Store commodity dynamic price determination method, electronic equipment and computer readable storage medium

Abstract

The invention discloses a store commodity dynamic price determination method, electronic equipment and a computer readable storage medium, and relates to the technical field of data processing. The method comprises the following steps: receiving store internal data and external data transmitted by a target store network manager; cleaning and standardizing the multi-dimensional data to obtain a standardized feature vector; determining the weight coefficient of each feature component through a weight adaptive adjustment strategy based on a pricing target; carrying out fusion processing to obtain a comprehensive feature vector; calculating an adaptation degree Q value corresponding to each pricing adjustment action through a reinforcement learning model; and based on the Q value, determining a target pricing action from the pricing action space, generating a dynamic price adjustment instruction corresponding to the target pricing action, and sending the dynamic price adjustment instruction to the target store webmaster to execute price adjustment. According to the scheme, through a reinforcement learning pricing mechanism of multi-dimensional data fusion and weight adaptive adjustment, accurate dynamic adaptation of store commodity prices can be realized, and pricing flexibility and market response efficiency are improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to methods for determining dynamic prices of goods in stores, electronic devices, and computer-readable storage media. Background Technology

[0002] As the retail industry accelerates its digital transformation, in-store product pricing is gradually moving away from the traditional single cost-plus model and towards a dynamic adjustment model that combines market supply and demand, inventory levels, and other factors to adapt to immediate changes in consumer demand and intensifying market competition. Currently, most stores' dynamic pricing schemes rely on preset fixed rules or simple statistical models. For example, they may manually set inventory thresholds and corresponding price adjustment ratios, implementing a fixed price reduction when product inventory exceeds the preset threshold, or periodically collect competitor prices and adjust selling prices based on a fixed difference. These solutions require no complex technical adaptations in practical applications and have low deployment costs, thus they are widely adopted.

[0003] However, existing dynamic pricing schemes have significant technical shortcomings, making it difficult to meet the actual needs of stores. Firstly, their data processing capabilities are limited, only able to handle a few types of internal, fixed data such as inventory and historical sales. They cannot handle various types of dynamic data in real time, including real-time customer traffic data collected by store flow sensors, rainfall / temperature data returned from weather platform interfaces, and user comments on social media. This results in a lack of crucial technical input support for pricing. Secondly, their decision-making logic is inflexible, employing fixed weight settings and simple algorithms. This fails to adjust the emphasis of different data based on store data collection (such as fluctuations in customer flow sensor data) or external data changes (such as sudden weather changes), and it cannot handle the complex relationships between these different types of data. Ultimately, this leads to slow and inaccurate pricing results. Therefore, how to make dynamic pricing schemes compatible with various types of dynamic data, while allowing the decision-making logic to adapt to data changes in a timely and flexible manner, has become a pressing technical problem to be solved. Summary of the Invention

[0004] This application provides a method for determining the dynamic price of goods in a store, an electronic device, and a computer-readable storage medium, which can achieve accurate, dynamic, and timely adjustment of goods prices based on a reinforcement learning mechanism that integrates multi-dimensional data fusion and adaptive weight adjustment.

[0005] In a first aspect, embodiments of this application provide a method for determining the dynamic price of goods in a store, characterized in that it is applied to a server, and the method includes: The system receives multi-dimensional data transmitted from the network administrator of the target store. This multi-dimensional data includes internal and external data. Internal data includes real-time operational data collected by the target store's inventory management system, POS system, and customer flow sensors, as well as category and behavioral information of current online visitors obtained from the target store's online service channels. External data includes real-time competitor price data and weather data related to the target store, obtained by the target store's network administrator through a third-party data service API interface. Multi-dimensional data is cleaned and standardized to obtain a standardized feature vector. The standardized feature vector includes feature components of multiple dimensions, and each feature component corresponds to the standardization result of a class of data. Based on the pricing target corresponding to the target store, the weight coefficients of each feature component in the standardized feature vector are determined through a weight adaptive adjustment strategy. Each feature component in the standardized feature vector is weighted by its corresponding weight coefficient, and then all weighted feature components are fused to obtain a comprehensive feature vector for inputting into the reinforcement learning model. The comprehensive feature vector is input into the preset reinforcement learning model, and the model calculates and outputs the fitness Q value corresponding to each pricing adjustment action in the pricing action space. Based on the adaptation Q value corresponding to each pricing adjustment action, the target pricing action is determined from the pricing action space, a dynamic price adjustment instruction corresponding to the target pricing action is generated, and the dynamic price adjustment instruction is sent to the network administrator of the target store. The network administrator then synchronously updates the product pricing information of the target store's offline POS system, shelf price tag display terminal, and online service channels.

[0006] This solution integrates multi-dimensional data from both inside and outside the store. After cleaning and standardization, it combines a weighted adaptive adjustment strategy with a reinforcement learning model to achieve dynamic pricing. The multi-dimensional data covers key influencing factors such as operations, visitors, competitors, and weather, ensuring the comprehensiveness of the pricing basis. The weighted adaptive adjustment strategy can dynamically optimize the weights of feature components according to the pricing target, improving the fit between pricing and the target. The reinforcement learning model quantifies the fit of pricing actions through Q-values, enabling scientific decision-making. Instructions are uniformly synchronized through the network management system to the offline POS system, price tag terminals, and online channels, avoiding price inconsistencies. Ultimately, it achieves accurate, timely, and synchronized dynamic pricing of goods, taking into account both store operational goals and market response needs.

[0007] In conjunction with the first aspect, in the first possible implementation of the first aspect, the pricing action space is a set of standardized pricing adjustment actions pre-configured and stored on the server, including at least two preset price adjustment directions and corresponding adjustment ranges.

[0008] This solution uses a pre-defined set of standardized pricing actions to clarify the direction and magnitude of price adjustments, avoiding arbitrariness and uncertainty in pricing actions. The standardized set ensures that pricing adjustments comply with store operation rules (such as avoiding price adjustments beyond the permitted range), while reducing the decision complexity of the reinforcement learning model and improving the efficiency of pricing calculations. Multiple pre-defined adjustment actions can cover different market scenarios, making pricing strategies more timely and flexible, and adapting to the diverse operational needs of stores.

[0009] In conjunction with the first aspect, the second possible implementation of the first aspect, the method for determining the dynamic price of store merchandise, also includes: Obtain the pricing target for the target store; the pricing target includes any one or more combinations of the following: (1) A fixed pricing target that is pre-configured on the server and stored in the local database and matches the business type of the target store; the business type of the target store is one of the preset business types, which includes: fresh food stores, fast-moving consumer goods convenience stores, brand specialty stores, and maternal and infant product stores. (2) The server receives, in real time, the temporary pricing target entered by the store manager through the back-end management terminal via a communication link established with the network management of the target store; (3) The server automatically generates a dynamic pricing target after cross-analysis of at least two types of data from the multi-dimensional data of the target store.

[0010] This solution offers three types of pricing targets and their combinations: fixed, temporary, and dynamic, adaptable to different store operating scenarios. Fixed pricing targets are precisely matched with the business type, meeting the core needs of different business formats such as fresh food and fast-moving consumer goods. Temporary pricing targets allow for flexible intervention by management personnel, adapting to special scenarios such as promotions and emergencies. Dynamic pricing targets are generated based on cross-analysis of multi-dimensional data, enabling real-time response to market changes. The combination of the three types of targets can take into account both the store's long-term strategy and short-term needs, solving the problem of insufficient adaptability of a single pricing target, and making the pricing strategy more targeted and flexible.

