Method for determining and adjusting replenishment strategy

Abstract

The embodiment of the invention relates to the field of data processing, in particular to a method for determining and adjusting a replenishment strategy. The method comprises the following steps: acquiring a replenishment strategy function for outputting a replenishment decision amount, wherein the replenishment strategy function has adjustable strategy parameters; acquiring state information; based on the state information, determining a replenishment decision amount through a replenishment strategy function, and outputting a replenishment instruction corresponding to the replenishment decision amount; and obtaining index information related to the replenishment instruction, and updating the strategy parameters based on the index information. Wherein strategy parameter updating is limited by updating constraint parameters, and the updating constraint parameters are adaptively determined along with changes of the state information, so that the updating constraint parameters are increased relative to the reference constraint parameters under the condition that the risk degree represented by the state information is larger than a risk threshold value, and the updating constraint parameters are increased relative to the reference constraint parameters under the condition that the risk degree is smaller than the risk threshold value. The updated constraint parameter is reduced relative to the reference constraint parameter.

Description

Methods for determining and adjusting replenishment strategies Technical Field

[0001] This application relates to the field of data processing, and in particular to methods for determining and adjusting replenishment strategies. Background Technology

[0002] Inventory management and replenishment decisions are fundamental issues in supply chain operations, widely applied in retail, e-commerce, manufacturing, and spare parts support scenarios. Common approaches in related technologies include economic order quantity (EOQ), safety stock, reorder point, (s, S) strategy, and the newsboy model, used to determine the timing and quantity of replenishment under specific conditions.

[0003] However, in actual business operations, demand and supply are often characterized by uncertainty and dynamic changes. Lead time, promotional activities, seasonality and external disturbances can affect inventory consumption and delivery rhythm, making replenishment decisions a matter of weighing multiple factors such as holding costs, stockout risks and ordering constraints.

[0004] Therefore, how to generate reasonable replenishment decisions based on inventory and related business information under multi-cycle operating conditions, and maintain the stability and adaptability of the strategy in a complex and volatile environment, remains a focus of attention in related technical fields. Summary of the Invention

[0005] The purpose of this application is to provide a method for determining and adjusting replenishment strategies, so as to achieve a more reasonable balance between factors such as inventory holding costs, stockout risks and ordering costs.

[0006] The objective of this application is achieved through the following technical solution: Firstly, embodiments of this application provide a method for determining and adjusting a replenishment strategy, comprising: obtaining a replenishment strategy function for outputting a replenishment decision quantity, the replenishment strategy function having adjustable strategy parameters; obtaining state information, the state information including at least one of inventory information and other information related to replenishment; determining a replenishment decision quantity based on the state information using the replenishment strategy function, and outputting a replenishment instruction corresponding to the replenishment decision quantity; obtaining indicator information related to the replenishment instruction, and updating the strategy parameters based on the indicator information; wherein the updating of the strategy parameters is subject to an update constraint parameter, the update constraint parameter being adaptively determined as the state information changes, such that when the risk level represented by the state information is greater than a risk threshold, the update constraint parameter increases relative to the baseline constraint parameter, and when the risk level is less than the risk threshold, the update constraint parameter decreases relative to the baseline constraint parameter.

[0007] In some embodiments, the other information related to replenishment includes at least one of the following: in-transit order information, demand forecast information, promotional information, or other external factor information; wherein the external factor information includes lead time information and / or cost parameter information.

[0008] In some embodiments, the status information includes at least two of the following: inventory information and other information related to replenishment.

[0009] In some embodiments, the indicator information represents at least two of the following: revenue, inventory holding cost, stockout penalty cost, and ordering cost; and / or, the indicator information also represents a replenishment decision quantity change item, which is related to the difference between the replenishment decision quantities corresponding to two adjacent replenishment instructions; in some embodiments, the replenishment decision quantity is a continuous value or a discrete value; when the replenishment decision quantity is a continuous value, the replenishment decision quantity is determined by the continuous variable output by the replenishment strategy function through a preset mapping relationship to satisfy a preset constraint.

[0010] In some embodiments, the update constraint parameter is adaptively determined as the state information changes, including: determining the update constraint parameter based on the deviation of the inventory information from a preset inventory threshold; wherein, when the inventory information is lower than the preset inventory threshold, the update constraint parameter increases relative to the baseline constraint parameter, and the update constraint parameter is positively correlated with the deviation.

[0011] In some embodiments, updating the strategy parameters based on the indicator information includes: applying a threshold limit or pruning limit to the difference measure of the replenishment strategy function before and after the update in the update target, wherein the threshold limit or pruning limit is determined by the update constraint parameter; wherein the difference measure includes the ratio of the probability or probability density of the same replenishment decision quantity before and after the update; and / or, the threshold limit or pruning limit is an asymmetric limit.

[0012] In some embodiments, determining the replenishment decision quantity based on the state information and the replenishment strategy function includes: generating a time-series feature representation based on historical sales sequences and / or demand forecast sequences; fusing the time-series feature representation with non-time-series feature representations corresponding to inventory information and / or orders in transit information to determine the replenishment decision quantity; wherein the replenishment decision quantity is a continuous value or a discrete value, and when the replenishment decision quantity is a continuous value, the replenishment decision quantity is determined to be a replenishment decision quantity that satisfies a preset constraint through a preset mapping relationship.

[0013] In a second aspect, embodiments of this application provide a computer storage medium for storing a program, wherein the program, when running, controls the device where the computer storage medium is located to execute the method described in any of the first aspects.

[0014] Thirdly, embodiments of this application provide a computer program product for implementing the method described in any of the first aspects.

[0015] This application provides a method for determining and adjusting replenishment strategies. This application addresses the technical problems of traditional inventory management methods, which are prone to falling into short-sightedness in single-cycle scenarios with multi-cycle demand fluctuations and supply uncertainties, and difficulty in balancing multiple objectives such as holding costs and stockout risks. Furthermore, existing PPOs have limitations in the scope of strategy updates and insufficient capture of long-cycle dependencies when used for replenishment. This application adopts a multi-cycle dynamic replenishment method that models the replenishment process as a Markov decision process and introduces a multi-stage long-view reward function, a state-dependent asymmetric clipping mechanism, and a strategy network architecture that integrates attention mechanisms into the PPO framework. Attached Figure Description

[0016] This application will be further described below with reference to the accompanying drawings and specific embodiments.

[0017] Figure 1 is a flowchart illustrating a method for determining and adjusting a replenishment strategy according to an embodiment of this application.

[0018] Figure 2 is the overall architecture for determining and adjusting a replenishment strategy according to an embodiment of this application.

[0019] Figure 3 is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the description of the embodiments of this application, it should be understood that 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. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0022] In related technologies, inventory management often employs methods such as the newsboy model, economic order quantity, safety stock, and reorder point. However, these methods are typically based on single-cycle assumptions and idealized distribution / fixed lead time assumptions, making them difficult to adapt to the multi-cycle and uncertain changing scenarios in real supply chains. These methods tend to focus more on the current cycle, which can easily lead to short-sighted decisions, resulting in local optima rather than global optima, and making it difficult to balance multiple objectives such as reducing holding costs, mitigating stockout risks, and addressing order delays.

