Digital store management method and system

Abstract

The invention discloses a digital store management method and system, and the method comprises the steps: obtaining multi-source data in real time according to customer behavior data, sales data, inventory data and environment data of a store, and generating a time-synchronized multi-source data set; inputting the multi-source data set into a behavior analysis model, extracting space-time mode characteristics of customer behaviors, and generating a customer behavior analysis report; generating an intelligent replenishment scheme according to the customer behavior analysis report and the inventory data; and inputting the customer behavior analysis report and the intelligent replenishment scheme into a visual platform based on augmented reality, generating a visual chart of store operation, and generating a final digital store management scheme through an interactive operation interface and a multi-dimensional screening function. According to the embodiment of the invention, an efficient, intelligent and flexible store management solution can be provided, and the whole-process intelligentization of data acquisition, behavior analysis, inventory optimization and visual decision is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of store management, and in particular to a digital store management method and system. Background Art

[0002] With the development of science and technology and advancements in artificial intelligence (AI), digital management has gradually become a crucial means of improving efficiency and operations across various industries. This is particularly true in the retail industry, where modern store management not only requires processing massive amounts of customer behavior, sales, inventory, and environmental data, but also requires real-time responses to market changes and customer demands to maintain competitiveness and profitability. Traditional store management methods often rely on manual record-keeping and human judgment, which is not only time-consuming and labor-intensive but also prone to data errors and decision-making delays, making it difficult to adapt to rapidly changing market environments and customer demands. Summary of the Invention

[0003] The purpose of the present invention is to provide a digital store management method and system to address the deficiencies in the existing technology, and to provide an efficient, intelligent and flexible store management solution to achieve full-process intelligence in data collection, behavior analysis, inventory optimization and visual decision-making.

[0004] An embodiment of the present application provides a digital store management method, the method comprising: Based on the store's customer behavior data, sales data, inventory data, and environmental data, an edge computing-based data collection framework is used to obtain multi-source data in real time. Through an adaptive data fusion algorithm, data from different sources is uniformly converted into a standardized data format to generate a time-synchronized multi-source data set. Inputting the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, and combines historical sales data to predict customer preferences and generate a customer behavior analysis report. Based on the customer behavior analysis report and inventory data, a deep reinforcement learning-based inventory management model is used to dynamically adjust inventory strategies based on sales trends and replenishment cycles. Through a multi-objective optimization algorithm, inventory costs and sales demand are balanced to generate intelligent replenishment plans. The customer behavior analysis report and the intelligent replenishment plan are input into a visualization platform based on augmented reality to generate a visualization chart of store operations. Through an interactive operation interface and multi-dimensional screening functions, managers are supported to monitor and adjust operation strategies in real time and generate the final digital store management plan.

[0005] Optionally, based on the store's customer behavior data, sales data, inventory data, and environmental data, a data collection framework based on edge computing is used to acquire multi-source data in real time. By using an adaptive data fusion algorithm, data from different sources is uniformly converted into a standardized data format to generate a time-synchronized multi-source data set, including: Based on the store's customer behavior data, sales data, inventory data, and environmental data, an edge computing-based data collection framework is adopted to obtain multi-source data in real time through distributed edge nodes. Each edge node is equipped with lightweight data caching technology to ensure the real-time and continuity of data collection. For the collected multi-source data, a deep learning-based format recognition model is used to automatically identify the format types of different data sources. An adaptive data cleaning algorithm is used to filter noise and fill missing values in the data to generate a preliminary standardized data set. For the preliminary standardized data set, a multi-source data synchronization algorithm based on timestamp alignment is used to eliminate the time differences between different data sources. The missing data is supplemented by interpolation and filling methods to generate a time-synchronized multi-source data set. For time-synchronized multi-source data sets, a standardized method based on an adaptive data fusion algorithm is adopted to uniformly convert data from different sources into a standardized data format. Through dynamic weight adjustment technology, the accuracy and consistency of data fusion are ensured to generate the final multi-source data set.

[0006] Optionally, the multi-source dataset is input into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, combine historical sales data, predict customer preferences, and generate a customer behavior analysis report, including: For multi-source datasets, we use a behavior analysis model based on a spatiotemporal graph convolutional network. We abstract customers, products, time, and space into nodes in a graph structure, and abstract the spatiotemporal relationships between nodes into edges. We then generate a preliminary spatiotemporal graph structure using dynamic graph structure learning technology. For the preliminary spatiotemporal graph structure, a feature extraction method based on multi-scale convolution kernels is used. This method combines the local details and global trends of customer behavior to extract multi-scale features. Adaptive convolution kernel adjustment technology is used to capture features at different scales and generate a preliminary multi-scale feature representation. For the preliminary multi-scale feature representation, a feature fusion method based on multi-head attention technology is used to perform weighted fusion of features of different scales. Customer preferences are predicted based on historical sales data to generate preliminary customer preference prediction results. For the preliminary customer preference prediction results, an analysis method based on the explainability module is used, combined with customer behavior data and sales trends to generate the final customer behavior analysis report.

[0007] Optionally, based on the customer behavior analysis report and inventory data, an inventory management model based on deep reinforcement learning is adopted to dynamically adjust inventory strategies in combination with sales trends and replenishment cycles. Through a multi-objective optimization algorithm, inventory costs and sales demand are balanced to generate an intelligent replenishment plan, including: Based on customer behavior analysis reports and inventory data, we adopted an inventory management model based on deep reinforcement learning. We combined sales trends and replenishment cycles to define inventory optimization goals and generated a preliminary optimization problem model using a multi-objective optimization framework. For the preliminary optimization problem model, we use a policy generation method based on the policy gradient method, combining inventory status and sales trends to dynamically adjust the inventory policy. Through adaptive learning rate adjustment technology, we optimize the accuracy and stability of the policy and generate a preliminary inventory management policy. For the initial inventory management strategy, a multi-objective optimization method based on the weighted summation method is used. The replenishment plan is dynamically adjusted based on inventory cost and sales demand. The dynamic weight allocation technology is used to balance the importance of different optimization objectives and generate a preliminary intelligent replenishment plan. For the preliminary intelligent replenishment plan, a verification method based on simulation is adopted, combining inventory status and sales trends to verify the feasibility and effectiveness of the replenishment plan. Through feedback correction technology, the replenishment plan is dynamically adjusted to generate the final intelligent replenishment plan.

[0008] Optionally, the customer behavior analysis report and the intelligent replenishment plan are input into an augmented reality-based visualization platform to generate a visualization chart of store operations. Through an interactive operation interface and multi-dimensional screening functions, managers can monitor and adjust operation strategies in real time, generating a final digital store management plan, including: For customer behavior analysis reports and intelligent replenishment plans, we use an augmented reality-based visualization platform, combined with store operation data, to generate initial visualization charts. Conditional constraint technology is used to ensure the accuracy and readability of the charts and generate preliminary visualization charts. For the preliminary visualization chart, we adopted a design method based on interactive operation interface, combined with the needs of managers, designed multi-dimensional screening functions and dynamic interactive operations, and optimized the interface layout and operation process through user behavior analysis to generate a preliminary interactive visualization solution; For the preliminary interactive visualization solution, a real-time update method based on streaming computing and incremental update technology was adopted. In combination with real-time operational data, the visualization content was dynamically updated. Through real-time monitoring technology, the timeliness and operability of the data were ensured, and a preliminary real-time update visualization solution was generated. For the preliminary real-time update visualization solution, an optimization method based on user feedback is adopted, combined with manager interaction data and real-time performance monitoring, to dynamically adjust the visualization content, and generate the final digital store management solution through feedback correction technology.

[0009] Another embodiment of the present application provides a digital store management system, the system comprising: The acquisition module is used to acquire multi-source data in real time based on the store's customer behavior data, sales data, inventory data, and environmental data using an edge computing-based data acquisition framework. Using an adaptive data fusion algorithm, it converts data from different sources into a standardized data format to generate a time-synchronized multi-source data set. An extraction module is configured to input the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, combine historical sales data, predict customer preferences, and generate a customer behavior analysis report. A generation module is used to use a deep reinforcement learning-based inventory management model based on the customer behavior analysis report and inventory data, dynamically adjust inventory strategies based on sales trends and replenishment cycles, and generate intelligent replenishment plans by balancing inventory costs and sales demand through a multi-objective optimization algorithm; The management module is used to input the customer behavior analysis report and the intelligent replenishment plan into the augmented reality-based visualization platform to generate a visualization chart of store operations. Through the interactive operation interface and multi-dimensional screening function, it supports managers to monitor and adjust operation strategies in real time and generate the final digital store management plan.

[0010] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when run.