[0011] In conjunction with the first aspect, in the third possible implementation of the first aspect, the matching rules between the fixed pricing objective and the business type include: If the target store is a fresh food store, then a fixed pricing target of "prioritizing inventory clearance + guaranteeing minimum gross profit ≥ first preset value" will be matched. "Prioritizing inventory clearance" means that the pricing will prioritize the efficiency of product inventory turnover. When the inventory turnover days are ≥ the preset turnover threshold, the price adjustment logic with the primary goal of reducing inventory will be triggered. "Guaranteing minimum gross profit ≥ first preset value" means that the actual gross profit of the product after the price adjustment will not be lower than the minimum gross profit threshold preset by the server for fresh food categories. If the target store is a fast-moving consumer goods convenience store, then a fixed pricing target of "high-frequency sales + gross profit balance ≥ second preset value" will be matched. "High-frequency sales" refers to focusing on essential goods with a weekly purchase frequency ≥ preset frequency threshold. Essential goods include at least one of the following: drinking water, snacks, daily necessities, toiletries, and convenience foods. Gross profit balance means ensuring that the weekly sales fluctuation of these goods is ≤ preset fluctuation threshold through pricing control, so as to avoid a sharp drop in sales due to excessively high pricing or insufficient inventory due to excessively low pricing. If the target store is a brand specialty store, then a fixed pricing target of "brand premium + member retention ≥ third preset value" will be matched. Here, "brand premium" means that the product price is higher than that of similar non-brand products by a percentage greater than or equal to the preset premium percentage. "Member retention ≥ third preset value" means that after the price adjustment takes effect, the server will calculate the member repurchase rate within the preset observation period, and the member repurchase rate will not be lower than the minimum member repurchase threshold preset by the server. The preset observation period is a time period that is pre-configured by the server and stored in the local database, and it is adapted to the consumption cycle of brand products. If the target store is a maternity and baby product store, a fixed pricing target of "safety endorsement + essential supply guarantee + gross profit ≥ fourth preset value" will be matched. "Safety endorsement" requires the price to be linked to product safety certification information, which includes at least one of the following: organic certification, infant product safety test report, raw material traceability certification, and production standard certification. Price adjustments will only be applied to products with safety endorsement. "Essential supply guarantee" means that when adjusting the price of essential maternity and baby products, the inventory depth must be simultaneously verified to be ≥ the preset supply guarantee inventory. Essential maternity and baby products include at least one of the following: milk powder, diapers, baby wipes, complementary food, and baby care products. "Gross profit ≥ fourth preset value" means that after the price adjustment, the gross profit of essential products will not be lower than the server's preset gross profit threshold for essential maternity and baby products. Among them, the first preset value, the second preset value, the third preset value, the fourth preset value, as well as the preset turnover threshold, the preset frequency threshold, the preset fluctuation threshold, the preset premium ratio, the preset supply guarantee inventory, and the preset observation period are all personalized thresholds that are pre-configured by the server and stored in the local database. They are specifically calculated and determined based on the industry average gross profit level, store operating cost structure, and preset operating strategies for the corresponding business type.

[0012] This solution sets differentiated fixed pricing targets and quantifiable rules for stores of different business types, achieving precise adaptation of pricing objectives: fresh food stores focus on "clearing inventory + guaranteeing gross profit" to balance loss control and profit needs; fast-moving consumer goods convenience stores emphasize "high-frequency sales + balanced gross profit" to ensure stable sales and reasonable profits for essential goods; brand specialty stores emphasize "brand premium + member retention" to maintain brand value and customer loyalty; and maternity and baby product stores highlight "safety endorsement + essential supply guarantee + gross profit guarantee" to meet the core demands of the maternity and baby category for safety and supply stability. The setting of personalized thresholds makes the pricing rules more in line with the actual operating scenarios of stores, avoiding a one-size-fits-all approach and improving the practicality and operability of fixed pricing targets.

[0013] In conjunction with the first aspect, in the fourth possible implementation of the first aspect, the server automatically generates a dynamic pricing target based on cross-analysis of at least two types of data from the multi-dimensional data of the target store, including at least one of the following cross-analysis scenarios: Cross-analyze inventory and sales data. When the inventory turnover days of a product are greater than or equal to the first cycle threshold and the sales volume decreases by more than or equal to the first proportion threshold within the preset number of days, a dynamic pricing target for "deep inventory clearance" is generated. Cross-analyze customer flow data and competitor price data. When the real-time customer flow growth rate is greater than or equal to the second proportion threshold and the price of the same category of competitor products is less than or equal to the current selling price of the target store multiplied by the first coefficient threshold, a dynamic pricing target for "seizing market share" is generated. Cross-analyze online visitor consumption data and product gross profit data. When the average order amount of online visitors within a preset statistical period is greater than or equal to a preset average order amount threshold, and the current gross profit of the product is less than or equal to the fifth preset value, a dynamic pricing target of "slight price increase + value-added services" is generated. The average order amount is calculated as the total cumulative order amount of online visitors within the preset statistical period divided by the total cumulative number of orders. Cross-analyze weather data and offline customer flow data. When the weather reaches a preset severe level and the offline customer flow decreases by more than or equal to the third proportion threshold, generate a dynamic pricing target of "online exclusive discount + offline discount". Among them, the first cycle threshold, the first proportion threshold, the second proportion threshold, the third proportion threshold, the first coefficient threshold, the preset number of days, the preset average order consumption amount threshold, the fifth preset value, and the preset severity level are all personalized thresholds that are pre-configured by the server and stored in the local database.

[0014] This solution generates dynamic pricing targets through multi-dimensional data cross-analysis, enabling precise responses to market changes: cross-analysis of inventory and sales data can promptly identify slow-moving risks and trigger deep inventory clearance strategies; cross-analysis of customer traffic and competitor pricing data can seize peak customer traffic opportunities and gain market share through pricing; cross-analysis of online consumption and gross profit data can optimize profitability while ensuring customer acceptance; cross-analysis of weather and offline customer traffic data can mitigate the impact of severe weather on store revenue; and the setting of quantitative thresholds makes the generation logic of dynamic pricing targets clearer and the triggering conditions more explicit, improving the real-time and targeted nature of pricing strategies and helping stores quickly adapt to complex and ever-changing market environments.

[0015] In conjunction with the first aspect, in the fifth possible implementation of the first aspect, when there are multiple combinations of pricing objectives, the method further includes: determining whether there is a conflict among the multiple combinations of pricing objectives, where the conflict includes: the existence of contradictory price adjustment constraints; If a conflict exists, the pricing objectives are determined from highest to lowest priority, with the constraints of higher-priority pricing objectives covering the conflict constraints of lower-priority pricing objectives. The pricing objective priorities are: fixed pricing objective priority ≥ temporary pricing objective priority ≥ dynamic pricing objective priority. If there is no conflict, the constraints of all pricing objectives are merged to generate a comprehensive pricing objective as the pricing objective for the target store. The merging method includes taking the intersection of the constraints of all pricing objectives.