[0023] Although deep reinforcement learning has been introduced in recent years, existing proximal policy optimization (PPO) algorithms still face challenges when used for replenishment, such as limited update range, insufficient adaptation to action space distribution, and insufficient capture of long-term dependencies, which affect the flexibility and stability of the policy.

[0024] To address this, this application models the replenishment problem as a Markov Decision Process (MDP) and implements multi-cycle dynamic replenishment decisions based on an improved PPO. This allows the agent to learn interactively in a simulation environment to adapt to demand fluctuations and data uncertainty. Specifically, the state vector incorporates information such as current inventory, orders in transit, demand forecasts, and promotional / external factors. The action output is the replenishment quantity (which can be continuous actions). Furthermore, a multi-stage long-view reward system integrates cross-cycle inventory accumulation costs, stockout penalties, and action smoothing terms into the optimization objective. Simultaneously, in low-inventory, high-risk areas, state-dependent asymmetric clipping is employed to improve strategy adjustment efficiency.

[0025] In addition, the strategy network incorporates a time-series multi-head attention mechanism to capture key demand characteristics from long-sequence historical sales and complex fluctuations, thereby improving robustness to non-stationary demand and supply delays.

[0026] Based on the above technical solutions, simulation / backtesting shows that total operating costs can be reduced by about 10% to 15%, holding costs by about 15% to 20%, and the stockout rate can be reduced by about 30% (e.g., from 4.0% to 2.8%). It also makes replenishment instructions smoother, reduces order volatility variance, improves turnover, and enhances training stability and generalization ability.

[0027] Referring to Figure 1, Figure 1 is a flowchart illustrating a method for determining and adjusting a replenishment strategy according to an embodiment of this application.

[0028] This application provides a method for determining and adjusting a replenishment strategy, including steps S101-S104: S101: Obtain a replenishment strategy function for outputting replenishment decision quantities, wherein the replenishment strategy function has adjustable strategy parameters.

[0029] S102: Obtain status information, which includes at least one of inventory information and other information related to replenishment.

[0030] S103: Based on the status information, determine the replenishment decision quantity through the replenishment strategy function, and output the replenishment instruction corresponding to the replenishment decision quantity.

[0031] S104: Obtain indicator information related to the replenishment instruction, and update the strategy parameters based on the indicator information.

[0032] The updating of the strategy parameters is subject to an update constraint parameter, which is adaptively determined as the state information changes, such that when the risk level represented by the state information is greater than a risk threshold, the update constraint parameter increases relative to the baseline constraint parameter, and when the risk level is less than the risk threshold, the update constraint parameter decreases relative to the baseline constraint parameter.

[0033] In some embodiments, the method for determining and adjusting replenishment strategies is used to generate replenishment decision quantities in a supply chain replenishment scenario and adjust the replenishment strategies in conjunction with feedback information during operation. The method for determining and adjusting replenishment strategies can be executed by a server, edge computing device, or business terminal, or it can be executed multiple times by the same computing device at different points in time to adapt to the replenishment needs of different replenishment cycles, different products, or different warehouses.

[0034] In some embodiments, the replenishment strategy function used to output the replenishment decision quantity is used to establish a mapping relationship between state information and replenishment decision quantity. Specifically, the replenishment strategy function can be in the form of a parameterized function, a lookup table function, a piecewise function, or a combined function, or it can be a strategy expression form composed of rules and parameters. For example, the replenishment strategy function can calculate the target replenishment quantity based on inventory information and demand forecast information, or it can adjust the determination logic of the target replenishment quantity when promotional information is triggered.

[0035] In some embodiments, the replenishment strategy function has adjustable strategy parameters that influence the output or distribution of the replenishment strategy function. Specifically, the adjustable strategy parameters may include weighting coefficients, threshold parameters, smoothing coefficients, lookup table index values, inflection point parameters of the piecewise function, mean parameters or variance parameters of the probability distribution, etc. The adjustable strategy parameters may also include a set of parameters that reflect business preferences, such as a parameter combination that is more biased towards reducing stockout risk or a parameter combination that is more biased towards reducing inventory holding costs.

[0036] In some embodiments, obtaining status information is used as input to form a replenishment strategy function. Specifically, obtaining status information may include reading data from an inventory management system, order management system, enterprise resource planning system, or data service interface, or it may include cleaning, aligning, and normalizing the read data to obtain status information. The status information can be constructed at the product granularity, warehouse granularity, or product-warehouse combination granularity so that the replenishment strategy function outputs replenishment decision quantities for different objects.

[0037] In some embodiments, the status information includes at least inventory information. Inventory information characterizes the current level of inventory available to meet demand. Specifically, inventory information may include one or more indicators such as existing inventory quantity, available inventory quantity, locked inventory quantity, safety stock quantity, days in stock, and inventory occupancy rate. Inventory information may also include statistics related to inventory turnover, such as historical average consumption rate or recent cycle consumption rate.

[0038] In some embodiments, the status information also includes at least one of other information related to replenishment. Specifically, other information related to replenishment may include in-transit order information, demand forecast information, promotional information, or other external factor information. For example, in-transit order information may include the number of orders not yet delivered, the estimated arrival time, or the batch number of deliveries. Demand forecast information may include a demand forecast sequence for multiple future time steps, a forecast confidence interval, or a forecast volatility measure. Promotional information may include the promotion start time, promotion duration, promotion intensity flag, or activity type flag. Other external factor information may include one or more of the following: lead time information, supply capacity information, price information, cost parameter information, holiday flags, or external disturbance information such as weather.

[0039] In some embodiments, based on state information, a replenishment decision quantity is determined through a replenishment strategy function to generate an executable replenishment decision result. Specifically, the replenishment decision quantity can be a quantity representing the replenishment scale, such as order quantity, transfer quantity, replenishment batch quantity, or replenishment frequency. The replenishment decision quantity can be a continuous value or a discrete value. For example, the replenishment decision quantity can take values ​​within a preset range or can be selected from multiple candidate replenishment quantity levels. The replenishment decision quantity can also be mapped or modified by business constraints, such as satisfying minimum order quantity constraints, full case multiple constraints, budget constraints, or warehouse capacity constraints.

[0040] In some embodiments, outputting a replenishment instruction corresponding to the replenishment decision quantity is used to transmit the replenishment decision quantity to the execution side. Specifically, the replenishment instruction may include the replenishment decision quantity and additional fields related to execution, such as product identifier, warehouse identifier, supplier identifier, expected arrival time, priority, or order batch information. The replenishment instruction can be output to the purchasing system, order system, or warehousing system via interface calls, message queues, or files to trigger ordering, allocation, or stock preparation operations.

[0041] In some embodiments, acquiring indicator information related to replenishment instructions is used to reflect the execution results or operational quality of the replenishment instructions. Specifically, the indicator information can come from actual business operation results, or from evaluation results in a simulation environment or historical playback environment. The indicator information can be generated according to the replenishment cycle and associated with replenishment decision quantities and status information for use in updating strategy parameters.

