Intelligent container operation optimization and inventory management method and system based on big data

Abstract

This invention relates to the field of smart vending machine operation and big data analysis technology, specifically disclosing a method and system for optimizing smart vending machine operation and managing inventory based on big data. This invention integrates multi-source heterogeneous data to generate spatiotemporal fusion data, and uses a weighted fusion method combining multi-model prediction and time-period product correlation matrix to obtain a comprehensive predicted demand. It then constructs a mixed-integer programming model with the goal of maximizing inventory turnover efficiency to solve replenishment decisions. This solves the problems of large demand forecasting deviations, delayed replenishment timing and quantity decisions, and the inability to dynamically adjust inventory configuration due to data silos in traditional inventory management models. It achieves improvements in the accuracy of smart vending machine demand forecasting, inventory turnover efficiency, and refined operational capabilities, while reducing operating costs.

Description

Technical Field

[0001] This invention relates to the field of smart vending machine operation and big data analysis technology, and in particular to a method and system for optimizing smart vending machine operation and inventory management based on big data. Background Technology

[0002] In the field of smart vending machine operation, traditional inventory management models mainly rely on manual experience and periodic inventory checks, making it difficult to cope with dynamic fluctuations in consumer behavior and real-time changes in demand. Existing systems generally lack the ability to effectively integrate multi-source heterogeneous data, resulting in data silos between sales data, vending machine status information, and external environmental factors. This leads to significant deviations in demand forecasting, delayed decisions on replenishment timing and quantity, and the frequent coexistence of out-of-stock best-selling items and stockpiles of slow-moving goods. Furthermore, existing technologies for inventory configuration optimization often employ fixed thresholds or simple rule engines, failing to make differentiated dynamic adjustments based on vending machine location, time characteristics, and user profiles. This results in low inventory turnover efficiency and rising operating costs, hindering the scalable profitability and refined operational capabilities of smart vending machine networks.

[0003] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for optimizing the operation and inventory management of smart vending machines based on big data.

[0005] In a first aspect, the present invention provides a method for optimizing the operation and inventory management of intelligent vending machines based on big data. The technical solution of this method is as follows: Acquire sales data, vending machine status data, and external environment data from multiple smart vending machines. The sales data includes the sales time and quantity of each product in each vending machine. The vending machine status data includes the current inventory of each product in each vending machine and the vending machine location information. The external environment data includes time period characteristics data and weather data of the location of each vending machine. The sales data, the container status data, and the external environment data are fused according to the time and space dimensions to generate multi-source fused data for each container in each time unit; Based on the multi-source fusion data, multi-model predictions are performed on each commodity in each container to obtain multiple candidate predicted demand quantities for each commodity in each container in the next time unit. For each container, corresponding user profile data is obtained based on the container location information, and a time-period product correlation matrix for each container in the next time unit is constructed based on the user profile data, the time period feature data, and the sales data. The multiple candidate predicted demand quantities are weighted and fused with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container; Based on the comprehensive forecasted demand for each item in each container, the current inventory level, the inventory capacity parameters of each container, and the logistics constraint parameters of each container, a mixed integer programming model is constructed with the goal of maximizing overall inventory turnover efficiency. Solve the mixed integer programming model to obtain the replenishment quantity decision and replenishment timing decision for each item in each container, generate replenishment task instructions for each container based on the replenishment quantity decision and the replenishment timing decision, and send the replenishment task instructions to the corresponding execution terminal.

[0006] Secondly, this invention provides a smart vending machine operation optimization and inventory management system based on big data. The technical solution of this system is as follows: The acquisition module is used to acquire sales data, vending machine status data, and external environment data of multiple smart vending machines. The sales data includes the sales time and sales quantity of each product in each vending machine. The vending machine status data includes the current inventory of each product in each vending machine and the vending machine location information. The external environment data includes time period characteristic data and weather data of the location of each vending machine. The generation module is used to integrate the sales data, the container status data, and the external environment data according to the time and space dimensions to generate multi-source fused data for each container in each time unit. The prediction module is used to perform multi-model predictions on each commodity in each container based on the multi-source fusion data, and obtain multiple candidate predicted demand quantities for each commodity in each container in the next time unit. The module is used to obtain corresponding user profile data for each container based on the container location information, and to construct a time-period product correlation matrix for each container in the next time unit based on the user profile data, the time period feature data, and the sales data. The fusion module is used to perform weighted fusion of the multiple candidate predicted demand quantities with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container; A module is established to construct a mixed integer programming model with the goal of maximizing overall inventory turnover efficiency, based on the comprehensive predicted demand for each product in each container, the current inventory, the inventory capacity parameters of each container, and the logistics constraint parameters of each container. The decision module is used to solve the mixed integer programming model to obtain the replenishment quantity decision and replenishment timing decision for each item in each container, and generate replenishment task instructions for each container based on the replenishment quantity decision and the replenishment timing decision, and send the replenishment task instructions to the corresponding execution terminal.

[0007] The technical solution of this invention integrates multi-source heterogeneous data to generate spatiotemporal fusion data, and uses a multi-model prediction and time-period commodity correlation matrix weighted fusion method to obtain comprehensive predicted demand. Then, a mixed integer programming model with the goal of maximizing inventory turnover efficiency is constructed to solve replenishment decisions. This solves the problems of large demand forecast deviations, delayed replenishment timing and replenishment quantity decisions, and inability to dynamically adjust inventory configuration caused by data silos in traditional inventory management models. It realizes the improvement of demand forecast accuracy, inventory turnover efficiency and refined operation capabilities of smart vending machines, and the reduction of operating costs.

