Edge computing resource allocation method and system based on multi-objective optimization

Abstract

The present invention provides an edge computing resource allocation method and system based on multi-objective optimization, which relates to the field of data processing. The method includes: determining the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores based on historical computing resource demand information of multiple unmanned retail stores, and then determining an edge computing deployment plan and setting up multiple edge computing devices; predicting the future computing resource demands of multiple unmanned retail stores based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores, and demand prediction auxiliary information; obtaining resource allocation auxiliary information; determining an edge computing resource allocation plan based on the future computing resource demands of multiple unmanned retail stores and the resource allocation auxiliary information, and allocating edge computing resources to multiple unmanned retail stores, which has the advantage of improving the flexibility of edge computing resource allocation in unmanned retail scenarios.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to an edge computing resource allocation method and system based on multi-objective optimization. Background Art

[0002] ‌Edge computing is a distributed computing model that aims to move computing power and storage resources closer to users or data sources to reduce response delays between devices and servers‌. It reduces dependence on centralized cloud servers by processing data on edge devices, sensors, and edge nodes, providing faster, more efficient, and more secure computing services. Edge computing also plays an important role in the retail industry. Unmanned retail stores use in-store cameras, sensors, and edge computing to identify customers and record the items they purchase to achieve a seamless shopping experience. This technology not only simplifies the shopping process, but also provides retailers with valuable data insights.

[0003] In the existing edge computing environment, the number of resources for edge computing applications is not sufficient due to various environmental conditions (such as cost, technology, hardware, etc.). In the actual application process, it is often necessary to consider the optimal allocation and reasonable scheduling of resources, that is, on the premise of meeting the resource load constraints of each edge computing application, select appropriate tasks from the request tasks and allocate them to the appropriate edge computing applications, so as to maximize the system resource utilization and achieve the best overall system performance. In the prior art, the resource allocation of edge computing applications is usually based on the existing experience of operation and maintenance personnel, such as configuring resources according to the number of users around the edge computing application, or configuring resources according to the type of edge computing application. For example, for instant applications, more resources are configured, but the user's access behavior is highly uncontrollable and variable. If the resource allocation strategy cannot be adjusted in time according to the task situation and the edge computing application status, it will seriously affect the performance of the edge computing application, causing the performance of the entire system to decrease.

[0004] Therefore, it is necessary to provide an edge computing resource allocation method and system based on multi-objective optimization to improve the flexibility of edge computing resource allocation in unmanned retail scenarios. Summary of the invention

[0005] The present invention provides an edge computing resource allocation method based on multi-objective optimization, including: obtaining historical computing resource demand information of multiple unmanned retail stores; determining the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores based on the historical computing resource demand information of the multiple unmanned retail stores; determining an edge computing deployment plan based on the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores; setting up multiple edge computing devices according to the edge computing deployment plan; obtaining demand prediction auxiliary information; predicting the future computing resource demands of multiple unmanned retail stores based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores and the demand prediction auxiliary information; obtaining resource allocation auxiliary information; determining an edge computing resource allocation plan based on the future computing resource demands of multiple unmanned retail stores and the resource allocation auxiliary information, and allocating edge computing resources to the multiple unmanned retail stores according to the edge computing resource allocation plan.

[0006] Furthermore, based on the historical computing resource demand information of multiple unmanned retail stores, the resource demand characteristics of each unmanned retail store and the resource demand association characteristics between any two unmanned retail stores are determined, including: determining multiple date types; for each date type, determining the resource demand characteristics of the unmanned retail store corresponding to the date type based on the historical computing resource demand information of the unmanned retail store; for any two unmanned retail stores, calculating the first demand similarity and demand complementarity coefficient of the two unmanned retail stores corresponding to each date type based on the resource demand characteristics of the two unmanned retail stores corresponding to each date type, wherein the resource demand association characteristics between any two unmanned retail stores at least include the first demand similarity and demand complementarity coefficient of the two unmanned retail stores corresponding to each date type.

[0007] Furthermore, multiple date types are determined, including: determining multiple initial date types; for each initial date type, determining a demand change curve corresponding to the initial date type for each unmanned retail store based on historical computing resource demand information of multiple unmanned retail stores, performing variational modal decomposition on the demand change curve corresponding to the initial date type for the unmanned retail store, determining component features of the initial date type corresponding to the unmanned retail store, and determining a first component feature matrix corresponding to the initial date type based on the component features of the initial date type corresponding to each unmanned retail store; for any two initial date types, calculating a merging coefficient of the two initial date types based on the first component feature matrices of the two initial date types; and determining multiple date types from the multiple initial date types based on the merging coefficient of the two initial date types.

