Intelligent commodity distribution method and system based on dynamic paving threshold and multi-level priority

Abstract

The application belongs to the technical field of intelligent warehousing and commodity distribution, and relates to an intelligent commodity distribution method and system based on a dynamic paving threshold and a multi-level priority. A dynamic paving threshold is obtained through weighted comprehensive calculation, dynamic weight adjustment and nonlinear smoothing processing by selecting commodity attribute factors, inventory level factors and environmental factors, replacing the traditional fixed threshold, realizing the precise adaptation of the paving store range to the dynamic changes of commodities and the environment, and solving the poor adaptability problem of the fixed threshold. Furthermore, a multi-level priority scoring model containing depth-first, breadth-first and additional dimensions is constructed, a comprehensive priority score is obtained through weighted integration, the limitations of traditional single-dimensional priority determination are broken through, and it is ensured that the distribution priority can fully cover the core demand dimension. Then, candidate stores are screened based on the dynamic paving threshold, multi-level sorting is performed in combination with the comprehensive priority, and dynamic correction is performed, realizing the precise matching of the distribution range and the priority.

Description

Technical Field

[0001] This application belongs to the field of intelligent warehousing and commodity distribution technology, and more specifically, relates to an intelligent commodity distribution method and system based on dynamic deployable thresholds and multi-level priorities. Background Technology

[0002] In the field of commodity circulation, a reasonable distribution strategy is the key to ensuring efficient circulation of goods, improving inventory utilization, and meeting the sales needs of terminal stores. At present, traditional commodity distribution methods rely heavily on manual experience or fixed distribution rules, which have many drawbacks.

[0003] On the one hand, traditional distribution often uses a fixed threshold for the range of stores that can be distributed, which cannot be adapted and adjusted according to the dynamic changes in product attributes (such as life cycle and price range), inventory levels and market environment (such as seasonal changes and promotional activities). This results in some products being fully distributed to potential stores when inventory is sufficient, while being over-distributed to non-core stores when inventory is tight, causing an imbalance in inventory distribution.

[0004] On the other hand, the criteria for prioritizing order allocation are too simplistic, often relying solely on store rating or historical sales volume for ranking. In order to comprehensively consider factors such as planned order volume, geographic coverage, and real-time sales trends, order allocation resources cannot accurately match high-demand, high-value store-product combinations. This leads to stockouts in some stores and inventory backlogs in others, reducing order allocation efficiency and overall sales revenue.

[0005] In summary, how to dynamically adapt to different product and environmental characteristics to determine the available distribution range, and at the same time accurately determine the priority of distribution based on multiple factors to achieve a reasonable allocation of inventory resources, has become a core technical problem that urgently needs to be solved in the current distribution technology field. Summary of the Invention

[0006] This invention provides an intelligent product allocation method and system based on dynamic availability thresholds and multi-level priorities. It aims to dynamically adapt to different product and environmental characteristics to determine the availability range, while comprehensively considering multiple factors to accurately determine allocation priorities, thereby achieving a reasonable allocation of inventory resources.

[0007] On one hand, this invention provides an intelligent product allocation method based on dynamic deployable thresholds and multi-level priorities, comprising the following steps: Data Acquisition and Preprocessing: Collect multi-dimensional basic data required for allocation decisions, and preprocess the collected multi-dimensional basic data to obtain a structured dataset; Dynamic scalability threshold calculation: Based on a structured dataset, product attribute factors, inventory level factors, and environmental factors are selected. The initial scalability threshold is obtained by weighted summation. Then, the dynamic scalability threshold for each product is obtained through dynamic weight adjustment and non-linear smoothing. Multi-level priority comprehensive score: Based on the structured dataset, a multi-priority scoring model is constructed that includes depth-first dimension, breadth-first dimension and additional dimensions. The comprehensive priority score of each store-product pair is calculated by weighted summation. Threshold filtering and priority sorting: Based on the dynamic scalable threshold, stores are filtered to obtain a set of candidate stores. Then, the set of candidate stores is sorted in multiple levels based on the comprehensive priority score. After dynamic correction based on real-time business data, a basic ordering candidate list is generated. Loan allocation rule embedding and decision output: Based on the generated basic loan allocation candidate list, preset business rules are embedded for filtering. After inventory verification and loan allocation quantity adjustment, the optimized loan allocation plan is output.

[0008] This invention selects product attribute factors, inventory level factors, and environmental factors, and obtains a dynamic available-store threshold through weighted comprehensive calculation, dynamic weight adjustment, and nonlinear smoothing, replacing the traditional fixed threshold. This achieves accurate adaptation of the available store range to dynamic changes in products and environment, solving the problem of poor adaptability of fixed thresholds. Furthermore, it constructs a multi-level priority scoring model including depth-first, breadth-first, and additional dimensions, and obtains a comprehensive priority score through weighted integration, breaking through the limitations of traditional single-dimensional priority determination and ensuring that the allocation priority can comprehensively cover core demand dimensions. Subsequently, candidate stores are screened based on the dynamic available-store threshold, and multi-level sorting and dynamic correction are performed in combination with the comprehensive priority, achieving accurate matching between allocation range and priority. This ensures the compliance and feasibility of allocation decisions and systematically solves the core technical problems of unbalanced inventory allocation and inaccurate priority determination in traditional allocation methods.

[0009] Preferably, the multi-dimensional basic data includes product attribute data, warehouse inventory data, external environment data, store data, and real-time sales data.