[0008] Compared to existing technologies that use fixed thresholds or simple rule engines for replenishment decisions, this invention integrates multi-source heterogeneous data along time and space dimensions, eliminating data silos between sales data, shelf status data, and external environment data. This allows demand forecasting to simultaneously reflect temporal evolution patterns and spatial distribution differences. By employing a weighted fusion of multi-model forecasting and a time-series commodity correlation matrix to calculate comprehensive forecast demand, the prediction results combine the advantages of different forecasting models and capture time-based correlation purchasing patterns between commodities, overcoming the shortcomings of single forecasting models in nonlinear consumption scenarios. By constructing a mixed-integer programming model aimed at maximizing overall inventory turnover efficiency and incorporating inventory space competition and historical stockout penalty terms into the objective function, replenishment decisions can meet inventory capacity and logistics resource constraints while suppressing space crowding caused by excessive replenishment of similar commodities and enhancing the replenishment priority of frequently stockout commodities. This achieves coordinated optimization and differentiated dynamic adjustment of replenishment quantity and timing decisions.

[0009] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0011] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1This is a flowchart illustrating an embodiment of the intelligent vending machine operation optimization and inventory management method based on big data according to the present invention. Figure 2 This is a schematic diagram of an embodiment of the intelligent vending machine operation optimization and inventory management system based on big data according to the present invention. Detailed Implementation

[0012] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0013] Figure 1 This diagram illustrates a flowchart of an embodiment of a smart vending machine operation optimization and inventory management method based on big data provided by the present invention, executed by a control terminal. Figure 1 As shown, it includes the following steps: S1. Acquire sales data, vending machine status data, and external environment data of multiple smart vending machines. The sales data includes the sales time and sales quantity of each product in each vending machine. The vending machine status data includes the current inventory of each product in each vending machine and the vending machine location information. The external environment data includes the time period characteristics data and weather data of the location of each vending machine.

[0014] Among them, smart vending machines refer to automated terminal equipment deployed in public places or commercial areas that have unattended vending functions. The equipment contains a product storage space, a payment recognition module, and a communication module, and is used to provide users with self-service purchase of goods. For example, a smart vending machine numbered C001 deployed in the lobby of Building D, Office Building C, District B, City A, contains two products: 500 ml bottles of purified water and original flavor potato chips. After the user scans the code to open the door and takes the goods, the payment is automatically deducted.

[0015] The sales time refers to the specific moment when a user purchases and completes payment for a product recorded by the smart vending machine; for example, smart vending machine number C001 recorded that a 500ml bottle of purified water was purchased and paid for by a user at 8:35 AM on January 1, 2026. The sales quantity refers to the total number of items purchased and paid for by the smart vending machine within the specified time range; for example, smart vending machine number C001 sold a total of 6 500ml bottles of purified water between 8:00 AM and 9:00 AM on January 1, 2026.

[0016] The current inventory refers to the actual remaining sellable quantity of a certain product in the smart vending machine after a real-time inventory check at a specified time. For example, at 9:00 AM on January 1, 2026, the current inventory of 500ml bottles of purified water in smart vending machine C001 is 24 bottles. Vending machine location information refers to the coordinate data or address description information identifying the geographical location of the smart vending machine. For example, the location information of smart vending machine C001 is X degrees Y minutes North latitude and M degrees N minutes East longitude, corresponding to the address: Lobby of Building D, Office Building C, District B, City A.

[0017] Among them, time-period characteristic data refers to pre-labeled time attribute information based on the division of different time periods within a day, used to characterize typical patterns of user consumption behavior within a time period; for example, 8:00 AM to 9:00 AM is labeled as the "weekday morning rush hour" time-period characteristic data, during which user purchasing behavior is mainly focused on quickly taking away bottled water. Weather data refers to meteorological conditions information of the location of the smart vending machine within a specified time interval, including weather conditions and temperature information; for example, at 8:00 AM on January 1, 2026, the weather data for the location of smart vending machine numbered C001 is "sunny, temperature 15 degrees Celsius".

[0018] S2. The sales data, the container status data, and the external environment data are fused according to the time and space dimensions to generate multi-source fused data for each container in each time unit.

[0019] The time dimension refers to the perspective of organizing and dividing data according to chronological order, reflecting the evolution of data over time; for example, dividing January 1, 2026 into hourly time units and counting the sales volume within each hourly unit. The spatial dimension refers to the perspective of differentiating and associating smart vending machines in different locations according to geographical location, reflecting the differences in data distribution across space; for example, counting the sales volume of smart vending machines numbered C001 and C002 within the same time unit and comparing them based on vending machine location information.

[0020] In this context, a time unit refers to a segment of continuous time of equal or unequal length, serving as the basic time granularity for data analysis and decision-making. For example, a day can be divided into 24 time units, each lasting one hour. Multi-source fusion data refers to a single structured data record generated by aligning and merging data from different data sources according to both time and spatial dimensions. This record simultaneously includes sales information, vending machine status information, and external environmental information. For example, the sales quantity of 6 bottles, current inventory of 24 bottles, vending machine location information "Lobby of Building D, Office Building C, District B, City A," time period characteristic data "Weekday morning rush hour," and weather data "Sunny, temperature 15 degrees Celsius" for smart vending machine C001 within the time unit of 8:00 AM to 9:00 AM on January 1, 2026, can be merged into a single multi-source fusion data record.

