Collaborative scheduling method and system for edge computing

Abstract

The present invention provides a collaborative scheduling method and system for edge computing, which relates to the field of edge computing. The method comprises: calculating computing demand consistency parameters of any two unmanned retail stores according to historical computing demands of multiple unmanned retail stores, grouping the multiple unmanned retail stores, obtaining grouping results, and determining multiple edge computing nodes; establishing a weighted graph model according to the computing demand consistency parameters of any two unmanned retail stores and the grouping results; predicting the computing demands of multiple unmanned retail stores at multiple consecutive future time points according to the computing demand consistency parameters of any two unmanned retail stores and the computing demands of multiple unmanned retail stores at multiple consecutive current time points; determining the dynamic weights of edges according to the computing demands of multiple unmanned retail stores at multiple consecutive future time points, updating the weighted graph model, and determining the optimal edge computing scheduling scheme, which has the advantage of improving the quality of edge computing scheduling.

Description

Technical Field

[0001] The present invention relates to the field of edge computing, and in particular to a collaborative scheduling method and system for edge computing. Background Art

[0002] With the development of the Internet of Everything, the contradiction of communication delay in the mode where cloud computing directly processes computing tasks from mobile devices has become increasingly prominent, and it is becoming increasingly difficult to meet high real-time requirements. Edge computing has emerged as the times require. Edge computing provides computing close to the data, which provides new possibilities for low latency and privacy and security protection. In the retail industry, edge computing also plays an important role. Unmanned retail stores use in-store cameras, sensors, and edge computing to identify customers and record the items purchased by customers to achieve a seamless shopping experience. This technology not only simplifies the shopping process, but also provides retailers with valuable data insights.

[0003] In edge scenarios, on the one hand, the dynamic changes of edge computing nodes and mobile devices make the implementation of edge computing more difficult; on the other hand, the computing power of edge computing nodes is limited, and the service capacity of a single edge node is very limited, but the scale is huge. Therefore, for edge scenarios, multi-edge node computing collaboration is an effective strategy to fully release edge computing power, greatly improving task processing capabilities and service experience. In the research of edge collaboration, collaborative scheduling is an important link and a current research hotspot. The computing power of edge nodes is often weak, and business requests also have the characteristics of temporal and spatial inhomogeneity. A single edge computing node cannot handle excessive computing tasks. Edge computing nodes need to transfer excessive computing tasks to cloud computing platforms or other edge computing nodes for processing. However, edge computing nodes cannot accurately predict the load conditions and load unloading decisions of other edge computing nodes. Once an edge computing node transfers computing tasks to an edge computing node with heavy load, it will not only fail to speed up the processing progress of computing tasks, but will prolong the processing progress of computing tasks, seriously affecting the user experience.

[0004] Therefore, it is necessary to provide a collaborative scheduling method and system for edge computing to improve the quality of edge computing scheduling. Summary of the invention

[0005] The present invention provides a collaborative scheduling method for edge computing, comprising: obtaining historical computing demands of a plurality of unmanned retail stores; calculating computing demand consistency parameters of any two unmanned retail stores according to the historical computing demands of the plurality of unmanned retail stores; grouping the plurality of unmanned retail stores according to the computing demand consistency parameters of any two unmanned retail stores to obtain grouping results; determining a plurality of edge computing nodes according to the grouping results; establishing a weighted graph model according to the computing demand consistency parameters of any two unmanned retail stores and the grouping results, wherein the weighted graph model comprises demand nodes representing unmanned retail stores and computing nodes representing edge computing nodes, and the demand nodes are associated with at least one node. The points are connected by edges, and the static weight of the edge is determined based on the transmission delay between the unmanned retail store and the edge computing node and the computing demand consistency parameter of any two unmanned retail stores; the computing demand of multiple unmanned retail stores at multiple consecutive current time points is obtained; based on the computing demand consistency parameter of any two unmanned retail stores and the computing demand of multiple unmanned retail stores at multiple consecutive current time points, the computing demand of multiple unmanned retail stores at multiple consecutive future time points is predicted; based on the computing demand of multiple unmanned retail stores at multiple consecutive future time points, the dynamic weight of the edge is determined, the weighted graph model is updated, and the optimal edge computing scheduling plan is determined based on the updated weighted graph model.

