An AI agent-based unmanned retail cabinet full-link operation management method

Abstract

The application relates to an AI intelligent agent-based unmanned retail cabinet full-link operation management method, and belongs to the technical field of unmanned retail and artificial intelligence, and comprises the following steps: S1, real-time collection of multi-source heterogeneous operation data: real-time collection of multi-source heterogeneous operation data; S2, intelligent sensing of operation states and understanding of situations; S3, AI intelligent agent full-link collaborative decision-making; S4, automatic execution and issuing of decision instructions; S5, execution effect monitoring and closed-loop feedback: real-time monitoring of the execution effect of the decision instructions, feedback of the execution result to the AI intelligent agent, and formation of a full-link automatic closed loop of "sensing-> decision-making-> execution-> feedback". The AI intelligent agent-based unmanned retail cabinet full-link operation management method, by constructing a unified AI intelligent agent as an operation center, integrates replenishment management, inventory management, commodity identification, equipment operation and maintenance, marketing strategies and other links into a unified decision-making framework, and realizes information sharing and collaborative optimization of each link.

Description

Technical Field

[0001] This invention relates to the fields of unmanned retail and artificial intelligence technology, specifically to a full-chain operation management method for unmanned retail cabinets based on AI intelligent agents, applicable to the entire chain operation scenarios of unmanned retail cabinets, including intelligent replenishment, product recognition, inventory management, sales forecasting, and equipment maintenance. Background Technology

[0002] With the rapid development of artificial intelligence, the Internet of Things, and mobile payment technologies, unmanned retail kiosks, as a new retail format, have been widely used in public places such as office buildings, residential communities, transportation hubs, and schools. Unmanned retail kiosks, operating 24 / 7 without human intervention, significantly reduce labor costs and meet consumers' needs for immediate and convenient shopping. However, the actual operation of unmanned retail kiosks involves multiple stages, including product replenishment, inventory management, sales forecasting, product identification, and equipment maintenance. These stages are interconnected and mutually influential, making the complexity of operation and management far higher than that of traditional retail.

[0003] The existing unmanned retail locker operation and management technologies have the following main shortcomings: First, the various operational processes are fragmented and lack overall collaborative optimization. In the existing system, replenishment decisions, inventory management, product identification, and equipment maintenance are mostly handled by independent modules or manually, lacking effective information sharing and collaborative decision-making mechanisms. For example, replenishment plans do not fully consider sales forecasts and inventory turnover, leading to untimely or excessive replenishment; equipment troubleshooting and product replenishment are independent of each other, resulting in low operational efficiency.

[0004] Secondly, decision-making relies heavily on human experience, resulting in a low level of intelligence. Existing systems still primarily depend on human experience for key operational decisions such as replenishment timing, quantity, and product display, lacking data-driven intelligent decision support. Differences in experience among different operational personnel lead to inconsistent decision quality, making it difficult to adapt to complex scenarios with a wide variety of products and frequent sales fluctuations.

[0005] Third, there is a lack of end-to-end automated closed-loop management capabilities. Existing technologies have achieved a certain degree of automation in individual links (such as automatic inventory and automatic payment), but have not yet formed a complete automated closed loop from status perception and intelligent decision-making to execution feedback. When abnormal situations such as stockouts, equipment failures, or misplaced goods occur, a large amount of manual intervention is still required.

