Storage cabinet grid matching method and device and storage medium

Abstract

The invention relates to a locker grid matching method and device and a storage medium, and the method comprises the steps: S1, receiving order information, and carrying out the analysis based on the order information, and obtaining a commodity feature, a user feature and a time feature; s2, based on the user features and the commodity features, performing preliminary screening on the grid types of the storage cabinets, and obtaining all or part of the grid types as first grid types through screening; s3, splicing the lattice features of the first lattice types with the commodity features, the user features and the time features obtained through analysis, and inputting the spliced features into a trained first model to obtain a first duration and a second duration corresponding to the lattice types; and S4, based on the first duration and the second duration of each lattice type, in combination with the idle lattice proportion of each lattice type, calculating to obtain an objective function value of each lattice type, and selecting the lattice type with the optimal objective function value as a matching result of the current order. Compared with the prior art, the method has the advantages of improving the use convenience of the user and the like.

Description

Technical Field

[0001] This invention relates to the field of locker compartment allocation, and in particular to a locker compartment matching method, apparatus and storage medium. Background Technology

[0002] With the rapid development of e-commerce and on-demand delivery services, smart lockers, including parcel lockers, self-service food delivery lockers, and smart retail lockers, have become a key infrastructure for solving the "last mile" delivery problem and improving the convenience of package pickup for users. Their core function is to automatically allocate a suitable compartment for temporary storage of different orders. Therefore, the level of intelligence in the compartment matching strategy directly affects the operational efficiency, space utilization, and user experience of the lockers.

[0003] Currently, existing methods for matching locker compartments can be mainly categorized as follows: 1. Matching method based on simple rules: This is the most traditional and common method. The system usually assigns matches based on preset static rules, for example: Sequential allocation: Allocation is carried out sequentially according to the grid number, which is simple to implement but highly prone to blindness.

[0004] Size matching: Prioritize selecting compartments whose size is closest to the product volume to avoid wasting space. This is the most basic optimization, but it only considers a single dimension.

[0005] Based on idle status: Prioritize the allocation of idle grids, or consider the balance of grid usage to a certain extent to prevent some grids from being overused while others remain idle for a long time.

[0006] 2. Distance-based matching method: In some application scenarios where multiple lockers are deployed or a locker has multiple areas, the system will try to allocate the locker closest to the user or the most easily accessible compartment in the locker based on the user's delivery address or real-time location.

[0007] However, these existing technologies have obvious flaws and limitations, and cannot meet the increasingly complex operational needs: The existing methods only focus on one or two explicit factors, such as the size of the compartment and the physical distance, and fail to comprehensively consider multi-dimensional information such as product characteristics, user profiles and time context.

[0008] This ultimately leads to low turnover efficiency of the lockers. Therefore, there is an urgent need in this field for a more intelligent locker compartment matching method that can overcome the above-mentioned shortcomings. This method can significantly improve the turnover efficiency of lockers and increase the time interval for customers to retrieve their items. Summary of the Invention

[0009] The purpose of this invention is to provide a method, device, and storage medium for matching locker compartments in order to solve the problems existing in the prior art.

[0010] The objective of this invention can be achieved through the following technical solutions: A method for matching locker compartments includes: Step S1: Receive order information, and parse the order information to obtain product characteristics, user characteristics, and time characteristics; Step S2: Based on user characteristics and product characteristics, perform a preliminary screening of the compartment types of each locker, and select all or some of the compartment types as the first compartment type. If the locker position, compartment height and compartment size of two compartments are the same, they correspond to the same compartment type. Conversely, if any one of the locker position, compartment height from the ground and compartment size of two compartments is different, they correspond to different compartment types. Step S3: After concatenating the compartment features of each first compartment type with the parsed product features, user features and time features, input the concatenation into the trained first model to obtain the first duration and second duration corresponding to each compartment type. The first duration represents the time interval from the time of entering the cabinet to the time of opening the cabinet, and the second duration represents the time interval from the time of opening the cabinet to the time when the cabinet door is closed by the user. Step S4: Based on the first and second durations of each grid type, and combined with the idle grid ratio of each grid type, calculate the objective function value of each grid type, and select the grid type with the optimal objective function value as the matching result for the current order.