[0021] S3. Based on the multi-source fusion data, perform multi-model predictions on each commodity in each container to obtain multiple candidate predicted demand quantities for each commodity in each container in the next time unit.

[0022] Among them, multi-model prediction refers to using multiple different types of prediction models to predict the demand for the same object, so as to combine the prediction advantages of different models. For example, time series prediction model, gradient boosting tree model and long short-term memory network model are used to predict the demand for 500 ml bottles of purified water in the smart vending machine numbered C001 in the next time unit.

[0023] The next time unit refers to the time unit following the current time unit, serving as the target period for demand forecasting and replenishment decisions. For example, if the current time unit is from 8:00 AM to 9:00 AM on January 1, 2026, then the next time unit is from 9:00 AM to 10:00 AM on January 1, 2026. The candidate predicted demand quantity refers to the demand forecast result for a specific product within a specified time unit, output by a single forecasting model. For example, the time series forecasting model outputs a candidate predicted demand of 7 bottles of 500ml purified water for smart vending machine C001 from 9:00 AM to 10:00 AM on January 1, 2026.

[0024] S4. For each container, obtain the corresponding user profile data based on the container location information, and construct the time period product correlation matrix for each container in the next time unit based on the user profile data, the time period feature data, and the sales data.

[0025] The user profile data refers to a set of feature vectors reflecting the consumption preferences of user groups at the location of the smart vending machine, generated based on historical consumption behavior statistics. For example, user profile data is generated based on historical sales records in the lobby of the office building where the smart vending machine with the number C001 is located. The user profile data shows that users at this location have a preference weight of 0.7 for purified water and a preference weight of 0.3 for potato chips during weekday morning peak hours. The time-slot product association matrix refers to a matrix constructed with products as the row and column dimensions. Each element in the matrix represents the probability or association strength of two products being purchased simultaneously by a user within a specified time unit. For example, for the time unit of 9:00 AM to 10:00 AM on January 1, 2026, for the smart vending machine with the number C001, the association value between 500ml bottles of purified water and original flavor potato chips is 0.4 in the time-slot product association matrix.

[0026] S5. The multiple candidate predicted demand quantities are weighted and fused with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container.

[0027] Among them, the comprehensive forecast demand refers to the final demand forecast value obtained by weighting and fusing multiple candidate forecast demand values ​​with the commodity correlation matrix of the time period, which is used for subsequent replenishment decisions; for example, the comprehensive forecast demand for 500 ml bottles of purified water in smart vending machine number C001 is 8 bottles from 9:00 to 10:00 am on January 1, 2026.

[0028] S6. Based on the comprehensive predicted demand for each item in each container, the current inventory, the inventory capacity parameters of each container, and the logistics constraint parameters of each container, construct a mixed integer programming model with the goal of maximizing overall inventory turnover efficiency.

[0029] The inventory capacity parameter refers to the maximum number of each type of product that can be stored in the smart vending machine, used to constrain the maximum inventory limit for replenishment decisions; for example, the inventory capacity parameter for 500ml bottles of purified water in smart vending machine C001 is 50 bottles. The logistics constraint parameter refers to the upper limit on the total number of products that can be transported in a single replenishment operation, used to constrain the feasibility of replenishment decisions; for example, the maximum single-operation capacity in the logistics constraint parameter is 200 items.

[0030] Maximizing overall inventory turnover efficiency means optimizing the average inventory turnover of all goods in the smart vending machine network to the highest level, so that inventory resources are fully utilized. For example, by optimizing replenishment decisions, the average inventory turnover of all goods in smart vending machines C001 and C002 can be increased from 1.2 times per day to 1.8 times per day.

[0031] Among them, the mixed integer programming model refers to a mathematical optimization model in which the objective function and constraints of the model include continuous variables and integer variables, and is used to solve the optimal combination of decision variables under multiple constraints; for example, the constructed mixed integer programming model includes replenishment quantity decision variables and replenishment timing decision variables, and the constraints include inventory capacity constraints and single transportation capacity constraints.

[0032] S7. Solve the mixed integer programming model to obtain the replenishment quantity decision and replenishment timing decision for each item in each container, generate replenishment task instructions for each container based on the replenishment quantity decision and the replenishment timing decision, and send the replenishment task instructions to the corresponding execution terminal.

[0033] The replenishment quantity decision refers to the decision on the quantity of inventory to be replenished for each item in each smart vending machine within a specific time unit. For example, the replenishment quantity decision determines that 15 bottles of 500ml purified water should be replenished in smart vending machine C001 between 9:00 AM and 10:00 AM on January 1, 2026. The replenishment timing decision refers to the binary decision on whether to perform a replenishment operation for each item in each smart vending machine within the current time unit. For example, the replenishment timing decision determines that the 500ml bottles of purified water in smart vending machine C001 should be replenished at 9:00 AM on January 1, 2026. The replenishment task instruction refers to the execution command generated by the control system, which includes the replenishment object, the replenishment item list, and the replenishment quantity, used to guide the execution terminal to complete the replenishment operation. For example, the replenishment task instruction includes information such as "smart vending machine C001, 15 bottles of 500ml purified water, 8 bags of original flavor potato chips". An execution terminal refers to a handheld device or automated device that receives replenishment task instructions and executes replenishment operations; for example, an execution terminal is a mobile terminal device held by warehouse managers, which displays replenishment task instructions and guides managers to complete the replenishment.