[0016] This solution addresses the contradictions in a combination of multiple pricing objectives through conflict assessment and priority rules, ensuring the consistency of pricing strategies: priority settings ensure that fixed pricing objectives (core store strategies) are implemented first, temporary pricing objectives (special needs) are flexibly supplemented, and dynamic pricing objectives (market response) adapt to changes; the overlap and integration of coverage rules in case of conflict and constraints in case of no conflict avoids confusion in price adjustment logic and integrates the core demands of multiple objectives, so that the pricing strategy not only conforms to the long-term operation direction of the store, but also takes into account short-term needs and market changes, improving the rationality and feasibility of pricing decisions in multiple scenarios.

[0017] In conjunction with the first aspect, in the sixth possible implementation of the first aspect, the method further includes: The server calculates the degree to which pricing targets have been achieved in real time; Identify the characteristic components corresponding to the pricing objective; If the achievement rate of the pricing target is less than the achievement rate threshold, the weight coefficients of the feature components corresponding to the pricing target will be adjusted in the direction of improving the achievement rate of the pricing target; If the achievement rate of the pricing target is greater than or equal to the achievement rate threshold, the initial weight coefficients of the feature components corresponding to the pricing target remain unchanged.

[0018] This solution dynamically adjusts the weights of feature components based on the achievement of pricing targets, achieving a closed-loop optimization with adaptive weights: real-time calculation of achievement can promptly detect deviations between pricing strategies and targets; by adjusting the weights of corresponding feature components (such as increasing the weight of inventory data when inventory clearance targets are not achieved), the pricing model becomes more focused on target achievement, improving the targeting of pricing strategies; weight maintenance rules can maintain pricing stability when targets are achieved, avoiding over-adjustment; the entire mechanism makes weight adjustments more purposeful and flexible, further improving the fit between dynamic pricing and store targets.

[0019] In conjunction with the first aspect, in the seventh possible implementation of the first aspect, the current online visitor category information includes any one of the following categories: new visitors, returning visitors, member visitors, and high-value visitors; new visitors include visitors with no historical consumption records; returning visitors include visitors with at least a preset consumption record within the most recent first preset time interval; member visitors include visitors who have activated store membership services; and high-value visitors include visitors whose cumulative consumption amount within the most recent second preset time interval is ≥ a value threshold.

[0020] The current online visitor behavior information includes one or more of the following: product categories viewed, product page viewing time, number of product categories viewed, number of times added to cart, number of products favorited, order conversion rate, historical repurchase frequency, and page dwell path.

[0021] This solution clearly defines the specific dimensions of online visitor categories and behavioral information, providing precise user-level basis for pricing decisions: visitor category segmentation can distinguish different value user groups (such as high-value visitors and members), making pricing strategies more aligned with users' willingness to pay; behavioral information covers the entire link of data such as browsing, adding to cart, and conversion, which can accurately identify user needs and consumption preferences; detailed information dimensions make the user-related components in the standardized feature vector more representative, improving the accuracy of weight calculation and reinforcement learning model inference, ultimately achieving personalized pricing based on user profiles, improving user acceptance and consumption conversion rates.

[0022] Secondly, embodiments of this application provide an electronic device, including: a processor and a memory; The processor is connected to the memory, which stores a computer program. The processor executes the computer program stored in the memory to cause the electronic device to perform the steps of the method for determining the dynamic price of store merchandise as described in the first aspect or any possible implementation of the first aspect.

[0023] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs a method for determining the dynamic price of store merchandise as described in the first aspect or any possible implementation thereof.

[0024] It is understood that the technical effects achieved by the electronic device described in the second aspect and the computer-readable storage medium described in the third aspect are similar to the technical effects achieved by the corresponding technical means in the matching method of the fields in the table described in the first aspect, and will not be repeated here. Attached Figure Description

[0025] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments 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.

[0026] Figure 1 This is a flowchart illustrating a method for determining the dynamic price of goods in a store, as described in one embodiment of this application. Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0028] It should be understood that the term "multiple" in this invention refers to two or more. In the description of this invention, unless otherwise stated, " / " indicates "or," for example, A / B can mean A or B; "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist, for example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, to facilitate a clear description of the technical solutions of this application, terms such as "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply differences.

[0029] In traditional brick-and-mortar store pricing models, businesses often use fixed prices or manual price adjustments based on experience. For example, fresh food stores rely on employees to "lower prices based on intuition" according to inventory, while fast-moving consumer goods convenience stores maintain a uniform selling price over a long period. This model has significant limitations: First, manual price adjustments cannot simultaneously consider multiple dimensions of information such as inventory, customer traffic, and competitor prices. For instance, a fresh food store employee might only look at inventory price reductions, ignoring the market opportunity of price increases for similar products from competitors that day. Second, pricing lacks dynamic adaptability. For example, when offline customer traffic drops sharply on rainy days, stores struggle to adjust online prices in a timely manner to attract consumers.

[0030] These problems lead to stores facing either a contradiction of "inventory backlog + profit loss" or missing market opportunities due to rigid pricing, making it difficult to accurately match business objectives and respond quickly to market changes. Therefore, how to achieve timely and dynamic pricing based on multi-dimensional data and adapted to business objectives has become an urgent technical problem to be solved.

[0031] The following is combined Figure 1 This application will now describe an embodiment of a method for determining the dynamic price of goods in a store, which is applied to a server. The method includes steps S101 to S106.

[0032] S101. Receive multi-dimensional data transmitted by the network management system of the target store. The multi-dimensional data includes: internal store data and external data.

[0033] The internal data of the store includes: real-time operational data collected by the target store's inventory management system, POS system, and customer flow sensors, as well as the category and behavior information of current online visitors obtained from the target store's online service channels; the external data includes: real-time competitor price data and weather data related to the target store obtained by the target store's network administrator through a third-party data service application programming interface (API).

[0034] In some possible implementations, online visitor category information includes new visitors, returning visitors, member visitors, and high-value visitors. New visitors are those with no historical purchase history; returning visitors are those with a pre-defined purchase history within the most recent first pre-defined time interval; member visitors are those who have activated store membership services; and high-value visitors are those whose cumulative purchase amount within the most recent second pre-defined time interval is greater than or equal to a value threshold. Online visitor behavior information includes browsed product categories, page browsing time, number of times added to cart, number of favorites, and order conversion rate. This setting can accurately capture the consumption preferences of different users, providing more granular user data support for pricing.

[0035] In some possible implementations, data transmission can be accomplished through various interfaces such as APIs, Software Development Kits (SDKs), and Message Queues (MQ). The network administrator can coordinate the update frequency of each data source, such as synchronizing competitor prices every 15 minutes or weather data every hour, while employing encrypted transmission methods to ensure data security. This broadens data access channels and balances data real-time performance with security.

[0036] For example, when a fast-moving consumer goods convenience store connects to external data, it pulls the prices of bottled beverages from three nearby supermarkets every 15 minutes through the "Competitor Price API". Developers use the "Store Data Acquisition SDK" to quickly integrate real-time transaction data from the POS system and pedestrian flow data from customer flow sensors into the transmission link. When the pedestrian flow data surges during the morning peak (9:00-10:00), more than 1,000 pedestrian flow records are temporarily stored through "MQ" to avoid data congestion and loss. The data is released in an orderly manner after the system's processing capacity is idle, ensuring stable data transmission.