[0006] Further, the historical computing needs of the unmanned retail stores include the computing needs of the unmanned retail stores in multiple historical time periods; based on the historical computing needs of multiple unmanned retail stores, the computing need consistency parameters of any two unmanned retail stores are calculated, including: for each historical time period, according to the computing needs of each unmanned retail store in the historical time period, the validity parameter of the historical time period is calculated; according to the validity parameter of each historical time period, the valid historical time period is screened from multiple historical time periods; for each unmanned retail store, according to the computing needs of the unmanned retail store in each valid historical time period, the mean computing need and the computing need fluctuation parameter of the unmanned retail store in each valid historical time period are calculated; for any two unmanned retail stores, according to the mean computing need of the two unmanned retail stores in each valid historical time period, the mean consistency parameter of the two unmanned retail stores is calculated, according to the computing need fluctuation parameter of the two unmanned retail stores in each valid historical time period, the fluctuation consistency parameter of the two unmanned retail stores is calculated, and according to the mean consistency parameter and fluctuation consistency parameter of the two unmanned retail stores, the computing need consistency parameter of the two unmanned retail stores is calculated.

[0007] Furthermore, according to the computing requirement consistency parameters of any two unmanned retail stores, multiple unmanned retail stores are grouped to obtain a grouping result, including: S11, for each unmanned retail store, the basic computing requirement of the unmanned retail store is determined according to the historical computing requirement of the unmanned retail store; S12, according to the computing requirement consistency parameters of any two unmanned retail stores, the number of groups M is initialized; S13, according to the number of groups M, M unmanned retail stores are selected from multiple unmanned retail stores as group center unmanned retail stores; S14, for each unmanned retail store, according to the computing requirement consistency parameters of the unmanned retail store and each group center unmanned retail store, multiple unmanned retail stores are sorted to generate a first sorting result; S15, according to the first sorting result, the unmanned retail store to be currently grouped is determined; S16, according to the computing requirement consistency parameters of the unmanned retail store to be currently grouped and each group center unmanned retail store, each unmanned retail store is sorted. The group computing requirements of the unmanned retail store group corresponding to the group center unmanned retail store and the maximum threshold of the group computing requirements are determined to determine the group center unmanned retail store that matches the unmanned retail store to be grouped currently, and the group computing requirements of the unmanned retail store group corresponding to the matched group center unmanned retail store are updated according to the basic computing requirements of the unmanned retail store, and S17 is executed; S17, whether all unmanned retail stores have completed the allocation, if so, execute S18, if not, execute S15; S18, according to the computing requirement consistency parameters of any two unmanned retail stores included in the unmanned retail store group, determine whether each unmanned retail store group meets the grouping requirements, if so, obtain the grouping result, if not, execute S19; S19, according to the computing requirement consistency parameters of any two unmanned retail stores included in the unmanned retail store group that does not meet the grouping requirements, add a group center unmanned retail store from the unmanned retail store group that does not meet the grouping requirements, and execute S14.

[0008] Further, according to the computing requirement consistency parameters between the current unmanned retail store to be grouped and each group center unmanned retail store, the group computing requirement of the unmanned retail store group corresponding to each group center unmanned retail store, and the maximum threshold of the group computing requirement, the group center unmanned retail store matching the current unmanned retail store to be grouped is determined, including: S161, according to the computing requirement consistency parameters between the current unmanned retail store to be grouped and each group center unmanned retail store, a plurality of group center unmanned retail stores are sorted to generate a second sorting result, wherein in the second sorting result, the group center unmanned retail store with a smaller computing requirement consistency parameter with the current unmanned retail store to be grouped is sorted higher; S162, according to the second sorting result, the current group center unmanned retail store is determined; S163, according to the basic computing requirements of the unmanned retail store, the current The group computing demand of the unmanned retail store group corresponding to the group center unmanned retail store and the maximum threshold of the group computing demand are used to determine whether the current group center unmanned retail store is the group center unmanned retail store that matches the current unmanned retail store to be grouped. If so, determine the group center unmanned retail store that matches the current unmanned retail store to be grouped, and execute S164. If not, execute S162. S164, determine whether all group center unmanned retail stores have completed the matching. If so, execute S165. S165, based on the basic computing demand of the unmanned retail store, update the group computing demand of the unmanned retail store group corresponding to each group center unmanned retail store, and use the group center unmanned retail store of the unmanned retail store group with the smallest difference between the updated group computing demand and the maximum threshold of the group computing demand as the group center unmanned retail store that matches the current unmanned retail store to be grouped.

[0009] Furthermore, based on the grouping results, multiple edge computing nodes are determined, including: for each unmanned retail store group, the location and configuration of the edge computing node corresponding to the unmanned retail store group are determined based on the location of each unmanned retail store included in the unmanned retail store group, the computing requirement consistency parameters of any two unmanned retail stores, and the basic computing requirements of each unmanned retail store.