[0042] In some embodiments, the metric information can represent at least two of the following: revenue, inventory holding costs, stockout penalty costs, and ordering costs. Specifically, revenue may be related to completed sales volume or service level; inventory holding costs may be related to average inventory level or days in stock; stockout penalty costs may be related to stockout quantity, stockout duration, or unmet demand; and ordering costs may be related to the number of orders placed, transportation costs, or procurement costs. The metric information can also represent a change in replenishment decision quantity, which is related to the difference between the replenishment decision quantities corresponding to two adjacent replenishment instructions, and is used to reflect the degree of fluctuation in replenishment decision quantity.

[0043] In some embodiments, adjusting strategy parameters based on indicator information are used to adjust the output behavior of the replenishment strategy function. Specifically, updating the adjustable strategy parameters based on indicator information can be done online, offline, or periodically. For example, updating the adjustable strategy parameters based on indicator information may include determining the update direction and magnitude based on the indicator information, or selecting or switching combinations of adjustable strategy parameter values ​​based on the indicator information. Updating the adjustable strategy parameters based on indicator information may also include smoothing the update amount during the update process to avoid update jitter.

[0044] In some embodiments, the updates to adjustable strategy parameters are constrained by update constraint parameters, which are adaptively determined as the state information changes. Specifically, update constraint parameters can be used to limit the update magnitude, update rate, or allowed range of strategy changes. Update constraint parameters can be expressed in the form of thresholds, pruning boundaries, upper limits on step size, or confidence ranges. The adaptive determination of update constraint parameters as the state information changes can be understood as the update constraint parameters not being fixed constants, but dynamically adjusted based on inventory information, demand forecast information, or lead time information, so as to apply update constraints of different strengths under different operating conditions.

[0045] In some embodiments, the risk level represented by the status information is used to reflect the stockout risk or inventory risk faced by replenishment decisions, and the risk threshold is used to distinguish between states with higher and lower risk levels. Specifically, the risk level can be determined based on the deviation of inventory information from safety stock, or it can be determined by combining in-transit order information and demand forecast information. For example, the risk level can increase as inventory information decreases, or it can increase when demand forecast information fluctuates more or lead time information becomes longer. The risk threshold can be a preset threshold or it can be configured based on historical statistics or business strategies.

[0046] In some embodiments, when the risk level is greater than a risk threshold, the updated constraint parameter increases relative to the baseline constraint parameter; when the risk level is less than the risk threshold, the updated constraint parameter decreases relative to the baseline constraint parameter. The baseline constraint parameter can be a pre-set default constraint level or a constraint level obtained statistically from historical stable states. Increasing the updated constraint parameter relative to the baseline constraint parameter can be understood as allowing a larger range of parameter changes or a faster update step size; decreasing the updated constraint parameter relative to the baseline constraint parameter can be understood as tightening the range of parameter changes or reducing the update step size. The specific value can be determined according to the definition of the updated constraint parameter.

[0047] In this way, based on the input status information containing inventory and replenishment information, replenishment decision quantities can be output and replenishment instructions can be generated. At the same time, the adjustable strategy parameters can be updated using the indicator information related to the replenishment instructions, and the update process can be constrained by the update constraint parameters that adapt to the status information. This maintains the adaptability and stability of strategy adjustment under different risk levels and helps to achieve a more reasonable trade-off between factors such as inventory holding costs, stockout risks and ordering costs.

[0048] In some embodiments, the other information related to replenishment includes at least one of the following: in-transit order information, demand forecast information, promotional information, or other external factor information; wherein the external factor information includes lead time information and / or cost parameter information.

[0049] In some embodiments, additional replenishment-related information in the status information is used to supplement and characterize supply and demand and fulfillment factors beyond inventory information, so that the replenishment strategy function can comprehensively consider delivery rhythm, demand changes, and business constraints when determining the replenishment decision quantity. Specifically, the additional replenishment-related information may include at least one of the following: in-transit order information, demand forecast information, promotional information, or other external factor information, or multiple of them may be used in combination depending on the business scenario.

[0050] In some embodiments, in-transit order information is used to reflect replenishment resources that have been placed but not yet received into inventory. Specifically, in-transit order information may include in-transit quantity, estimated arrival time, arrival batch, order priority, order cancellation or delay flags, etc. For example, in-transit order information can be maintained at the product granularity or product-warehouse combination granularity so that the replenishment strategy function can distinguish between the current available inventory and the short-term upcoming quantity when determining the replenishment decision quantity.

[0051] In some embodiments, demand forecasting information is used to reflect the level or trend of demand changes over a future period. Specifically, demand forecasting information may include a forecasted demand sequence for multiple future time steps, a forecast confidence range, and a forecast volatility level. For example, demand forecasting information can be generated from historical sales data, seasonal characteristics, and holiday characteristics, and can also be revised by incorporating promotional information. Promotional information is used to reflect the pull or disturbance of demand caused by promotional activities. Specifically, promotional information may include the promotion start time, promotion duration, promotion intensity marker, activity type marker, channel marker, etc.

[0052] In some embodiments, other external factor information is used to reflect external constraints or disturbances that affect replenishment decisions, other than those on the demand and order sides. This other external factor information includes at least one of lead time information and cost parameter information. Specifically, lead time information characterizes the time characteristics from the issuance of a replenishment order to the arrival of replenishment resources in the warehouse. Lead time information may include expected lead time, minimum lead time, maximum lead time, lead time volatility, lead time delay markers, etc. For example, lead time information can be derived from supplier historical fulfillment data, logistics timeliness statistics, or current transportation status. Cost parameter information characterizes replenishment-related costs and constraints. Cost parameter information may include one or more of the following: inventory holding cost parameters, stockout penalty cost parameters, ordering cost parameters, transportation cost parameters, capital occupation cost parameters, etc., and may also include parameterized information related to warehouse capacity, budget, or purchase discounts.

[0053] To address this, by incorporating information on orders in transit, demand forecasts, promotions, or other external factors, and further by introducing lead time and cost parameters into these external factors, the replenishment strategy function can obtain a more complete state profile when determining replenishment decision quantities. This helps to make replenishment decisions that are more in line with business constraints in scenarios where demand fluctuations, delivery uncertainties, and cost constraints coexist. At the same time, it provides an input basis that is closer to actual operating conditions for subsequent strategy parameter updates.

[0054] In some embodiments, the status information includes at least two of the following: inventory information and other information related to replenishment.

[0055] In some embodiments, status information is used as input to a replenishment strategy function so that the replenishment strategy function can simultaneously consider inventory levels and multidimensional business factors when determining replenishment decision quantities. Specifically, status information may include at least two of the following: inventory information and other information related to replenishment, so that the status information reflects not only existing inventory information, but also factors affecting replenishment decisions, such as the pace of supply-side deliveries and the characteristics of demand-side fluctuations.

[0056] As an example, status information may include inventory information, in-transit order information, and demand forecast information. In some embodiments, in-transit order information can be used to characterize replenishment resources for which replenishment instructions have been issued but have not yet been received, and demand forecast information can be used to characterize the predicted demand level or demand change trend over multiple future time steps. For example, when inventory information is at a low level and in-transit order information indicates insufficient short-term arrivals, the replenishment strategy function can output a relatively large replenishment decision amount; when inventory information is at a moderate level and in-transit order information indicates sufficient short-term arrivals, the replenishment strategy function can output a relatively small replenishment decision amount.