[0008] Furthermore, based on the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores, an edge computing deployment plan is determined, including: grouping multiple unmanned retail stores according to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type, and determining multiple first unmanned retail store groups; for each first unmanned retail store group, determining the number of edge computing devices corresponding to the first unmanned retail store group according to the resource demand characteristics of each unmanned retail store included in the first unmanned retail store group, and determining the edge computing center location corresponding to the first unmanned retail store group according to the resource demand characteristics and location information of each unmanned retail store included in the first unmanned retail store group, wherein the edge computing deployment plan includes the number of edge computing devices and the edge computing center location corresponding to each first unmanned retail store group.

[0009] Furthermore, according to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type, multiple unmanned retail stores are grouped to determine multiple first unmanned retail store groups, including: calculating the grouping distance of the two unmanned retail stores according to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type; and grouping the multiple unmanned retail stores according to the grouping distance of any two unmanned retail stores through the K-means algorithm to determine multiple first unmanned retail store groups.

[0010] Furthermore, obtaining demand forecasting auxiliary information includes: obtaining historical meteorological information corresponding to historical computing resource demand information of multiple unmanned retail stores; determining multiple meteorological factors; for each date type, determining the demand-related meteorological factors corresponding to the date type based on the historical computing resource demand information and historical meteorological information of multiple unmanned retail stores; obtaining meteorological information based on the demand-related meteorological factors corresponding to each date type, wherein the demand forecasting auxiliary information at least includes meteorological information.

[0011] Furthermore, based on the resource demand characteristics of each unmanned retail store, the resource demand association characteristics between any two unmanned retail stores, and the demand prediction auxiliary information, the future computing resource demands of multiple unmanned retail stores are predicted, including: for each unmanned retail store, according to the component characteristics of the unmanned retail store corresponding to each date type, a second component feature matrix corresponding to the unmanned retail store is determined; for any two unmanned retail stores, according to the second component feature matrices corresponding to the two unmanned retail stores and the first demand similarity corresponding to each date type, the second demand similarity of the two unmanned retail stores is calculated; according to the second demand similarity of any two unmanned retail stores, multiple unmanned retail stores are grouped to determine multiple second unmanned retail store groups; for each second unmanned retail store group, a demand prediction model is established and trained, and the initial future computing resource demand of the second unmanned retail store group is predicted through the demand prediction model according to the resource demand characteristics of the unmanned retail stores and the demand prediction auxiliary information; according to the resource demand association characteristics between any two unmanned retail stores and the initial future computing resource demand of each second unmanned retail store group, the future computing resource demand of multiple unmanned retail stores is determined.

[0012] Furthermore, the resource allocation auxiliary information includes at least the operating status information of each edge computing device and the data transmission distance between any unmanned retail store and any edge computing device; according to the future computing resource requirements of multiple unmanned retail stores and the resource allocation auxiliary information, the edge computing resource allocation plan is determined, including: S11, establishing a multi-objective optimization function; S12, sorting multiple unmanned retail stores according to the future computing resource requirements of multiple unmanned retail stores, and generating a sorting result; S13, determining the current unmanned retail store according to the sorting result; S14, determining the candidate edge computing device of the current unmanned retail store according to the future computing resource requirements of the current unmanned retail store and the resource allocation auxiliary information; S15, determining the priority value of the candidate edge computing device based on the multi-objective optimization function; S16, determining the optimal edge computing device corresponding to the current unmanned retail store based on the priority value of the candidate edge computing device; S17, updating the operating status information of each edge computing device according to the future computing resource requirements of the current unmanned retail store, and executing S13.

[0013] Furthermore, the multi-objective optimization function is related to the demand complementarity coefficient of any two unmanned retail stores corresponding to each date type, the operating status information of each edge computing device, and the data transmission distance between any unmanned retail store and any edge computing device.

[0014] The present invention provides an edge computing resource allocation system based on multi-objective optimization, which applies the above-mentioned edge computing resource allocation method based on multi-objective optimization, including: a data acquisition module, which is used to acquire historical computing resource demand information of multiple unmanned retail stores; a data analysis module, which is used to determine the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores based on the historical computing resource demand information of the multiple unmanned retail stores; a layout optimization module, which is used to determine the edge computing layout plan based on the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores; an edge computing module, which includes multiple edge computing devices set according to the edge computing layout plan; an information acquisition module, which is used to acquire demand prediction auxiliary information; a demand prediction module, which is used to predict the future computing resource requirements of multiple unmanned retail stores based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores and the demand prediction auxiliary information; the information acquisition module is also used to acquire resource allocation auxiliary information; a resource allocation module, which is used to determine the edge computing resource allocation plan based on the future computing resource requirements of the multiple unmanned retail stores and the resource allocation auxiliary information, and allocate edge computing resources to the multiple unmanned retail stores according to the edge computing resource allocation plan.