[0034] The technical solution of this embodiment integrates multi-source heterogeneous data to generate spatiotemporal fusion data, and uses a multi-model prediction and time-period commodity correlation matrix weighted fusion to obtain a comprehensive predicted demand. Then, a mixed integer programming model with the goal of maximizing inventory turnover efficiency is constructed to solve the replenishment decision. This solves the problems of large demand forecast deviation, delayed replenishment timing and replenishment quantity decisions, and inability to dynamically adjust inventory configuration caused by data silos in the traditional inventory management model. It realizes the improvement of demand forecast accuracy, inventory turnover efficiency and refined operation capabilities of smart vending machines, and the reduction of operating costs.

[0035] In one optional approach, the step of fusing the sales data, the container status data, and the external environment data according to time and space dimensions to generate multi-source fused data for each container in each time unit includes: The sales data, status data, and external environment data of each container within each time unit are aligned according to timestamps and merged into a single multi-source fused data.

[0036] Timestamp alignment refers to matching and aligning data from different data sources according to their recording time, so that data at the same point in time or within the same time unit can be associated together. For example, the sales data of smart vending machine C001 at 8:35 am on January 1, 2026, the vending machine status data reported by the smart vending machine at 8:30 am on the same day, and the weather data at 8:00 am on the same day can be aligned according to their timestamps.

[0037] In the above-mentioned optional methods, sales data, container status data and external environment data are further merged into a single multi-source fusion data after being aligned according to timestamps. This breaks down the barriers between multi-source heterogeneous data, achieves unified data organization in the spatiotemporal dimensions, provides a structurally standardized and dimensionally consistent input foundation for subsequent multi-model prediction, and improves the quality of data fusion and model usability.

[0038] In one optional approach, the step of performing multi-model predictions on each item in each container based on the multi-source fusion data to obtain multiple candidate predicted demand quantities for each item in each container in the next time unit includes: The demand for each commodity in each container is predicted using a time series forecasting model, a gradient boosting tree model, and a long short-term memory network model, and the prediction results output by each model are used as the candidate predicted demand.

[0039] Among them, time series forecasting models refer to mathematical models that predict future values ​​by analyzing the time dependence of historical data based on the patterns of changes over time; for example, an autoregressive moving average model can be used to model the hourly sales volume of 500ml bottles of purified water in smart vending machine C001 over the past 7 days to predict the demand in the next time unit. Gradient boosting tree models refer to machine learning models that improve prediction accuracy by iteratively training multiple decision trees and accumulating the prediction results of each decision tree; for example, a gradient boosting tree model can be trained using historical sales data, time period feature data, and weather data as input features to predict the demand for 500ml bottles of purified water in smart vending machine C001. Long Short-Term Memory (LSTM) network models refer to deep learning models that are variants of recurrent neural networks, effectively learning long-term dependencies in time series data by introducing gating mechanisms; for example, an LTM network model can be used to learn the historical demand sequence of 500ml bottles of purified water in smart vending machine C001 to capture the differences in demand patterns between weekdays and weekends.

[0040] The prediction result refers to the numerical value of the demand output by the prediction model after inputting feature data; for example, the prediction result output by the time series prediction model is 7 bottles, the prediction result output by the gradient boosting tree model is 8 bottles, and the prediction result output by the long short-term memory network model is 6 bottles.

[0041] Among the above-mentioned optional methods, time series forecasting models, gradient boosting tree models, and long short-term memory network models are further used to predict the demand for each commodity in parallel, capturing the linear trend, nonlinear characteristics, and time-series dependence of demand, generating multi-perspective candidate predicted demand, expanding the coverage and reliability of prediction information, and enhancing the robustness of demand forecasting.

[0042] In one optional approach, the step of constructing a time-period product correlation matrix for each display case in the next time unit based on the user profile data, the time-period feature data, and the sales data includes: Based on the product preference vector in the user profile data and the historical purchase combination relationship corresponding to the time period feature data, calculate the correlation value of any two products in each shelf in the next time unit, and arrange the correlation values ​​according to the product dimension to form the product correlation matrix of the time period.

[0043] Among them, the product preference vector refers to a vector constructed with products as the dimension, where each element of the vector represents the degree of preference or purchase probability of a user group for the corresponding product; for example, the product preference vector in the user profile data of smart vending machine number C001 is [0.7, 0.3], where 0.7 corresponds to the preference for 500ml bottles of purified water and 0.3 corresponds to the preference for original flavor potato chips. Historical purchase combination relationship refers to the frequency and pattern of two products being purchased simultaneously in the same transaction, obtained from historical sales data; for example, based on the historical sales data of smart vending machine number C001, it was found that within the time unit of 8:00 AM to 9:00 AM, the proportion of times 500ml bottles of purified water and original flavor potato chips were purchased simultaneously in the same transaction was 0.4% of all transactions.

[0044] The correlation value refers to a numerical value that quantifies the synergistic relationship between two products in purchasing behavior. The higher the value, the higher the probability that the two products are purchased simultaneously. For example, the correlation value between 500ml bottles of purified water and original-flavored potato chips is 0.4 in the time unit of 8:00 AM to 9:00 AM. The product dimension refers to using product type as the basis for dividing the matrix into rows and columns to organize the correlation relationships between products. For example, using 500ml bottles of purified water as the first row and first column, and original-flavored potato chips as the second row and second column, a 2-row, 2-column product correlation matrix is ​​constructed.

[0045] In the above-mentioned optional methods, the correlation value of any two products is calculated and arranged to form a time-based product correlation matrix based on the historical purchase combination relationship corresponding to the product preference vector in the user profile and the time period characteristics. This quantifies the implicit consumption relationship between products, reveals the time-based purchase pattern, and provides a correlation basis for demand forecasting correction and inventory linkage configuration.