[0011] Yet another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0012] Compared with the existing technology, the digital store management method provided by the present invention obtains multi-source data in real time based on the store's customer behavior data, sales data, inventory data and environmental data to generate a time-synchronized multi-source data set; inputs the multi-source data set into a behavior analysis model, extracts the spatiotemporal pattern characteristics of customer behavior, and generates a customer behavior analysis report; generates an intelligent replenishment plan based on the customer behavior analysis report and inventory data; inputs the customer behavior analysis report and the intelligent replenishment plan into an augmented reality-based visualization platform to generate a visualization chart of store operations, and generates a final digital store management plan through an interactive operation interface and multi-dimensional screening functions, thereby providing an efficient, intelligent and flexible store management solution, realizing the full process intelligence of data collection, behavior analysis, inventory optimization and visualization decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A hardware structure block diagram of a computer terminal for a digital store management method provided by an embodiment of the present invention; Figure 2 A flowchart of a digital store management method provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a digital store management system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0015] The embodiment of the present invention first provides a digital store management method, which can be applied to electronic devices such as computer terminals, specifically ordinary computers.

[0016] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal for a digital store management method provided by an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0017] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, which, when executed, can cause the processor to execute any one of the digital store management methods.

[0018] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0019] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any digital store management method.

[0020] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0021] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0022] See also Figure 2 , an embodiment of the present invention provides a digital store management method, which may include the following steps: S201 uses an edge computing-based data collection framework to acquire multi-source data in real time based on the store's customer behavior data, sales data, inventory data, and environmental data. Using an adaptive data fusion algorithm, the data from different sources is uniformly converted into a standardized data format to generate a time-synchronized multi-source data set. This method uses an edge computing-based data collection framework to acquire real-time store customer behavior, sales, inventory, and environmental data. This process involves multiple distributed edge nodes that collect data in real time from different store locations, ensuring the timeliness and accuracy of the data. Using an adaptive data fusion algorithm, the system can uniformly convert data from different sources into a standardized data format. This standardization process not only improves data availability but also provides a reliable foundation for subsequent data analysis and decision-making. Ultimately, the resulting time-synchronized multi-source dataset provides a comprehensive perspective for subsequent customer behavior analysis and inventory management.

[0023] The implementation of this step has significantly enhanced the intelligence of store management. By acquiring and standardizing multi-source data in real time, managers gain a more comprehensive understanding of customer behavior and market dynamics, enabling more accurate decision-making. Standardized data sets provide high-quality input for subsequent analytical models, ensuring the accuracy and reliability of analytical results. This not only helps improve customer satisfaction but also optimizes inventory management, reduces operating costs, and ultimately increases store profitability.

[0024] Specifically, we can use an edge computing-based data collection framework to acquire multi-source data in real time through distributed edge nodes based on the store's customer behavior data, sales data, inventory data, and environmental data. Each edge node is equipped with lightweight data caching technology to ensure the real-time and continuity of data collection. During this phase, stores leverage distributed edge nodes within an edge computing architecture to acquire real-time data from various sources. This data includes customer behavior, sales figures, inventory status, and environmental factors such as temperature and humidity. The deployment of edge nodes allows data collection to be performed closer to the data source, reducing latency and improving data processing efficiency. The application of lightweight data caching technology ensures timely data storage and processing during peak traffic times, preventing data loss due to network latency. The core of this step is to achieve real-time data collection and processing, providing a foundation for subsequent data analysis and decision-making. Through the deployment of distributed edge nodes, stores can quickly respond to market changes and customer needs, improving operational efficiency. Furthermore, the acquisition of real-time data enables stores to better monitor customer behavior and sales trends, thereby optimizing inventory management and marketing strategies.

[0025] First, stores deploy multiple edge computing nodes at key locations (such as entrances, checkout counters, and shelves). These nodes collect and process information from various data sources. For example, edge nodes at entrances can monitor customer entry and exit in real time, using cameras or sensors to record visit frequency and peak hours. Nodes on shelves can monitor the sales and inventory status of specific products. Each node serves as the first layer of data collection, enabling local preliminary data collation and preprocessing, reducing the burden on central servers.

[0026] Secondly, the lightweight data caching technology equipped in edge nodes allows data to be cached for a short period of time, ensuring data is not lost, especially in high-traffic situations. For example, during a promotional event, store traffic can surge. Edge nodes can store customer scans and purchase data locally and upload it to the central server after the network load decreases or data traffic stabilizes. This mechanism ensures real-time and continuous data collection, providing a solid foundation for subsequent data analysis.

[0027] Finally, the edge computing architecture offers the advantage of low-latency data processing. When a customer makes a purchase, the relevant data can be processed immediately and used to update inventory information, without having to wait for the data to be uploaded to the cloud and then returned. This real-time feedback mechanism enables stores to respond quickly to customer needs, optimizing inventory management, improving the customer experience, and boosting sales.

[0028] For the collected multi-source data, a deep learning-based format recognition model is used to automatically identify the format types of different data sources. An adaptive data cleaning algorithm is used to filter noise and fill missing values in the data to generate a preliminary standardized data set. During this phase, the store utilizes deep learning technology to identify the format of the collected multi-source data, automatically determining its source and type. This process ensures data accuracy and consistency. Subsequently, an adaptive data cleaning algorithm is used to process the data, removing noise and filling missing values, thereby generating a preliminary standardized dataset. This standardization process ensures that subsequent data analysis can be conducted in a unified format, avoiding analytical errors caused by inconsistent data formats. This step is crucial for improving data quality and usability. Through automated format recognition and data cleaning, the store can reduce manual intervention and improve data processing efficiency and accuracy. This standardized dataset provides a reliable foundation for subsequent analysis and decision-making, ensuring the validity of the data analysis results.

[0029] After data collection is complete, stores use deep learning format recognition models to analyze data from various sources. Specifically, trained convolutional neural network (CNN) models can identify the format of customer behavior data, sales data, and inventory data. For example, customer behavior data may include information such as timestamps, customer IDs, and product IDs, while sales data may include sales volume and sales date. Once input into the model, the deep learning system automatically identifies the structure of this data, laying the foundation for subsequent data cleaning and processing.

[0030] Next, an adaptive data cleaning algorithm is used to process the initially identified data. This algorithm automatically identifies noise and missing values in the data. For example, a customer's purchase record may lack product ID information, or business data from a specific time period may not be recorded due to a system failure. The data cleaning algorithm then fills in these missing values. Common strategies include using mean or mode imputation, or extrapolating the imputation based on other similar data. This process reduces manual intervention through automation, improving the efficiency and accuracy of data cleaning.

[0031] Finally, the cleaned data is uniformly converted into a standardized format, forming a preliminary standardized dataset. For example, stores can specify a unified data format that includes all required fields and their data types. This standardization process ensures that subsequent data analysis can be performed in the same format, allowing data from different sources to be seamlessly combined to form a complete database, providing reliable data support for subsequent analysis.

[0032] For the preliminary standardized data set, a multi-source data synchronization algorithm based on timestamp alignment is used to eliminate the time differences between different data sources. The missing data is supplemented by interpolation and filling methods to generate a time-synchronized multi-source data set. During this phase, stores synchronize data from different sources using a timestamp alignment algorithm to eliminate time discrepancies caused by different data collection times. Interpolation fills in missing data, ensuring complete data records at every point in time. This process ensures data consistency across time, enabling subsequent analysis to be based on accurate time series data. The core of this step is ensuring data temporal consistency and avoiding analytical bias caused by time discrepancies. Time-synchronized datasets provide a reliable foundation for subsequent customer behavior analysis and sales forecasting, resulting in more accurate and effective analytical results.

[0033] When processing the initial standardized dataset, the store first applies a multi-source data synchronization algorithm based on timestamp alignment. Specifically, the data from each data source is sorted by timestamp and then aligned across different time dimensions. For example, sales data might be timestamped by the hour, while customer behavior data might be recorded by the minute. This ensures that all data is compared and analyzed on the same time dimension, eliminating errors caused by different collection times.

[0034] Next, for missing data that arises during the alignment process, such as incomplete sales data for a certain time period, stores can use interpolation to fill in missing values. Specifically, stores can infer missing values based on data from previous and subsequent time points. For example, if customer behavior data indicates a peak in purchases at a certain point in time, but sales data shows a significant drop in sales, the interpolation algorithm will fill in the missing values based on sales trends at adjacent time points. This approach not only ensures data continuity but also ensures its authenticity and legitimacy.

[0035] Finally, the time-synchronized, multi-source dataset generated through this process can serve as the foundation for further analysis, ensuring reliable records at every point in time. This is crucial for decision support, such as customer behavior analysis and sales forecasting. For example, when evaluating the effectiveness of promotional campaigns, stores can accurately compare customer in-store behavior with sales data, enabling more informed operational decisions.

[0036] For time-synchronized multi-source data sets, a standardized method based on an adaptive data fusion algorithm is adopted to uniformly convert data from different sources into a standardized data format. Through dynamic weight adjustment technology, the accuracy and consistency of data fusion are ensured to generate the final multi-source data set.