[0010] Preferably, the dynamic layable threshold calculation includes the following steps: First, the selected product attribute factors, inventory level factors, and environmental factors are standardized. Then, the initial threshold for the proportion of stores that can be scalable is obtained by weighted summation. The weights of each factor are obtained by linear regression analysis of historical data and are dynamically updated. Then, the initial threshold for the proportion of shops that can be scalable is non-linearly smoothed using the exponential smoothing method to eliminate threshold abrupt changes caused by data fluctuations, thus obtaining the final dynamic threshold for scalable shops.

[0011] Preferably, the product attribute factors are determined based on the product life cycle stage and price range; the inventory level factors are calculated based on the inventory tightness index; and the environmental factors are determined by combining seasonal adjustment factors and promotional activity adjustment factors.

[0012] Preferably, the depth-first dimension is obtained by weighted summation based on planned order volume and historical sales performance; the breadth-first dimension is obtained by weighted summation based on store level and historical geographic coverage; and the additional dimension is obtained by weighted summation based on real-time sales trends and product attributes. Then, a learning model is used to optimize the weights of the depth-first dimension, the breadth-first dimension, and the additional dimensions, and a comprehensive priority score is obtained by weighted summation of the scores of each dimension.

[0013] Preferably, the basic cargo allocation candidate list is obtained based on the following steps: Calculate the current store coverage rate for each product, compare the store coverage rate with the dynamic available store threshold, and filter the candidate store set by combining the priority correction factor. A three-level sorting mechanism is used to sort the candidate store set. The first level is sorted in descending order by comprehensive priority score. The second level is sorted in descending order by store level when the comprehensive priority scores are the same. The third level is sorted in descending order by inventory matching degree when the comprehensive priority scores and store levels are the same. The sorting results are dynamically corrected in combination with real-time inventory warnings, marketing activities and logistics cost optimization needs to generate a basic order fulfillment candidate list.

[0014] Preferably, the inventory matching degree is obtained by weighted calculation of the inventory fit index and the matching index, wherein the inventory fit index is determined based on the degree of match between the store's historical sales capacity and the current inventory level, and the matching index is determined by combining the store's historical sales stability and store weight.

[0015] Preferably, the preset business rules include rules for not picking up items with missing sizes and rules for not picking up items with incomplete sets; the business rules are applied sequentially to filter the store-product combinations in the basic order picking candidate list; Introducing rule-based pre-detection factors, the risk of code shortage and incomplete sets is estimated based on historical data. The rule-based pre-detection factors are then fed back into the dynamic playable threshold calculation process to adjust the dynamic playable threshold. Perform inventory verification on the filtered store-product combination, adjust the allocation quantity to meet the inventory limit and minimum allocation requirements, and output an optimized allocation plan that includes the target store, allocated products, allocation quantity, and execution time.

[0016] Preferably, the rule for not fulfilling orders for out-of-size items is that when the sales cost is divided by the average inventory level, and the inventory of any size of the product does not meet the preset requirements, it is determined to be out of stock, and the product is removed from the corresponding store's order fulfillment candidate list. The rule of not including items in a set is that if the inventory of any component of a set does not meet the preset ordering requirements, the set is deemed not to be a set and is removed from the corresponding store's ordering candidate list.

[0017] On the other hand, an intelligent goods distribution system based on dynamic deployable thresholds and multi-level priorities is provided, including: The multi-source data acquisition and preprocessing module is used to collect multi-dimensional basic data required for allocation decisions, and to preprocess the collected multi-dimensional basic data to generate a structured dataset. The dynamic scalable threshold calculation module is connected to the multi-source data acquisition and preprocessing module. It is used to select product attribute factors, inventory level factors and environmental factors based on the structured dataset, calculate the initial scalable store ratio threshold through weighted summation, and then obtain the dynamic scalable threshold for each product through dynamic weight adjustment and nonlinear smoothing. The multi-level priority scoring module is connected to the multi-source data acquisition and preprocessing module. It is used to construct a multi-level priority scoring model based on the structured dataset, which includes depth-first dimension, width-first dimension and additional dimensions. The comprehensive priority score of each store-product pair is obtained by weighted integration calculation. The filtering and sorting module is connected to the dynamic deployable threshold calculation module and the multi-level priority scoring module, respectively. It then performs multi-level sorting based on the comprehensive priority score and dynamically corrects it in combination with real-time business data to generate a basic cargo allocation candidate list.

[0018] The beneficial effects of the invention include: This invention selects product attribute factors, inventory level factors, and environmental factors, and obtains a dynamic available-store threshold through weighted comprehensive calculation, dynamic weight adjustment, and nonlinear smoothing, replacing the traditional fixed threshold. This achieves accurate adaptation of the available store range to dynamic changes in products and environment, solving the problem of poor adaptability of fixed thresholds. Furthermore, it constructs a multi-level priority scoring model including depth-first, breadth-first, and additional dimensions, and obtains a comprehensive priority score through weighted integration, breaking through the limitations of traditional single-dimensional priority determination and ensuring that the allocation priority can comprehensively cover core demand dimensions. Subsequently, candidate stores are screened based on the dynamic available-store threshold, and multi-level sorting and dynamic correction are performed in combination with the comprehensive priority, achieving accurate matching between allocation range and priority. Finally, preset business rules are embedded for filtering and inventory verification and quantity adjustment, outputting an optimized allocation plan. This ensures the compliance and feasibility of allocation decisions and systematically solves the core technical problems of unbalanced inventory allocation and inaccurate priority determination in traditional allocation methods.