[0006] Fourth, it lacks self-learning and self-optimization capabilities. Existing systems cannot automatically optimize decision-making strategies based on historical operational data. As operating time progresses, the effectiveness of these decisions cannot continuously improve, making it difficult to adapt to dynamic environments such as changes in product structure, seasonal fluctuations, and evolving consumer habits. Therefore, there is an urgent need for a solution that can achieve intelligent, collaborative, and automated operation management of unmanned retail cabinets across the entire chain. This invention aims to provide a method for the full-chain operation management of unmanned retail cabinets based on AI intelligent agents, in order to solve the problems of fragmented processes, low levels of intelligence, and lack of self-optimization capabilities in existing technologies. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a full-chain operation management method for unmanned retail cabinets based on AI intelligent agents. It has the advantages of full-chain intelligent collaboration, data-driven decision-making, and closed-loop automatic optimization, and solves the problems of fragmented operation links and lack of collaboration, reliance on human experience and low level of intelligence, and lack of end-to-end closed-loop management capabilities in existing technologies.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for end-to-end operation and management of unmanned retail cabinets based on AI intelligent agents includes the following steps: S1. Real-time collection of multi-source heterogeneous operational data: Real-time collection of multi-source heterogeneous operational data through sensor networks deployed in unmanned retail cabinets and external data interfaces; S2. Intelligent perception and situational understanding of operational status: The multi-source heterogeneous operational data is input into the perception module of the AI ​​agent to perform data fusion and status recognition, and output multi-dimensional operational situational perception results. S3. AI Intelligent Agent Full-Link Collaborative Decision-Making: Based on the multi-dimensional operational situation perception results, the AI ​​intelligent agent's decision-making module generates a collaborative decision-making scheme that includes replenishment decisions, product management decisions, equipment operation and maintenance decisions, and marketing strategy decisions; S4. Automatic execution and issuance of decision instructions: The collaborative decision-making scheme is converted into executable instructions and automatically issued to the corresponding execution unit of the unmanned retail cabinet; S5. Execution effect monitoring and closed-loop feedback: Real-time monitoring of the execution effect of decision instructions, and feedback of the execution results to the AI ​​agent, forming a full-link automatic closed loop of "perception → decision → execution → feedback".

[0009] Furthermore, the AI ​​agent includes a perception module, a decision-making module, and an execution scheduling module; the perception module is used to perform fusion analysis and state recognition on multi-source heterogeneous data; the decision-making module includes a replenishment decision submodule, an inventory optimization submodule, an equipment operation and maintenance submodule, and a marketing strategy submodule; the execution scheduling module is used to convert decision schemes into execution instructions and perform task scheduling.

[0010] Furthermore, the operational data in S1 includes product inventory data, sales transaction data, equipment operating status data, product image data, environmental perception data, and external environment data.

[0011] Furthermore, the multi-dimensional operational situational awareness results in S2 include at least: product shortage warning level, inventory health score, equipment health score, sales anomaly detection results, and replenishment demand urgency score.

[0012] Furthermore, the replenishment decision in S3 includes determining the replenishment timing, generating a replenishment product list, and calculating the replenishment quantity; the product management decision includes optimizing product display and suggesting the elimination of slow-moving products; the equipment operation and maintenance decision includes triggering equipment self-checks and predicting faults; and the marketing strategy decision includes dynamic pricing and generating promotional plans.

[0013] Furthermore, the AI ​​agent employs a learning mechanism that combines reinforcement learning and deep neural networks, training decision-making strategies using historical operational data and continuously learning and optimizing during actual operation.

[0014] Furthermore, in S4, for unmanned retail cabinets equipped with replenishment robots, the replenishment decision automatically generates replenishment robot path planning and replenishment operation instructions; for unmanned retail cabinets not equipped with replenishment robots, the replenishment decision pushes replenishment guidance to the operators via mobile terminals.

[0015] Furthermore, the method also includes an active handling step for abnormal events: when the AI ​​agent detects equipment failure, product stagnation, or payment anomalies, it automatically executes a preset emergency handling strategy and evaluates the handling effect after the handling is completed.

[0016] Furthermore, the method also includes an automatic operation report generation step: the AI ​​agent automatically generates an operation analysis report according to a preset cycle, the report including sales analysis, inventory analysis, equipment operation analysis and optimization suggestions.

[0017] Furthermore, the execution unit includes a replenishment robot or replenishment guidance terminal, a product identification system, an equipment self-inspection and repair system, and a pricing and promotion system.