[0015] Compared with the prior art, the edge computing resource allocation method and system based on multi-objective optimization provided by the present invention have at least the following beneficial effects:

[0016] 1. By obtaining historical computing resource demand information of unmanned retail stores, we can accurately understand the resource usage of each store. Based on this information, we can determine the layout plan of edge computing devices to ensure the reasonable allocation and efficient use of resources. By predicting future computing resource requirements, we can further optimize resource allocation and avoid idle or overused resources. Edge computing pushes data processing and analysis to the "edge" of the network, where the data is generated, reducing the need to transmit data to the central data center, thereby reducing bandwidth costs. Through accurate resource allocation, unnecessary equipment purchase and maintenance costs can be avoided, further reducing operating costs. Edge computing devices are close to the data source and can process and analyze data in real time, so as to quickly respond to the operational needs of the store. This is particularly important for unmanned retail stores, which can ensure a smooth experience for customers during the shopping process. Considering the correlation characteristics of resource requirements between any two unmanned retail stores helps to consider the mutual influence between stores when allocating resources. Through collaborative resource allocation, the stability and reliability of the entire system can be enhanced to ensure good performance during peak hours or emergencies. Edge computing devices can collect and process large amounts of data, providing strong support for the intelligent operation of unmanned retail stores. For example, by analyzing data such as customer shopping behavior and inventory status, more accurate marketing strategies and inventory management plans can be formulated. Edge computing enhances data privacy and security and reduces the risk of data leakage by placing data processing and analysis close to the data source. This is especially important for unmanned retail stores, as sensitive information such as customer information and transaction data needs to be processed and analyzed in a secure environment.

[0017] 2. By determining multiple date types (such as weekdays, weekends, holidays, etc.) and analyzing the historical computing resource demand information of unmanned retail stores for each date type, the resource demand characteristics of each store under different date types can be accurately identified. This helps to allocate the most appropriate edge computing resources to the store based on its actual operation. By calculating the first demand similarity and demand complementarity coefficient for each date type between any two unmanned retail stores, the method can evaluate the complementarity of resource demands between stores. This helps to consider the mutual influence between stores when allocating resources and achieve collaborative optimization of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0019] Figure 1 It is a flowchart of an edge computing resource allocation method based on multi-objective optimization according to some embodiments of this specification;

[0020] Figure 2 It is a module diagram of an edge computing resource allocation system based on multi-objective optimization as shown in some embodiments of this specification. DETAILED DESCRIPTION

[0021] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0022] Figure 1 is a flow chart of an edge computing resource allocation method based on multi-objective optimization according to some embodiments of this specification, such as Figure 1 As shown, the edge computing resource allocation method based on multi-objective optimization may include the following steps.

[0023] Step 110, obtaining historical computing resource demand information of multiple unmanned retail stores.

[0024] Among them, the historical computing resource demand information of the unmanned retail store may include the computing resource demand (for example, memory demand, video memory demand, and CPU core number demand, etc.) of the unmanned retail store at multiple historical time points in multiple historical periods (for example, one day).

[0025] It is understandable that unmanned retail stores collect customers' shopping behaviors through cameras and sensors and complete shopping settlement. Therefore, the computing resource demand of unmanned retail stores at a certain point in time is directly related to the customer flow of the unmanned retail stores at that point in time.

[0026] Step 120: Determine the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores based on the historical computing resource demand information of the plurality of unmanned retail stores.

[0027] In some embodiments, step 120 specifically includes:

[0028] Determine various date types, such as weekdays, weekends, holidays, and days adjacent to holidays (e.g., the day before a holiday and the day after a holiday, etc.);

[0029] For each date type, determine the resource demand characteristics of the unmanned retail store corresponding to the date type based on the historical computing resource demand information of the unmanned retail store;

[0030] For any two unmanned retail stores, the first demand similarity and demand complementarity coefficient of the two unmanned retail stores corresponding to each date type are calculated based on the resource demand characteristics of the two unmanned retail stores corresponding to each date type, wherein the resource demand association characteristics between any two unmanned retail stores at least include the first demand similarity and demand complementarity coefficient of the two unmanned retail stores corresponding to each date type.

[0031] Specifically, the historical computing resource demand information of the unmanned retail store can be classified according to multiple date types to determine multiple data classes, where one data class corresponds to one date type. Each data class can include the computing resource demand of the unmanned retail store at multiple historical time points in multiple historical periods with a date type corresponding to the date type of the data class.