[0046] In one optional approach, the step of weightedly fusing the multiple candidate predicted demand quantities with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container includes: The fusion weights of each candidate predicted demand are determined based on the prediction errors of each prediction model within the historical period. The candidate predicted demand is then weighted and summed according to the fusion weights to obtain the initial fusion demand. Finally, the initial fusion demand is dynamically corrected based on the correlation values ​​between the product and other products in the product correlation matrix of the time period.

[0047] Prediction error refers to the deviation between the predicted demand output by the prediction model and the actual demand. For example, a time series prediction model predicts that the demand for 500ml bottles of purified water in smart vending machine C001 will be 7 bottles between 8:00 AM and 9:00 AM on January 1, 2026, while the actual demand is 6 bottles, resulting in a prediction error of 1 bottle. Fusion weight refers to the coefficient assigned to each candidate predicted demand when merging multiple candidate predicted demands into a single comprehensive predicted demand, reflecting the reliability of each prediction model. For example, the fusion weight for the time series prediction model is 0.3, the fusion weight for the gradient boosting tree model is 0.4, and the fusion weight for the long short-term memory network model is 0.3.

[0048] The initial fused demand refers to the preliminary demand forecast obtained by weighting and summing multiple candidate predicted demands according to their respective fusion weights, before adjusting for product association. For example, multiplying 7 bottles by 0.3, 8 bottles by 0.4, and 6 bottles by 0.3 and then summing them yields an initial fused demand of 7.1 bottles. Dynamic adjustment refers to the process of adjusting the initial fused demand based on the association relationships between products and other products in the product association matrix for a given time period. This adjustment process dynamically changes as the association matrix changes. For example, based on the association value of 0.4 between 500ml bottles of purified water and original flavor potato chips, and the predicted demand for original flavor potato chips, the initial fused demand of 7.1 bottles is dynamically adjusted to 8 bottles.

[0049] In the above-mentioned optional methods, the fusion weights are further determined based on the historical prediction errors of each model, and the candidate predicted demand is weighted and summed to obtain the initial fused demand. Then, it is dynamically corrected based on the commodity correlation matrix of the time period, so as to realize the synergistic use of the prediction advantages of multiple models and commodity correlation information, and improve the accuracy and adaptability of the comprehensive predicted demand.

[0050] In one alternative approach, the step of determining the fusion weights of each candidate forecast demand based on the forecast errors of each forecast model over a historical period includes: For each item in each container, the prediction error of each prediction model in the previous time unit is calculated. The reciprocal of the prediction error is normalized and used as the fusion weight of each candidate predicted demand. The fusion weight is updated in each time unit.

[0051] Reciprocal normalization refers to a mathematical transformation method that involves taking the reciprocal of a set of values, dividing each reciprocal by the sum of all reciprocals, so that the sum of the transformed values ​​equals 1. For example, if the prediction errors of three prediction models are 1 bottle, 0.5 bottles, and 2 bottles, taking the reciprocals yields 1, 2, and 0.5, and normalization yields fusion weights of 0.29, 0.57, and 0.14.

[0052] In the above-mentioned optional methods, the prediction error of each model in the previous time unit is further calculated, and the inverse of the error is normalized as the fusion weight and updated in each time unit. This allows the weight allocation to be dynamically adjusted as the model performance changes, maintaining the timeliness and optimality of the multi-model fusion strategy and ensuring the stability and reliability of long-term prediction.

[0053] In one optional approach, the mixed-integer programming model includes replenishment quantity decision variables and replenishment timing decision variables for each item in each container. The constraints of the mixed-integer programming model include that the inventory quantity of each item in each container after replenishment does not exceed the inventory capacity parameter, the total replenishment quantity of each container in a single replenishment does not exceed the single transport capacity limit in the logistics constraint parameters, and the sum of the replenishment timing decision variables for each item within a time unit does not exceed the preset replenishment frequency limit.

[0054] In this mixed-integer programming model, the replenishment quantity decision variable is a continuous variable representing the quantity of each product that should be replenished. The variable's value is constrained by inventory capacity and logistics resources. For example, the replenishment quantity decision variable $Q_{1,1}$ represents the replenishment quantity of 500ml bottles of purified water in smart vending machine C001, with a value ranging from 0 to 50. The replenishment timing decision variable is a 0-1 integer variable representing whether to perform a replenishment operation. A value of 1 indicates that replenishment will be performed, and a value of 0 indicates that replenishment will not be performed. For example, a replenishment timing decision variable $\delta_{1,1}$ with a value of 1 indicates that a replenishment operation will be performed on January 1, 2026 at 9:00 AM for the 500ml bottles of purified water in smart vending machine C001.

[0055] The post-replenishment inventory refers to the total inventory quantity after adding the replenishment quantity to the current inventory quantity, used to determine whether the inventory capacity limit has been exceeded; for example, if the current inventory quantity is 24 bottles and the replenishment quantity is 15 bottles, the post-replenishment inventory quantity is 39 bottles. The total quantity for a single replenishment refers to the sum of the quantities of all items replenished for the same smart vending machine in a single replenishment operation; for example, replenishing smart vending machine C001 with 15 bottles of 500ml purified water and 8 bags of original flavor potato chips results in a total of 23 items for a single replenishment.