[0057] As another example, status information may include inventory information, as well as demand forecast information and promotional information. In some embodiments, promotional information can be used to reflect potential demand increases or disturbances caused by promotional activities, and demand forecast information can be adjusted by incorporating promotional information when generating the forecast demand sequence. For example, if promotional information indicates that a promotion is about to begin and demand forecast information indicates a future increase in demand, the replenishment strategy function can output a higher replenishment decision amount to meet potential demand.

[0058] As another example, status information may include inventory information, in-transit order information, and other external factor information. In some embodiments, other external factor information may include at least one of lead time information and cost parameter information. For example, if lead time information indicates a longer or more volatile delivery cycle, the replenishment strategy function may adjust the replenishment decision quantity to reduce the stockout risk caused by delivery uncertainty; if cost parameter information indicates a high stockout penalty cost, the replenishment strategy function may prioritize replenishment quantity when inventory information is low, and if cost parameter information indicates a high inventory holding cost, the replenishment strategy function may reduce replenishment quantity when inventory information is high.

[0059] It is understandable that the selection and combination of other replenishment-related information in the status information can be determined based on the business object and business scenario. In some embodiments, for high-turnover goods, demand forecast information and promotional information can be prioritized; for goods with unstable supply, in-transit order information and lead time information can be prioritized; and for cost-sensitive businesses, cost parameter information and inventory information can be used together to construct the status information. The status information can also be constructed at the product granularity, warehouse granularity, or product and warehouse combination granularity, so that the replenishment strategy function can output replenishment decision quantities for different objects.

[0060] This application embodiment introduces at least two of the following into the state information: inventory information and other information related to replenishment. This allows the replenishment strategy function to obtain richer input basis when determining the replenishment decision quantity, thereby making the generation of the replenishment decision quantity more in line with the comprehensive situation of supply and demand changes, delivery rhythm and business constraints, and providing a more sufficient state representation basis for strategy parameter updates.

[0061] In some embodiments, the indicator information represents at least two of the following: revenue, inventory holding costs, stockout penalty costs, and ordering costs; and / or, the indicator information also represents a replenishment decision quantity change item, which is related to the difference in replenishment decision quantities corresponding to two adjacent replenishment instructions.

[0062] In some embodiments, the indicator information can be used to reflect the operational results or quality corresponding to replenishment instructions, and serve as the basis for updating adjustable strategy parameters. In some specific implementations, the indicator information can represent at least two of the following: revenue, inventory holding costs, stockout penalty costs, and ordering costs, so that strategy parameter updates can comprehensively consider the impact of replenishment decisions on business objectives and cost constraints.

[0063] In this context, revenue can be used to characterize the value contribution of replenishment orders to meeting demand and achieving sales. In some embodiments, revenue can be related to actual sales volume, or it can be related to service level, satisfaction rate, or order fulfillment rate. For example, revenue can be calculated by replenishment cycle, or by product granularity or product and warehouse combination granularity. Revenue calculation can be based on actual sales data, outbound data, or order completion data.

[0064] In some embodiments, inventory holding costs can be used to characterize the costs associated with holding inventory. Specifically, inventory holding costs can be related to average inventory levels, ending inventory levels, or days in stock, or they can be related to warehousing costs, capital costs, and loss risks. For example, inventory holding costs can be determined based on inventory information and preset inventory holding cost parameters, which can be configured according to product attributes, warehousing costs, or capital costs.

[0065] In some embodiments, stockout penalty costs can be used to characterize the losses or risks associated with stockouts. Specifically, stockout penalty costs can be related to the amount of stockout, the amount of unmet demand, the duration of the stockout, or the amount of delayed delivery, or they can be related to a decline in customer experience, order cancellations, or losses from alternative purchases. For example, stockout penalty costs can be determined based on the difference between actual demand and available supply, or based on stockout event flags and preset penalty parameters.

[0066] In some embodiments, ordering costs can be used to characterize the ordering and fulfillment costs associated with replenishment activities. Specifically, ordering costs may include order processing costs, transportation costs, procurement costs, or fixed and variable costs associated with order batches. Ordering costs may be related to the replenishment decision quantity, or to the number of replenishment orders placed or the order batches. For example, in the presence of minimum order quantity constraints or full-case multiple constraints, the calculation of ordering costs may consider the actual order quantity mapped from the replenishment decision quantity.

[0067] In some embodiments, the indicator information can also characterize the replenishment decision quantity change item, which reflects the degree of fluctuation in the replenishment decision quantity over a period of time. Specifically, the replenishment decision quantity change item is related to the difference between the replenishment decision quantities corresponding to two adjacent replenishment instructions. The replenishment decision quantity change item can be determined using the absolute value of the difference, the square of the difference, or the normalized difference. For example, the replenishment decision quantity change item can compare the replenishment decision quantities of the same product in adjacent replenishment cycles, or it can aggregate the replenishment decision quantity changes of multiple products in the same warehouse.

[0068] In some embodiments, the combination of indicator information can be configured according to business objectives. Specifically, when the business objective is more focused on ensuring supply, the indicator information can prioritize including stockout penalty costs and benefits; when the business objective is more focused on reducing inventory holding, the indicator information can prioritize including inventory holding costs and ordering costs; when replenishment fluctuations in the business scenario affect the stability of the supply chain, the indicator information can further include replenishment decision quantity changes, so that strategy parameter updates consider both costs and benefits while taking into account the fluctuation constraints of replenishment decision quantities.

[0069] In this regard, the embodiments of this application use indicators to represent at least two of the following: revenue, inventory holding costs, stockout penalty costs, and ordering costs. Optionally, they can further represent the change in replenishment decision quantity. This can provide a more comprehensive measurement basis for updating strategy parameters, enabling the adjustment of the replenishment strategy function to be closer to the actual business operation goals and constraints, and helping to make a more reasonable trade-off between the stability of replenishment decision quantity and cost benefits.

[0070] In some embodiments, the replenishment decision quantity is a continuous value or a discrete value; when the replenishment decision quantity is a continuous value, the replenishment decision quantity is determined by the continuous variable output by the replenishment strategy function through a preset mapping relationship to satisfy the preset constraints.

[0071] The replenishment decision quantity can be used to characterize the replenishment scale or replenishment intensity. In some embodiments, a continuous replenishment decision quantity can be understood as the replenishment decision quantity output by the replenishment strategy function being able to take any real number value within a preset range, so as to adjust the replenishment scale more finely. A discrete replenishment decision quantity can be understood as the replenishment decision quantity being selected from a preset number of candidate values, such as from multiple order tiers, multiple replenishment levels, or multiple fixed batch options.

[0072] In some embodiments, when the replenishment decision quantity is a continuous value, the replenishment strategy function outputs a continuous variable, which represents the original output of the replenishment decision quantity. The continuous variable can be a normalized continuous variable or an unnormalized continuous variable. For example, the continuous variable can fall within a preset interval, or it can be output in any real number form, and the actual replenishment decision quantity is subsequently obtained through a preset mapping relationship.