[0032] The resource demand characteristics of unmanned retail stores corresponding to date types can be determined according to the following process:

[0033] S21. For each historical period included in the data class corresponding to the date type, calculate the single-period resource demand mean and single-period computing resource demand standard deviation of the computing resource demand of the unmanned retail store at multiple historical time points in the historical period;

[0034] S22, averaging the single-cycle computing resource demand average corresponding to each historical cycle included in the data class corresponding to the date type to obtain the multi-cycle computing resource demand average;

[0035] S23, calculating the mean value of the single-period computing resource requirements corresponding to each historical period included in the data class corresponding to the date type, and obtaining the mean standard deviation of the multi-period computing resource requirements;

[0036] S24, calculating the average of the single-period computing resource requirement standard deviations corresponding to each historical period included in the data class corresponding to the date type, to obtain the average of the multi-period computing resource requirement standard deviations;

[0037] S25. A method is provided for calculating the standard deviation of single-cycle computing resource requirements corresponding to each historical cycle included in the data class corresponding to the date type, and a multi-cycle computing resource requirement standard deviation is obtained, wherein the resource requirement characteristics of the unmanned retail store corresponding to the date type may at least include the multi-cycle computing resource requirement mean of the unmanned retail store corresponding to the date type, the standard deviation of the multi-cycle computing resource requirement mean, the mean of the cycle computing resource requirement standard deviation, and the multi-cycle computing resource requirement standard deviation.

[0038] The similarity of the first demand of two unmanned retail stores for each date type can be calculated according to the following formula:

[0039]

[0040] in, is the first demand similarity of the corresponding date type between the i-th unmanned retail store and the j-th unmanned retail store, and is the preset weight, and greater than 0, , Calculate the resource demand mean for the multi-period corresponding date type of the i-th unmanned retail store, Calculate the mean resource demand for multiple periods of the j-th unmanned retail store corresponding to the date type, is the mean standard deviation of resource demand for the i-th unmanned retail store corresponding to the date type in multiple periods, is the mean standard deviation of resource demand for the jth unmanned retail store corresponding to the date type in multiple periods, is the mean of the standard deviation of resource requirements for multiple periods corresponding to the date type of the i-th unmanned retail store, is the mean of the standard deviation of resource requirements for multiple periods corresponding to the date type of the j-th unmanned retail store, Calculate the standard deviation of resource requirements for multiple periods of the i-th unmanned retail store corresponding to the date type, Calculate the standard deviation of resource requirements for multiple periods of the j-th unmanned retail store corresponding to the date type, and is the preset parameter, and Greater than 0.

[0041] The demand complementarity coefficient of two unmanned retail stores can be calculated according to the following formula:

[0042]

[0043] in, is the demand complementarity coefficient between the i-th unmanned retail store and the j-th unmanned retail store, and is the preset weight, and greater than 0, , Calculate the average resource demand for the i-th unmanned retail store in the n-th historical period included in the date type. Calculate the average resource demand for the j-th unmanned retail store in the n-th historical period included in the date type, The total number of historical periods included for the date type, is the preset parameter, greater than 0, Calculate the standard deviation of resource requirements for the i-th unmanned retail store in the n-th historical period included in the date type, Calculate the standard deviation of resource requirements for the j-th unmanned retail store in a single period for the date type.

[0044] In some embodiments, multiple date types are determined, including:

[0045] Determine various initial date types, such as weekdays, weekends, holidays, days adjacent to holidays, end of month, mid-month, etc.;

[0046] For each initial date type, based on historical computing resource demand information of multiple unmanned retail stores, determine a demand change curve corresponding to the initial date type for each unmanned retail store, perform variational modal decomposition on the demand change curve corresponding to the initial date type for the unmanned retail store, determine component features of the unmanned retail store corresponding to the initial date type, and determine a first component feature matrix corresponding to the initial date type based on the component features of the initial date type for each unmanned retail store;

[0047] For any two initial date types, the merging coefficient of the two initial date types is calculated according to the first component feature matrix of the two initial date types;

[0048] A plurality of date types are determined from the plurality of initial date types according to a merging coefficient of the two initial date types.

[0049] Specifically, the demand change curve of the unmanned retail store corresponding to the initial date type may be composed of a line connecting the single-cycle computing resource demand averages of the unmanned retail store in the historical cycle included in the initial date type.

[0050] The component characteristics of the unmanned retail store corresponding to the initial date type may include the center frequency of each component, the bandwidth of each component, the amplitude of each component, the phase of each component and the correlation coefficient between any two components.