[0037] During this phase, stores utilize adaptive data fusion algorithms to standardize time-synchronized, multi-source datasets. Dynamic weight adjustment technology allows stores to flexibly adjust the weights of data fusion based on the importance and reliability of different data sources, ensuring the final dataset achieves optimal accuracy and consistency. This step is crucial for efficient data fusion, ensuring that data from different sources can be analyzed under unified standards. Dynamic weight adjustment allows stores to optimize the data fusion process based on actual conditions, improve the accuracy and reliability of data analysis, and provide a solid data foundation for subsequent customer behavior analysis and inventory management.

[0038] After generating a time-synchronized, multi-source dataset, the store employs a standardized approach based on an adaptive data fusion algorithm. Specifically, the store can use a trained model to assess the importance of different data sources (such as customer behavior data, sales data, and environmental data). For example, during one time period, customer behavior data may be more indicative of sales trends, while inventory data may be more important during another. The adaptive data fusion algorithm automatically calculates the weighting of each data source to better reflect current business priorities in the final dataset.

[0039] Furthermore, dynamic weighting technology continuously optimizes the data fusion process based on real-time data and contextual information. For example, when a store experiences a seasonal promotion, the importance of relevant data sources may change. The system can promptly adjust the weighting to make the impact of customer behavior data during the promotion more significant. This flexibility ensures that the data fusion results accurately reflect the store's current operating status and market dynamics.

[0040] Finally, the resulting multi-source dataset, after adaptive data fusion, is used for subsequent analysis and decision support. Stores can generate more comprehensive and realistic datasets to guide operations. For example, by combining customer purchasing behavior with real-time inventory dynamics, stores can better formulate replenishment strategies and promotional plans. This data-driven decision-making approach significantly improves store operational efficiency and customer satisfaction.

[0041] S202: Inputting the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, and combines historical sales data to predict customer preferences and generate a customer behavior analysis report. In this step, the store inputs the collected multi-source datasets into a behavior analysis model based on a spatiotemporal graph convolutional network. The model is designed to represent the complex relationships between customers, products, time, and space through a graph structure. Specifically, customer behavior data is abstracted as nodes in the graph, and the spatiotemporal relationships between nodes are represented by edges. Through dynamic graph structure learning, the model can dynamically capture changes in customer behavior and identify customer behavior patterns under different temporal and spatial conditions. This approach can not only extract local details, such as customers' purchasing preferences within a specific time period, but also identify global trends, such as seasonal sales fluctuations, thereby providing a comprehensive perspective for subsequent analysis.

[0042] The core significance of this step lies in improving the understanding and prediction of customer behavior through deep learning technology. By extracting spatiotemporal pattern features, stores can more accurately grasp customer purchasing habits and preferences, thereby formulating more precise marketing strategies. For example, if the model identifies a significant increase in purchase frequency of a particular product on weekends, the store can launch a corresponding promotion before the weekend to increase sales. Furthermore, by analyzing historical sales data, stores can better predict future customer demand, optimize inventory management, and reduce stockouts and overstocks, ultimately achieving both sales and customer satisfaction.

[0043] Specifically, we can use a behavior analysis model based on a spatiotemporal graph convolutional network for multi-source data sets, abstracting customers, products, time, and space into nodes in a graph structure, and abstracting the spatiotemporal relationships between nodes into edges. Through dynamic graph structure learning technology, we can generate a preliminary spatiotemporal graph structure. In this step, customer behavior, product information, timestamps, and spatial locations in the multi-source datasets must first be organized and labeled. Each customer, product, and time point is considered a node in the graph, and the purchase relationship between customers and products, and the relationship between time and customer behavior, are represented by edges. For example, if customer A purchases product X at a certain point in time, an edge connecting customer A and product X will be formed in the graph. In this way, the model can effectively capture customer behavior patterns and product sales dynamics. The introduction of dynamic graph structure learning technology enables the model to dynamically update the graph structure based on changes in real-time data, ensuring that it reflects the latest trends in customer behavior.

[0044] The core significance of this step lies in improving the understanding and analysis of customer behavior through the formal representation of a graph structure. By abstracting customers, products, time, and space into a graph structure, the model can better capture the complexity and diversity of customer behavior. For example, if the model identifies that a customer frequently purchases a certain type of product within a specific time period, the store can develop personalized marketing strategies based on this information to improve the customer's shopping experience and satisfaction. Furthermore, the flexibility of the graph structure enables the model to adapt to different business scenarios, helping stores maintain competitiveness in a rapidly changing market environment.

[0045] In this step, multi-source datasets must first be preprocessed to ensure data quality and consistency. Data preprocessing involves cleaning and standardizing customer behavior data, product information, timestamps, and spatial locations. For example, a customer's purchase history may contain missing values or outliers. During the preprocessing phase, interpolation or mean-filling methods are used to address these missing values and ensure the completeness of each customer's behavior data. Next, this data is converted into a graph structure, with customers, products, time, and space as nodes. Relationships between nodes are represented by edges. For example, if customer A purchases product X at time T, an edge is formed in the graph connecting customer A and product X.

[0046] Next, dynamic graph structure learning techniques are used to generate a preliminary spatiotemporal graph structure. Dynamic graph structure learning allows the model to dynamically update the graph structure based on real-time data changes. For example, when customer A makes a purchase within the store, the system updates the edge between customer A and item X in the graph in real time to reflect the latest purchase. This dynamic update mechanism enables the model to promptly capture changes in customer behavior, ensuring the timeliness and accuracy of analysis results. Furthermore, the introduction of graph convolutional networks enables the model to learn features within the graph structure, extracting the spatiotemporal patterns of customer behavior, providing a foundation for subsequent analysis.

[0047] Finally, the generated spatiotemporal graph structure is trained using a graph convolutional network to extract the spatiotemporal patterns of customer behavior. The model optimizes the graph convolution parameters through multiple iterations, enabling it to better capture both local details and global trends in customer behavior. This approach not only enables the model to identify customer purchasing habits but also analyzes the sales performance of different products under different temporal and spatial conditions, providing strong support for subsequent predictions of customer preferences.

[0048] For the preliminary spatiotemporal graph structure, a feature extraction method based on multi-scale convolution kernels is used. This method combines the local details and global trends of customer behavior to extract multi-scale features. Adaptive convolution kernel adjustment technology is used to capture features at different scales and generate a preliminary multi-scale feature representation. In this step, based on the preliminary spatiotemporal graph structure, multi-scale convolution kernels are used for feature extraction. The design of multi-scale convolution kernels allows the model to capture characteristics of customer behavior at different scales. For example, smaller convolution kernels can capture customers' rapid purchases over a short period of time, while larger convolution kernels can identify customers' long-term purchasing trends. Through adaptive convolution kernel adjustment technology, the model can dynamically adjust the size and shape of the convolution kernel based on the characteristics of the data to better adapt to different feature extraction requirements. This process not only improves the accuracy of feature extraction but also enhances the model's ability to recognize complex customer behavior patterns.

[0049] The significance of this step lies in comprehensively improving customer behavior analysis capabilities through multi-scale feature extraction. By combining local details with global trends, the model can more accurately identify customer purchasing habits and preferences. For example, if the model finds that a customer frequently purchases a certain type of product within a specific time period, the store can develop personalized marketing strategies based on this information to enhance the customer's shopping experience and satisfaction. Furthermore, multi-scale feature extraction can help stores identify potential market opportunities, optimize product mix and promotional strategies, and thus improve overall sales performance.

[0050] In this step, we first need to design multi-scale convolution kernels to extract customer behavior features at different scales. These multi-scale convolution kernels can include both small-scale and large-scale kernels, designed to capture short-term and long-term customer behavior patterns, respectively. For example, a small-scale kernel can capture a customer's rapid purchases within a specific time period, while a large-scale kernel can identify long-term purchasing trends. This allows the model to comprehensively analyze customer behavior and identify underlying purchasing habits.

[0051] Next, adaptive convolution kernel adjustment technology is used to dynamically adjust the size and shape of the convolution kernel based on the characteristics of customer behavior data. For example, during peak hours, when customers may be purchasing more frequently, the model can use a smaller convolution kernel to capture rapidly changing purchasing behavior; during stable periods, the model can use a larger convolution kernel to identify long-term trends. This dynamic adjustment mechanism enables the model to better adapt to different feature extraction requirements and improve feature extraction accuracy.

[0052] Finally, multi-scale feature extraction generates preliminary multi-scale feature representations. These feature representations provide the basis for subsequent customer preference predictions. The model can input these extracted multi-scale features into subsequent analysis modules and, combined with historical sales data, predict customer preferences. For example, if the model identifies a customer who frequently purchases a certain type of product within a specific time period, the store can develop personalized marketing strategies based on this information to enhance the customer's shopping experience and satisfaction.