[0019] This invention achieves dynamic adaptation and precise optimization of allocation strategies through the collaborative design of dynamic availability thresholds and multi-level priority scoring, effectively improving the scientific and rational nature of allocation decisions. The dynamic availability thresholds can respond in real time to changes in products, inventory, and the environment, avoiding inventory imbalances caused by fixed thresholds and significantly improving inventory utilization. The multi-level priority scoring mechanism comprehensively covers multiple core needs, enabling precise matching of allocation resources to high-value store-product combinations, reducing stockouts and overstocking. Simultaneously, the embedding of business rules ensures allocation compliance, ultimately achieving the goals of improving overall sales revenue and optimizing product turnover efficiency. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 The above is a flowchart illustrating the overall steps of an embodiment of the present invention.

[0022] Figure 2 The flowchart for dynamic layable threshold iterative correction provided in the embodiments of the present invention is shown.

[0023] Figure 3 The flowchart for threshold filtering and priority sorting provided in the embodiments of the present invention is shown.

[0024] Figure 4 This is a system framework diagram provided for an embodiment of the present invention. Detailed Implementation

[0025] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0026] Example 1

[0027] This embodiment provides an intelligent product allocation method based on dynamic deployable thresholds and multi-level priorities. See [link to relevant documentation]. Figure 1 As shown, it includes the following steps: Data Acquisition and Preprocessing: By collecting multi-dimensional basic data required for allocation decisions and ensuring data quality, consistency and standardization through preprocessing, a structured dataset is formed. The multi-dimensional basic data mentioned above includes: Product attribute data includes product life cycle stage (pre-divided into four categories: introduction, growth, maturity, and decline), price range (such as high, medium, and low), and category information; Warehouse inventory data includes current inventory levels (real-time inventory quantity) and safety stock thresholds (minimum inventory levels set based on historical demand). External environmental data includes seasonal factors (such as spring, summer, autumn and winter classification) and promotional activity plans (such as promotion time and intensity). Store data includes store rating (e.g., A, B, C), historical sales performance (e.g., sales in the last 30 days), and geographical location. Real-time sales data includes recent sales trends (such as time series of daily sales volume) and stockout rate (calculated as the ratio of stockout occurrences to total queries).

[0028] The preprocessing includes: Data cleaning includes missing value handling, outlier detection and handling, and noise reduction. The missing value handling includes filling missing values ​​for logarithmic data (such as inventory levels) using linear interpolation; and filling missing values ​​for categorical data (such as product lifecycle stages) using the mode imputation method. The outlier detection and processing includes using the Z-score method to identify outliers. When |Z-score| is greater than 3, it is determined to be an outlier, and it is corrected to a reasonable range by truncation (replaced with the reasonable upper and lower limits of the index, such as the threshold range determined based on the 3σ principle). The denoising process targets time-series data and applies a moving average method for smoothing. The size of the moving window is dynamically set according to the data collection frequency (for example, a 7-day window is used for daily sales data and a 4-week window is used for weekly data) to eliminate random fluctuations.

[0029] Normalization and standardization conversion: including normalization processing, standardization index calculation, and classification data coding; The normalization process involves using the minimum-maximum normalization method on numerical data to map values ​​to the [0,1] interval. The standardized indicator is calculated by converting inventory levels into inventory turnover rate, which is the cost of sales divided by the average inventory level; where cost of sales refers to the cost of sales within a specified period, such as the total cost of sales over 30 days. Categorical data encoding: Non-numerical data is one-hot encoded and converted into binary vectors.

[0030] After the above preprocessing, all data is integrated into a structured dataset, stored in tabular form and containing timestamp identifiers.

[0031] Dynamically calculable threshold calculation: See Figure 2 As shown, the dynamic scalable threshold in this embodiment includes four steps: factor selection and standardization, calculation of the initial scalable store ratio threshold, dynamic weight adjustment, and non-linear smoothing, as detailed below: Three key factors were selected: product attribute factor, inventory level factor, and environmental factor. All of them were standardized to a range of 0 to 1. In this embodiment, the Product Attribute Factor (PAF) is determined based on the product lifecycle stage and price band. An exemplary standardization rule is: PAF = 0.2 for the introduction stage, 0.8 for the growth stage, 0.6 for the maturity stage, and 0.4 for the decline stage; the PAF increases by 0.1 for high price bands and decreases by 0.1 for low price bands. Since products in the growth stage need to be widely distributed to seize the market, while products in the decline stage need to shrink their distribution scope to reduce inventory backlog, the PAF value for the growth stage is much larger than the PAF value for the decline stage. Furthermore, the price band adjustment further refines the differences in product distribution strategies.

[0032] The Inventory Level Factor (ILF) is calculated based on the Inventory Tightness Index (ISI), specifically: ISI = 1 - (Inventory Turnover Rate / Target Turnover Rate); ILF = 1 - ISI; In the above calculation formula, the inventory turnover rate is obtained based on the standardized indicators of the previous steps, while the target turnover rate is a reasonable turnover rate threshold preset based on the historical sales data of the product, the characteristics of the product category, and the operational goals. When the ISI is high (low inventory turnover rate, tight inventory), the ILF decreases, and the subsequent threshold increases to prioritize the core stores. When the ISI is low (high inventory turnover rate, ample inventory), the ILF is high, and the subsequent threshold decreases to expand the store coverage.