[0051] A row vector of the first component feature matrix is ​​a component feature of an initial date type. The cosine similarity of the first component feature matrices of two initial date types can be calculated as the merging coefficient of the two initial date types. If there are two initial date types whose merging coefficients are greater than the merging coefficient threshold, the two initial date types can be merged.

[0052] Step 130: determine an edge computing deployment plan based on the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores.

[0053] In some embodiments, step 130 specifically includes:

[0054] According to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type, the plurality of unmanned retail stores are grouped to determine a plurality of first unmanned retail store groups;

[0055] For each first unmanned retail store group, the number of edge computing devices corresponding to the first unmanned retail store group is determined based on the resource demand characteristics of each unmanned retail store included in the first unmanned retail store group, and the edge computing center location corresponding to the first unmanned retail store group is determined based on the resource demand characteristics and location information of each unmanned retail store included in the first unmanned retail store group, wherein the edge computing deployment plan includes the number of edge computing devices and the edge computing center location corresponding to each first unmanned retail store group.

[0056] In some embodiments, according to the demand complementarity coefficient of two unmanned retail stores corresponding to each date type, a plurality of unmanned retail stores are grouped to determine a plurality of first unmanned retail store groups, including:

[0057] According to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type, the grouping distance between the two unmanned retail stores is calculated;

[0058] The multiple unmanned retail stores are grouped according to the grouping distance between any two unmanned retail stores by using a K-means algorithm to determine multiple first unmanned retail store groups.

[0059] Specifically, the grouping distance between two unmanned retail stores can be calculated according to the following formula:

[0060]

[0061] in, is the grouping distance between the i-th unmanned retail store and the j-th unmanned retail store, is the preset parameter, greater than 0, is the total number of date types, is the demand complementarity coefficient of the i-th unmanned retail store and the j-th unmanned retail store corresponding to the m-th date type.

[0062] For each first unmanned retail store group, the number of edge computing devices corresponding to the first unmanned retail store group can be determined through a quantity prediction model based on the resource demand characteristics of each unmanned retail store included in the first unmanned retail store group, wherein the quantity prediction model can be a convolutional neural network model.

[0063] For each first unmanned retail store group, the location weight of each unmanned retail store can be determined according to the resource demand characteristics of each unmanned retail store included in the first unmanned retail store group, and the location information of each unmanned retail store included in the first unmanned retail store group can be weighted summed according to the location weight of each unmanned retail store to obtain the edge computing center location.

[0064] For example, the location weight of each unmanned retail store can be calculated according to the following formula:

[0065]

[0066] in, is the location weight of the i-th unmanned retail store, is the total number of unmanned retail stores included in the first unmanned retail store group to which the i-th unmanned retail store belongs, is the multi-period computing resource demand mean of the j-th unmanned retail store included in the first unmanned retail store group to which the i-th unmanned retail store belongs, is the multi-period computing resource demand mean standard deviation of the j-th unmanned retail store included in the first unmanned retail store group to which the i-th unmanned retail store belongs.

[0067] Step 140, set up multiple edge computing devices according to the edge computing deployment plan.

[0068] Specifically, for each first unmanned retail store group, edge computing devices corresponding to the first unmanned retail store group may be set at the unmanned retail store closest to the edge computing center location corresponding to the first unmanned retail store group.

[0069] Step 150: Obtain demand forecast auxiliary information.

[0070] In some embodiments, step 150 specifically includes:

[0071] Obtain historical meteorological information corresponding to historical computing resource demand information of multiple unmanned retail stores;

[0072] Determine various meteorological factors (e.g., temperature, humidity, pollutant concentration, light intensity, etc.);

[0073] For each date type, determine the demand-related meteorological factors corresponding to the date type based on historical computing resource demand information and historical meteorological information of multiple unmanned retail stores;

[0074] Meteorological information is obtained according to demand-related meteorological factors corresponding to each date type, wherein the demand forecast auxiliary information at least includes meteorological information.

[0075] Specifically, for each date type, the correlation coefficient between the unmanned retail store and the meteorological factor can be calculated based on the single-cycle resource demand average of multiple historical periods included in the date type and the average of the meteorological factor in multiple historical periods included in the date type. The correlation coefficient between each unmanned retail store and the meteorological factor is averaged, and the mean correlation coefficient corresponding to the meteorological factor is calculated. When the mean correlation coefficient corresponding to the meteorological factor is greater than the correlation coefficient mean threshold, the meteorological factor is used as the demand-related meteorological factor corresponding to the date type.

[0076] Step 160 , predicting the future computing resource demands of multiple unmanned retail stores based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores, and the demand prediction auxiliary information.