[0053] For the preliminary multi-scale feature representation, a feature fusion method based on multi-head attention technology is used to perform weighted fusion of features of different scales. Customer preferences are predicted based on historical sales data to generate preliminary customer preference prediction results. In this step, the model inputs the preliminary multi-scale feature representation into a feature fusion module based on a multi-head attention mechanism. This multi-head attention mechanism allows the model to focus on the importance of different features when fusing them, thereby achieving weighted fusion. For example, the model can dynamically adjust the weights of individual features based on the influence of historical sales data, resulting in more accurate predictions of customer preferences. This approach comprehensively considers both short-term and long-term customer behavior to generate preliminary predictions of customer preferences. This result not only provides data support for subsequent marketing strategies but also helps stores make more informed decisions regarding product recommendations and promotions.

[0054] The significance of this step lies in improving the accuracy and reliability of customer preference predictions through feature fusion. By combining multi-scale features, the model can more comprehensively reflect customer purchasing behavior and preferences. For example, if the model identifies a customer's preference for a certain product category within a specific time period, the store can develop personalized marketing strategies based on this information to enhance the customer's purchasing experience and satisfaction. Furthermore, accurate customer preference predictions can help stores optimize inventory management, reduce out-of-stocks and overstocks, and thus improve overall operational efficiency.

[0055] In this step, the initial multi-scale feature representation is first input into a feature fusion module based on a multi-head attention mechanism. This multi-head attention mechanism allows the model to focus on the importance of different features when fusing them, thereby achieving weighted fusion. For example, the model can dynamically adjust the weights of various features based on the influence of historical sales data, resulting in more accurate predictions of customer preferences. In this way, the model comprehensively considers both short-term and long-term customer behavior to generate preliminary predictions of customer preferences.

[0056] Next, the model performs a weighted fusion of features at different scales. Specifically, the model assigns a weight to each feature, the size of which depends on its importance in predicting customer preferences. For example, if a customer's past purchase history indicates a preference for a certain type of product, the model will assign a higher weight to that feature, thereby improving prediction accuracy. Through this weighted fusion, the model can more comprehensively reflect customers' purchasing behavior and preferences.

[0057] Finally, based on the integrated features, the model generates preliminary predictions of customer preferences. This result not only provides data support for subsequent marketing strategies but also helps stores make more informed decisions regarding product recommendations and promotions. For example, if the model predicts that a customer is likely to be interested in a certain product category in the future, the store can prepare relevant promotions in advance to enhance the customer's shopping experience and satisfaction.

[0058] For the preliminary customer preference prediction results, an analysis method based on the explainability module is used, combined with customer behavior data and sales trends to generate the final customer behavior analysis report.

[0059] In this step, the model combines preliminary customer preference predictions with customer behavior data and sales trends, and uses the analytical methods of the explainability module to generate a final customer behavior analysis report. The introduction of the explainability module enables the model to provide explanations for the predictions, helping managers understand the reasons behind customer behavior. For example, the model can analyze why a customer prefers a specific brand, perhaps due to a promotional event or personal preference. In this way, managers can better understand customer needs and formulate appropriate marketing strategies.

[0060] The significance of this step lies in enhancing the practicality and guidance of customer behavior analysis reports through interpretable analysis. By combining customer behavior data with sales trends, the model can provide managers with in-depth insights, helping them develop more precise marketing strategies. For example, if a report reveals a customer's preference for a certain product category during a specific time period, the store can develop personalized promotions based on this information to enhance the customer's shopping experience and satisfaction. Furthermore, in-depth analysis results can help stores identify potential market opportunities, optimize their product mix and promotional strategies, and ultimately improve overall sales performance.

[0061] In this step, the initial customer preference predictions must be combined with customer behavior data and sales trends. The inclusion of an explainability module enables the model to provide explanations for the predictions, helping managers understand the underlying reasons behind customer behavior. For example, the model can analyze why a customer prefers a particular brand, perhaps due to a promotion or personal preference. This allows managers to better understand customer needs and formulate appropriate marketing strategies.

[0062] Next, the model generates a detailed customer behavior analysis report. This report includes information such as the customer's purchasing habits, preferred products, and purchase frequency, and provides explanations based on sales trends. For example, if the report reveals a customer's preference for a certain product category during a specific time period, the store can develop personalized promotions based on this information to enhance the customer's shopping experience and satisfaction. Furthermore, the report can provide predictions about changes in customer behavior, helping managers prepare for future responses.

[0063] Finally, managers can formulate marketing strategies based on customer behavior analysis reports. Through in-depth analysis, managers can identify potential market opportunities, optimize product mixes and promotional strategies, and ultimately boost overall sales performance. For example, if a report reveals a customer's preference for a particular product category within a specific time period, stores can prepare relevant promotional activities in advance to enhance customer shopping experience and satisfaction.

[0064] S203, based on the customer behavior analysis report and inventory data, using a deep reinforcement learning-based inventory management model, combined with sales trends and replenishment cycles, dynamically adjust inventory strategies, and generate intelligent replenishment plans by balancing inventory costs and sales demand through a multi-objective optimization algorithm; This approach first analyzes customer behavior reports to identify purchasing patterns and preferences. This information, combined with real-time sales data, predicts future sales trends. Inventory data also provides information on current inventory levels and replenishment cycles. A deep reinforcement learning model continuously learns from historical data and real-time feedback to optimize inventory management strategies, minimizing inventory costs and maximizing sales demand. Using a multi-objective optimization algorithm, the model balances multiple objectives, ensuring that customer needs are met while reducing inventory holding costs and the risk of stockouts.

[0065] The core significance of this approach lies in improving store operational efficiency and customer satisfaction through intelligent inventory management. By dynamically adjusting inventory strategies, stores can better respond to changes in market demand, reduce inventory overhangs and stockouts, and thus improve capital turnover and sales. Furthermore, inventory management models based on deep reinforcement learning can be continuously optimized. As data accumulates, the model's predictive capabilities and decision-making efficiency will continue to improve, helping stores maintain their advantage in a highly competitive market environment and ultimately achieve higher profitability and customer loyalty.

[0066] Specifically, we can use a deep reinforcement learning-based inventory management model based on customer behavior analysis reports and inventory data, combine sales trends and replenishment cycles, define inventory optimization goals, and generate a preliminary optimization problem model through a multi-objective optimization framework. In this step, customer behavior analysis reports must first be integrated with inventory data to provide comprehensive information for the inventory management model. By analyzing customer purchasing behavior, the model can identify which products are in high demand during specific time periods. Combined with inventory data, it can determine whether current inventory can meet future sales demand. Next, the inventory management model based on deep reinforcement learning defines inventory optimization objectives, such as maximizing sales, minimizing inventory costs, and reducing out-of-stock risk. Using a multi-objective optimization framework, the model transforms these objectives into a preliminary optimization problem model, laying the foundation for subsequent inventory policy adjustments. The significance of this step lies in improving the efficiency and accuracy of inventory management through scientific inventory optimization goal setting. By combining customer behavior analysis with inventory data, the model can better predict future sales trends and formulate more appropriate inventory strategies. This not only helps reduce inventory overstocks and out-of-stock situations, but also improves capital turnover, ultimately achieving higher sales and customer satisfaction.

[0067] At this stage, customer behavior analysis reports and inventory data must be integrated to provide comprehensive information for inventory management models. Customer behavior analysis reports typically include data such as customer purchase frequency, preferred products, and seasonal purchasing trends, while inventory data includes current inventory levels, replenishment cycles, and historical sales data. Through this data integration, the model can identify which products are in high demand during specific time periods. For example, if analysis reveals that sales of a certain brand of beverage increase significantly during the summer, the model will incorporate this information into the setting of inventory optimization targets.

[0068] Next, the deep reinforcement learning-based inventory management model defines multiple inventory optimization objectives. These might include maximizing sales, minimizing inventory costs, and reducing the risk of out-of-stocks. To achieve these goals, the model employs a multi-objective optimization framework, transforming them into a preliminary optimization problem model. For example, the model might set a goal of minimizing inventory holding costs while ensuring customer demand. This approach allows the model to balance multiple objectives, ensuring that diverse needs are properly balanced in practice.

[0069] Finally, the resulting preliminary optimization problem model lays the foundation for subsequent inventory policy adjustments. Through algorithmic analysis, the model prioritizes various objectives and assigns appropriate weights to each. For example, if inventory levels are low and sales are rising, the model may prioritize replenishing stock to meet customer demand, while if inventory is sufficient, it may prioritize reducing inventory costs. This dynamic adjustment allows the model to flexibly respond to customer demand and inventory management challenges in an ever-changing market environment.