[0033] The environmental factor (EF) is calculated by weighting and integrating seasonal adjustment factors and promotional activity adjustment factors, specifically: Calculate the seasonal adjustment factor : =(Current quarter sales - Historical average sales for the same period) / Historical average sales for the same period; in Theoretically, the range is [-1, +∞), but in this embodiment when When the sales amount exceeds 1, take 1 (sales exceeding twice the historical amount are counted as twice the amount). Use -0.5 (for sales declines exceeding 50%, calculate as 50%) to avoid interference from extreme values; right Normalize to [0,1]: Sales revenue decreased by 50% )correspond Sales increased by 100% )correspond ; Calculate the adjustment factor for promotional activities : =Promotional strength × Promotional impact coefficient; where the promotional strength is preset to [0, 0.5] (e.g., 0.05 for 5% strength, 0.5 for 50% strength), and the promotional impact coefficient is the coefficient of the [0, 2] interval fitted from historical data (e.g., 2 when the promotion significantly increases sales, 1 when the impact is small), and finally... Constrained in [0,1] (if the value exceeds 1, take the value 1); Weighted fusion : ,in Indicates seasonal weighting, Indicates promotional weight, all are positive numbers and In this embodiment, the focus is primarily on seasonality, therefore the default is... ; In this embodiment, Reflecting the strength of seasonal demand, Reflects the strength of demand driven by promotions, weighted Naturally falling within the [0,1] range, no forced truncation is required; when seasonal demand is strong (high EF), the dynamic layable threshold is lowered to expand coverage, and when demand is weak (low EF), the threshold is raised to shrink coverage, ensuring dynamic adaptability.

[0034] Based on the obtained product attribute factors, inventory level factors, and environmental factors, the initial threshold for the proportion of stores that can be scalable is calculated using a dynamic weighted summation formula. : ; In the formula: , and These are the weights of the product attribute factor, inventory level factor, and environmental factor, respectively, and the sum of the three weights is 1. In this embodiment, based on historical data, linear regression is used to dynamically optimize the above weights. For example, the regression equation is as follows: ; In the formula: Represents the intercept term; , , These are the regression coefficients of each factor; Indicates the error term; In this embodiment, regression analysis is used to identify the impact layer of each factor on the actual store coverage, and the weights are adaptively adjusted as the data changes (i.e., each weight...). , and (New weights are obtained by normalizing the regression coefficients) to avoid subjective bias.

[0035] For the obtained threshold of scalable stores In this embodiment, further smoothing processing is performed to eliminate threshold abrupt changes caused by data fluctuations: ; In the formula: This represents the smoothed threshold for the current time period; This indicates the initial threshold value being calculated. This represents the smoothed threshold value from the previous time period. This represents the smoothing coefficient, with a value between 0 and 1.

[0036] In this embodiment, through multi-factor fusion and dynamic optimization, the generated dynamic distribution threshold can respond to changes in product attributes, inventory status, and market environment. Compared with traditional fixed thresholds, it can significantly improve the adaptability and accuracy of distribution strategies.

[0037] Multi-level priority comprehensive score: In this embodiment, a comprehensive priority score is calculated for each store-product pair to balance the needs of depth-first and breadth-first priorities. By establishing a multi-level priority scoring model that includes depth-first, breadth-first, and additional dimensions, the final score is obtained through weighted integration and weight optimization. The Depth-First Dimension (DPD) reflects the store's urgency for product demand and is obtained by a weighted sum of the Order Planned Quantity Score (OPS) and Historical Sales Performance Score (HSS).

[0038] In the formula: Indicates the weighting of the planned order volume score; Indicates the weighting of historical sales performance scores; ; The planned order volume based on the current store-product combination is obtained through minimum-maximum normalization; The data was obtained using minimum-maximum normalization based on sales figures over the past 30 days.

[0039] The width-first dimensional (WPD) reflects a store's coverage needs and potential, and is obtained by a weighted sum of the Store Level Score (SLS) and Geographic Coverage Score (GLS): ; In the formula: As a weighting factor in store rating; Indicates the weight of the geographic location coverage score; ; Determined based on a preset grading table (e.g., Grade A 0.9, Grade B 0.6, Grade C 0.3); Calculated based on regional coverage (number of covered stores / total number of stores).

[0040] The additional dimension (AD) encompasses other key factors influencing allocation decisions, and is obtained by a weighted sum of the Real-Time Sales Trend Score (RSTS) and Product Attribute Score (PAS): ; In the formula: , Each represents the corresponding weight, and the sum is 1; where The projected 7-day sales growth rate is obtained based on the moving average forecast. The value is determined based on the commodity's value (e.g., high value 0.8, ordinary 0.5, low value 0.2).

[0041] Then, a weighted sum is performed based on the obtained depth-first dimension value, width-first dimension value, and additional dimension value to obtain a comprehensive priority score. The weights in the weighted sum are optimized based on a lightweight neural network (such as a single-layer perceptron) to optimize the weights of the depth-first dimension, width-first dimension, and additional dimension. Specifically, the depth-first dimension value, width-first dimension value, and additional dimension value are used as inputs, and the training objective is to minimize the error between the actual sales revenue of historical distribution and the predicted sales revenue. The optimized weights are then output.

[0042] In this embodiment, depth, width, and additional multi-dimensional factors are integrated, and the weights are optimized through neural network learning. Compared with the traditional fixed-weight scoring method, it can better adapt to complex business scenarios, and the priority scoring can more accurately reflect the actual value of the store-product combination.

