment quantity of specific SKU combinations. Loading and shipping sequence The system adapts to transportation routes and target vehicle / container IDs; and pushes scheduling instructions to the WMS warehouse management system and capacity locking and electronic lock opening / closing instructions to the TMS logistics management system.

9. The intelligent replenishment system for pet supplies based on sales rate prediction according to claim 1, characterized in that: The instruction execution and feedback interface module establishes a bidirectional long-term connection channel to achieve positive instruction issuance and reverse data closed-loop feedback, specifically including the following: The system connects to the ERP system, WMS warehouse management system, and TMS logistics management system through a pre-defined standardized open interface. It receives end-to-end integrated scheduling instruction data packets and pushes them to the back-end ERP system to lock purchase accounts and discount unit prices, to the WMS warehouse management system for automatic scheduling of picking and reservation of warehouse capacity, and to the TMS logistics management system to lock time slots and draw tracks. Establish a two-way long connection channel to capture the physical node feedback data stream containing WMS actual picking data, TMS vehicle real-time trajectory and ERP warehouse entry confirmation status; after normalization and cleaning, when picking shortage or TMS dispatch timeout is detected, send a local recalculation request to the multi-objective collaborative decision optimization module. The cleaned execution feedback data is then injected back into the upstream module's dataset to achieve incremental closed-loop updates of parameters across the entire process. Specifically, this includes: The time difference between the actual factory departure time and the agreed delivery time of this order is used as a new sample and pushed into the supplier's fulfillment and delivery history dataset N. The oldest batch of data is automatically discarded, keeping window N constant. The expected value of the supplier's factory departure delay is dynamically updated. ,in, For the updated supplier delivery delay expectations, The previous supplier's expected delivery delay was used. This is the time difference between the actual manufacturing time and the agreed delivery time for this order. If delivery is delayed, When delivered on time, ; To smooth the update factor; Based on real-time capacity utilization and delivery time feedback from the TMS logistics management system, the three-dimensional time slot feature matrix is ​​analyzed. Perform real-time dimensionality reduction and coverage updates, including: Capacity status dimension Z: based on actual loading volume The remaining available volume of the corresponding road segment and time slot will be deducted. With remaining available load capacity ; Time window dimension Y: Dynamically adjust the baseline of travel time for future time slots on the transportation route based on the actual travel time of vehicles.

Images