[0077] In some embodiments, step 160 specifically includes:

[0078] For each unmanned retail store, according to the component features of the unmanned retail store corresponding to each date type, determine a second component feature matrix corresponding to the unmanned retail store, wherein a row vector of the second component feature matrix corresponds to a date type, and a row vector of the second component feature matrix is ​​a component feature of the unmanned retail store corresponding to a date type;

[0079] For any two unmanned retail stores, the second demand similarity of the two unmanned retail stores is calculated according to the second component feature matrix corresponding to the two unmanned retail stores and the first demand similarity corresponding to each date type. For example, the cosine similarity of the second component feature matrix corresponding to the two unmanned retail stores can be calculated, and the first demand similarity of the two unmanned retail stores corresponding to each date type can be averaged to obtain the first demand similarity mean, and the cosine similarity of the second component feature matrix corresponding to the two unmanned retail stores and the first demand similarity mean are weightedly summed to calculate the second demand similarity of the two unmanned retail stores;

[0080] According to the similarity of the second demands of any two unmanned retail stores, multiple unmanned retail stores are grouped to determine multiple second unmanned retail store groups. For example, the multiple unmanned retail stores can be grouped according to the similarity of the second demands of any two unmanned retail stores by using a K-means algorithm;

[0081] For each second unmanned retail store group, establish and train a demand prediction model, and predict the initial future computing resource demand of the second unmanned retail store group according to the resource demand characteristics of the unmanned retail stores and the demand prediction auxiliary information through the demand prediction model, wherein the demand prediction model may be a convolutional neural network model;

[0082] The future computing resource requirements of the plurality of unmanned retail stores are determined according to the resource requirement association characteristics between any two unmanned retail stores and the initial future computing resource requirements of each second unmanned retail store group.

[0083] Specifically, the future computing resource requirements of multiple unmanned retail stores can be determined through a demand correction model based on the resource demand correlation characteristics between any two unmanned retail stores and the initial future computing resource requirements of each second unmanned retail store group, wherein the demand correction model can be a convolutional neural network model.

[0084] Step 170: Obtain resource allocation auxiliary information.

[0085] In some embodiments, the resource allocation auxiliary information includes at least the operating status information of each edge computing device and the data transmission distance between any unmanned retail store and any edge computing device.

[0086] Specifically, the operating status information of the edge computing device may include at least CPU usage, memory occupancy, and whether it is in a fault state.

[0087] Step 180, determine an edge computing resource allocation plan based on the future computing resource requirements of multiple unmanned retail stores and resource allocation auxiliary information, and allocate edge computing resources to the multiple unmanned retail stores according to the edge computing resource allocation plan.

[0088] In some embodiments, step 180 specifically includes:

[0089] S11, establishing a multi-objective optimization function;

[0090] S12. Sort the multiple unmanned retail stores according to their future computing resource requirements, and generate a sorting result. For example, the greater the future computing resource requirements of the unmanned retail stores, the higher the unmanned retail store is ranked.

[0091] S13. Determine the current unmanned retail store according to the ranking result. For example, the unmanned retail store with the highest ranking and for which the optimal edge computing device has not been determined is used as the current unmanned retail store.

[0092] S14. Determine the candidate edge computing devices of the current unmanned retail store according to the future computing resource demand of the current unmanned retail store and the resource allocation auxiliary information. For example, an edge computing device whose remaining computing resources are greater than the future computing resource demand of the current unmanned retail store, whose data transmission distance is less than the data transmission distance threshold and which is in normal operation may be used as a candidate edge computing device.

[0093] S15. Determine the priority value of the candidate edge computing device based on the multi-objective optimization function;

[0094] S16. Determine the optimal edge computing device corresponding to the current unmanned retail store based on the priority values ​​of the candidate edge computing devices. For example, use the candidate edge computing device with the largest priority value as the optimal edge computing device corresponding to the current unmanned retail store.

[0095] S17. Update the operating status information of each edge computing device according to the future computing resource requirements of the current unmanned retail store, and execute S13.

[0096] In some embodiments, the multi-objective optimization function is related to the demand complementarity coefficient of any two unmanned retail stores for each date type, the operating status information of each edge computing device, and the data transmission distance between any unmanned retail store and any edge computing device.

[0097] Specifically, the multi-objective optimization function can be:

[0098]

[0099] in, is the priority value of the i-th candidate edge computing device, is the total number of unmanned retail stores that have been assigned to the i-th candidate edge computing device, is the demand complementarity coefficient between the qth unmanned retail store assigned to the i-th candidate edge computing device and the current unmanned retail store, is the remaining computing resources of the ith candidate edge computing device, is the data transmission distance between the current unmanned retail store and the i-th candidate edge computing device, , , is the weight coefficient, , , greater than 0, , , and is the attenuation coefficient, , and Greater than 0.