[0037] The multi-dimensional data here forms the "information foundation" for pricing: for example, internal data for a fresh food store might include "200 catties of vegetables in stock (inventory management system data), 50 customers visited the store within one hour (customer flow sensor data), and 30 new online visitors viewed the vegetables page (online service channel data)." External data might include "competitor's vegetables are priced at 2.5 yuan / catties (obtained from a third-party API), and there was moderate rain that day (weather data)." This data covers key dimensions such as store operations, users, and the market environment, avoiding the problem of "single information" in traditional pricing.

[0038] From the perspective of data transmission logic, having the network administrator collect and transmit multi-dimensional data in a unified manner can ensure the centralization and authority of the data source: the network administrator can perform preliminary screening of data from various systems, eliminating obviously abnormal and invalid data (such as a sudden jump of 1,000 people data from a passenger flow sensor), while coordinating the data update frequency of online service channels and third-party APIs (such as synchronizing competitor prices every 15 minutes and synchronizing weather data every hour), ensuring the real-time nature of the data and providing "fresh" information input for subsequent dynamic pricing.

[0039] S102. Clean and standardize the multi-dimensional data to obtain a standardized feature vector. The standardized feature vector includes feature components of multiple dimensions, and each feature component corresponds to the standardization result of a class of data.

[0040] A standardized feature vector consists of multiple feature components, each corresponding to a standardized result for a class of data. Data cleaning removes invalid information (such as incorrect inventory figures), while standardization unifies the scale of data in different formats: for example, "200 catties of inventory" can be transformed into "3 days of inventory turnover," and "competitor's price of 2.5 yuan" can be transformed into "10% lower than our current price." The resulting feature vector resembles "[3 days of inventory turnover, 50 customers, 10% lower competitor price, rainy day]," allowing subsequent calculations to be based on a unified standard, which helps improve pricing accuracy.

[0041] In some possible implementations, data processing can utilize streaming platforms such as Flink and Kafka Streams, supporting real-time processing as data is generated and enabling cleaning and standardization to be completed in milliseconds. This can significantly improve data processing efficiency and meet the high timeliness requirements of real-time pricing.

[0042] Furthermore, data cleaning also addresses data redundancy and format conflicts: for example, when there are discrepancies between the sales records of the same product in the POS system and the inventory system, the actual transaction data from the POS system will prevail, and the deviation in the inventory system will be marked to remind subsequent verification; for text-based data (such as weather data like "moderate rain" or "heavy rain"), it can be converted into numerical codes (such as 1 and 2) to facilitate inclusion in feature vectors for calculation. Standardized feature vectors not only reduce the interference of outliers on the pricing model but also allow the model to more efficiently capture the correlations between data across different dimensions—for example, clearly identifying the potential link between "rainy days" and "increased online visitor volume," thereby outputting pricing recommendations that are more aligned with the actual scenario.

[0043] S103. Based on the pricing target corresponding to the target store, determine the weight coefficients of each feature component in the standardized feature vector through a weight adaptive adjustment strategy.

[0044] Pricing objectives determine the "importance" of each data point: for example, if the goal of a fresh food store is "prioritizing inventory clearance," then the weight of "inventory turnover days" will be increased (e.g., set to 0.4), while the weight of "competitor prices" will be decreased (e.g., set to 0.2); if the goal is "profit priority," then the weight of "gross profit data" will be increased. This adaptive adjustment makes pricing more focused on business objectives, avoiding a "one-size-fits-all" weight allocation.

[0045] Adaptive weight adjustment can be combined with optimization algorithms such as gradient descent or genetic algorithms. For example, using the achievement of pricing goals as the loss function, the contribution of each feature component to goal achievement is calculated iteratively, and the weight coefficients are dynamically adjusted. If the "inventory turnover days" decreases by one day and the achievement of inventory clearance goals increases by 15%, then the weight of that feature will be finely adjusted in the direction of improving goal achievement. At the same time, weight adjustment boundaries can be preset for different industry characteristics. For example, the weight of "competitor prices" for fast-moving consumer goods convenience stores should not be lower than 0.15, to avoid overemphasizing a certain feature and causing pricing to deviate from market reality, ensuring that weight adjustment is within a reasonable range while taking into account both goal orientation and market adaptability.

[0046] In some possible implementations, a step of obtaining the pricing target corresponding to the target store may be included before step S103. The pricing target includes any one or more combinations of the following: (1) A fixed pricing target that is pre-configured on the server and stored in the local database and matches the business type of the target store; the business type of the target store is one of the preset business types, which includes: fresh food stores, fast-moving consumer goods convenience stores, brand specialty stores, and maternal and infant product stores. (2) The server receives, in real time, the temporary pricing target entered by the store manager through the back-end management terminal via a communication link established with the network management of the target store; (3) The server automatically generates a dynamic pricing target after cross-analysis of at least two types of data from the multi-dimensional data of the target store.

[0047] For example, a brand's fixed pricing target for a specialty store is "brand premium + member retention ≥ 80%" (pre-configured to match high-end business formats); before Double 11, the management personnel enter "temporary pricing target: sales increase of 30%" through the backend; on the day of the event, after the system analyzes "member add-to-cart rate decrease of 15% + competitor discount increase of 20%", it automatically generates "dynamic pricing target: member exclusive discount increase of 5%". The three factors are combined to form the final pricing target, which both protects the brand and boosts sales.

[0048] In some possible implementations, the server pre-configures and stores in a local database a fixed pricing target that matches the operating type of the target store. This can be based on the matching rules between the fixed pricing target and the operating type, including: if the operating type of the target store is a fresh food store, then a fixed pricing target of "prioritizing inventory clearance + guaranteeing minimum gross profit ≥ first preset value" is matched; where "prioritizing inventory clearance" means that the pricing prioritizes ensuring the efficiency of product inventory turnover. When the inventory turnover days are ≥ a preset turnover threshold, a price adjustment logic with the primary goal of reducing inventory is triggered. "Guaranteed minimum gross profit ≥ first preset value" means that the actual gross profit of the product after the price adjustment is not lower than the minimum gross profit threshold preset by the server for fresh food categories. If the target store is a fast-moving consumer goods convenience store, then a fixed pricing target of "high-frequency sales + gross profit balance ≥ second preset value" will be matched. "High-frequency sales" refers to focusing on essential goods with a weekly purchase frequency ≥ preset frequency threshold. Essential goods include at least one of the following: drinking water, snacks, daily necessities, toiletries, and convenience foods. Gross profit balance means ensuring that the weekly sales fluctuation of these goods is ≤ preset fluctuation threshold through pricing control, avoiding a sharp drop in sales due to excessively high pricing or insufficient inventory due to excessively low pricing. If the target store is a brand specialty store, a fixed pricing target of "brand premium + member retention ≥ third preset value" will be matched. Here, "brand premium" refers to the percentage increase in price compared to similar non-branded products being greater than or equal to the preset premium percentage. "Member retention ≥ third preset value" means that after the price adjustment takes effect, the server will calculate the member repurchase rate within a preset observation period, and the member repurchase rate will not be lower than the server's preset minimum member repurchase threshold. The preset observation period is a time period pre-configured by the server and stored in the local database, and it is adapted to the consumption cycle of branded products. If the target store is a maternity and baby product store, a target of "safety endorsement + essential supply + gross profit ≥ fourth preset value" will be matched. The pricing target is defined as follows: "Value" refers to a fixed price target; "Safety Endorsement" requires the price to be linked to product safety certification information, which includes at least one of the following certifications: organic certification, infant product safety testing report, raw material traceability certification, and production standard certification. Price adjustments are only applied to products with safety endorsements. "Essential Supply Guarantee" means that when adjusting prices for essential baby and maternity products, inventory depth must be simultaneously verified to be ≥ a preset supply guarantee inventory. Essential baby and maternity products include at least one of the following: milk powder, diapers, baby wipes, complementary foods, and baby care products. "Gross Profit ≥ Fourth Preset Value" means that after price adjustment, the gross profit of essential baby and maternity products is not lower than the preset gross profit threshold for essential baby and maternity product categories on the server. The first, second, third, and fourth preset values, as well as preset turnover threshold, preset frequency threshold, preset fluctuation threshold, preset premium ratio, preset supply guarantee inventory, and preset observation period are all personalized thresholds pre-configured by the server and stored in the local database. These thresholds are specifically calculated based on the industry average gross profit level for the corresponding business type, the store's operating cost structure, and preset operating strategies.