[0070] For the preliminary optimization problem model, we use a policy generation method based on the policy gradient method, combining inventory status and sales trends to dynamically adjust the inventory policy. Through adaptive learning rate adjustment technology, we optimize the accuracy and stability of the policy and generate a preliminary inventory management policy. In this step, the policy gradient method is used to generate an inventory management policy based on the preliminary optimization problem model. The policy gradient method is a reinforcement learning technique that, by continuously iteratively optimizing the policy, enables the model to make more accurate decisions in a dynamic environment. The model dynamically adjusts the inventory policy based on current inventory status and sales trends. For example, when sales of a particular product increase, the model automatically increases the replenishment quantity of that product to avoid stock-outs. Furthermore, adaptive learning rate adjustment technology is used to ensure a stable and efficient policy optimization process, avoiding policy fluctuations caused by excessively large or small learning rates. The significance of this step lies in improving the flexibility and responsiveness of inventory management through dynamic inventory policy adjustment. By incorporating real-time inventory status and sales trends, the model can promptly adjust replenishment plans to ensure that inventory meets customer demand and reduce stock-outs and overstocks. Furthermore, the accuracy and stability of the optimization policy directly impacts inventory management effectiveness, thereby improving overall operational efficiency.

[0071] During this phase, a policy gradient method is used to generate an inventory management policy based on the preliminary optimization problem model. Policy gradient methods are a reinforcement learning technique that iteratively optimizes policies, enabling the model to make more accurate decisions in dynamic environments. Specifically, the model evaluates different inventory management strategies based on current inventory status and sales trends. For example, if sales of a particular product are increasing, the model may consider increasing the replenishment quantity of that product to avoid stock-outs.

[0072] Integrating inventory status, the model monitors inventory levels and sales data in real time and dynamically adjusts inventory strategies. For example, if a newly released electronic product experiences exceptionally strong sales during a given week, the model will automatically identify this trend and recommend increasing the restock quantity for that product. Furthermore, adaptive learning rate adjustment technology ensures a stable and efficient strategy optimization process. By dynamically adjusting the learning rate, the model can flexibly adjust the strategy update speed to accommodate varying market conditions, avoiding strategy fluctuations caused by excessively high or low learning rates.

[0073] Finally, after multiple iterations, the model generates a preliminary inventory management policy. This policy not only considers current inventory status and sales trends, but also incorporates learnings from historical data. For example, if the model discovers that a particular product generally sells well during a particular season, it will automatically adjust replenishment plans to ensure customer demand is met during peak periods. In this way, the model can make accurate inventory management decisions in a dynamic environment, improving overall operational efficiency.

[0074] For the initial inventory management strategy, a multi-objective optimization method based on the weighted summation method is used. The replenishment plan is dynamically adjusted based on inventory cost and sales demand. The dynamic weight allocation technology is used to balance the importance of different optimization objectives and generate a preliminary intelligent replenishment plan. In this step, a weighted summation method is used to perform multi-objective optimization based on the preliminary inventory management strategy. This method dynamically adjusts the replenishment plan by assigning weights to each optimization objective, taking into account inventory costs and sales demand. For example, if sales demand for a particular product is high, the model will assign a higher weight to that objective, thereby increasing the replenishment quantity of that product. Simultaneously, dynamic weight assignment technology enables the model to flexibly adjust the weights of various objectives based on real-time data changes, ensuring that customer needs are met while reducing inventory holding costs. The significance of this step lies in improving the scientific and rational nature of inventory management through multi-objective optimization. By dynamically adjusting the replenishment plan, the model can better balance inventory costs and sales demand, reduce inventory backlogs and stock-outs, and thus improve capital turnover and sales. Furthermore, the flexible weight assignment mechanism enables the model to adapt to market changes, ensuring the effectiveness and timeliness of inventory management strategies.

[0075] During this phase, a weighted summation approach is used to optimize multiple objectives based on the initial inventory management strategy. This approach dynamically adjusts replenishment plans by assigning weights to each optimization objective, taking into account inventory costs and sales demand. First, the model analyzes current inventory status and sales trends to identify products with high demand and low inventory levels. For example, if a popular product is nearing depletion, the model will prioritize replenishment for that product.

[0076] The model then assigns weights to each optimization objective, dynamically adjusting replenishment plans based on real-time sales data and inventory status. For example, if demand for a particular product increases significantly during a certain period, the model will automatically increase the replenishment quantity for that product and make adjustments within the replenishment cycle to ensure that inventory can meet customer demand. Furthermore, dynamic weighting technology enables the model to flexibly adjust the weights of various objectives based on real-time data changes. For example, during peak sales periods, the model may prioritize meeting customer demand, while during periods of stable sales, it may focus more on reducing inventory costs.

[0077] Finally, after multiple rounds of optimization, the model generates a preliminary intelligent replenishment plan. This plan not only considers current inventory levels and sales trends, but also incorporates learnings from historical data. For example, if the model discovers that a particular product generally sells well during a particular season, it will automatically adjust the replenishment plan to ensure customer demand is met during peak periods. In this way, the model can make accurate inventory management decisions in a dynamic environment, improving overall operational efficiency.

[0078] For the preliminary intelligent replenishment plan, a verification method based on simulation is adopted, combining inventory status and sales trends to verify the feasibility and effectiveness of the replenishment plan. Through feedback correction technology, the replenishment plan is dynamically adjusted to generate the final intelligent replenishment plan.

[0079] In this step, simulation technology is used to verify the preliminary intelligent replenishment plan based on the plan. By building a simulation model and combining it with current inventory status and sales trends, the model can simulate the performance of different replenishment plans in actual operations. For example, the model can simulate the impact of replenishment plans on inventory levels and customer satisfaction under different sales scenarios. In this way, managers can evaluate the feasibility and effectiveness of the replenishment plan to ensure that it achieves the expected results in actual application. The significance of this step is to verify the feasibility of the replenishment plan through simulation and reduce risks in actual operations. By verifying the replenishment plan, managers can promptly identify potential problems and make adjustments, thereby ensuring the effectiveness and reliability of the inventory management strategy. In addition, the application of feedback correction technology enables the model to dynamically adjust the replenishment plan based on the simulation results, improving the flexibility and adaptability of inventory management.

[0080] In this phase, simulation technology is used to validate the preliminary intelligent replenishment plan. First, a simulation model is constructed that combines current inventory status and sales trends to simulate the performance of different replenishment plans in real-world operations. For example, the model can simulate the impact of replenishment plans on inventory levels and customer satisfaction under different sales scenarios. This allows managers to assess the feasibility and effectiveness of the replenishment plan, ensuring that it will achieve the expected results in real-world applications.

[0081] The model then evaluates the feasibility and effectiveness of the replenishment plan and identifies potential issues. For example, if the model finds that a replenishment plan cannot meet customer demand under a specific sales scenario, managers can promptly adjust the replenishment plan to ensure that inventory can meet customer demand. Furthermore, the model dynamically adjusts the replenishment plan through feedback correction technology to ensure the flexibility and adaptability of the replenishment plan. For example, if the simulation finds that the replenishment quantity of a particular product is too high, resulting in inventory backlogs, the model will automatically adjust the replenishment plan to reduce the replenishment quantity of that product.

[0082] Finally, after multiple rounds of verification and adjustment, the final intelligent replenishment plan is generated. Based on the simulation results, managers can optimize replenishment strategies to ensure they achieve the desired results in actual operations. For example, if the model recommends increasing the replenishment quantity of a certain product during a specific time period, managers can implement this recommendation to improve customer satisfaction and sales. In this way, the model can make accurate inventory management decisions in a dynamic environment, improving overall operational efficiency.

[0083] S204: Input the customer behavior analysis report and the intelligent replenishment plan into an augmented reality-based visualization platform to generate a visualization chart of store operations. Through an interactive operation interface and multi-dimensional screening functions, managers are supported to monitor and adjust operation strategies in real time, thereby generating a final digital store management plan.

[0084] The customer behavior analysis report and the intelligent replenishment plan are input into an augmented reality-based visualization platform. This platform uses advanced visualization technology to transform complex data into easy-to-understand charts and information displays. The platform integrates key data from the customer behavior analysis report, such as customer preferences and purchasing trends, with the inventory management strategies from the intelligent replenishment plan in real time to generate visual charts of store operations. These charts not only display current sales and inventory levels but also dynamically update to reflect real-time market changes. Through the interactive interface, managers can easily view data from different dimensions for in-depth analysis and decision-making.

[0085] The core significance of this process lies in improving managers' understanding of store operations and their decision-making capabilities. Through visual charts, managers can quickly identify sales trends, customer preferences, and inventory status, allowing them to make timely adjustments and optimizations. For example, if a visual chart shows a sharp increase in sales of a particular product, managers can immediately take measures to increase replenishment of that product to avoid stock-outs. In addition, the interactive operation interface and multi-dimensional filtering function enable managers to flexibly adjust operational strategies based on different needs and scenarios, ultimately achieving higher operational efficiency and customer satisfaction.