[0011] The product features include product size classification and product type; the user features include recipient address, last four digits of recipient's mobile phone number, and order address; and the time features include order time, order date, and date category.

[0012] The product categories include documents, takeout food, and merchandise, and the date categories include weekdays and holidays.

[0013] Step S2 includes: Step S2-1: Filter out grid types with insufficient grid size based on product size classification; Step S2-2: Filter out some grid types that are too far away based on the recipient's address; Step S2-3: If the date category is a holiday or the order time is during off-peak hours, filter out grid types whose grid height from the ground exceeds the set height threshold.

[0014] Step S2 includes: Step S2-1: Filter out grid types with insufficient grid size based on product size classification; Step S2-2: Filter out some grid types that are too far away based on the recipient's address; Step S2-3: If the idle ratio of all grids is lower than the first pre-configured threshold, filter out grid types whose ground height exceeds the set height threshold.

[0015] The mathematical expression for the objective function value is: in: Let k be the objective function value of the grid type. The first duration of grid type k, The second duration is for grid type k. To utilize the weighting coefficients of the balance term in the grid, The total number of grids of grid type k. This indicates the idle status of the i-th specific grid in grid type k. The average idle ratio of all grids in the system. γ An adjustable power exponent is used to control the idle ratio. R k Sensitivity to the impact on the final objective function value For the ideal first duration, Let be the standard deviation of the Gaussian decay function for the first duration.

[0016] The ideal first duration is obtained based on Gaussian decay modeling.

[0017] The first model is a neural network model.

[0018] A locker compartment matching device includes a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the method described above.

[0019] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. By designing a multi-source information fusion method, the trained first model is used to predict a first duration representing the time interval from the time the item is placed in the cabinet to the time it is opened, and a second duration representing the time interval from the time the cabinet is opened to the time the cabinet door is closed by the user. The first duration can, to some extent, express the actual distance and preference between the user and the type of compartment, while the second duration can, to some extent, express the difficulty of retrieving the item between the user and the type of compartment, including the matching degree between the user's height and the height of the compartment from the ground. Based on the above information, the objective function value is obtained, which can be used to filter and match the optimal type of compartment, ultimately improving the overall turnover efficiency and the user's retrieval experience.

[0021] 2. Product features include product size classification and product type; user features include recipient address, last four digits of recipient's mobile phone number, and order address; time features include order time, order date, and date category. Product categories include documents, takeout food, and goods; date categories include weekdays and holidays. This provides sufficient multi-source feature information to improve the prediction accuracy of the first and second time durations.

[0022] 3. If the idle ratio of all grid cells is lower than the first pre-configured threshold, then filter out grid cell types whose ground height exceeds the set height threshold. Determine whether to call the grid cells based on the unique constraint of the idle ratio of all grid cells, thereby improving the utilization rate of high-quality grid cells during idle periods.

[0023] 4. By designing a suitable method for calculating the objective function value, the stability and reliability of the matching results can be improved. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the main steps of the method of the present invention. Detailed Implementation

[0025] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0026] A method for matching locker compartments, such as Figure 1 As shown, it includes: Step S1: Receive order information, and parse the order information to obtain product characteristics, user characteristics, and time characteristics; In this embodiment, product features include product size classification and product type; user features include recipient address, last four digits of recipient's mobile phone number and order address; and time features include order time, order date and date category.

[0027] Product categories include documents, takeout food, and merchandise; date categories include weekdays and holidays.

[0028] Since it is impossible to directly obtain the detailed information of the person who placed the order, it is necessary to indirectly build a user profile through the above information. In this embodiment, the recipient address and the order address are not exactly the same. The order address is collected when placing an order through the APP, while the recipient address is obtained from the order structured data. This application mainly considers the case of ordering by oneself. The recipient address is more associated with department information, while the order address is associated with the geographical information of the actual workstation.

[0029] Step S2: Based on user characteristics and product characteristics, perform a preliminary screening of the compartment types of each locker, and select all or some of the compartment types as the first compartment type. If the locker position, compartment height and compartment size of two compartments are the same, they correspond to the same compartment type. Conversely, if any one of the locker position, compartment height from the ground and compartment size of two compartments is different, they correspond to different compartment types. In one embodiment, step S2 includes: Step S2-1: Filter out grid types with insufficient grid size based on product size classification; Generally, the compartments can be divided into multiple levels, including three levels: large, medium, and small. Of course, in some embodiments, there may only be two levels: large and small. Thus, based on the size of the product, it can be determined whether each type of compartment can be stored. Unless otherwise specified, the product size classification is generally assumed to be applicable to all compartment sizes. Large items are usually noted in the remarks column. Alternatively, in other embodiments, photos of the items can be collected, and AI can be used to identify whether they should be stored in the compartments of each level.