[0100] Figure 2 is a module schematic diagram of an edge computing resource allocation system based on multi-objective optimization according to some embodiments of this specification, such as Figure 2 As shown, the edge computing resource allocation system based on multi-objective optimization may include a data acquisition module, a data analysis module, a layout optimization module, an edge computing module, an information acquisition module, a demand prediction module and a resource allocation module.

[0101] A data acquisition module, used to obtain historical computing resource demand information of multiple unmanned retail stores;

[0102] A data analysis module, for determining the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores based on historical computing resource demand information of multiple unmanned retail stores;

[0103] A deployment optimization module is used to determine the edge computing deployment plan based on the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores;

[0104] An edge computing module, including a plurality of edge computing devices arranged according to an edge computing deployment scheme;

[0105] An information acquisition module is used to obtain auxiliary information for demand forecasting;

[0106] A demand prediction module, used to predict the future computing resource requirements of multiple unmanned retail stores based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores, and demand prediction auxiliary information;

[0107] The information acquisition module is also used to obtain auxiliary information on resource allocation;

[0108] The resource allocation module is used to determine the edge computing resource allocation plan based on the future computing resource requirements of multiple unmanned retail stores and resource allocation auxiliary information, and allocate edge computing resources to multiple unmanned retail stores according to the edge computing resource allocation plan.

[0109] The edge computing resource allocation system based on multi-objective optimization can be used to execute the edge computing resource allocation method based on multi-objective optimization.

[0110] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. The edge computing resource allocation method based on multi-objective optimization is characterized by: include: Obtain historical computing resource demand information for multiple unmanned retail stores; Determine the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores based on historical computing resource demand information of multiple unmanned retail stores; Determine the edge computing deployment plan based on the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores; Arrange a plurality of edge computing devices according to the edge computing deployment scheme; Obtain auxiliary information for demand forecasting; Predict the future computing resource requirements of multiple unmanned retail stores based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores, and demand prediction auxiliary information; Obtain auxiliary information for resource allocation; Determine an edge computing resource allocation scheme based on future computing resource requirements of the plurality of unmanned retail stores and resource allocation auxiliary information, and allocate edge computing resources to the plurality of unmanned retail stores according to the edge computing resource allocation scheme; The resource allocation auxiliary information includes at least the operating status information of each edge computing device and the data transmission distance between any unmanned retail store and any edge computing device; Based on the future computing resource requirements and resource allocation auxiliary information of multiple unmanned retail stores, determine the edge computing resource allocation plan, including: S11. Establish a multi-objective optimization function, wherein the multi-objective optimization function is related to the demand complementarity coefficient of any two unmanned retail stores corresponding to each date type, the operating status information of each edge computing device, and the data transmission distance between any unmanned retail store and any edge computing device; S12. Sort the multiple unmanned retail stores according to future computing resource requirements of the multiple unmanned retail stores, and generate a sorting result; S13. Determine the current unmanned retail store according to the sorting result; S14. Determine a candidate edge computing device for the current unmanned retail store based on the future computing resource requirements and resource allocation auxiliary information of the current unmanned retail store; S15. Determine the priority value of the candidate edge computing device based on the multi-objective optimization function; S16. Determine the optimal edge computing device corresponding to the current unmanned retail store based on the priority values ​​of the candidate edge computing devices; S17. Update the operating status information of each edge computing device according to the future computing resource requirements of the current unmanned retail store, and execute S13.

2. The edge computing resource allocation method based on multi-objective optimization according to claim 1 is characterized in that: Based on the historical computing resource demand information of multiple unmanned retail stores, the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores are determined, including: Identify multiple date types; For each date type, determining the resource demand characteristics of the unmanned retail store corresponding to the date type according to the historical computing resource demand information of the unmanned retail store; For any two unmanned retail stores, the first demand similarity and demand complementarity coefficient of the two unmanned retail stores corresponding to each date type are calculated based on the resource demand characteristics of the two unmanned retail stores corresponding to each date type, wherein the resource demand association characteristics between the any two unmanned retail stores at least include the first demand similarity and demand complementarity coefficient of the two unmanned retail stores corresponding to each date type.