[0018] Compared with existing technologies, this invention provides a full-chain operation and management method for unmanned retail cabinets based on AI intelligent agents, which has the following beneficial effects: 1. This AI-based full-chain operation and management method for unmanned retail cabinets constructs a unified AI agent as the operation hub, incorporating replenishment management, inventory management, product identification, equipment maintenance, marketing strategies, and other aspects into a unified decision-making framework. This enables information sharing and collaborative optimization across all aspects. The AI ​​agent makes decisions based on global optimization rather than local optimization, effectively avoiding the operational efficiency losses caused by the fragmentation of various aspects in the traditional model, and significantly improving the overall operational efficiency of unmanned retail cabinets.

[0019] 2. This AI-based unmanned retail cabinet full-chain operation management method automatically identifies the operational status and generates the optimal decision-making solution through real-time fusion analysis of multi-source heterogeneous operational data by the AI ​​agent. This transforms the traditional decision-making mode that relies on human experience into a data-driven intelligent decision-making mode. Key operational decisions such as replenishment timing, replenishment quantity, product display, and dynamic pricing are all automatically completed by the AI ​​agent. Operation personnel only need to perform replenishment operations and handle abnormal situations that the system cannot handle automatically, which greatly reduces labor costs and decision-making bias.

[0020] 3. This AI-based unmanned retail cabinet full-chain operation management method adopts a learning mechanism that combines reinforcement learning and deep neural networks. The AI ​​agent continuously collects execution feedback data during actual operation, and gradually improves decision-making quality through continuous trial and error and optimization. As the operation time extends, key indicators such as replenishment accuracy, inventory turnover rate, equipment utilization rate and sales conversion rate of the system continue to improve, realizing the automatic evolution of operation management effect and effectively adapting to dynamic environments such as changes in product structure, seasonal fluctuations and changes in consumption habits. Attached Figure Description

[0021] Figure 1 This is an overall flowchart of the whole-chain operation and management method of unmanned retail cabinet based on AI intelligent agent according to the present invention.

[0022] Figure 2 This is an AI agent module architecture diagram of an AI agent-based full-link operation management method for unmanned retail cabinets according to the present invention.

[0023] Figure 3 This is a schematic diagram of the closed-loop management data flow of the whole-link operation management method for unmanned retail cabinets based on AI intelligent agents according to the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1 to 3 .

[0026] Example 1: System Overall Architecture and AI Agent Deployment This embodiment provides a full-chain operation and management method for unmanned retail cabinets based on AI intelligent agents. It is deployed on the intelligent operation and management platform of a large chain unmanned retail enterprise to manage 500 unmanned retail cabinets distributed in various areas of the city.

[0027] System overall architecture: The system consists of four parts: the terminal layer, the data layer, the intelligent decision-making layer, and the execution layer. The terminal layer includes a sensor network (weight sensors, photoelectric sensors, cameras, temperature and humidity sensors, current sensors, etc.) deployed in each unmanned retail cabinet, intelligent product channels, payment modules, and communication modules. The data layer is responsible for the aggregation, cleaning, standardized storage, and real-time streaming processing of multi-source heterogeneous data. The intelligent decision-making layer, or AI agent, runs on a cloud server cluster (configured with multi-core CPUs, GPU accelerator cards, and large-capacity storage) and is the core decision-making hub of the system. The execution layer includes replenishment robots, automatic replenishment guidance terminals, a product recognition system, an equipment self-checking system, and a price display system.

[0028] Module composition of AI intelligent agents: AI agents comprise the following core modules: (1) Perception Module: Responsible for the fusion analysis and status identification of multi-source heterogeneous data. The perception module has built-in a commodity image recognition model based on convolutional neural network (CNN) (used to analyze commodity display images in the cabinet and identify out-of-stock locations and misplaced commodities), a sales trend prediction model based on long short-term memory network (LSTM) (used to predict sales in the next 24 hours), and an anomaly detection model based on statistical model (used to identify sales anomalies and equipment anomalies). The perception module completes a full status scan every 5 minutes and outputs the commodity out-of-stock warning level (red / yellow / green three levels), inventory health score (0-100 points), equipment health score (0-100 points), and replenishment demand urgency score (0-100 points) for each cabinet and each aisle.