[0030] Step S2-2: Filter out some grid types that are too far away based on the recipient's address; Generally, using the recipient's address can quickly avoid some lockers that are very far away. In addition, this step chooses to use the recipient's address instead of the order address because the recipient's address has certain structured information, which makes it easier to filter quickly and reduce the computational cost.

[0031] Step S2-3: If the date category is a holiday or the order time is during off-peak hours, filter out grid types whose grid height from the ground exceeds the set height threshold.

[0032] Therefore, during holidays or off-peak hours, the utilization rate of lockers is relatively low, so the use of individual compartments can be minimized, and compartments with suitable height can be used first.

[0033] Furthermore, in some other embodiments, step S2 includes: Step S2-1: Filter out grid types with insufficient grid size based on product size classification; Step S2-2: Filter out some grid types that are too far away based on the recipient's address; Step S2-3: If the idle ratio of all grids is lower than the first pre-configured threshold, filter out grid types whose ground height exceeds the set height threshold.

[0034] In this way, replacing other metrics with a single indicator—that the idle ratio of all grids is below the first pre-configured threshold—can be a simpler and more efficient approach.

[0035] Step S3: After concatenating the compartment features of each first compartment type with the parsed product features, user features and time features, input the concatenation into the trained first model to obtain the first duration and second duration corresponding to each compartment type. The first duration represents the time interval from the time of entering the cabinet to the time of opening the cabinet, and the second duration represents the time interval from the time of opening the cabinet to the time when the cabinet door is closed by the user. Step S4: Based on the first and second durations of each grid type, and combined with the idle grid ratio of each grid type, calculate the objective function value of each grid type, and select the grid type with the optimal objective function value as the matching result for the current order.

[0036] In this embodiment, the mathematical expression for the objective function value is: in: Let k be the objective function value of the grid type. The first duration of grid type k, The second duration is for grid type k. To utilize the weighting coefficients of the balance term in the grid, The total number of grids of grid type k. This indicates the idle status of the i-th specific grid in grid type k. The average idle ratio of all grids in the system. γ An adjustable power exponent is used to control the idle ratio. R k Sensitivity to the impact on the final objective function value For the ideal first duration, Let be the standard deviation of the Gaussian decay function for the first duration.

[0037] The ideal first duration is obtained based on Gaussian decay modeling and can be understood as the expected ideal first duration. The greater the deviation of the duration from this value, the greater the penalty.

[0038] Therefore, using the above objective function value has the following advantages compared to some conventional forms: 1. By introducing and These parameters allow system administrators to define ideal operating conditions and tolerance ranges, enabling policy-driven intelligent matching. For example, a smaller tolerance range can be set during peak periods. To achieve rapid turnover; larger capacity can be set during off-peak periods. To protect equipment and enhance the user experience.

[0039] 2. By using log(1+T1T2), the rate of increase in negative effects decreases as the waiting time increases, which aligns better with reality. The increase in anxiety from waiting 1 minute to 2 minutes is far greater than from 60 minutes to 61 minutes. Optimization decisions are more aligned with user experience. This avoids the system making over-optimization decisions that offer negligible improvements to the actual experience in pursuit of extremely short mathematical durations.

[0040] 3. The exponent γ in the numerator can adjust the intensity of the preference for high idle ratios. A grid utilization balancing term is introduced into the denominator to penalize situations where there are excessive differences in grid utilization within the same type. This improves the overall system utilization and stability, promotes balanced consumption of grid resources, avoids local overload, extends the overall lifespan of equipment, and enhances the system's resilience to peak loads.

[0041] 4. Possesses self-learning and adaptive capabilities. This is a crucial step from "static rules" to "intelligent decision-making systems," greatly enhancing the patent creativity and technological depth of the solution.