[0049] For example, if a maternity and baby product store's fixed pricing target is "safety endorsement + guaranteed supply of essential goods + gross profit ≥ 12%", it will prioritize verifying the validity of the milk powder's "safety test report" when setting prices, ensuring a supply of ≥ 50 cans in stock. At the same time, when adjusting prices, it will ensure that the gross profit is not less than 12%, which not only meets parents' core demand for "safety" but also protects the store's profits.

[0050] In some possible implementations, the server automatically generates a dynamic pricing target based on cross-analysis of at least two types of data from the target store's multi-dimensional data. This includes at least one of the following cross-analysis scenarios: cross-analysis of inventory data and sales data; when the inventory turnover days are ≥ the first period threshold and the month-on-month sales decrease within a preset number of days is ≥ the first proportion threshold, a "deep inventory clearance" dynamic pricing target is generated; cross-analysis of customer traffic data and competitor price data; when the real-time customer traffic increases month-on-month by ≥ the second proportion threshold and the price of competing products in the same category is ≤ the target store's current selling price × the first coefficient threshold, a "market share grabbing" dynamic pricing target is generated; cross-analysis of online visitor consumption data and product gross profit data; when the average order value of online visitors within a preset statistical period... When the total cost of goods sold is greater than or equal to the preset average order cost threshold and the current gross profit of the goods is less than or equal to the fifth preset value, a dynamic pricing target of "slight price increase + value-added services" is generated. The average order cost is calculated as the total cumulative order amount of online visitors within the preset statistical period ÷ the total cumulative number of orders. Weather data and offline customer flow data are cross-analyzed. When the weather reaches a preset severe level and the offline customer flow decreases by more than or equal to the third proportional threshold, a dynamic pricing target of "online exclusive discounts + offline discounts" is generated. The first period threshold, first proportional threshold, second proportional threshold, third proportional threshold, first coefficient threshold, preset number of days, preset average order cost threshold, fifth preset value, and preset severe level are all personalized thresholds pre-configured by the server and stored in the local database.

[0051] For example, if a fast-moving consumer goods convenience store's carbonated beverage inventory turnover days reach 7 days (the first cycle threshold is 5 days), and sales have decreased by 25% in the past 3 days compared to the previous period (the first proportion threshold is 20%), the system generates a "deep inventory clearance" target after cross-analysis; if weekend real-time customer traffic increases by 40% compared to the previous period (the second proportion threshold is 30%), and the price of the same beverage from competitors is ≤ the store's price × 0.95 (the first coefficient threshold is 0.95), then a "market share grabbing" target is generated.

[0052] S104. Each feature component in the standardized feature vector is weighted by its corresponding weight coefficient, and then all weighted feature components are fused to obtain a comprehensive feature vector for inputting into the reinforcement learning model.

[0053] Weighted fusion transforms "data + weights" into signals that the model can recognize: for example, "inventory turnover of 3 days (weight 0.4)" is calculated as 1.2, and "competitor price 10% lower (weight 0.2)" is calculated as 0.2. After fusion, a comprehensive vector is input into the model. The reinforcement learning model acts like an "intelligent advisor," scoring each pricing action (Q-score): for example, "price reduction of 15%" has a Q-score of 80 (suitable for clearing inventory), and "maintaining the original price" has a Q-score of 50, providing a quantitative basis for pricing decisions.

[0054] Taking the "high-frequency sales" pricing target of fast-moving consumer goods (FMCG) convenience stores as an example, the goal is to improve the turnover efficiency and repurchase frequency of essential goods in the store through reasonable pricing, allowing goods to circulate quickly and reducing inventory backlog. This goal needs to ensure that the price of goods meets the expectations of mass consumers to attract high-frequency purchases, while also taking into account the store's basic gross profit, achieving a balance between sales volume and profit. Assuming the standardized feature vector is [sales ranking share 0.6 (weight 0.3), member add-to-cart rate 20% (weight 0.3), competitor price is the same (weight 0.2), high temperature weather (coded as 0.8, weight 0.2)], the sales ranking share of 0.6 means that the relative share of the product's sales among the same category of products in the store is 60%. For example, if the store has 10 types of bottled water, and this bottled water's sales are second only to the top one, ranking among the top in the same category, 0.6 is the standardized quantitative result of its sales competitiveness (the range is usually 0-1, the closer to 1, the higher the sales ranking). A 20% add-to-cart rate means that within a statistical period (e.g., one week), 20% of the store's member visitors added the product to their shopping cart. For example, if 100 member visitors viewed a snack at a fast-moving consumer goods convenience store within a week, and 20 of them added it to their cart, then the add-to-cart rate for this snack is 20%. This data directly reflects the member group's interest in the product and their purchase intention. "Price parity with competitors" means that the current price of the product at the target store is consistent with the real-time price of the same or similar substitute products from competitors in the same area (e.g., other convenience stores or supermarkets nearby), without being higher or lower than the competitors. For example, if a bottled beverage at a fast-moving consumer goods convenience store is priced at 3 yuan per bottle at the target store, and the same beverage from three nearby competitors is also priced at 3 yuan per bottle, this situation falls under "price parity with competitors." This status will be used as a reference feature for pricing decisions, combined with a weight of 0.2 (i.e., a 20% influence percentage) in the overall pricing judgment. "High Temperature Weather (coded as 0.8)" refers to converting the non-numerical weather category information of "high temperature" into a model-calcifiable value of 0.8 through standardized coding. This is a routine operation of "numericalizing classification features" in data processing. Specifically, a correspondence rule between weather levels and codes is first preset (e.g., low temperature = 0.2, normal temperature = 0.5, high temperature = 0.8, extreme high temperature = 1.0). "High temperature" corresponds to the code 0.8, which quantifies the degree of influence of weather on consumer behavior (the higher the value, the stronger the correlation between weather and pricing) and allows non-numerical weather information to be integrated into the feature vector to participate in subsequent weighted calculations and model inference, ensuring the consistency of multi-dimensional data format.

[0055] After weighted calculation, the values ​​of each component are 0.6×0.3=0.18, 20%×0.3=0.06, 0×0.2=0, and 0.8×0.2=0.16, respectively. The resulting composite feature vector is [0.18, 0.06, 0, 0.16]. This vector is then input into the reinforcement learning model.