[0029] (2) Decision Module: This module includes a replenishment decision submodule, an inventory optimization submodule, an equipment operation and maintenance submodule, and a marketing strategy submodule. The decision module is based on a deep reinforcement learning (DRL) framework. It takes the operational status output by the perception module as input and aims to minimize overall operating costs and maximize sales revenue to generate collaborative decision-making solutions.

[0030] (3) Execution scheduling module: responsible for breaking down the decision scheme into executable instructions and sending them to each execution unit through the API interface, while tracking the execution progress and status of the instructions.

[0031] Example 2: Specific Implementation of Replenishment Decisions This embodiment details the complete process of an AI agent generating replenishment decisions.

[0032] Status Awareness: At 8:00 AM on a certain workday, the sensing module performed a routine status scan of an unmanned retail kiosks in an office building and identified the following statuses: In Warehouse A (beverages, 12 SKUs), 3 SKUs had inventory below the safety stock level (red warning level); in Warehouse B (snacks, 8 SKUs), 1 SKU had inventory below the safety stock level (yellow warning level); the overall replenishment urgency score was 78 points (out of 100, a score above 60 triggers a replenishment decision). The sensing module also acquired the weather data for the area that day (high temperature 35℃), holiday data (non-holidays), and historical sales data (beverage sales at this kiosks increased by approximately 40% compared to normal days under similar high-temperature weather conditions).

[0033] Intelligent Decision Making: The replenishment decision submodule of the decision-making module performs decision reasoning based on the above perception results and in conjunction with a reinforcement learning policy network. Replenishment timing determined: The decision module judges that the urgency of the current replenishment demand (78 points) has exceeded the trigger threshold (60 points), and the current time (8:00 am) is a low customer flow period, which is suitable for replenishment and will not significantly affect sales. The replenishment timing is determined to be 8:00 am to 9:00 am on the same day.

[0034] Replenishment item list generation: The decision module generates a replenishment item list based on the stockout warning level (red priority), inventory health (the extent to which it is below the safety stock) and sales forecast (a 40% increase in beverage sales during hot weather). The replenishment items are: 12 units of each of the 3 beverage SKUs (8 units of regular replenishment + 4 units of additional replenishment due to high temperature), and 6 units of 1 snack SKU (regular replenishment quantity).

[0035] Intelligent replenishment quantity calculation: The replenishment quantity is calculated by the reinforcement learning policy network based on the following factors: current inventory level, historical sales rate, predicted sales volume for the next 24 hours, safety stock threshold, product shelf life, channel capacity, and replenishment cost. In this embodiment, after tens of thousands of simulation training sessions, the reinforcement learning policy network has learned to proactively increase beverage replenishment during hot weather, effectively reducing the risk of stockouts.

[0036] Execution scheduling: The execution scheduling module transforms replenishment decisions into replenishment guidelines: For sites equipped with replenishment robots, it generates robot path planning and retrieval instructions; for sites without robots (in this embodiment), it pushes replenishment task orders through the operator's mobile APP. The task order includes: cabinet number, cargo channel number, product name, suggested replenishment quantity, estimated replenishment time, and optimal replenishment route.

[0037] Feedback on Results: After replenishment is completed, the perception module rescans and confirms the inventory status, feeding back the replenishment results (actual replenishment quantity, replenishment time, and improvement in stockout status) to the decision-making module for subsequent optimization of the reinforcement learning strategy network. Statistics show that after implementing AI-powered replenishment decision-making, the stockout rate of this cabinet decreased from 8.5% to 2.1%, inventory turnover increased by 35%, and sales losses due to stockouts decreased by approximately 72%.

[0038] Example 3: Specific Implementation of Product Identification and Anomaly Detection This embodiment details the application of AI agents in product recognition and anomaly detection.

[0039] Product recognition process: The unmanned retail cabinet is equipped with four high-definition wide-angle cameras, which automatically capture images of the product display inside the cabinet each time the door closes (4-8 images per cabinet per session). The images are uploaded to the cloud in real time via 4G / 5G network, where the AI ​​agent's perception module performs image analysis. Image preprocessing: Distortion correction, illumination correction, and size normalization are performed on the uploaded image.