3. The edge computing resource allocation method based on multi-objective optimization according to claim 2 is characterized in that: Determines multiple date types, including: Determine various initial date types; For each initial date type, based on historical computing resource demand information of multiple unmanned retail stores, determine a demand change curve corresponding to the initial date type for each unmanned retail store, perform variational modal decomposition on the demand change curve corresponding to the initial date type for the unmanned retail store, determine component features of the unmanned retail store corresponding to the initial date type, and determine a first component feature matrix corresponding to the initial date type based on the component features of the initial date type for each unmanned retail store; For any two initial date types, the merging coefficient of the two initial date types is calculated according to the first component feature matrix of the two initial date types; A plurality of date types are determined from the plurality of initial date types according to a merging coefficient of the two initial date types.

4. The edge computing resource allocation method based on multi-objective optimization according to claim 2 or 3, characterized in that: According to the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores, the edge computing deployment plan is determined, including: According to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type, the plurality of unmanned retail stores are grouped to determine a plurality of first unmanned retail store groups; For each first unmanned retail store group, the number of edge computing devices corresponding to the first unmanned retail store group is determined based on the resource demand characteristics of each unmanned retail store included in the first unmanned retail store group, and the edge computing center location corresponding to the first unmanned retail store group is determined based on the resource demand characteristics and location information of each unmanned retail store included in the first unmanned retail store group, wherein the edge computing deployment plan includes the number of edge computing devices and the edge computing center location corresponding to each first unmanned retail store group.

5. The edge computing resource allocation method based on multi-objective optimization according to claim 4 is characterized in that: According to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type, the multiple unmanned retail stores are grouped to determine multiple first unmanned retail store groups, including: According to the demand complementarity coefficient of the two unmanned retail stores corresponding to each date type, the grouping distance between the two unmanned retail stores is calculated; The multiple unmanned retail stores are grouped according to the grouping distance between any two unmanned retail stores using the K-means algorithm to determine multiple first unmanned retail store groups.

6. The edge computing resource allocation method based on multi-objective optimization according to claim 3 is characterized in that: Obtain auxiliary information for demand forecasting, including: Obtain historical meteorological information corresponding to historical computing resource demand information of multiple unmanned retail stores; Determine various meteorological factors; For each date type, determine the demand-related meteorological factors corresponding to the date type based on historical computing resource demand information and historical meteorological information of multiple unmanned retail stores; Meteorological information is acquired according to demand-related meteorological factors corresponding to each date type, wherein the demand forecast auxiliary information at least includes meteorological information.

7. The edge computing resource allocation method based on multi-objective optimization according to claim 6 is characterized in that: Based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores, and demand forecasting auxiliary information, the future computing resource demand of multiple unmanned retail stores is predicted, including: For each unmanned retail store, determining a second component feature matrix corresponding to the unmanned retail store according to the component features of the unmanned retail store corresponding to each date type; For any two unmanned retail stores, the second demand similarity of the two unmanned retail stores is calculated according to the second component feature matrices corresponding to the two unmanned retail stores and the first demand similarity corresponding to each date type; According to the similarity of the second demands of any two unmanned retail stores, the plurality of unmanned retail stores are grouped to determine a plurality of second unmanned retail store groups; For each second unmanned retail store group, establishing and training a demand prediction model, and predicting the initial future computing resource demand of the second unmanned retail store group according to the resource demand characteristics of the unmanned retail stores and the demand prediction auxiliary information through the demand prediction model; The future computing resource requirements of the plurality of unmanned retail stores are determined according to the resource requirement association characteristics between any two unmanned retail stores and the initial future computing resource requirements of each second unmanned retail store group.

8. The edge computing resource allocation system based on multi-objective optimization is characterized by: The edge computing resource allocation method based on multi-objective optimization described in any one of claims 1 to 7 comprises: A data acquisition module, used to obtain historical computing resource demand information of multiple unmanned retail stores; A data analysis module, for determining the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores based on historical computing resource demand information of multiple unmanned retail stores; A deployment optimization module is used to determine the edge computing deployment plan based on the resource demand characteristics of each unmanned retail store and the resource demand correlation characteristics between any two unmanned retail stores; An edge computing module, comprising a plurality of edge computing devices arranged according to the edge computing deployment scheme; An information acquisition module is used to obtain auxiliary information for demand forecasting; A demand prediction module, used to predict the future computing resource requirements of multiple unmanned retail stores based on the resource demand characteristics of each unmanned retail store, the resource demand correlation characteristics between any two unmanned retail stores, and demand prediction auxiliary information; The information acquisition module is also used to acquire resource allocation auxiliary information; A resource allocation module is used to determine an edge computing resource allocation plan based on future computing resource requirements of multiple unmanned retail stores and resource allocation auxiliary information, and allocate edge computing resources to the multiple unmanned retail stores according to the edge computing resource allocation plan.

Images