[0043] Threshold filtering and priority sorting: See Figure 3 As shown, in this embodiment, a basic allocation candidate list is generated through three stages: dynamic availability threshold screening of stores, multi-level priority sorting, and real-time dynamic correction. This ensures that allocation resources are tilted towards high-value stores and responds to real-time business changes, as detailed below: First, based on the number of stores already stocked and the total number of stores, for each product, calculate the percentage of stores currently covered relative to the total number of stores to obtain the current coverage rate; when the current coverage rate is less than or equal to the threshold of stores that can be stocked. If the current coverage rate is greater than the threshold for eligible stores, the store will be retained in the candidate store set; if the current coverage rate is greater than the threshold for eligible stores. If that happens, the store will no longer be included in the candidate store set for that product.

[0044] As a further implementation of this embodiment, a Priority Correction Factor (PCF) is introduced to avoid the rigid threshold from unfairly disqualifying high-value stores. The PCF is calculated based on a weighted fusion of the store's grade coefficient and priority score contribution value. Specifically: Fixed coefficients are assigned based on preset store levels (A / B / C), with the coefficients positively correlated with the store level. For example, the coefficient for level A stores is 0.15; for level B stores, it's 0.1; and for level C stores, it's 0.05. Then, a comprehensive priority score is taken and multiplied by a fixed weight of 0.1 (to control the impact of priority on PCF and avoid excessive bias), i.e., priority score contribution value = comprehensive priority score × 0.1. The PCF is obtained by summing the priority score contribution value and the store level coefficient. Simultaneously, the PCF value range is constrained to [0.05, 0.2]. Values ​​below 0.05 are set to 0.05 to ensure minimum flexibility; values ​​above 0.2 are set to 0.2 to avoid excessively relaxed thresholds leading to inventory dispersion. The corrected effective threshold = × (1 + PCF); Replace the layable store threshold used in the above coverage comparison with the corrected effective threshold. That's all; In a further embodiment of this example, a linkage mechanism is used to enable high-priority stores to operate even if their coverage rate slightly exceeds the threshold. It can also enter the candidate pool at times.

[0045] The multi-level priority sorting is a three-level sorting mechanism, which refines the ranking of stores in the candidate pool, as follows: Level 1: Sort by comprehensive priority score in descending order, with higher scores ranking higher, to ensure that high-value stores and product combinations are given priority in allocation; Level 2: When the overall priority scores are the same, stores are arranged in descending order of their level (Level A > Level B > Level C), with priority given to core level stores. Level 3: When the overall priority score and store level are the same, they are sorted in descending order by inventory matching degree. The inventory matching degree is calculated by weighting the Inventory Suitability Index (ISI) and the Matching Index (MI), as follows: ; ; ; In the formula: Indicates inventory matching degree; This represents the store's historical weekly average sales volume, calculated as an arithmetic mean based on the store-product corresponding weekly sales data for the past 30 (or 6) calendar weeks. This indicates the current inventory level of a product, referring to the real-time available inventory of that product at the time of the allocation decision. The available inventory refers to the inventory level after deducting the locked inventory. This indicates the store's highest historical weekly sales. This indicates the store's weight, which is preset according to the store's level. For example, level A stores have a weight of 1.2; level B stores have a weight of 1; and level C stores have a weight of 0.8.

[0046] After obtaining the initial candidate list for order fulfillment based on the above three-level sorting mechanism, the sorting results are dynamically corrected in conjunction with real-time inventory alerts, marketing activities, and logistics cost optimization needs to generate a basic candidate list for order fulfillment, as follows: The system queries the central warehouse's inventory level in real time. If the available inventory of a store's required goods is lower than the safety stock threshold, the store's position in the list will be automatically lowered. The specific position to be lowered can be based on preset values, such as lowering the ranking by 2 positions according to a fixed preset value, to avoid orders that cannot be delivered.

[0047] If a store is conducting an official promotional activity related to the target product, the priority of the corresponding store will be increased to ensure sufficient supply during the activity and maximize the benefits of the activity. In this embodiment, increasing the priority means increasing the ranking of the store, which can move the store to the top of the ranking of stores with the same priority. For example, A, B, and C belong to the same priority. At this time, C is ranked at the end. However, since store C is conducting a promotional activity, C will be ranked at the front, resulting in the ranking of C, A, B.

[0048] Secondly, identify clusters of stores in geographical proximity, sort stores with similar priorities together, and merge delivery routes to reduce unit logistics costs.

[0049] In this embodiment, threshold filtering and priority sorting are deeply coupled rather than processed independently. At the same time, real-time dynamic correction is introduced, so that the final candidate list has adaptive adjustment capabilities. Compared with traditional static sorting, it can better meet actual business needs and improve the efficiency of goods allocation and resource utilization.

[0050] Cargo allocation rule embedding and decision output: For each store-product combination in the basic order fulfillment candidate list, the rules of not fulfilling orders for missing sizes and not fulfilling orders for incomplete sets are applied sequentially for filtering. If either rule is triggered, the product is removed from the candidate list of the corresponding store. The rule of not picking up out-of-stock items is that when the inventory of any size of a product does not meet the preset requirements, it is determined to be out of stock and the product is removed from the corresponding store's order list. The rule of not including items in a set is that if the inventory of any component of a set does not meet the preset ordering requirements, the set is deemed not to be a set and is removed from the corresponding store's ordering candidate list.