[0086] Specifically, we can use an augmented reality-based visualization platform to generate initial visualization charts for customer behavior analysis reports and intelligent replenishment plans, combined with store operation data. Through conditional constraint technology, we can ensure the accuracy and readability of the charts and generate preliminary visualization charts. At this stage, the customer behavior analysis reports and intelligent replenishment solutions must first be integrated into an augmented reality-based visualization platform. This platform presents complex data graphically, helping managers more intuitively understand the underlying data. Using conditional constraints, the system ensures that the generated visualizations meet high standards of accuracy and readability. For example, if a product's sales data shows an anomaly, the system automatically flags that data point, allowing managers to quickly identify and take appropriate action.

[0087] Next, the platform generates initial visualizations using store operational data. These charts may include sales trends, inventory levels, and customer traffic heat maps. By combining these charts, managers gain a comprehensive understanding of store operations. For example, a sales trend chart can reveal sales fluctuations over a specific time period, while an inventory level chart can help managers determine whether restocking is necessary. This allows managers to quickly access critical information and make timely decisions.

[0088] Finally, the resulting preliminary visualizations provide a foundation for subsequent decision-making. Managers can conduct in-depth analysis based on these charts, identifying potential issues and developing appropriate strategies. For example, if a heat map shows a sharp drop in customer traffic during a certain period, managers can consider adjusting promotional strategies or optimizing merchandise displays to attract more customers. This data-driven decision-making approach can significantly improve store operational efficiency.

[0089] The significance of this step lies in using augmented reality technology to present complex data in an intuitive manner, helping managers better understand and analyze store operations. By generating accurate and easy-to-read visualizations, managers can quickly identify potential issues and take appropriate measures, thereby improving decision-making efficiency and accuracy. Furthermore, real-time visualization of data can enhance managers' sensitivity to market changes, enabling them to adjust operational strategies in a timely manner, thereby improving customer satisfaction and sales performance.

[0090] In practical implementation, the system first inputs customer behavior analysis reports and intelligent replenishment plans into an augmented reality visualization platform. This platform then uses advanced graphics processing technology to transform the data into visual charts. For example, sales data can be displayed as a bar chart or line graph, while inventory data can be presented as a pie chart or heat map. This allows managers to clearly see the changes in various indicators.

[0091] The system then applies conditional constraints to ensure the generated visualizations meet high standards of accuracy and readability. For example, if a product's sales volume falls below a preset threshold, the system automatically marks that product's portion of the chart in red to draw managers' attention. The system also offers data filtering capabilities, allowing managers to filter by time period, product category, or customer group, generating more targeted visualizations.

[0092] Finally, the generated preliminary visualizations are integrated into an interactive interface that managers can operate via touchscreens or other interactive devices. This allows managers to monitor store operations in real time and quickly respond based on the visualized data. For example, if inventory levels for a particular product are low, managers can immediately adjust replenishment plans to ensure a continuous supply. This dynamic visualization management significantly improves store operational efficiency and customer satisfaction.

[0093] For the preliminary visualization chart, we adopted a design method based on interactive operation interface, combined with the needs of managers, designed multi-dimensional screening functions and dynamic interactive operations, and optimized the interface layout and operation process through user behavior analysis to generate a preliminary interactive visualization solution; During this phase, the system further optimizes the initially generated visualizations, employing a design approach based on interactive interfaces. Through in-depth communication with managers, the system understands their specific needs during data analysis and designs an interactive interface that better suits their actual use cases. For example, a manager might want to quickly view sales data for a specific time period or conduct in-depth analysis of sales for a specific product category.

[0094] Next, the system will design a multi-dimensional filtering function, allowing managers to filter data based on different dimensions. These dimensions may include time, product category, customer group, etc. This way, managers can quickly obtain the information they need without having to browse through large amounts of data. For example, if a manager wants to view the sales of a specific product over the past month, they simply select the corresponding product and time period in the filtering interface, and the system will automatically update the visualization chart to display the relevant data.

[0095] Finally, the system optimizes interface layout and operational processes through user behavior analysis. By analyzing managers' usage habits, the system can identify which functions are most frequently used and which steps can be simplified. For example, if managers frequently need to view sales and inventory data, the system can place these two functions prominently on the interface for quick access. This optimization makes the interactive visualization solution more consistent with managers' usage habits and improves work efficiency.

[0096] The significance of this step lies in improving the manager's experience during data analysis by optimizing the interactive interface. By designing multi-dimensional filtering functions and dynamic interactive operations, managers can more easily access the information they need, thereby improving the efficiency and accuracy of decision-making. Furthermore, the optimized interface layout and operation process will reduce manager operation time, allowing them to focus more on strategic decision-making, further improving store operational efficiency.

[0097] In its implementation, the system first communicates with managers to understand their specific needs and usage habits during data analysis. Based on this information, the system designs an interactive interface that meets the manager's needs. For example, the interface might include multiple tabs displaying sales data, inventory data, and customer behavior data, allowing managers to quickly switch views by clicking on different tabs.

[0098] Next, the system implements a multi-dimensional filtering function, allowing managers to filter data based on different dimensions. For example, managers can select a specific time period, product category, or customer group, and the system will automatically update the visualization chart based on the selection to display the relevant data. This way, managers can quickly obtain the information they need without having to browse through large amounts of data.

[0099] Finally, the system optimizes interface layout and operational processes through user behavior analysis. By analyzing managers' usage habits, the system can identify which functions are most frequently used and which steps can be simplified. For example, if managers frequently need to view sales and inventory data, the system can place these two functions prominently on the interface for quick access. This optimization makes the interactive visualization solution more consistent with managers' usage habits and improves work efficiency.

[0100] For the preliminary interactive visualization solution, a real-time update method based on streaming computing and incremental update technology was adopted. In combination with real-time operational data, the visualization content was dynamically updated. Through real-time monitoring technology, the timeliness and operability of the data were ensured, and a preliminary real-time update visualization solution was generated. During this phase, the system will update the preliminary interactive visualization solution in real time to ensure data timeliness and operability. By utilizing streaming computing and incremental update technologies, the system can access store operational data in real time and dynamically update the visualization content. For example, if the sales data of a particular product changes within a short period of time, the system will immediately reflect this change in the visualization chart, ensuring that managers have timely access to the latest information.

[0101] Next, the system dynamically updates the visualizations based on real-time operational data. This real-time data might include sales, inventory levels, customer traffic, and more. By combining this data with the visualizations, managers can monitor store operations in real time. For example, if the inventory level of a particular item falls below a preset threshold, the system automatically marks the item in the visualization, alerting managers to restock it promptly.

[0102] Finally, the system uses real-time monitoring technology to ensure data timeliness and operability. By monitoring the system's operational status, managers can promptly identify potential issues and take appropriate measures. For example, if the system detects a failure in a data source, managers can immediately investigate and ensure data accuracy and completeness. This real-time update allows managers to respond quickly in a dynamic environment, improving store operational efficiency.

[0103] The significance of this step lies in ensuring that managers have timely access to the latest operational data through real-time updates of visualization content, enabling them to respond quickly. By integrating real-time operational data, managers can monitor store operations in real time, identify potential issues promptly, and take appropriate measures. Furthermore, the application of real-time monitoring technology will enhance managers' trust in the data, improve the accuracy and effectiveness of decision-making, and further enhance store operational efficiency.

[0104] In its implementation, the system first establishes a streaming computing framework to acquire real-time store operational data. This data may include sales, inventory levels, customer traffic, and more. Through streaming computing, the system processes data the instant it is generated and updates the results to the visualization charts in real time. For example, if sales of a particular product increase significantly within a short period of time, the system will immediately reflect this change in the visualization charts.

[0105] The system then dynamically updates the visualizations based on real-time operational data. By combining real-time data with visualizations, managers can monitor store operations in real time. For example, if the inventory level of a particular product falls below a preset threshold, the system automatically marks the item in the visualization, alerting managers to restock it promptly. This allows managers to quickly access critical information and make timely decisions.

[0106] Finally, the system uses real-time monitoring technology to ensure data timeliness and operability. By monitoring the system's operational status, managers can promptly identify potential issues and take appropriate measures. For example, if the system detects a failure in a data source, managers can immediately investigate and ensure data accuracy and completeness. This real-time update allows managers to respond quickly in a dynamic environment, improving store operational efficiency.

[0107] For the preliminary real-time update visualization solution, an optimization method based on user feedback is adopted, combined with manager interaction data and real-time performance monitoring, to dynamically adjust the visualization content, and generate the final digital store management solution through feedback correction technology.

[0108] During this phase, the system will further optimize the initial real-time visualization solution, using an optimization method based on user feedback. By collecting data on managers' interactions with the visualization platform, the system can identify which functions are most commonly used and which information is of greatest interest. For example, if a manager frequently views sales data for a particular product, the system can optimize the placement of that data for easier access. Furthermore, real-time performance monitoring technology will help the system assess the responsiveness and stability of the visualization platform, ensuring smooth operation even under high load.