[0040] Object detection: A YOLOv8-based object detection model is used to identify the location and category of goods in the image. The model has been trained on a product dataset containing more than 100,000 labeled images, and the recognition accuracy reaches 98.5%.

[0041] Quantity verification: The identification results are compared with the inventory data recorded by the system item by item to identify out-of-stock locations, misplaced goods, and abnormal goods.

[0042] Anomaly detection process: The AI ​​agent simultaneously performs multi-dimensional anomaly detection: (1) Sales Anomaly Detection: The sales forecasting model based on LSTM predicts the expected sales volume of each product. When the deviation between the actual sales volume and the predicted value exceeds 3 standard deviations, a sales anomaly alarm is triggered. For example, if the actual sales volume of a certain beverage drops by 65% ​​compared to the predicted value on a certain day, the system will automatically mark it as a "sales anomaly". After analysis, it is found that the product was misplaced in the lower aisle (visual blind spot). The AI ​​agent will automatically generate a "product display adjustment" suggestion and push it to the operations staff.

[0043] (2) Equipment Anomaly Detection: By analyzing data from current sensors, temperature and humidity sensors, and network status, the AI ​​agent monitors the operating status of the refrigeration system, payment module, and communication module in real time. When an abnormal increase in power consumption of a cabinet's refrigeration system is detected (25% higher than the normal operating value and lasting for more than 2 hours), the system determines it as a "premonition of refrigeration equipment anomaly," automatically triggers a self-test command, and pushes a "refrigeration equipment maintenance reminder" (including equipment number, abnormal parameters, suggested maintenance time, and maintenance steps) to the maintenance personnel. After implementing AI agent equipment anomaly detection, the early detection rate of equipment failures has increased from less than 10% to over 75%.

[0044] (3) Payment anomaly detection: When multiple consecutive payment failures or payment timeouts are detected, the system automatically judges the status of the payment module. If the payment module is confirmed to be abnormal, the system automatically switches the cabinet to the "scan code payment only" mode and pushes a maintenance work order. At the same time, it sends temporary goods transfer suggestions to other nearby cabinets to disperse the flow of customers.

[0045] Example 4: Specific Implementation of Dynamic Pricing and Marketing Strategies This embodiment details the application of AI agents in marketing strategy decision-making.

[0046] Dynamic pricing mechanism: The AI ​​agent's marketing strategy submodule automatically adjusts product prices based on real-time sales data, inventory levels, competitor pricing, and external environmental data (weather, holidays, etc.). When the inventory of a product is higher than the safety stock limit and the sales rate in the past 7 days is less than 60% of the historical average, the system judges that the product has a risk of slow sales and automatically triggers a price reduction promotion strategy (the price reduction range is 5%-15%, dynamically calculated based on inventory pressure and sales elasticity).

[0047] When the inventory of a certain product is below the safety stock limit and the replenishment cost is high, the system determines that the product is at risk of stockout and automatically triggers a moderate price increase strategy (price increase of 3%-8%) to adjust demand and reduce stockout losses.

[0048] When high temperatures are detected in the surrounding area, the system automatically implements a "high-temperature special offer" strategy (discount of 5%-10%) for beverage products to increase sales and customer satisfaction.

[0049] Promotional Plan Generation: The AI ​​agent automatically generates personalized promotional plans based on sales data, inventory data, and user profile data. Bundling promotions for slow-moving products: Combining slow-moving products with fast-moving products for promotional purposes (e.g., "Buy best-selling beverage A + slow-moving snack B and enjoy a bundled discount") to increase the turnover rate of slow-moving products.

[0050] Time-based promotions: Automatically generate time-based promotional plans such as "afternoon tea specials" for low-traffic periods (such as 2-4 pm) to stimulate consumer demand.

[0051] Performance Tracking and Optimization: After each marketing strategy is executed, the AI ​​agent tracks the strategy's effectiveness (sales increase, profit changes, inventory turnover improvement, etc.), inputs the performance data into the reinforcement learning strategy network, and continuously optimizes the decision-making quality of subsequent marketing strategies. Statistics show that after implementing AI agent dynamic pricing, the average gross profit margin of goods increased by 8.5%, the turnover rate of slow-moving goods increased by 65%, and the ROI (Return on Investment) of promotional activities increased by approximately 40%.