[0056] The single-trip capacity limit refers to the maximum total number of goods that can be transported in a single replenishment operation as specified in the logistics constraint parameters; for example, if the single-trip capacity limit is 200 pieces, the total number of goods replenished in a single operation shall not exceed 200 pieces. The preset replenishment frequency limit refers to the maximum number of times a replenishment operation is allowed to be performed on the same product in the same smart vending machine within a specified time unit; for example, if the preset replenishment frequency limit is 2 times per day, that is, the same product can be replenished a maximum of 2 times per day.

[0057] In the above-mentioned optional methods, a mixed integer programming model is further constructed with replenishment quantity decision variables and replenishment timing decision variables. With inventory capacity, single transportation capacity limit and replenishment frequency limit as constraints, the replenishment decision is transformed into a mathematical optimization problem with hard constraints, realizing the collaborative decision-making of replenishment quantity and replenishment timing and the limitation of feasible region.

[0058] In one alternative approach, when weighted and fused with the multiple candidate predicted demand quantities and the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity for each commodity in each container, the comprehensive predicted demand quantity is calculated according to the following formula: ; in, This represents the overall projected demand for the j-th type of product in the i-th container in the next time unit. Indicates the number of prediction models. This represents the fusion weight of the j-th item in the i-th container under the k-th prediction model, dynamically updated based on the prediction error of the previous time unit. This represents the candidate predicted demand quantity of the j-th type of product in the i-th container, obtained under the k-th prediction model. This represents the dynamic adjustment coefficient for correlation. This represents the set of goods that are associated with the j-th type of goods. This represents the correlation value between the j-th type of product and the m-th type of product in the i-th container in the next time unit. Let represent the weighted average of multiple candidate predicted demand quantities for the m-th type of product in the i-th container. Let represent the arithmetic mean of the candidate predicted demand for the m-th type of commodity in the i-th container under each prediction model. This represents the demand difference adjustment coefficient. This represents the weighted average of multiple candidate predicted demand quantities for the j-th type of commodity in the i-th container.

[0059] It should be noted that the above formula is based on the idea of ​​combining multi-model prediction weighted fusion with dynamic correction of product correlation. The weighted fusion part dynamically calculates the fusion weight using the prediction errors of each prediction model in the previous time unit, and then sums the candidate predicted demand outputs from different models. The dynamic correlation correction part introduces an exponential decay function to adjust the impact of demand differences on the correction magnitude, based on the correlation values ​​between products in the product correlation matrix for the time period and the difference between the predicted demand of a product and the average predicted demand. This constructs a comprehensive predicted demand calculation formula that simultaneously integrates the advantages of multi-model prediction and the collaborative purchasing relationship between products. This formula ensures that the final comprehensive predicted demand output not only combines the prediction accuracy advantages of multiple prediction models but also adaptively corrects the predicted values ​​according to the related purchasing patterns between products under the characteristics of the time period, thereby improving the accuracy of demand prediction in scenarios where products exhibit related consumption patterns.

[0060] In the above-mentioned optional methods, the comprehensive predicted demand is further calculated according to the formula of weighted summation of fusion weights and candidate predicted demand plus correlation correction term, so as to realize the quantitative fusion of multi-model prediction results and commodity correlation information, provide accurate demand prediction input for replenishment optimization, and support data-driven inventory decision-making.

[0061] In an alternative approach, when constructing a mixed-integer programming model that aims to maximize overall inventory turnover efficiency, the objective function of the mixed-integer programming model is expressed by the following formula: ; in, Indicates the total number of containers. Indicates the total number of product categories. Let represent the decision variable for replenishing the j-th type of product in the i-th container. This represents the overall projected demand for the j-th type of product in the i-th container in the next time unit. This represents the current inventory of the j-th type of product in the i-th container at the current time unit. This represents the positive part of the difference between replenishment demand and current inventory. This represents the inventory capacity parameter for the j-th type of product in the i-th container. This represents the competition coefficient for inventory space. Let represent the set of goods that compete with the j-th type of goods. Let represent the inventory competition intensity coefficient between the j-th product and the m-th product in the i-th container. Let represent the decision variable for replenishing the m-th type of product in the i-th container. This represents the replenishment quantity saturation adjustment coefficient. This represents the historical average replenishment quantity of the j-th item in the i-th container. This indicates the out-of-stock penalty coefficient. Let represent the decision variable for replenishing the j-th type of product in the i-th container. This represents the predicted stockout loss coefficient for the j-th item in the i-th container during the period without replenishment. This represents the cumulative penalty coefficient based on historical stockouts. Indicates the number of backtracking time units. This represents the current inventory of the j-th type of product in the i-th container at the th time unit. This represents the comprehensive predicted demand for the j-th type of commodity in the i-th container at the t-h+1 time unit.

[0062] It should be noted that the above formula is based on the objective function of maximizing inventory turnover efficiency. By introducing replenishment quantity decision variables and replenishment timing decision variables, the numerator is designed as the product of replenishment quantity and replenishment demand gap to represent the turnover contribution brought by replenishment. The denominator is designed as the sum of inventory capacity and inventory space competition term to reflect the impact of replenishment on inventory capacity constraints and competition among products. At the same time, a stockout penalty term is introduced into the objective function. This penalty term includes the replenishment timing decision variable, the predicted stockout loss coefficient, and the cumulative penalty factor based on historical stockout records. Thus, a mixed integer programming objective function is constructed that simultaneously considers inventory turnover contribution, inventory space competition, stockout loss, and the cumulative effect of historical stockouts. The above formula enables the replenishment quantity decision and replenishment timing decision to maximize overall inventory turnover efficiency while satisfying inventory capacity and logistics resource constraints. At the same time, the inventory competition term inhibits the space crowding caused by excessive replenishment of similar products, and the historical stockout cumulative penalty term enhances the replenishment priority of frequently stockout products, thereby achieving refined and differentiated replenishment decisions.