[0051] As a further implementation of this embodiment, a rule-based pre-detection factor (RIF) is introduced to pre-judge the code and the risk of incomplete sets, and feeds it back to the dynamic layable threshold calculation process to achieve forward-looking adjustment, as detailed below: Discontinued Size Risk Assessment (DRP): Based on real-time inventory data for each size of the product, historical sales consumption rate, and future sales forecasts, calculate the probability of discontinuing a size within a future preset period (e.g., 7 days, matching the order fulfillment plan execution cycle). An example of the steps is as follows: Extract the real-time inventory Stock(s) and the average daily sales SalesAvg(s) of each size of the product; use the time series ARIMA model (input is the daily sales data of the past 30 days) to output the predicted sales SalesPred(s); calculate the number of days that the inventory can support for each size SupportDays(s) = Stock(s) / SalesAvg(s). If SupportDays(s) < 7 and SalesPred(s) ≥ Stock(s), then the risk indicator for the size being out of stock is 1; otherwise, it is 0. Where DRP = number of sizes at risk of running out of stock / total number of sizes in the product, with a value range of [0,1].

[0052] SRP (Supply Requirement Planning): For packaged products, based on the inventory status of each component, historical consumption trends, and component dependencies, predict the probability of insufficient component inventory leading to incomplete sets within a preset period. An example of the steps is as follows: Extract the component list (ComponentList) of the packaged products, the real-time inventory Stock(c) of each component, the average daily sales of the past 30 days SalesAvg(c) and the preset minimum allocation quantity MinAllocationQty; use the ARIMA model consistent with DRP, input the daily sales data of each component in the past 30 days, and output the predicted sales SalesPred(c). Calculate the number of days each component's inventory can support (SupportDays(c)) and its fulfillment capacity (fulfillment coefficient): SupportDays(c)=Stock(c) / SalesAvg(c); The fulfillment factor for stock allocation is calculated as: Stock(c) / (MinAllocationQty + SalesPred(c)). When the allocation satisfaction coefficient is less than 1, it means that the component inventory may not be able to simultaneously meet the current allocation demand and future sales demand. If SupportDays(c) < 7 and the fulfillment coefficient < 1, then the component's stockout risk is marked as 1; otherwise, it is marked as 0. Wherein, SRP = number of components at risk of stockout / total number of components in the kit, with a value range of [0,1]; Based on the obtained results of the Discontinuation Risk Assessment (DRP) and the Incomplete Set Risk Prediction (SRP), the pre-detection factor is calculated by weighted summation. The weight of the weighted summation can be adjusted according to the actual situation. For example, if more attention is paid to the discontinuation risk, the weight corresponding to the discontinuation risk will be higher. It is only necessary to ensure that the sum of the weight corresponding to the discontinuation risk and the weight corresponding to the incomplete set risk prediction is 1.

[0053] The calculated rule-based pre-detection factor is then substituted into the formula for calculating the initial salable store ratio threshold. After the initial salable store ratio threshold is calculated, the rule-based pre-detection factor is used for adjustment. The specific adjustment formula is as follows: ; In the formula: This represents the initial threshold for the percentage of shops that can be vacated after adjustment by rule-based pre-inspection factors. Indicates the rule-based pre-detection factor; In this embodiment, the adjusted By incorporating the threshold smoothing process from the preceding steps, if the risk of missing sizes or incomplete sets of goods is high, lowering the initial threshold before smoothing allows for a more dynamic and playable threshold obtained after subsequent weighted dynamic adjustment and non-linear smoothing. Synchronous reduction, thereby reducing allocation in advance during candidate list generation and avoiding frequent adjustments to subsequent rule filtering; after adjustment We will continue to participate in the dynamic optimization of weights and the non-linear smoothing process, and finally output a dynamic layable threshold that adapts to the risk level.

[0054] Perform a final inventory check on the store-product combinations after rule filtering to ensure that the order quantity meets the inventory limits and minimum order quantity requirements. If the suggested order quantity is greater than the available inventory, adjust the order quantity to the available inventory. If the adjusted suggested order quantity is less than the minimum order quantity, it is determined to be an invalid order and the combination is removed.

[0055] In this embodiment, invalid order fulfillment is avoided by filtering with hard rules, the adaptability of the overall strategy is improved by adjusting the linkage between rules and thresholds, and the feasibility of the order fulfillment plan is ensured by inventory verification.

[0056] Example 2

[0057] See Figure 4As shown, this embodiment provides an intelligent goods distribution system based on dynamic deployable thresholds and multiple priorities, including: Multi-source data acquisition and preprocessing module: It is connected to the dynamic layable threshold calculation module and the multi-level priority scoring module respectively, and synchronizes the generated structured dataset to the two modules as the basic input for subsequent calculations.

[0058] The multi-source data acquisition and preprocessing module is used to collect multi-dimensional basic data required for distribution decisions and to preprocess the collected multi-dimensional basic data to generate a structured dataset. The multi-dimensional basic data includes product attribute data, warehouse inventory data, external environment data, store data, and real-time sales data.

[0059] Specifically, the multi-source data acquisition and preprocessing module integrates ETL tools to incrementally acquire multi-source data from the enterprise database and external systems on a daily schedule; it then sequentially performs preprocessing operations such as data cleaning (missing value imputation, outlier handling, and noise reduction), normalization and standardization transformation (normalization of numerical data, calculation of standardized indicators, and one-hot encoding of categorical data); finally, it outputs a structured dataset to ensure data quality, consistency, and standardization.