[0052] Example 5: Complete Workflow of End-to-End Closed-Loop Management This embodiment uses a complete end-to-end closed-loop management process as an example for illustration.

[0053] Scenario: Due to a nearby night market event, a community's unmanned retail kiosks experienced a three-fold increase in customer traffic compared to usual, resulting in a much faster turnover of goods than normal. Simultaneously, the refrigeration compressor in the kiosks exhibited intermittent abnormal noise, posing a potential malfunction risk.

[0054] Step S1, Data Acquisition: The sensor network collects real-time sales data (the hourly sales of beverages surged from 12 to 45 units on average), inventory data (the inventory level of beverages dropped rapidly), equipment status data (the compressor vibration sensor detected abnormal vibration frequency), and external environment data (information on surrounding night market activities was obtained through an external data interface).

[0055] Step S2, Status Perception: The AI ​​intelligent agent perception module performs fusion analysis on multi-source data and outputs multi-dimensional situational perception results: the urgency score for beverage replenishment needs is 92 points (extremely high), the health score for the refrigeration system is 65 points (attention status), and the sales anomaly detection result is "abnormal surge in sales (activity-driven)".

[0056] Step S3, Collaborative Decision-Making: The AI ​​agent's decision-making module integrates all perception results to generate a collaborative decision-making scheme. Replenishment Decision: Due to the impending depletion of beverage inventory, an emergency replenishment order is automatically generated. The replenishment quantity is 2.5 times the normal replenishment quantity, and the replenishment priority is set to the highest, requiring the replenishment to be completed within 1 hour.

[0057] Equipment operation and maintenance decision: Due to abnormal vibration of the refrigeration compressor, a pre-inspection instruction and spare parts preparation suggestion are generated. Technical personnel will be arranged to conduct on-site inspection within 2 hours after the replenishment is completed.

[0058] Marketing strategy decision: To avoid frequent restocking due to a surge in sales, the price of beverage products will be automatically increased by 5% to stabilize demand.

[0059] Step S4, Execution and Issuance: Emergency replenishment orders are pushed to the two nearest operations personnel via the APP; equipment pre-inspection orders are pushed to the regional maintenance engineer; price adjustment orders are automatically issued to the electronic price tag system of the cabinet to complete the price update.

[0060] Step S5, Feedback Closed Loop: After replenishment is completed, the sensing module confirms that the inventory has returned to normal levels; after the equipment pre-inspection is completed, the maintenance engineer uploads the inspection results (confirming that the compressor bearing is worn and has been replaced on site), and the system records the maintenance history and cost; the sales data after the price adjustment is fed back to the AI ​​agent to optimize subsequent decision-making strategies.

[0061] Results: Through the full-link collaborative management of the AI ​​intelligent agent, the cabinet successfully coped with the sudden surge in customer traffic brought about by the night market activities, without any stockouts, and the equipment malfunctions were repaired before they could worsen, achieving stable and efficient operation.

[0062] Comparative Example 1: Traditional Manual Decision-Making Management The operation of unmanned retail cabinets is managed using traditional manual decision-making methods: each link operates independently, replenishment is determined by the experience of the operators, equipment inspections are carried out on a fixed cycle, there is no predictive maintenance, and merchandise management relies on manual inspection.

[0063] Comparative Results: In the operation of 500 unmanned retail kiosks of the same scale, the traditional method resulted in an average daily stockout rate of 8.5%, an average equipment failure repair time of 4.5 hours, and a manpower requirement of 12 people. After adopting the AI ​​intelligent agent management system of this invention, the average daily stockout rate decreased to 2.1%, the failure repair time was shortened to 1.8 hours, and the manpower requirement was reduced to 5 people. This invention has significant advantages in operational efficiency, cost control, and user satisfaction.