[0109] Next, the system dynamically adjusts the visualization content based on the manager's interaction data and real-time performance monitoring results. This means the system can automatically adjust the presentation of charts and the priority of information based on the manager's usage habits and needs. For example, if customer traffic surges during a certain period, the system can automatically promote relevant sales data and inventory status to a prominent position in the visualization interface, allowing managers to respond quickly. Furthermore, the system can automatically update visualizations based on real-time data changes, ensuring that managers always have the latest information.

[0110] Finally, using feedback correction technology, the system generates a final digital store management plan. This plan not only takes into account manager feedback but also incorporates real-time data changes to ensure the accuracy and practicality of the visualization content. For example, if a manager suggests an improvement to a certain feature during use, the system will record this feedback and optimize it in subsequent versions. In this way, the final digital store management plan will better meet actual operational needs, improving the efficiency and accuracy of managers' decision-making.

[0111] The significance of this step lies in ensuring that the visualization platform continues to meet managers' needs through optimization methods based on user feedback. By combining manager interaction data with real-time performance monitoring, the system can dynamically adjust visualization content to improve information usability and readability. Furthermore, the application of feedback correction technology will strengthen managers' trust in the system, enabling them to more effectively use visualization tools for decision-making, thereby improving overall store operational efficiency.

[0112] In practice, the system first establishes a user feedback collection mechanism to record manager interaction data when using the visualization platform. This data may include the types of charts managers view, the frequency of their operations, and the level of attention they pay to specific information. For example, if a manager frequently checks the sales trends of a particular product, the system will mark the relevant data for that product as "high priority" and highlight it in the visualization interface.

[0113] The system then evaluates the visualization platform's responsiveness and stability, combined with real-time performance monitoring results. By monitoring the system's operational status, managers can promptly identify potential issues and take appropriate action. For example, if the system detects a delay in a data source, managers can immediately investigate to ensure data accuracy and integrity. Furthermore, the system automatically updates visualization charts based on real-time data changes, ensuring managers always have the latest information.

[0114] Finally, the system uses feedback correction technology to generate the final digital store management plan. Manager feedback is incorporated into the system's optimization process to ensure the accuracy and practicality of the visualization content. For example, if a manager suggests an improvement to a specific feature, the system will record this feedback and optimize it in subsequent versions. In this way, the final digital store management plan will better meet actual operational needs, improving the efficiency and accuracy of managers' decision-making.

[0115] It can be seen that based on the store's customer behavior data, sales data, inventory data and environmental data, multi-source data is acquired in real time to generate a time-synchronized multi-source data set; the multi-source data set is input into the behavior analysis model to extract the spatiotemporal pattern characteristics of customer behavior and generate a customer behavior analysis report; based on the customer behavior analysis report and inventory data, an intelligent replenishment plan is generated; the customer behavior analysis report and the intelligent replenishment plan are input into the augmented reality-based visualization platform to generate a visualization chart of the store operation, and through the interactive operation interface and multi-dimensional screening function, the final digital store management plan is generated, thereby providing an efficient, intelligent and flexible store management solution, realizing the full process intelligence of data collection, behavior analysis, inventory optimization and visualization decision-making.

[0116] Another embodiment of the present invention provides a digital store management system, see Figure 3 , the system may include: Acquisition module 301 is used to acquire multi-source data in real time based on the store's customer behavior data, sales data, inventory data, and environmental data using an edge computing-based data acquisition framework. It then uses an adaptive data fusion algorithm to convert data from different sources into a standardized data format, generating a time-synchronized multi-source data set. Extraction module 302 is configured to input the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, combine historical sales data, predict customer preferences, and generate a customer behavior analysis report. Generation module 303 is used to adopt an inventory management model based on deep reinforcement learning based on the customer behavior analysis report and inventory data, combine sales trends and replenishment cycles, dynamically adjust inventory strategies, and generate intelligent replenishment plans by balancing inventory costs and sales demand through a multi-objective optimization algorithm; Management module 304 is used to input the customer behavior analysis report and the intelligent replenishment plan into the augmented reality-based visualization platform to generate a visualization chart of store operations. Through the interactive operation interface and multi-dimensional screening function, it supports managers to monitor and adjust operation strategies in real time and generate the final digital store management plan.

[0117] It can be seen that based on the store's customer behavior data, sales data, inventory data and environmental data, multi-source data is acquired in real time to generate a time-synchronized multi-source data set; the multi-source data set is input into the behavior analysis model to extract the spatiotemporal pattern characteristics of customer behavior and generate a customer behavior analysis report; based on the customer behavior analysis report and inventory data, an intelligent replenishment plan is generated; the customer behavior analysis report and the intelligent replenishment plan are input into the augmented reality-based visualization platform to generate a visualization chart of the store operation, and through the interactive operation interface and multi-dimensional screening function, the final digital store management plan is generated, thereby providing an efficient, intelligent and flexible store management solution, realizing the full process intelligence of data collection, behavior analysis, inventory optimization and visualization decision-making.

[0118] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.

[0119] Specifically, in this embodiment, the above-mentioned storage medium may be configured to store a computer program for performing the following steps: S201 uses an edge computing-based data collection framework to acquire multi-source data in real time based on the store's customer behavior data, sales data, inventory data, and environmental data. Using an adaptive data fusion algorithm, the data from different sources is uniformly converted into a standardized data format to generate a time-synchronized multi-source data set. S202: Inputting the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, and combines historical sales data to predict customer preferences and generate a customer behavior analysis report. S203, based on the customer behavior analysis report and inventory data, using a deep reinforcement learning-based inventory management model, combined with sales trends and replenishment cycles, dynamically adjust inventory strategies, and generate intelligent replenishment plans by balancing inventory costs and sales demand through a multi-objective optimization algorithm; S204: Input the customer behavior analysis report and the intelligent replenishment plan into an augmented reality-based visualization platform to generate a visualization chart of store operations. Through an interactive operation interface and multi-dimensional screening functions, managers are supported to monitor and adjust operation strategies in real time, thereby generating a final digital store management plan.

[0120] It can be seen that based on the store's customer behavior data, sales data, inventory data and environmental data, multi-source data is acquired in real time to generate a time-synchronized multi-source data set; the multi-source data set is input into the behavior analysis model to extract the spatiotemporal pattern characteristics of customer behavior and generate a customer behavior analysis report; based on the customer behavior analysis report and inventory data, an intelligent replenishment plan is generated; the customer behavior analysis report and the intelligent replenishment plan are input into the augmented reality-based visualization platform to generate a visualization chart of the store operation, and through the interactive operation interface and multi-dimensional screening function, the final digital store management plan is generated, thereby providing an efficient, intelligent and flexible store management solution, realizing the full process intelligence of data collection, behavior analysis, inventory optimization and visualization decision-making.

[0121] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0122] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0123] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program: S201 uses an edge computing-based data collection framework to acquire multi-source data in real time based on the store's customer behavior data, sales data, inventory data, and environmental data. Using an adaptive data fusion algorithm, the data from different sources is uniformly converted into a standardized data format to generate a time-synchronized multi-source data set. S202: Inputting the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, and combines historical sales data to predict customer preferences and generate a customer behavior analysis report. S203, based on the customer behavior analysis report and inventory data, using a deep reinforcement learning-based inventory management model, combined with sales trends and replenishment cycles, dynamically adjust inventory strategies, and generate intelligent replenishment plans by balancing inventory costs and sales demand through a multi-objective optimization algorithm; S204: Input the customer behavior analysis report and the intelligent replenishment plan into an augmented reality-based visualization platform to generate a visualization chart of store operations. Through an interactive operation interface and multi-dimensional screening functions, managers are supported to monitor and adjust operation strategies in real time, thereby generating a final digital store management plan.

[0124] It can be seen that based on the store's customer behavior data, sales data, inventory data and environmental data, multi-source data is acquired in real time to generate a time-synchronized multi-source data set; the multi-source data set is input into the behavior analysis model to extract the spatiotemporal pattern characteristics of customer behavior and generate a customer behavior analysis report; based on the customer behavior analysis report and inventory data, an intelligent replenishment plan is generated; the customer behavior analysis report and the intelligent replenishment plan are input into the augmented reality-based visualization platform to generate a visualization chart of the store operation, and through the interactive operation interface and multi-dimensional screening function, the final digital store management plan is generated, thereby providing an efficient, intelligent and flexible store management solution, realizing the full process intelligence of data collection, behavior analysis, inventory optimization and visualization decision-making.