[0056] In the fusion processing stage, different fusion strategies can be adopted according to data types: for numerical features (such as inventory turnover days and customer traffic), a weighted summation method is used; for categorical features (such as weather "rainy day" and "sunny day"), one-hot encoding is performed first, and then combined with weights to ensure that the fused comprehensive feature vector has a unified dimension and complete information. The reinforcement learning model can also be continuously trained by combining historical pricing cases of stores. For example, when the inventory turnover target is achieved by "reducing the price by 15%" multiple times in the "clearing inventory" scenario, the model will strengthen the correlation between the action and the target, making the Q-value calculation more accurate in the subsequent same scenario; at the same time, the model will regularly remove outdated historical data (such as promotional pricing records from six months ago) to avoid decision-making bias caused by changes in the market environment, and ensure the timeliness and reliability of the output Q-value.

[0057] S105. Input the comprehensive feature vector into the preset reinforcement learning model, and calculate and output the fitness Q value corresponding to each pricing adjustment action in the pricing action space through the reinforcement learning model.

[0058] The pricing action space is a pre-configured set of standardized pricing adjustment actions stored on the server, including at least two preset price adjustment directions and corresponding adjustment magnitudes. Examples include {+5%,+2%,0%,-2%,-5%} or "maintain original price," "price reduction of 5%," and "buy two get one free." This standardizes pricing actions, reduces arbitrariness in decision-making, and improves model computational efficiency.

[0059] Taking the example in step S104, after fusing to obtain the comprehensive feature vector [0.18, 0.06, 0, 0.16], it is input into the reinforcement learning model. The model will compare the three actions in the pricing action space: "maintain original price", "reduce price by 5%" and "buy two get one free". "Maintain original price" may affect sales due to high temperature weather, with a Q value of 65; "reduce price by 5%" can increase sales but slightly reduce profits, with a Q value of 75; "buy two get one free" fits the high-frequency sales target and has high user acceptance, with a Q value of 85.

[0060] In some possible implementations, the model can employ an ε-Greedy Policy to select actions. This involves exploring new actions randomly with a certain probability, and selecting the action with the highest Q-value with the remaining probability, while dynamically adjusting the exploration rate. This balances the stability and innovation of the pricing strategy, avoiding getting trapped in local optima. For example, a convenience store introduces a new product, "low-sugar soda," with an initial exploration rate ε = 0.5 (50% probability of random action selection). The model will select "maintain the original price," which has the highest Q-value, and will also randomly try new actions such as "8% price reduction" and "buy one get one free." After one week, having accumulated sufficient sales data, the exploration rate ε drops to 0.2 (20% probability of random exploration). The model then prioritizes the "5% price reduction," an action with a proven high Q-value, ensuring the new product quickly finds its optimal price while avoiding long-term reliance on a single strategy.

[0061] S106. Based on the adaptation Q value corresponding to each pricing adjustment action, determine the target pricing action from the pricing action space, generate a dynamic price adjustment instruction corresponding to the target pricing action, and send the dynamic price adjustment instruction to the network administrator of the target store. The network administrator will then synchronously update the product pricing information of the target store's offline POS system, shelf price tag display terminal, and online service channels.

[0062] In some possible implementations, price adjustment instructions are transmitted via the Message Queuing Telemetry Transport (MQTT) protocol, with an end-to-end response latency of less than 50 milliseconds, supporting real-time updates of Electronic Shelf Labels (ESLs). This enables instantaneous pricing and rapid response to market changes, making it particularly suitable for scenarios such as flash sales and limited-time offers. For example, if an e-commerce platform is conducting a flash sale, and a competitor suddenly reduces the price of the same phone from 3000 yuan to 2800 yuan one second before the sale begins, the system will push the "price reduction of 200 yuan" instruction to the ESL terminals of over 100 offline experience stores nationwide within 20 milliseconds via the MQTT protocol. Simultaneously, the price will be synchronized with the online mini-program price, ensuring that consumers see the same price as the competitor when the flash sale begins, thus preventing customer loss.

[0063] Taking the example in step S105, since the Q value corresponding to "buy two get one free" is at most 85, after determining "buy two get one free" as the target pricing action from the pricing action space, the server will generate a dynamic price adjustment instruction containing the product name (e.g., a certain brand of potato chips), the activity rule (buy two get one free), the execution time (effective immediately until the store closes that day), and the applicable channel (all channels), and send it to the target store's network administrator through an encrypted communication link. After receiving the instruction, the network administrator will first verify the integrity and legality of the instruction, and then simultaneously trigger the offline POS system to update the settlement rules (automatically identifying the second free item during settlement), the shelf price tag display terminal to pop up the "buy two get one free" activity logo, and the online mini-program / APP product details page to update the activity copy and purchase options, ensuring that consumers can enjoy the same pricing activity when shopping offline or ordering online, avoiding consumer disputes caused by information asynchrony between channels, and ensuring the efficient implementation of the "high-frequency sales" target.

[0064] Based on the Q-value corresponding to each pricing adjustment action, the target pricing action is determined from the pricing action space. A dynamic price adjustment instruction corresponding to the target pricing action is generated and sent to the network administrator of the target store. The network administrator then synchronously updates the product pricing information in the target store's offline POS system, shelf price tag display terminals, and online service channels. Here, the action with the highest Q-value (such as "15% price reduction") is selected, and the instruction is generated and sent to the network administrator. Upon receiving the instruction, the network administrator will simultaneously change the price of vegetables in the POS system to 2 yuan / jin, display 2 yuan on the shelf price tag, and update the online mini-program to 2 yuan. This ensures consistency between offline and online prices, avoiding the contradiction of "lower prices online and higher prices offline" for consumers and improving the consumer experience.

[0065] Among some possible implementations, incremental learning and online updating mechanisms can be employed to optimize model parameters and weight coefficients in real time based on feedback data such as sales volume, profit, and inventory after price adjustments. This allows pricing strategies to continuously adapt to market changes and improve decision-making accuracy. For example, a convenience store's summer beverage pricing model was trained in June based on data of "high temperatures + high customer traffic," with a weight of 0.3 for the "high temperature" feature. In July, with the arrival of the rainy season, feedback data showed a 20% decrease in beverage sales on "rainy days." The system used incremental learning to adjust the weight of the "weather" feature to 0.4 in real time (with the weight of "rainy day" encoded as 0.6 being increased), while simultaneously optimizing the λinv coefficient of the reward function, making pricing more suitable for the inventory turnover needs of the rainy season without retraining the entire model.

[0066] In some possible implementations, the system supports modular access to new data sources, such as social media trends and macroeconomic data, without requiring large-scale modifications to the existing architecture. This improves system scalability, continuously enriching the information dimensions for pricing decisions as business grows. For example, a popular snack brand discovered that "social media trends" affect sales (e.g., a snack's sales surge after receiving over 100,000 likes on Douyin). Through the system's modular API interface, it only took 3 days to integrate "Douyin trend data," incorporating it as a new feature dimension (e.g., "trend code 0.9") into the state vector. Subsequent pricing will prioritize assigning higher "customer traffic" weight to high-trend products, quickly capturing traffic dividends without needing to reconstruct the original data collection and model framework.

[0067] This embodiment solves the problems of traditional pricing, such as limited information, delayed decision-making, and subjectivity, through a complete process of "multi-dimensional data collection → standardized processing → adaptive weight adjustment → reinforcement learning decision-making → synchronized price adjustment across all channels." Multi-dimensional data provides a more comprehensive basis for pricing, adaptive weights ensure pricing focuses on business objectives, reinforcement learning makes decisions more scientific, and synchronized pricing across all channels avoids price confusion. Ultimately, it achieves accurate, timely, and efficient dynamic pricing, helping stores match inventory and market demands while improving the consistency of consumers' perceived prices.