[0064] Comparative Example 2: Single-stage automation (no end-to-end collaboration) Automation is only implemented in the replenishment process (automatic inventory monitoring + manual replenishment), but replenishment decisions are not linked to sales forecasts, equipment status, or marketing strategies, and each process remains independent.

[0065] Comparison results: While basic automatic replenishment functions can be achieved, replenishment efficiency is low (due to the lack of consideration for predictive replenishment and dynamic pricing to adjust demand), equipment failures still require passive manual response, and overall operational efficiency is significantly lower than that of this invention. Single-stage automation cannot achieve global optimization, and the contradictions and conflicts between different stages remain unresolved.

[0066] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for end-to-end operation and management of unmanned retail cabinets based on AI intelligent agents, characterized in that, Includes the following steps: S1. Real-time collection of multi-source heterogeneous operational data: Real-time collection of multi-source heterogeneous operational data through sensor networks deployed in unmanned retail cabinets and external data interfaces; S2. Intelligent perception and situational understanding of operational status: The multi-source heterogeneous operational data is input into the perception module of the AI ​​agent to perform data fusion and status recognition, and output multi-dimensional operational situational perception results. S3. AI Intelligent Agent Full-Link Collaborative Decision-Making: Based on the multi-dimensional operational situation perception results, the AI ​​intelligent agent's decision-making module generates a collaborative decision-making scheme that includes replenishment decisions, product management decisions, equipment operation and maintenance decisions, and marketing strategy decisions; S4. Automatic execution and issuance of decision instructions: The collaborative decision-making scheme is converted into executable instructions and automatically issued to the corresponding execution unit of the unmanned retail cabinet; S5. Execution effect monitoring and closed-loop feedback: Real-time monitoring of the execution effect of decision instructions, and feedback of the execution results to the AI ​​agent, forming a full-link automatic closed loop of "perception → decision → execution → feedback".

2. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The AI ​​agent comprises a perception module, a decision-making module, and an execution scheduling module; the perception module is used to perform fusion analysis and state recognition of multi-source heterogeneous data; the decision-making module comprises a replenishment decision submodule, an inventory optimization submodule, an equipment operation and maintenance submodule, and a marketing strategy submodule; the execution scheduling module is used to convert decision schemes into execution instructions and perform task scheduling.

3. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The operational data in S1 includes product inventory data, sales transaction data, equipment operating status data, product image data, environmental perception data, and external environment data.

4. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The multi-dimensional operational situational awareness results in S2 include at least: product shortage warning level, inventory health score, equipment health score, sales anomaly detection results, and replenishment demand urgency score.

5. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The replenishment decision in S3 includes determining the replenishment timing, generating a replenishment product list, and calculating the replenishment quantity; the product management decision includes optimizing product display and suggesting the elimination of slow-moving products; the equipment operation and maintenance decision includes triggering equipment self-checks and predicting faults; and the marketing strategy decision includes dynamic pricing and generating promotional plans.

6. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The AI ​​agent employs a learning mechanism that combines reinforcement learning and deep neural networks. It trains decision-making strategies using historical operational data and continuously learns and optimizes them during actual operation.

7. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, In step S4, for unmanned retail cabinets equipped with replenishment robots, the replenishment decision automatically generates replenishment robot path planning and replenishment operation instructions; for unmanned retail cabinets not equipped with replenishment robots, the replenishment decision pushes replenishment guidance to the operators via mobile terminals.

8. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The method also includes an active handling step for abnormal events: when the AI ​​agent detects equipment failure, product stagnation, or payment abnormality, it automatically executes a preset emergency handling strategy and evaluates the handling effect after the handling is completed.

9. The method for full-chain operation and management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The method also includes an automatic operation report generation step: the AI ​​agent automatically generates an operation analysis report according to a preset cycle, the report including sales analysis, inventory analysis, equipment operation analysis and optimization suggestions.

10. The method for full-chain operation management of unmanned retail cabinets based on AI intelligent agents according to claim 1, characterized in that, The execution unit includes a replenishment robot or replenishment guidance terminal, a product identification system, an equipment self-inspection and repair system, and a pricing and promotion system.

Images