[0073] In some embodiments, a preset mapping relationship can be used to convert continuous variables into replenishment decision quantities that satisfy preset constraints. Specifically, preset constraints may include one or more of the following: minimum order quantity constraint, maximum order quantity constraint, full container load constraint, minimum order amount constraint, budget constraint, warehouse capacity constraint, shipping batch constraint, or supply capacity constraint. The preset mapping relationship may include forms such as truncated mapping, scaling mapping, segmented mapping, quantization mapping, or combined mapping.

[0074] In other embodiments, the preset mapping relationship may include a truncated mapping. Specifically, a truncated mapping may restrict the original replenishment quantity corresponding to a continuous variable to between the minimum replenishment quantity and the maximum replenishment quantity. For example, a truncated mapping may map an original replenishment quantity less than the minimum replenishment quantity to the minimum replenishment quantity, and an original replenishment quantity greater than the maximum replenishment quantity to the maximum replenishment quantity, while keeping the original replenishment quantity falling within the range unchanged or adjusting it proportionally.

[0075] In some other embodiments, the preset mapping relationship may include quantization mapping. Specifically, quantization mapping can map the original replenishment quantity to a replenishment decision quantity that satisfies a full-case multiple constraint or a fixed batch size constraint. For example, quantization mapping can round up the original replenishment quantity to the replenishment quantity corresponding to a full-case multiple, or it can map the original replenishment quantity to the closest candidate replenishment quantity level.

[0076] In some embodiments, the preset mapping relationship may include a combined mapping based on multiple constraints. Specifically, when both budget constraints and storage capacity constraints exist, the preset mapping relationship can further limit the replenishment decision quantity to not exceed the range allowed by the budget limit and storage capacity limit, while satisfying the minimum order quantity constraint. When supply capacity constraints exist, the preset mapping relationship can adjust the replenishment decision quantity in conjunction with supply capacity information to ensure that the replenishment decision quantity does not exceed the supply capacity limit.

[0077] This application embodiment sets the replenishment decision quantity to a continuous or discrete value, and obtains the replenishment decision quantity that satisfies the preset constraints through a preset mapping relationship when the replenishment decision quantity is a continuous value. This enables the output of the replenishment strategy function to have both continuous adjustment capability and adaptability to common order batch and resource constraints in actual business, thereby improving the feasibility and consistency of replenishment instructions on the execution side.

[0078] In some embodiments, updating the strategy parameters based on the indicator information includes: applying a threshold limit or pruning limit to the difference measure of the replenishment strategy function before and after the update in the update target, wherein the threshold limit or pruning limit is determined by the update constraint parameter.

[0079] Wherein, the difference metric includes the ratio of the probability or probability density of the same replenishment decision before and after the update; and / or, the threshold limit or pruning limit is an asymmetric limit.

[0080] In some embodiments, updating strategy parameters based on indicator information is used to adjust the output behavior of the replenishment strategy function. Specifically, the strategy parameter update can determine the update direction and magnitude based on indicator information, or it can select the adjustment amount or adjustment ratio of the strategy parameters based on indicator information. To make the strategy parameter update process more in line with the stability requirements of business operations, constraints on the magnitude of strategy changes can be introduced into the update target.

[0081] In some embodiments, updating strategy parameters based on indicator information includes applying threshold limits or pruning limits to the difference measure of the replenishment strategy function before and after the update in the update objective. The difference measure is used to characterize the degree of change between the replenishment strategy function before and after the update. Specifically, the threshold limit or pruning limit can limit the difference measure to not exceed a preset range, or it can truncate the difference measure to the boundary of a preset range when the difference measure exceeds the preset range. The threshold limit or pruning limit is determined by the update constraint parameter, which can be understood as the boundary value of the threshold limit or pruning limit changing with the update constraint parameter. For example, when the update constraint parameter increases, the allowable range corresponding to the threshold limit or pruning limit increases; when the update constraint parameter decreases, the allowable range corresponding to the threshold limit or pruning limit decreases.

[0082] In some embodiments, the difference metric includes the ratio of the probability or probability density of the same replenishment decision before and after the update. Specifically, the replenishment strategy function can output probability or probability density distribution parameters of the replenishment decision. Before the update, the replenishment strategy function can provide a first probability or a first probability density for the same replenishment decision under the same state information. After the update, the replenishment strategy function can provide a second probability or a second probability density for the same replenishment decision under the same state information. The difference metric can be determined based on the ratio of the second probability to the first probability or based on the ratio of the second probability density to the first probability density. Using the ratio of probability or probability density as the difference metric allows for the measurement of the degree of strategy change brought about by the update of the replenishment strategy function while keeping the replenishment decision consistent.

[0083] In some embodiments, the threshold limit or pruning limit is an asymmetric limit. An asymmetric limit can be understood as applying different limit boundaries when the difference metric falls within different ranges. Specifically, an asymmetric limit may apply a first limit range when the difference metric is greater than a certain reference value, and a second limit range when the difference metric is less than that reference value, with different boundary values ​​for the first and second limit ranges. For example, an asymmetric limit can allow different allowable ranges for the direction of increase and decrease in the difference metric, so as to apply update limits of different strengths under different business risks or different policy change directions.

[0084] In some embodiments, the specific value of the threshold limit or pruning limit can be determined by the update constraint parameter combined with the deviation of the inventory information from a preset inventory threshold. Specifically, when the inventory information is lower than the preset inventory threshold and the risk level is high, the update constraint parameter can take a relatively large value, thereby making the allowable range of the threshold limit or pruning limit relatively large, so that the replenishment strategy function can adjust more quickly; when the inventory information is not lower than the preset inventory threshold and the risk level is low, the update constraint parameter can take a relatively small value, thereby making the allowable range of the threshold limit or pruning limit relatively small, so that the replenishment strategy function maintains a more stable change.

[0085] To address this, by imposing a threshold or pruning constraint determined by the update constraint parameter on the difference measure of the replenishment strategy function before and after the update in the update objective, and optionally adopting a difference measure based on probability or probability density ratio and an asymmetric constraint method, the update of the strategy parameter can be subject to adaptive constraints of the update constraint parameter while referencing indicator information. This makes the adjustment of the replenishment strategy function more in line with the update needs under different risk states and helps to reduce the volatility of the strategy update process.

[0086] In some embodiments, determining the replenishment decision quantity based on the state information and the replenishment strategy function includes: generating a time-series feature representation based on historical sales sequences and / or demand forecast sequences; and fusing the time-series feature representation with non-time-series feature representations corresponding to inventory information and / or orders in transit information to determine the replenishment decision quantity.

[0087] Wherein, the replenishment decision quantity is a continuous value or a discrete value, and when the replenishment decision quantity is a continuous value, the replenishment decision quantity is determined to be a replenishment decision quantity that satisfies a preset constraint through a preset mapping relationship.

[0088] In some embodiments, determining the replenishment decision quantity based on state information through a replenishment strategy function may include jointly representing and fusing time-series and non-time-series information. Specifically, the replenishment strategy function may first generate a time-series feature representation based on at least one of the historical sales series and demand forecast series, and then fuse the time-series feature representation with the non-time-series feature representations corresponding to inventory information and in-transit order information to determine the replenishment decision quantity.