[0060] The dynamic scalable threshold calculation module has the following input: it connects to the multi-source data acquisition and preprocessing module to obtain a structured dataset; the output connects to the filtering and sorting module to synchronize the dynamic scalable threshold TS to the filtering and sorting module as the benchmark for store selection; at the same time, it has a feedback connection with the allocation decision output module to receive the rule pre-detection factor RIF from the allocation decision output module for dynamic adjustment of the threshold.

[0061] The dynamic scalability threshold calculation module is based on the structured dataset output by the multi-source data acquisition and preprocessing module. It selects product attribute factors, inventory level factors, and environmental factors, and calculates the initial scalability store ratio threshold through weighted summation. Then, through dynamic weight adjustment and non-linear smoothing, the dynamic scalability threshold of each product is obtained.

[0062] Specifically, the dynamic layable threshold calculation module includes a factor standardization unit, an initial threshold calculation unit, a weight optimization unit, and a smoothing unit. The factor standardization unit standardizes three key factors and outputs standardized values ​​of PAF, ILF, and EF. The initial threshold calculation unit calculates the initial threshold T based on a weighted summation formula. The weight optimization unit dynamically updates the weights of each factor through linear regression analysis of historical data. The smoothing unit uses exponential smoothing to smooth the initial threshold and outputs the dynamic layable threshold. .

[0063] Multi-level priority scoring module: The input end connects to the multi-source data acquisition and preprocessing module to obtain a structured dataset; the output end connects to the filtering and sorting module to synchronize the comprehensive priority score of each store-product pair to the filtering and sorting module as the core basis for sorting.

[0064] The multi-level priority scoring module is based on the structured dataset output by the multi-source data collection and preprocessing module. It constructs a multi-level priority scoring model that includes depth-first dimension, breadth-first dimension and additional dimensions. The comprehensive priority score of each store-product pair is obtained through weighted integration calculation.

[0065] The multi-level priority scoring module incorporates a dimension scoring unit and a weight learning unit. The dimension scoring unit calculates the scores for the three dimensions DPD, WPD, and AD respectively. The weight learning unit uses a single-layer perceptron neural network, taking the scores of the three dimensions DPD, WPD, and AD as input, and aims to minimize the error between the actual sales revenue of historical distribution and the predicted sales revenue. It learns and optimizes the weights corresponding to the three dimensions and calculates the comprehensive priority score by weighted summation.

[0066] The filtering and sorting module has two input ends: one connected to the dynamic availability threshold calculation module and the other to the multi-level priority scoring module to obtain the dynamic availability threshold and the comprehensive priority score. It also connects to an external real-time business database (inventory, promotion, logistics) to obtain real-time data. The other end connects to the allocation decision output module to synchronize the basic allocation candidate list to the allocation decision output module.

[0067] The candidate store set is obtained by filtering stores based on the dynamic availability threshold output by the dynamic availability threshold calculation module. Then, the candidate store set is sorted in multiple levels based on the comprehensive priority score output by the multi-level priority scoring module. After dynamic correction based on real-time business data, a basic ordering candidate list is generated.

[0068] The filtering and sorting module includes a built-in store filtering unit, a multi-level sorting unit, and a dynamic correction unit. The store filtering unit calculates the current product coverage rate and, combined with the dynamic availability threshold and priority correction factor (PCF), filters to obtain a set of candidate stores. The multi-level sorting unit uses a three-level sorting mechanism (comprehensive priority score, store level, and inventory matching degree) to sort the candidate store set and generate an initial allocation candidate list. The dynamic correction unit accesses real-time inventory data, store promotional activity data, and logistics data to dynamically correct the initial list and output a basic allocation candidate list.

[0069] The order fulfillment decision output module has the following inputs: it connects to the filtering and sorting module to obtain a basic order fulfillment candidate list; it also connects to the multi-source data acquisition and preprocessing module to obtain product attribute data and warehouse inventory data to obtain the data required for rule judgment and inventory verification; the output connects to the external logistics execution system to synchronize the optimized order fulfillment plan to the logistics execution system and guide the order fulfillment execution; it also has a feedback connection with the dynamic deployable threshold calculation module to output RIF for threshold adjustment.

[0070] This module filters the basic order fulfillment candidate list output by the screening and sorting module by embedding preset business rules. After inventory verification and order fulfillment quantity adjustment, it outputs an optimized order fulfillment plan.

[0071] Specifically, the product allocation decision output module has a built-in rule filtering unit, linkage adjustment unit, and inventory verification unit. The rule filtering unit applies the rule of not allocating products with missing codes and the rule of not allocating products with incomplete sets to filter store-product combinations. The linkage adjustment unit calculates the rule pre-detection factor (RIF) and feeds it back to the dynamic availability threshold calculation module to adjust the dynamic availability threshold. The inventory verification unit queries the real-time available inventory, adjusts the allocation quantity to meet the inventory limit and minimum allocation quantity requirements, and generates and outputs an optimized product allocation plan.

[0072] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A smart product allocation method based on dynamic deployable thresholds and multi-level priorities, characterized in that, Includes the following steps: Data Acquisition and Preprocessing: Collect multi-dimensional basic data required for allocation decisions, and preprocess the collected multi-dimensional basic data to obtain a structured dataset; Dynamic scalability threshold calculation: Based on a structured dataset, product attribute factors, inventory level factors, and environmental factors are selected. The initial scalability threshold is obtained by weighted summation. Then, the dynamic scalability threshold for each product is obtained through dynamic weight adjustment and non-linear smoothing. Multi-level priority comprehensive score: Based on the structured dataset, a multi-priority scoring model is constructed that includes depth-first dimension, breadth-first dimension and additional dimensions. The comprehensive priority score of each store-product pair is calculated by weighted summation. Threshold filtering and priority sorting: Based on the dynamic scalable threshold, stores are filtered to obtain a set of candidate stores. Then, the set of candidate stores is sorted in multiple levels based on the comprehensive priority score. After dynamic correction based on real-time business data, a basic ordering candidate list is generated. Loan allocation rule embedding and decision output: Based on the generated basic loan allocation candidate list, preset business rules are embedded for filtering. After inventory verification and loan allocation quantity adjustment, the optimized loan allocation plan is output.

2. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority as described in claim 1, characterized in that, The multi-dimensional basic data includes product attribute data, warehouse inventory data, external environment data, store data, and real-time sales data.

3. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority according to claim 1, characterized in that, The calculation of the dynamic layable threshold includes the following steps: First, the selected product attribute factors, inventory level factors, and environmental factors are standardized. Then, the initial threshold for the proportion of stores that can be scalable is obtained by weighted summation. The weights of each factor are obtained by linear regression analysis of historical data and are dynamically updated. Then, the initial threshold for the proportion of shops that can be scalable is non-linearly smoothed using the exponential smoothing method to eliminate threshold abrupt changes caused by data fluctuations, thus obtaining the final dynamic threshold for scalable shops.

4. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority according to claim 3, characterized in that, The product attribute factors are determined based on the product life cycle stage and price range; the inventory level factors are calculated based on the inventory tightness index; and the environmental factors are determined by combining seasonal adjustment factors and promotional activity adjustment factors.

5. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority according to claim 1, characterized in that, The depth-first dimension is obtained by weighted summation based on planned order volume and historical sales performance; the breadth-first dimension is obtained by weighted summation based on store level and historical geographic coverage; the additional dimension is obtained by weighted summation based on real-time sales trends and product attributes. Then, a learning model is used to optimize the weights of the depth-first dimension, the breadth-first dimension, and the additional dimensions, and a comprehensive priority score is obtained by weighted summation of the scores of each dimension.

6. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority according to claim 1, characterized in that, The basic cargo allocation candidate list is obtained based on the following steps: Calculate the current store coverage rate for each product, compare the store coverage rate with the dynamic available store threshold, and filter the candidate store set by combining the priority correction factor. A three-level sorting mechanism is used to sort the candidate store set. The first level is sorted in descending order by comprehensive priority score. The second level is sorted in descending order by store level when the comprehensive priority scores are the same. The third level is sorted in descending order by inventory matching degree when the comprehensive priority scores and store levels are the same. The ranking results are dynamically revised based on real-time inventory alerts, marketing campaigns, and logistics cost optimization needs to generate a basic candidate list for order fulfillment.

7. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority according to claim 6, characterized in that, The inventory matching degree is calculated by weighting the inventory fit index and the matching index. The inventory fit index is determined based on the degree of match between the store's historical sales capacity and the current inventory level, while the matching index is determined by combining the store's historical sales stability and the store's weight.

8. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority as described in claim 1, characterized in that, The preset business rules include the rule of not including items with missing sizes and the rule of not including items with incomplete sets; the business rules are applied sequentially to filter the store-product combinations in the basic ordering candidate list; Introducing rule-based pre-detection factors, the risk of code shortage and incomplete sets is estimated based on historical data. The rule-based pre-detection factors are then fed back into the dynamic playable threshold calculation process to adjust the dynamic playable threshold. Perform inventory verification on the filtered store-product combination, adjust the allocation quantity to meet the inventory limit and minimum allocation requirements, and output an optimized allocation plan that includes the target store, allocated products, allocation quantity, and execution time.

9. The intelligent commodity allocation method based on dynamic deployable threshold and multi-level priority as described in claim 8, characterized in that, The rule of not picking up out-of-stock items is that when the inventory of any size of a product does not meet the preset requirements, it is determined to be out of stock and the product is removed from the corresponding store's order list. The rule of not including items in a set is that if the inventory of any component of a set does not meet the preset ordering requirements, the set is deemed not to be a set and is removed from the corresponding store's ordering candidate list.

10. An intelligent goods distribution system based on dynamic deployable thresholds and multi-level priorities, characterized in that: include: The multi-source data acquisition and preprocessing module is used to collect multi-dimensional basic data required for allocation decisions, and to preprocess the collected multi-dimensional basic data to generate a structured dataset. The dynamic scalable threshold calculation module is connected to the multi-source data acquisition and preprocessing module. It is used to select product attribute factors, inventory level factors and environmental factors based on the structured dataset, calculate the initial scalable store ratio threshold through weighted summation, and then obtain the dynamic scalable threshold for each product through dynamic weight adjustment and nonlinear smoothing. The multi-level priority scoring module is connected to the multi-source data acquisition and preprocessing module. It is used to construct a multi-level priority scoring model based on the structured dataset, which includes depth-first dimension, width-first dimension and additional dimensions. The comprehensive priority score of each store-product pair is obtained by weighted integration calculation. The filtering and sorting module is connected to the dynamic deployable threshold calculation module and the multi-level priority scoring module, respectively. It then performs multi-level sorting based on the comprehensive priority score and dynamically corrects it in combination with real-time business data to generate a basic cargo allocation candidate list. The allocation decision output module, connected to the filtering and sorting module, is used to embed preset business rules to filter the basic allocation candidate list, and output the optimized allocation plan after inventory verification and allocation quantity adjustment.

Images