[0125] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. A digital store management method, characterized in that: The method comprises: Based on the store's customer behavior data, sales data, inventory data, and environmental data, an edge computing-based data collection framework is used to obtain multi-source data in real time. Through an adaptive data fusion algorithm, data from different sources is uniformly converted into a standardized data format to generate a time-synchronized multi-source data set. Inputting the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, and combines historical sales data to predict customer preferences and generate a customer behavior analysis report. Based on the customer behavior analysis report and inventory data, a deep reinforcement learning-based inventory management model is used to dynamically adjust inventory strategies based on sales trends and replenishment cycles. Through a multi-objective optimization algorithm, inventory costs and sales demand are balanced to generate intelligent replenishment plans. The customer behavior analysis report and the intelligent replenishment plan are input into a visualization platform based on augmented reality to generate a visualization chart of store operations. Through an interactive operation interface and multi-dimensional screening functions, managers are supported to monitor and adjust operation strategies in real time and generate the final digital store management plan.

2. The method according to claim 1, characterized in that Based on the store's customer behavior data, sales data, inventory data, and environmental data, the edge computing-based data collection framework is used to acquire multi-source data in real time. Through an adaptive data fusion algorithm, data from different sources is uniformly converted into a standardized data format to generate a time-synchronized multi-source data set, including: Based on the store's customer behavior data, sales data, inventory data, and environmental data, an edge computing-based data collection framework is adopted to obtain multi-source data in real time through distributed edge nodes. Each edge node is equipped with lightweight data caching technology to ensure the real-time and continuity of data collection. For the collected multi-source data, a deep learning-based format recognition model is used to automatically identify the format types of different data sources. An adaptive data cleaning algorithm is used to filter noise and fill missing values in the data to generate a preliminary standardized data set. For the preliminary standardized data set, a multi-source data synchronization algorithm based on timestamp alignment is used to eliminate the time differences between different data sources. The missing data is supplemented by interpolation and filling methods to generate a time-synchronized multi-source data set. For time-synchronized multi-source data sets, a standardized method based on an adaptive data fusion algorithm is adopted to uniformly convert data from different sources into a standardized data format. Through dynamic weight adjustment technology, the accuracy and consistency of data fusion are ensured to generate the final multi-source data set.

3. The method according to claim 2, characterized in that The multi-source dataset is input into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, combines historical sales data, predicts customer preferences, and generates a customer behavior analysis report, including: For multi-source datasets, we use a behavior analysis model based on a spatiotemporal graph convolutional network. We abstract customers, products, time, and space into nodes in a graph structure, and abstract the spatiotemporal relationships between nodes into edges. We then generate a preliminary spatiotemporal graph structure using dynamic graph structure learning technology. For the preliminary spatiotemporal graph structure, a feature extraction method based on multi-scale convolution kernels is used. This method combines the local details and global trends of customer behavior to extract multi-scale features. Adaptive convolution kernel adjustment technology is used to capture features at different scales and generate a preliminary multi-scale feature representation. For the preliminary multi-scale feature representation, a feature fusion method based on multi-head attention technology is used to perform weighted fusion of features of different scales. Customer preferences are predicted based on historical sales data to generate preliminary customer preference prediction results. For the preliminary customer preference prediction results, an analysis method based on the explainability module is used, combined with customer behavior data and sales trends to generate the final customer behavior analysis report.

4. The method according to claim 3, characterized in that Based on the customer behavior analysis report and inventory data, the inventory management model based on deep reinforcement learning is adopted. In combination with sales trends and replenishment cycles, the inventory strategy is dynamically adjusted. Through a multi-objective optimization algorithm, inventory costs and sales demand are balanced to generate an intelligent replenishment plan, including: Based on customer behavior analysis reports and inventory data, we adopted an inventory management model based on deep reinforcement learning. We combined sales trends and replenishment cycles to define inventory optimization goals and generated a preliminary optimization problem model using a multi-objective optimization framework. For the preliminary optimization problem model, we use a policy generation method based on the policy gradient method, combining inventory status and sales trends to dynamically adjust the inventory policy. Through adaptive learning rate adjustment technology, we optimize the accuracy and stability of the policy and generate a preliminary inventory management policy. For the initial inventory management strategy, a multi-objective optimization method based on the weighted summation method is used. The replenishment plan is dynamically adjusted based on inventory cost and sales demand. The dynamic weight allocation technology is used to balance the importance of different optimization objectives and generate a preliminary intelligent replenishment plan. For the preliminary intelligent replenishment plan, a verification method based on simulation is adopted, combining inventory status and sales trends to verify the feasibility and effectiveness of the replenishment plan. Through feedback correction technology, the replenishment plan is dynamically adjusted to generate the final intelligent replenishment plan.

5. The method according to claim 4, characterized in that The customer behavior analysis report and the intelligent replenishment plan are input into the augmented reality-based visualization platform to generate a visualization chart of store operations. Through the interactive operation interface and multi-dimensional screening function, managers can monitor and adjust operation strategies in real time and generate the final digital store management plan, including: For customer behavior analysis reports and intelligent replenishment plans, we use an augmented reality-based visualization platform, combined with store operation data, to generate initial visualization charts. Conditional constraint technology is used to ensure the accuracy and readability of the charts and generate preliminary visualization charts. For the preliminary visualization chart, we adopted a design method based on interactive operation interface, combined with the needs of managers, designed multi-dimensional screening functions and dynamic interactive operations, and optimized the interface layout and operation process through user behavior analysis to generate a preliminary interactive visualization solution; For the preliminary interactive visualization solution, a real-time update method based on streaming computing and incremental update technology was adopted. In combination with real-time operational data, the visualization content was dynamically updated. Through real-time monitoring technology, the timeliness and operability of the data were ensured, and a preliminary real-time update visualization solution was generated. For the preliminary real-time update visualization solution, an optimization method based on user feedback is adopted, combined with manager interaction data and real-time performance monitoring, to dynamically adjust the visualization content, and generate the final digital store management solution through feedback correction technology.

6. A digital store management system, characterized in that: The system comprises: The acquisition module is used to acquire multi-source data in real time based on the store's customer behavior data, sales data, inventory data, and environmental data using an edge computing-based data acquisition framework. Using an adaptive data fusion algorithm, it converts data from different sources into a standardized data format to generate a time-synchronized multi-source data set. An extraction module is configured to input the multi-source dataset into a behavior analysis model based on a spatiotemporal graph convolutional network to extract spatiotemporal pattern features of customer behavior. The behavior analysis model uses dynamic graph structure learning and multi-scale feature fusion technology to capture local details and global trends of customer behavior, combine historical sales data, predict customer preferences, and generate a customer behavior analysis report. A generation module is used to use a deep reinforcement learning-based inventory management model based on the customer behavior analysis report and inventory data, dynamically adjust inventory strategies based on sales trends and replenishment cycles, and generate intelligent replenishment plans by balancing inventory costs and sales demand through a multi-objective optimization algorithm; The management module is used to input the customer behavior analysis report and the intelligent replenishment plan into the augmented reality-based visualization platform to generate a visualization chart of store operations. Through the interactive operation interface and multi-dimensional screening function, it supports managers to monitor and adjust operation strategies in real time and generate the final digital store management plan.

7. The system according to claim 6, characterized in that The acquisition module is specifically used to: Based on the store's customer behavior data, sales data, inventory data, and environmental data, an edge computing-based data collection framework is adopted to obtain multi-source data in real time through distributed edge nodes. Each edge node is equipped with lightweight data caching technology to ensure the real-time and continuity of data collection. For the collected multi-source data, a deep learning-based format recognition model is used to automatically identify the format types of different data sources. An adaptive data cleaning algorithm is used to filter noise and fill missing values in the data to generate a preliminary standardized data set. For the preliminary standardized data set, a multi-source data synchronization algorithm based on timestamp alignment is used to eliminate the time differences between different data sources. The missing data is supplemented by interpolation and filling methods to generate a time-synchronized multi-source data set. For time-synchronized multi-source data sets, a standardized method based on an adaptive data fusion algorithm is adopted to uniformly convert data from different sources into a standardized data format. Through dynamic weight adjustment technology, the accuracy and consistency of data fusion are ensured to generate the final multi-source data set.

8. The system according to claim 7, characterized in that The extraction module is specifically used to: For multi-source datasets, we use a behavior analysis model based on a spatiotemporal graph convolutional network. We abstract customers, products, time, and space into nodes in a graph structure, and abstract the spatiotemporal relationships between nodes into edges. We then generate a preliminary spatiotemporal graph structure using dynamic graph structure learning technology. For the preliminary spatiotemporal graph structure, a feature extraction method based on multi-scale convolution kernels is used. This method combines the local details and global trends of customer behavior to extract multi-scale features. Adaptive convolution kernel adjustment technology is used to capture features at different scales and generate a preliminary multi-scale feature representation. For the preliminary multi-scale feature representation, a feature fusion method based on multi-head attention technology is used to perform weighted fusion of features of different scales. Customer preferences are predicted based on historical sales data to generate preliminary customer preference prediction results. For the preliminary customer preference prediction results, an analysis method based on the explainability module is used, combined with customer behavior data and sales trends to generate the final customer behavior analysis report.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 5 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 5.

Images