[0089] In some embodiments, historical sales sequences are used to characterize the sales changes of a target object over multiple historical time steps. Specifically, historical sales sequences may include sales data statistically analyzed daily, weekly, or according to other preset periods, and may also include derived sequences such as moving averages, growth rates, or volatility of sales. Demand forecast sequences are used to characterize the predicted demand values ​​for multiple future time steps. Specifically, demand forecast sequences may be generated from historical sales sequences and external factor information, or may be directly provided by the business system, and may include predicted demand values ​​for multiple future time steps.

[0090] In some embodiments, generating a time-series feature representation is used to extract features that reflect trends, seasonality, and short-term fluctuations from historical sales sequences and demand forecast sequences. Specifically, generating a time-series feature representation may include weighted aggregation of historical sales sequences, weighted aggregation of demand forecast sequences, or joint weighted aggregation of historical sales sequences and demand forecast sequences. For example, a replenishment strategy function may assign higher weights to historical sales more recent than the current time step, or to demand forecasts more recent than the current time step. Generating a time-series feature representation may also involve generating features from the historical sales sequences and demand forecast sequences separately and then combining them to form a time-series feature representation.

[0091] In some embodiments, the non-time-series feature representations corresponding to inventory information and in-transit order information are used to characterize the inventory and arrival status directly related to the current replenishment cycle. Specifically, the non-time-series feature representations may include one or more features such as existing inventory quantity, available inventory quantity, safety stock quantity, in-transit quantity, estimated arrival time, and arrival batch, or may include derived features obtained by combining the above features, such as inventory coverage days or expected stockout window.

[0092] In some embodiments, temporal feature representations and non-temporal feature representations are fused to form a comprehensive feature representation for determining replenishment decision quantities. Specifically, the fusion may include methods such as concatenation fusion, weighted fusion, or gated fusion, or it may include normalizing the temporal and non-temporal feature representations separately before fusion. The fused comprehensive feature representation can be used as input to a replenishment strategy function to output replenishment decision quantities, or to output intermediate variables used to determine replenishment decision quantities.

[0093] In some embodiments, the replenishment decision quantity is a continuous or discrete value. When the replenishment decision quantity is a continuous value, it is determined to satisfy preset constraints through a preset mapping relationship. The specific content of the preset mapping relationship and preset constraints can be found in the description of the above embodiments. For example, the preset mapping relationship can map the continuous variable output by the replenishment strategy function to a replenishment decision quantity that satisfies minimum order quantity constraints, maximum order quantity constraints, full-case multiple constraints, or budget constraints. When the replenishment decision quantity is a discrete value, it can be selected from multiple candidate replenishment quantity levels and further verified whether it satisfies preset constraints.

[0094] This application embodiment generates a time-series feature representation based on historical sales sequences and demand forecast sequences, and integrates the time-series feature representation with the non-time-series feature representation corresponding to inventory information and in-transit order information to determine the replenishment decision quantity. This allows the replenishment strategy function to consider both historical demand evolution and current inventory arrival status when generating the replenishment decision quantity, thereby making the determination of the replenishment decision quantity more in line with the supply and demand changes and fulfillment rhythm under multi-cycle operation scenarios, and helping to maintain the executability and consistency of replenishment instruction output.

[0095] Please refer to Figure 2, which illustrates the overall architecture for determining and adjusting a replenishment strategy. This architecture consists of three parts: a data preprocessing layer, an interactive closed-loop layer, and a strategy optimization layer. State vectors, actions, rewards, and trajectory data flow between these three parts to form a closed-loop process oriented towards multiple replenishment cycles.

[0096] In the data preprocessing layer, historical sales / inventory data serves as the basic input. This data enters both the demand forecasting model to generate future demand forecast sequences and the feature engineering module to extract replenishment-related features. The feature engineering module integrates demand forecast sequences, current inventory levels, lead time distribution information corresponding to historical order fulfillment, and external factors such as inventory cost parameters to further construct a high-dimensional state vector. This state vector can serve as the initial input for the supply chain simulation environment, forming the initial state information s_0, or it can form the current state information s_t in subsequent replenishment cycles.

[0097] In the interactive closed-loop layer, the supply chain simulation environment serves as the simulation platform for the replenishment process, receiving replenishment actions output by the agent and simulating real business processes. After obtaining the current state information s_t, the agent outputs action a_t. Action a_t can correspond to a replenishment decision quantity or a parameterized expression of the replenishment decision quantity, and is used to drive the supply chain simulation environment to simulate processes such as order consumption, logistics delays, and inventory deductions. Under the action a_t, the supply chain simulation environment outputs the next state information s_{t+1} and simultaneously calculates the reward r_t. The reward r_t can be comprehensively determined based on factors such as current holding costs, stockout penalties, ordering costs, and changes in replenishment decision quantities, reflecting the operational results and indicator information of the replenishment action within the current replenishment cycle.

[0098] In the policy optimization layer, the experience replay buffer stores trajectory data generated by the interaction between the agent and the supply chain simulation environment. This trajectory data can include sampling results such as (s_t, a_t, r_t, s_{t+1}). The parameter update module can sample data from the experience replay buffer to obtain a batch of sampled data, calculate the update target based on the batch, and perform parameter updates. Parameter updates can be implemented using gradient descent methods; for example, some implementations may use a near-policy update approach. The parameter update module outputs the updated parameters and writes them back to the agent, updating the policy network parameters and value network parameters within the agent. After the update, the agent continues to output new actions a_t based on the updated policy network parameters under subsequent state information, thus forming a continuously iterative interactive closed loop with the supply chain simulation environment.

[0099] Through the three-layer structure shown in Figure 2, historical sales / inventory data are used to obtain state vectors through demand forecasting and feature engineering. The agent outputs replenishment actions based on the state vectors. The supply chain simulation environment returns the next state information and reward. Trajectory data enters the experience replay buffer and is sampled for parameter updates. The updated strategy network parameters and value network parameters are fed back to the agent, thereby completing the process of determining and adjusting the replenishment strategy in multiple replenishment cycles.

[0100] The following describes the method mentioned in the embodiments of this application in some specific application scenarios, in conjunction with the above embodiments. Specifically, it may include steps 1-6: In some embodiments, the replenishment strategy determination and adjustment method is geared towards daily-level replenishment decision execution for a single product and a single warehouse, and can also be extended to parallel execution for multiple products or multiple warehouses. In each replenishment cycle, the replenishment strategy determination and adjustment method can output a replenishment decision quantity based on status information and form a replenishment instruction. Simultaneously, it updates the strategy parameters of the replenishment strategy function based on the indicator information related to the replenishment instruction. The update of the strategy parameters is limited by update constraint parameters, which are adaptively determined as the status information changes.

[0101] In some embodiments, step 1 may be used to construct a multimodal temporal state space. Data sources include streaming data from enterprise resource planning (ERP) systems, order systems, and inventory systems, as well as daily snapshot data. First, heterogeneous data cleaning and temporal alignment are performed to align data from different sources on the same daily granularity timeline. For example, non-daily granularity event data is resampled and aggregated into a daily granularity sequence, and missing values ​​are padded forward or zero-filled to ensure daily states have usable input. Then, feature engineering is performed to obtain a state vector, which can be constructed across all dimensions from past to present to future, as exemplarily represented as: .