[0063] Among the above-mentioned optional methods, an objective function is further constructed according to a formula that includes replenishment demand and the positive part of the inventory difference, inventory competition term and stockout history penalty term. The optimization direction is to maximize the overall inventory turnover efficiency, so as to achieve a balance and coordination between inventory turnover speed, space utilization efficiency and service level, and improve the overall operational efficiency.

[0064] Figure 2 This diagram illustrates the structure of an embodiment of a smart vending machine operation optimization and inventory management system based on big data, provided by the present invention. Figure 2 As shown, this big data-based intelligent vending machine operation optimization and inventory management system includes: The acquisition module 201 is used to acquire sales data, vending machine status data and external environment data of multiple smart vending machines. The sales data includes the sales time and sales quantity of each product in each vending machine. The vending machine status data includes the current inventory of each product in each vending machine and vending machine location information. The external environment data includes time period characteristic data and weather data of the location of each vending machine. The generation module 202 is used to integrate the sales data, the container status data and the external environment data according to the time dimension and the spatial dimension to generate multi-source fused data of each container in each time unit; Prediction module 203 is used to perform multi-model prediction on each commodity in each container based on the multi-source fusion data, and obtain multiple candidate predicted demand quantities for each commodity in each container in the next time unit. The construction module 204 is used to obtain corresponding user profile data for each container based on the container location information, and construct a time-period product correlation matrix for each container in the next time unit based on the user profile data, the time period feature data and the sales data. The fusion module 205 is used to perform weighted fusion of the multiple candidate predicted demand quantities with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container; Module 206 is established to construct a mixed integer programming model with the goal of maximizing overall inventory turnover efficiency, based on the comprehensive predicted demand of each commodity in each container, the current inventory, the inventory capacity parameters of each container, and the logistics constraint parameters of each container. The decision module 207 is used to solve the mixed integer programming model to obtain the replenishment quantity decision and replenishment timing decision for each commodity in each container, and to generate replenishment task instructions for each container based on the replenishment quantity decision and the replenishment timing decision, and send the replenishment task instructions to the corresponding execution terminal.

[0065] The technical solution of this embodiment integrates multi-source heterogeneous data to generate spatiotemporal fusion data, and uses a multi-model prediction and time-period commodity correlation matrix weighted fusion to obtain a comprehensive predicted demand. Then, a mixed integer programming model with the goal of maximizing inventory turnover efficiency is constructed to solve the replenishment decision. This solves the problems of large demand forecast deviation, delayed replenishment timing and replenishment quantity decisions, and inability to dynamically adjust inventory configuration caused by data silos in the traditional inventory management model. It realizes the improvement of demand forecast accuracy, inventory turnover efficiency and refined operation capabilities of smart vending machines, and the reduction of operating costs.

[0066] Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0067] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0068] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and do not imply a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0069] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for optimizing the operation and managing inventory of smart vending machines based on big data, characterized in that, The method includes: Acquire sales data, vending machine status data, and external environment data from multiple smart vending machines. The sales data includes the sales time and quantity of each product in each vending machine. The vending machine status data includes the current inventory of each product in each vending machine and the vending machine location information. The external environment data includes time period characteristics data and weather data of the location of each vending machine. The sales data, the container status data, and the external environment data are fused according to the time and space dimensions to generate multi-source fused data for each container in each time unit; Based on the multi-source fusion data, multi-model predictions are performed on each commodity in each container to obtain multiple candidate predicted demand quantities for each commodity in each container in the next time unit. For each container, obtain the corresponding user profile data based on the container location information, and construct the time-period product correlation matrix for each container in the next time unit based on the user profile data, the time period feature data, and the sales data. The multiple candidate predicted demand quantities are weighted and fused with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container; Based on the comprehensive forecasted demand for each item in each container, the current inventory level, the inventory capacity parameters of each container, and the logistics constraint parameters of each container, a mixed integer programming model is constructed with the goal of maximizing overall inventory turnover efficiency. Solve the mixed integer programming model to obtain the replenishment quantity decision and replenishment timing decision for each item in each container, generate replenishment task instructions for each container based on the replenishment quantity decision and the replenishment timing decision, and send the replenishment task instructions to the corresponding execution terminal.

2. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 1, characterized in that, The steps of fusing the sales data, the container status data, and the external environment data according to time and space dimensions to generate multi-source fused data for each container in each time unit include: The sales data, status data, and external environment data of each container within each time unit are aligned according to timestamps and merged into a single multi-source fused data.

3. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 2, characterized in that, The steps of performing multi-model predictions on each commodity in each container based on the multi-source fusion data to obtain multiple candidate predicted demand quantities for each commodity in each container in the next time unit include: The demand for each commodity in each container is predicted using a time series forecasting model, a gradient boosting tree model, and a long short-term memory network model, and the prediction results output by each model are used as the candidate predicted demand.

4. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 1, characterized in that, The steps of constructing a time-period product correlation matrix for each display case in the next time unit based on the user profile data, the time period feature data, and the sales data include: Based on the product preference vector in the user profile data and the historical purchase combination relationship corresponding to the time period feature data, calculate the correlation value of any two products in each cabinet in the next time unit, and arrange the correlation values ​​according to the product dimension to form the product correlation matrix of the time period.

5. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 4, characterized in that, The step of weightedly fusing the multiple candidate predicted demand quantities with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container includes: The fusion weights of each candidate predicted demand are determined based on the prediction errors of each prediction model within the historical period. The candidate predicted demand is then weighted and summed according to the fusion weights to obtain the initial fusion demand. Finally, the initial fusion demand is dynamically corrected based on the correlation values ​​between the product and other products in the product correlation matrix of the time period.

6. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 5, characterized in that, The steps for determining the fusion weights of each candidate forecast demand based on the forecast errors of each forecast model over historical periods include: For each item in each container, the prediction error of each prediction model in the previous time unit is calculated. The reciprocal of the prediction error is normalized and used as the fusion weight of each candidate predicted demand. The fusion weight is updated in each time unit.

7. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 1, characterized in that, The mixed integer programming model includes replenishment quantity decision variables and replenishment timing decision variables for each commodity in each container. The constraints of the mixed integer programming model include that the inventory of each commodity after replenishment in each container does not exceed the inventory capacity parameter, the total replenishment amount of each container in a single replenishment does not exceed the single transport capacity limit in the logistics constraint parameters, and the sum of the replenishment timing decision variables of each commodity in the time unit does not exceed the preset replenishment frequency limit.

8. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 5, characterized in that, When weighted and fused with the multiple candidate predicted demand quantities and the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container, the comprehensive predicted demand quantity is calculated according to the following formula: ； in, This represents the overall projected demand for the j-th type of product in the i-th container in the next time unit. Indicates the number of prediction models. This represents the fusion weight of the j-th item in the i-th container under the k-th prediction model, dynamically updated based on the prediction error of the previous time unit. This represents the candidate predicted demand quantity of the j-th type of product in the i-th container, obtained under the k-th prediction model. This represents the dynamic adjustment coefficient for correlation. This represents the set of goods that are associated with the j-th type of goods. This represents the correlation value between the j-th type of product and the m-th type of product in the i-th container in the next time unit. Let represent the weighted average of multiple candidate predicted demand quantities for the m-th type of product in the i-th container. Let represent the arithmetic mean of the candidate predicted demand for the m-th type of commodity in the i-th container under each prediction model. This represents the demand difference adjustment coefficient. This represents the weighted average of multiple candidate predicted demand quantities for the j-th type of commodity in the i-th container.

9. The method for optimizing the operation and managing inventory of smart vending machines based on big data according to claim 7, characterized in that, When constructing a mixed-integer programming model with the objective of maximizing overall inventory turnover efficiency, the objective function of the mixed-integer programming model is expressed by the following formula: ； in, Indicates the total number of containers. Indicates the total number of product categories. Let represent the decision variable for replenishing the j-th type of product in the i-th container. This represents the overall projected demand for the j-th type of product in the i-th container in the next time unit. This represents the current inventory of the j-th type of product in the i-th container at the current time unit. This represents the positive part of the difference between replenishment demand and current inventory. This represents the inventory capacity parameter for the j-th type of product in the i-th container. This represents the competition coefficient for inventory space. Let represent the set of goods that compete with the j-th type of goods. Let represent the inventory competition intensity coefficient between the j-th product and the m-th product in the i-th container. Let represent the decision variable for replenishing the m-th type of product in the i-th container. This represents the replenishment quantity saturation adjustment coefficient. This represents the historical average replenishment quantity of the j-th item in the i-th container. This indicates the out-of-stock penalty coefficient. Let represent the decision variable for replenishing the j-th type of product in the i-th container. This represents the predicted stockout loss coefficient for the j-th item in the i-th container during the period without replenishment. This represents the cumulative penalty coefficient based on historical stockouts. Indicates the number of backtracking time units. This represents the current inventory of the j-th type of product in the i-th container at the th time unit. This represents the comprehensive predicted demand for the j-th type of commodity in the i-th container at the t-h+1 time unit.

10. A smart vending machine operation optimization and inventory management system based on big data, characterized in that, The system includes: The acquisition module is used to acquire sales data, vending machine status data, and external environment data of multiple smart vending machines. The sales data includes the sales time and sales quantity of each product in each vending machine. The vending machine status data includes the current inventory of each product in each vending machine and the vending machine location information. The external environment data includes time period characteristic data and weather data of the location of each vending machine. The generation module is used to integrate the sales data, the container status data, and the external environment data according to the time and space dimensions to generate multi-source fused data for each container in each time unit. The prediction module is used to perform multi-model predictions on each commodity in each container based on the multi-source fusion data, and obtain multiple candidate predicted demand quantities for each commodity in each container in the next time unit. The module is used to obtain corresponding user profile data for each container based on the container location information, and to construct a time-period product correlation matrix for each container in the next time unit based on the user profile data, the time period feature data, and the sales data. The fusion module is used to perform weighted fusion of the multiple candidate predicted demand quantities with the commodity correlation matrix of the time period to obtain the comprehensive predicted demand quantity of each commodity in each container; A module is established to construct a mixed integer programming model with the goal of maximizing overall inventory turnover efficiency, based on the comprehensive predicted demand for each product in each container, the current inventory, the inventory capacity parameters of each container, and the logistics constraint parameters of each container. The decision module is used to solve the mixed integer programming model to obtain the replenishment quantity decision and replenishment timing decision for each item in each container, and generate replenishment task instructions for each container based on the replenishment quantity decision and the replenishment timing decision, and send the replenishment task instructions to the corresponding execution terminal.

Images