[0068] In some possible implementations, when there are multiple combinations of pricing objectives, the method further includes: determining whether there are conflicts among the multiple combinations of pricing objectives, including: the existence of contradictory price adjustment constraints; if there are conflicts, determining the pricing objectives according to their priority from high to low, with the constraints of high-priority pricing objectives covering the conflict constraints of low-priority pricing objectives; the pricing objective priorities include: fixed pricing objective priority ≥ temporary pricing objective priority ≥ dynamic pricing objective priority; if there are no conflicts, integrating the constraints of all pricing objectives to generate a comprehensive pricing objective as the pricing objective corresponding to the target store, the integration method including: taking the intersection of the constraints of all pricing objectives.

[0069] For example, a fresh food store has a fixed pricing target of "guaranteed gross profit ≥ 5%", a temporary pricing target of "10% price reduction to clear inventory", and a dynamic pricing target of "15% price reduction". If the gross profit is 4% after reducing the price by 10% according to the temporary target (lower than the fixed target of 5%), a conflict is triggered. The fixed target constraint takes priority, and the price reduction is adjusted to 8%, which satisfies the need to clear inventory while ensuring that the gross profit is not lower than 5%.

[0070] Pricing objectives determine the "importance" of each data point: for example, if the goal of a fresh food store is "prioritizing inventory clearance," then the weight of "inventory turnover days" will be increased (e.g., set to 0.4), while the weight of "competitor prices" will be decreased (e.g., set to 0.2); if the goal is "profit priority," then the weight of "gross profit data" will be increased. This adaptive adjustment makes pricing more focused on business objectives, avoiding a "one-size-fits-all" weight allocation.

[0071] In some possible implementations, the method for determining the dynamic price of store goods also includes: the server calculating the achievement degree of the pricing target in real time; determining the feature components corresponding to the pricing target; if the achievement degree of the pricing target is less than the achievement degree threshold, adjusting the weight coefficient of the feature component corresponding to the pricing target in the direction of improving the achievement degree of the pricing target; if the achievement degree of the pricing target is greater than or equal to the achievement degree threshold, maintaining the initial weight coefficient of the feature component corresponding to the pricing target unchanged.

[0072] For example, a fresh food store's "inventory clearance" target achievement threshold is 70%, with an initial "inventory turnover days" weight of 0.4. One day after pricing is implemented, the achievement rate is only 50% (not reaching the threshold). The system increases the "inventory turnover days" weight to 0.5 while reducing the "customer traffic" weight to 0.2. Subsequent pricing will focus more on inventory data. Three days later, the achievement rate increases to 80%, and the initial weight is restored.

[0073] In some possible implementations, reinforcement learning models employ Deep Q-Networks (DQNs) to approximate the Q-function through a neural network, while simultaneously setting up an experience replay buffer to store historical experiences (states, actions, rewards, and new states) for model training. This effectively handles high-dimensional state spaces and improves the optimization capability of pricing strategies.

[0074] For example, in the DQN model of a fresh food store, the experience replay buffer stores more than 100,000 experience records from the past month, such as "state (200 catties of inventory + 50 customers + competitor price 2.5 yuan), action (10% price reduction), reward (8% profit increase + 30% inventory decrease), and new state (140 catties of inventory + 60 customers + competitor price 2.5 yuan)". When training a new model, 1,000 experience records are randomly selected from the buffer for iterative optimization to avoid the model relying too much on the latest data and improve decision robustness.

[0075] In some possible implementations, the model's reward function can be designed as Rt = λprofit・ΔProfitt + λinv・ΔInventoryt + λcomp・ΔCompetitivenesst, where ΔProfitt is the change in profit, ΔInventoryt is the change in inventory, ΔCompetitivenesst is the price difference with competitors, and λ is a dynamic weight. This quantifies diversified objectives, allowing pricing to simultaneously consider inventory, profit, and market competitiveness. For example, under the "clearing inventory" goal of a fresh food store, λprofit=0.2, λinv=0.6, and λcomp=0.2; if on a certain day, after a price reduction, ΔProfitt=-5% (profit slightly decreases), ΔInventoryt=-30% (inventory decreases significantly), and ΔCompetitivenesst=0 (price matches competitors), then Rt=0.2×(-5%)+0.6×(-30%)+0.2×0=-19% (here, the negative sign indicates that "inventory decrease" meets the inventory clearance goal, and the actual reward is converted to a positive score based on the absolute value). The model will determine that this action is suitable for the inventory clearance goal and will prioritize recommending similar actions in the future.

[0076] This application also provides an electronic device, including: a processor and a memory; the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device executes any of the embodiments of the previous method for determining the dynamic price of goods in stores.

[0077] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs any of the embodiments in the previous embodiments of the method for determining the dynamic price of goods in a store.

[0078] It is understood that the beneficial effects achieved by the electronic devices and computer-readable storage media provided above can be referred to the beneficial effects described in the method embodiments, and will not be repeated here.

[0079] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., Digital Versatile Disc (DVD)), or a semiconductor medium (e.g., Solid State Disk (SSD)).

[0080] The above-described embodiments are optional embodiments provided by this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the technical scope disclosed in this application should be included within the protection scope of this application.

Claims

1. A method for determining the dynamic price of goods in a store, characterized in that, Applied to a server, the method includes: The system receives multi-dimensional data transmitted from the network administrator of the target store. This multi-dimensional data includes internal store data and external data. The internal store data includes real-time operational data collected by the target store's inventory management system, POS system, and customer flow sensors, as well as category and behavioral information of current online visitors obtained from the target store's online service channels. The external data includes real-time competitor price data and weather data related to the target store, obtained by the target store's network administrator through a third-party data service API interface. The multi-dimensional data is cleaned and standardized to obtain a standardized feature vector. The standardized feature vector includes feature components of multiple dimensions, and each feature component corresponds to the standardization result of a class of data. Based on the pricing target corresponding to the target store, the weight coefficients of each feature component in the standardized feature vector are determined by a weight adaptive adjustment strategy. Each feature component in the standardized feature vector is weighted by its corresponding weight coefficient, and then all weighted feature components are fused to obtain a comprehensive feature vector for inputting into the reinforcement learning model. The comprehensive feature vector is input into a preset reinforcement learning model, and the model calculates and outputs the fitness Q value corresponding to each pricing adjustment action in the pricing action space. Based on the adaptation Q value corresponding to each of the pricing adjustment actions, a target pricing action is determined from the pricing action space, a dynamic price adjustment instruction corresponding to the target pricing action is generated, and the dynamic price adjustment instruction is sent to the network administrator of the target store, who then synchronously updates the product pricing information of the target store's offline POS system, shelf price tag display terminal, and online service channels.

2. The method for determining the dynamic price of store merchandise according to claim 1, characterized in that, The pricing action space is a set of standardized pricing adjustment actions pre-configured and stored on the server, including at least two preset price adjustment directions and corresponding adjustment ranges.