[0102] The physical state features can include current inventory quantity, backlog and outstanding orders, and in-transit order information. To represent future delivery schedules, in-transit order information can be constructed as an in-transit vector of length K: V t =[v t,1 ,v t,2 ,…,v t,K ]. Where, v t,i This can represent the quantity expected to arrive on the i-th day. Historical lag features can use the actual sales sequence of the past H days as a sliding window input to capture short-term fluctuations and unusual spikes. Probabilistic prediction features can use quantiles or variances of the demand forecast sequence to express future demand uncertainty; for example, inputting P10 / P50 / P90 quantiles facilitates more robust replenishment when demand fluctuates significantly. Contextual features can include embedded encodings of day of the week, month, whether it's a holiday, and discrete attributes such as product category and warehouse level to differentiate replenishment preferences under different scenarios.

[0103] In some embodiments, step 2 can be used to build a high-fidelity supply chain simulator. The simulator runs daily, internally maintaining inventory status transfer logic, arrival logic, and business constraint logic. Each daily cycle can run in the order of arrival and warehousing—demand fulfillment—replenishment execution. In the arrival and warehousing stage, the daily arrival quantity is added to inventory based on the in-transit vector and order arrival time. In the demand fulfillment stage, the daily demand quantity is generated and deducted from the outbound quantity; when inventory is insufficient, the backlog is handled according to the business model. For example, using the existing inventory quantity I... t Daily arrival volume A t On the same day, the sales volume of S was met. t Constructing inventory updates: I t+1 =I t +A t S t Quantity of goods and invoices in short supply B t Updates can be made in the "outstanding goods to outstanding orders" mode, for example: B t+1 =max(0, B t +Dt (I t +A t D) t This indicates the demand for the day.

[0104] The replenishment execution phase receives replenishment decisions and generates replenishment instructions, while simultaneously applying a reality constraint layer to ensure feasibility. For example, a full-case multiple constraint can constrain the original replenishment decision quantity 'a'. t Mapped to integer multiples of the wrapper unit p: The minimum order quantity constraint q can be further modified. t For example, when q t When the order quantity is below the minimum order quantity, adjust to 0 or the minimum order quantity. Lead time information can be implemented through randomized lead time. For example, the lead time L is sampled independently for each replenishment order; the lead time can be derived from a historical experience distribution or a parameterized distribution, such as: L Γ(k,θ). The sampled lead time determines the delivery date corresponding to the replenishment order, thereby updating the in-transit vector V. t It also reflects long-tail latency and order disorder. The simulator outputs the next state information s in each daily cycle. t+1 With reward r t， Reward r t As a form of indicator information.

[0105] In some embodiments, the indicator information represents at least two of the following: revenue, inventory holding costs, stockout penalty costs, and ordering costs, and optionally represents a change in replenishment decision quantity to suppress replenishment oscillations. For example, the reward r t It can be constructed using a multi-objective weighted approach: The coefficients can be provided or configured from cost parameter information to balance holding costs, stockout risk, and ordering costs.

[0106] In some embodiments, step 3 is used for the cold start and initialization of the replenishment strategy function. To reduce the high cost risk caused by initial random exploration, pre-training is performed based on offline historical replenishment records before interactive updates. Offline data can come from real replenishment decision records of the enterprise resource planning system over a period of time, forming sample pairs. Pre-training can minimize the error between predicted replenishment decisions and historical replenishment decisions, for example, by using the mean squared error form: .

[0107] Meanwhile, the internal parameters of the replenishment strategy function can be orthogonally initialized to improve gradient propagation stability, and the temporal input can be encoded with positional information to explicitly provide temporal location information, making it easier for the model to learn the decay rule of "higher weight for near-end information". The positional encoding can take a sine or cosine form, for example: Where pos represents the time step position and d represents the encoding dimension.

[0108] In some embodiments, step 4 can be a tiered learning training process. To reduce the convergence difficulties caused by sparse rewards in complex supply chain environments, the training progresses progressively in difficulty. The first stage uses deterministic environmental configurations, such as fixed lead times and stable demand, to enable the replenishment strategy function to quickly learn the basic rules for maintaining inventory levels. The second stage introduces stochastic lead times and non-stationary demand, enabling the replenishment strategy function to learn adaptive strategies under demand fluctuations and supply uncertainties. Furthermore, the reward structure can emphasize penalties for understocking to encourage a more proactive replenishment tendency in low-inventory, high-risk areas. The third stage introduces domain randomization and stress testing to significantly perturb environmental parameters, such as simulating demand surges and supply disruptions, enabling the replenishment strategy function to develop robust recovery capabilities under extreme boundary conditions.

[0109] In some embodiments, step 5 may be an online policy iteration based on multi-objective rewards. The agent runs in the simulator using the current replenishment policy function, collects trajectory data, and writes it to the trajectory buffer. The trajectory data may contain (s t ,a t ,r t ,s t+1 This includes information such as action probability. The parameter update module samples data from the trajectory buffer to obtain a batch of sampled data and calculates an advantage estimate to reduce update variance. For example, temporal difference residuals are used: δ t =r t +γV(s t+1 ) V(s t And use generalized dominance estimation: Where V(.) represents the value estimation function, γ represents the discount factor, and λ represents the coefficients of the balance deviation and variance.

[0110] When updating strategy parameters, a threshold or pruning restriction is applied to the difference measure of the replenishment strategy function before and after the update. The difference measure can be the ratio of the probability or probability density of the same replenishment decision before and after the update. And then perform cropping in the update target, for example: .

[0111] in and Asymmetric boundary conditions can be used. Updating constraint parameters can be used to determine... and The range of values ​​is used to achieve adaptive risk updates. The level of risk can be determined based on the deviation of inventory information from a preset inventory threshold, for example: When the risk level is greater than the risk threshold, the updated constraint parameters are increased relative to the baseline constraint parameters, and the clipping boundary is widened; when the risk level is less than the risk threshold, the updated constraint parameters are decreased relative to the baseline constraint parameters, and the clipping boundary is narrowed. For example, the clipping boundary can be: And set accordingly. and This allows for faster strategy adjustments in low-inventory, high-risk areas, and more stable strategy evolution in areas with relatively healthy inventory. The value estimation function can be updated synchronously, for example, by minimizing the error between the return target and the value estimate. Optionally, an entropy regularization term can be introduced to maintain exploration, or a replenishment decision quantity change term can be introduced to suppress replenishment oscillations and keep replenishment instructions smoother.

[0112] In some embodiments, step 6 is used for production environment deployment and virtual-to-physical migration. After training, the replenishment strategy function first runs in shadow mode. In shadow mode, it outputs suggested replenishment decisions but does not directly send them to the actual procurement execution chain. The system records a comparison log between the suggested replenishment decisions and the actual replenishment decisions executed manually. Subsequently, difference measurement and confidence assessment are performed. For example, in multiple scenarios, the differences between the suggested replenishment decisions and the manual strategy in terms of simulated operating costs and simulated turnover days are statistically analyzed. After meeting the preset stability and revenue conditions, it enters a small-scale gray-scale phase. During the gray-scale phase, a small number of core products are selected as pilots to give the replenishment strategy function real write permissions. During the production data closed-loop phase, real trajectory data is continuously collected and used for simulator calibration, such as calibrating lead time information distribution parameters and demand noise parameters. Based on the calibrated simulator, the replenishment strategy function is continuously fine-tuned to reduce the difference between the simulation environment and the real environment and maintain the long-term effectiveness of the strategy.