[0042] 5. Logarithmic and exponential decay functions are insensitive to extreme values ​​such as extremely short or extremely long prediction durations, and can smoothly handle errors in the prediction model. This improves the stability and reliability of the matching results and reduces suboptimal allocations caused by accidental fluctuations in the prediction model.

[0043] Furthermore, in this embodiment, the first model can be built using a neural network. Its input is the result of concatenating grid features with the parsed product features, user features, and time features. The output includes a first duration and a second duration, and its loss function is as follows: in: The predicted value for the first duration of the grid type corresponding to sample j. Let j be the predicted value for the second duration of the grid type corresponding to sample j. These are hyperparameters used to adjust the importance of the ideal duration guide. This is a hyperparameter used to adjust the importance of resource-aware regularization terms. Let N be the average idle ratio of the grid type to which sample i belongs during training data collection, and N be the number of training samples.

[0044] In the above formula, the ideal duration leading term is: The resource-aware regularization term is: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A locker compartment matching method, characterized by, Comprising: Step S1: receiving order information, and obtaining commodity characteristics, user characteristics and time characteristics based on the order information; Step S2: performing preliminary screening on the types of each locker compartment based on the user characteristics and the commodity characteristics, and selecting all or part of the types of the compartments as first types of compartments, wherein if the positions of the lockers, the heights of the compartments and the sizes of the compartments of two compartments are all the same, the two compartments correspond to the same type of compartment, otherwise, if any of the positions of the lockers, the heights of the compartments and the sizes of the compartments of two compartments are different, the two compartments correspond to different types of compartments; Step S3: inputting the characteristics of each first type of compartment and the commodity characteristics, the user characteristics and the time characteristics obtained in the step S1 into a trained first model after splicing, to obtain a first time length and a second time length corresponding to each type of compartment, wherein the first time length represents a time interval from a locker-in time to a locker-opening time, and the second time length represents a time interval from the locker-opening time to a locker-door-closing time by a user; Step S4: calculating target function values of each type of compartment based on the first time length and the second time length of each type of compartment and the idle compartment proportion of each type of compartment, and selecting a type of compartment with an optimal target function value as a matching result of the current order.

2. The method of claim 1, wherein, The commodity characteristics include commodity size classification and commodity type, the user characteristics include consignee address, consignee mobile phone tail number and order address, and the time characteristics include order time, order date and date category.

3. The method of claim 2, wherein, The commodity types include file, take-out food and commodity, and the date categories include weekday and holiday.

4. The method of claim 3, wherein, The step S2 comprises: Step S2-1: filtering part of the types of the compartments with insufficient sizes of the compartments based on the commodity size classification; Step S2-2: filtering part of the types of the compartments that are too far away based on the consignee address; Step S2-3: if the date category is holiday or the order time is a non-peak period, filtering part of the types of the compartments with heights of the compartments exceeding a set height threshold.

5. The method of claim 3, wherein, The step S2 comprises: Step S2-1: filtering part of the types of the compartments with insufficient sizes of the compartments based on the commodity size classification; Step S2-2: filtering part of the types of the compartments that are too far away based on the consignee address; Step S2-3: if the idle proportions of all the compartments are lower than a first preconfigured threshold, filtering part of the types of the compartments with heights of the compartments exceeding a set height threshold.

6. The method of claim 1, wherein, The mathematical expression of the target function value is: wherein: is the objective function value for the bin type k, is the first duration for the bin type k, is the second duration for the bin type k, is the weight coefficient for the bin utilization balancing term, is the total number of bins for the bin type k, is the idle status indication for the i-th specific bin in the bin type k, is the average idle ratio for all bins in the system, The ideal first time length is obtained based on Gaussian decay modeling. is the adjustable power index for controlling the idle ratio R k is the sensitivity of the impact on the final objective function value, is the ideal first duration, is the standard deviation of the Gaussian decay function for the first duration.

7. A method of matching a cabinet door to a cabinet drawer according to claim 6, wherein, The first model is a neural network model.

8. The method of claim 1, wherein, The processor implements the method of any one of claims 1-8 when executing the program.

9. A locker compartment matching apparatus comprising a memory, a processor, and a program stored in the memory, wherein, The program is executed to implement the method of any one of claims 1-8.

10. A storage medium having stored thereon a program, characterized by ​

Images