3. The method for determining the dynamic price of store merchandise according to claim 1, characterized in that, Also includes: Obtain the pricing target corresponding to the target store; the pricing target includes any one or more combinations of the following: (1) The server pre-configures and stores in a local database a fixed pricing target that matches the operating type of the target store; The target store's business type is one of a preset set of business types, which includes: fresh food stores, fast-moving consumer goods convenience stores, brand specialty stores, and maternity and baby product stores. (2) The server receives, in real time, the temporary pricing target entered by the store manager through the back-end management terminal via a communication link established with the network management of the target store; (3) The server automatically generates a dynamic pricing target after cross-analyzing at least two types of data in the multi-dimensional data of the target store.

4. The method for determining the dynamic price of store merchandise according to claim 3, characterized in that, The matching rules between the fixed pricing objective and the business type include: If the target store is a fresh food store, then a fixed pricing target of "prioritizing inventory clearance + guaranteeing minimum gross profit ≥ first preset value" will be matched; where "prioritizing inventory clearance" means that the pricing will prioritize ensuring the efficiency of product inventory turnover. When the inventory turnover days are ≥ the preset turnover threshold, the price adjustment logic with the primary goal of reducing inventory will be triggered. "Guaranteed minimum gross profit ≥ first preset value" means that the actual gross profit of the product after the price adjustment will not be lower than the minimum gross profit threshold preset by the server for fresh food categories. If the target store is a fast-moving consumer goods convenience store, then a fixed pricing target of "high-frequency sales + gross profit balance ≥ second preset value" will be matched. Here, "high-frequency sales" refers to focusing on essential goods with a weekly purchase frequency ≥ preset frequency threshold. Essential goods include at least one of the following: drinking water, snacks, daily necessities, toiletries, and convenience foods. Gross profit balance means ensuring that the weekly sales fluctuation of these goods is ≤ preset fluctuation threshold through pricing control, so as to avoid a sharp drop in sales due to excessively high pricing or insufficient inventory due to excessively low pricing. If the target store is a brand specialty store, then a fixed pricing target of "brand premium + member retention ≥ third preset value" will be matched; wherein, "brand premium" refers to the percentage increase in the price of the product compared to similar non-brand products ≥ preset premium percentage, and "member retention ≥ third preset value" refers to the member repurchase rate within a preset observation period after the price adjustment takes effect, wherein the member repurchase rate is not lower than the minimum member repurchase threshold preset by the server; the preset observation period is a time period pre-configured by the server and stored in the local database, and is adapted to the consumption cycle of brand products; If the target store is a maternity and baby product store, a fixed pricing target of "safety endorsement + essential supply guarantee + gross profit ≥ fourth preset value" will be matched. The "safety endorsement" price must be linked to product safety certification information, which includes at least one of the following certifications: organic certification, infant product safety test report, raw material traceability certification, and production standard certification. Price adjustments will only be applied to products with safety endorsement. "Essential supply guarantee" means that when adjusting the price of essential maternity and baby products, the inventory depth must be simultaneously verified to be ≥ the preset supply guarantee inventory. Essential maternity and baby products include at least one of the following: milk powder, diapers, baby wipes, complementary food, and baby care products. "Gross profit ≥ fourth preset value" means that after the price adjustment, the gross profit of essential products will not be lower than the gross profit threshold preset by the server for essential maternity and baby products. Among them, the first preset value, the second preset value, the third preset value, the fourth preset value, the preset turnover threshold, the preset frequency threshold, the preset fluctuation threshold, the preset premium ratio, the preset supply guarantee inventory, and the preset observation period are all personalized thresholds pre-configured by the server and stored in the local database, and are specifically calculated and determined based on the industry average gross profit level, store operating cost structure, and preset operating strategy of the corresponding business type.

5. The method for determining the dynamic price of store merchandise according to claim 3, characterized in that, The server automatically generates a dynamic pricing target based on cross-analysis of at least two types of data from the multi-dimensional data of the target store, including at least one of the following cross-analysis scenarios: Cross-analyze inventory and sales data. When the inventory turnover days of a product are greater than or equal to the first cycle threshold and the sales volume decreases by more than or equal to the first proportion threshold within the preset number of days, a dynamic pricing target for "deep inventory clearance" is generated. Cross-analyze customer flow data and competitor price data. When the real-time customer flow growth rate is greater than or equal to the second proportion threshold and the price of the same category of competitor products is less than or equal to the current selling price of the target store multiplied by the first coefficient threshold, a dynamic pricing target for "seizing market share" is generated. Cross-analyze online visitor consumption data and product gross profit data. When the average order amount of online visitors within a preset statistical period is greater than or equal to a preset average order amount threshold, and the current gross profit of the product is less than or equal to the fifth preset value, a dynamic pricing target of "slight price increase + value-added services" is generated. The average order amount is calculated as the total cumulative order amount of online visitors within the preset statistical period divided by the total cumulative number of orders. Cross-analyze weather data and offline customer flow data. When the weather reaches a preset severe level and the offline customer flow decreases by more than or equal to the third proportion threshold, generate a dynamic pricing target of "online exclusive discount + offline discount". Among them, the first cycle threshold, the first proportion threshold, the second proportion threshold, the third proportion threshold, the first coefficient threshold, the preset number of days, the preset average order consumption amount threshold, the fifth preset value, and the preset severity level are all personalized thresholds pre-configured by the server and stored in the local database.

6. The method for determining the dynamic price of store merchandise according to claim 3, characterized in that, When the pricing objectives are multiple combinations, the method further includes: Determine whether there is a conflict among multiple combinations of the pricing objectives, including: the existence of contradictory price adjustment constraints; If a conflict exists, the pricing objectives are determined from highest to lowest priority, with the constraints of higher-priority pricing objectives covering the conflict constraints of lower-priority pricing objectives. The pricing objective priorities include: fixed pricing objective priority ≥ temporary pricing objective priority ≥ dynamic pricing objective priority. If there is no conflict, the constraints of all pricing objectives are merged to generate a comprehensive pricing objective as the pricing objective corresponding to the target store. The merging method includes taking the intersection of the constraints of all pricing objectives.

7. The method for determining the dynamic price of store merchandise according to any one of claims 1-6, characterized in that, The method further includes: The server calculates the degree to which the pricing target has been achieved in real time; Determine the feature components corresponding to the pricing objective; If the achievement rate of the pricing target is less than the achievement rate threshold, the weight coefficients of the feature components corresponding to the pricing target will be adjusted in the direction of improving the achievement rate of the pricing target; If the achievement rate of the pricing target is greater than or equal to the achievement rate threshold, the initial weight coefficients of the feature components corresponding to the pricing target remain unchanged.

8. The method for determining the dynamic price of store merchandise according to any one of claims 1 to 6, characterized in that, The current online visitor category information includes any one of the following categories: new visitor, returning visitor, member visitor, and high-value visitor; new visitor includes visitors with no historical consumption record; returning visitor includes visitors with at least a preset consumption record within the most recent first preset time interval; member visitor includes visitors who have activated store membership services; and high-value visitor includes visitors whose cumulative consumption amount within the most recent second preset time interval is ≥ a value threshold. The current online visitor behavior information includes one or more of the following: product categories viewed, product page browsing time, number of product categories viewed, number of times added to cart, number of products favorited, order conversion rate, historical repurchase frequency, and page dwell path.

9. An electronic device, characterized in that, include: Processor and memory; The processor is connected to the memory, which stores a computer program. The processor executes the computer program stored in the memory to cause the electronic device to perform the method for determining the dynamic price of store merchandise as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the method for determining the dynamic price of store merchandise as described in any one of claims 1-8.

Images