[0113] This application also provides a computer storage medium, characterized in that the computer storage medium is used to store a program, wherein, when the program is running, it controls the device where the computer storage medium is located to execute the method described in any of the above-mentioned embodiments.

[0114] This application also provides a computer program product, characterized in that the computer program product is used to implement the method described in any of the above embodiments.

[0115] Referring to Figure 3, Figure 3 is a structural block diagram of an electronic device provided in an embodiment of this application.

[0116] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the control method of the DC-DC converter provided in the above embodiments.

[0117] The electronic device may include: a memory 110, a processor 120, and a communication interface 130. The memory 110, the processor 120, and the communication interface 130 are connected through internal connection paths.

[0118] The memory 110 is used to store computer programs, which in some implementations may include code for implementing the methods of the embodiments of this application.

[0119] The processor 120 executes the computer program stored in the memory 110 to control the communication interface 130 to receive input data and information, and output operation results and other data. In some implementations, when the solutions of the embodiments of this application are implemented by software or firmware, the computer program used to implement the solutions of the embodiments of this application can be stored in the processor 120 and executed by the processor 120.

[0120] The memory 110 may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM). It should be noted that the memory 110 described herein is intended to include, but is not limited to, any memory of these and other suitable types. As an example, the memory 110 includes random access memory (RAM), cache memory, and read-only memory (ROM). The memory 110 stores a computer program that can be executed by processor 120, causing processor 120 to implement the steps of any of the methods described above.

[0121] The processor 120 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 120 can be any conventional processor.

[0122] In implementation, each step of the above method can be completed by the integrated logic circuitry of the hardware in the processor 120 or by instructions in software form. The method disclosed in the embodiments of this application can be directly implemented by the hardware processor, or by a combination of hardware and software modules in the processor 120. The software modules can be located in mature storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in the memory 110, and the processor 120 reads the information in the memory 110 and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are not provided here.

[0123] In some implementations, in addition to the hardware units described above, electronic devices may also include software modules, such as operating systems, basic input / output systems (BIOS), and application software.

[0124] An operating system is used to manage one or more hardware and software resources of an electronic device; it is the kernel and foundation of the electronic device. The operating system handles fundamental tasks such as managing and configuring memory, determining the priority of system resource allocation and demand, controlling input and output devices, operating networks, and managing file systems. To facilitate user operation, most operating systems provide a user interface for interaction with the system.

[0125] The BIOS is used to perform hardware initialization during the power-on boot phase and to provide runtime services for the operating system and applications. In some implementations, the BIOS can also monitor and display processor temperature and execute temperature protection strategies.

[0126] Application software, also known as an application program, can be understood as software written for a specific user application purpose, and is one of the main categories of computer software. For example, application software can be a program used to achieve purposes such as power control and temperature management.

[0127] It is understood that the specific examples in this application are only intended to help those skilled in the art better understand the implementation of this application, and are not intended to limit the scope of protection of this application.

[0128] It is understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application.

[0129] It is understood that the various implementation methods described in this application can be implemented individually or in combination, and this application does not limit them.

[0130] Unless otherwise stated, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "one or more" as used in this application includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0131] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes and beneficial effects of the embodiments described above can be referred to the corresponding processes and beneficial effects in other embodiments, and will not be repeated here.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0134] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the technical solution in this application, depending on actual needs.

[0135] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0136] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, essentially, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for determining and adjusting a replenishment strategy, characterized in that, include: Obtain a replenishment strategy function for outputting replenishment decision quantities, the replenishment strategy function having adjustable strategy parameters; The system acquires status information, which includes at least one of inventory information and other information related to replenishment. Based on the status information, it determines a replenishment decision quantity through the replenishment strategy function and outputs a replenishment instruction corresponding to the replenishment decision quantity. It acquires indicator information related to the replenishment instruction and updates the strategy parameters based on the indicator information. The update of the strategy parameters is subject to an update constraint parameter, which is adaptively determined as the status information changes, such that when the risk level represented by the status information is greater than a risk threshold, the update constraint parameter increases relative to the baseline constraint parameter, and when the risk level is less than the risk threshold, the update constraint parameter decreases relative to the baseline constraint parameter.

2. The method according to claim 1, characterized in that, The other information related to replenishment includes at least one of the following: in-transit order information, demand forecast information, promotional information, or other external factor information; wherein the external factor information includes lead time information and / or cost parameter information.

3. The method according to claim 1, characterized in that, The status information includes at least two of the following: inventory information and other information related to replenishment.

4. The method according to claim 1, characterized in that, The indicator information represents at least two of the following: revenue, inventory holding costs, stockout penalty costs, and ordering costs; and / or, the indicator information also represents a replenishment decision quantity change item, which is related to the difference in replenishment decision quantities corresponding to two adjacent replenishment instructions.

5. The method according to claim 1, characterized in that, The replenishment decision quantity can be a continuous value or a discrete value; when the replenishment decision quantity is a continuous value, the replenishment decision quantity is determined by the continuous variable output by the replenishment strategy function through a preset mapping relationship to satisfy the preset constraints.

6. The method according to claim 1, characterized in that, The update constraint parameter is adaptively determined as the state information changes, including: determining the update constraint parameter based on the deviation of the inventory information from a preset inventory threshold; wherein, when the inventory information is lower than the preset inventory threshold, the update constraint parameter increases relative to the baseline constraint parameter, and the update constraint parameter is positively correlated with the deviation.

7. The method according to claim 1, characterized in that, The step of updating the strategy parameters based on the indicator information includes: applying a threshold limit or pruning limit to the difference measure of the replenishment strategy function before and after the update in the update target, wherein the threshold limit or pruning limit is determined by the update constraint parameter; wherein the difference measure includes the ratio of the probability or probability density of the same replenishment decision quantity before and after the update; and / or, the threshold limit or pruning limit is an asymmetric limit.

8. The method according to claim 1, characterized in that, The step of determining the replenishment decision quantity based on the state information and through the replenishment strategy function includes: generating a time-series feature representation based on historical sales sequences and / or demand forecast sequences; fusing the time-series feature representation with non-time-series feature representations corresponding to inventory information and / or orders in transit information to determine the replenishment decision quantity; wherein the replenishment decision quantity is a continuous value or a discrete value, and when the replenishment decision quantity is a continuous value, the replenishment decision quantity is determined to be a replenishment decision quantity that satisfies a preset constraint through a preset mapping relationship.

9. A computer storage medium, characterized in that, The computer storage medium is used to store a program, wherein, when the program is running, it controls the device where the computer storage medium is located to execute the method according to any one of claims 1-8.

10. A computer program product, characterized in that, The computer program product is used to implement the method as described in any one of claims 1-8.

Images