[0010] Furthermore, a weighted graph model is established based on the computing demand consistency parameters and grouping results of any two unmanned retail stores, including: for each unmanned retail store, the candidate edge computing nodes of the unmanned retail store are determined based on the transmission delay between the unmanned retail store and each edge computing node; the static weights of the edges connecting the demand nodes representing the unmanned retail stores and the computing nodes representing the candidate edge computing nodes are calculated based on the transmission delay between the unmanned retail store and the candidate edge computing nodes and the computing demand consistency parameters of each unmanned retail store included in the unmanned retail store group corresponding to the unmanned retail store and the candidate edge computing nodes; the weighted graph model is established based on the candidate edge computing nodes of each unmanned retail store and the static weights of the edges connecting the demand nodes representing the unmanned retail stores and the computing nodes representing the candidate edge computing nodes.

[0011] Furthermore, based on the computing demand consistency parameters of any two unmanned retail stores and the computing demands of the multiple unmanned retail stores at multiple consecutive current time points, the computing demands of the multiple unmanned retail stores at multiple consecutive future time points are predicted, including: for each unmanned retail store, based on the computing demand consistency parameters of any two unmanned retail stores, a reference unmanned retail store is determined, and the initial computing demands of the unmanned retail store at multiple consecutive future time points are predicted based on the computing demands of the unmanned retail store at multiple consecutive current time points through a demand prediction model, and the initial computing demands of the unmanned retail store at multiple consecutive future time points are corrected based on the initial computing demands of the reference unmanned retail store at multiple consecutive future time points through a demand correction model to predict the computing demands of the unmanned retail store at multiple consecutive future time points.

[0012] Furthermore, according to the computing requirements of multiple unmanned retail stores at multiple consecutive future time points, the dynamic weights of the edges are determined, the weighted graph model is updated, and based on the updated weighted graph model, the optimal edge computing scheduling scheme is determined, including: S21, according to the computing requirements of multiple unmanned retail stores at multiple consecutive future time points and the configuration of the edge computing nodes corresponding to the unmanned retail store group, the unmanned retail stores to be scheduled are determined; S22, according to the computing requirements of the unmanned retail stores to be scheduled at multiple consecutive future time points, the unmanned retail stores to be scheduled are sorted to generate a third sorting result; S23, according to the computing requirements of other unmanned retail stores at multiple consecutive future time points, the computing loads of multiple edge computing nodes at multiple consecutive future time points are determined; S24, according to the third sorting result, the current unmanned retail store to be scheduled is determined; S25, according to multiple The computing load of the edge computing node at multiple consecutive future time points is used to determine the dynamic weights of the edges connecting the demand nodes representing the current unmanned retail store to be scheduled and the computing nodes representing the candidate edge computing nodes of the current unmanned retail store to be scheduled, and update the weighted graph model; S26. According to the static weights and dynamic weights of the edges of the computing nodes representing the candidate edge computing nodes of the current unmanned retail store to be scheduled, the optimal edge computing node of the current unmanned retail store to be scheduled is determined; S27. According to the computing demand of the current unmanned retail store to be scheduled at multiple consecutive future time points, the computing load of the optimal edge computing node of the current unmanned retail store to be scheduled at multiple consecutive future time points is updated; S28. Determine whether all the unmanned retail stores to be scheduled have completed scheduling. If so, generate the optimal edge computing scheduling plan. If not, execute S24.

[0013] Furthermore, according to the computational loads of multiple edge computing nodes at multiple consecutive future time points, the dynamic weights of the edges connecting the demand nodes representing the current unmanned retail store to be scheduled and the computing nodes representing the candidate edge computing nodes of the current unmanned retail store to be scheduled are determined, including: according to the computational loads of multiple edge computing nodes at multiple consecutive future time points and the computational demands of the current unmanned retail store to be scheduled at multiple consecutive future time points, valid candidate edge computing nodes and invalid candidate edge computing nodes of the current unmanned retail store to be scheduled are determined, and the dynamic weights of the invalid candidate edge computing nodes are assigned to 0; for the valid candidate edge computing nodes of the current unmanned retail store to be scheduled, the dynamic weights of the edges connecting the demand nodes representing the current unmanned retail store to be scheduled and the computing nodes representing the candidate edge computing nodes of the current unmanned retail store to be scheduled are determined according to the computational loads of the valid candidate edge computing nodes of the current unmanned retail store to be scheduled at multiple consecutive future time points.

[0014] The present invention provides a collaborative scheduling system for edge computing, which is applied to the above-mentioned collaborative scheduling method for edge computing, including: a data acquisition module, which is used to obtain the historical computing requirements of multiple unmanned retail stores; a data analysis module, which is used to calculate the computing requirement consistency parameters of any two unmanned retail stores based on the historical computing requirements of the multiple unmanned retail stores; a store grouping module, which is used to group the multiple unmanned retail stores according to the computing requirement consistency parameters of any two unmanned retail stores to obtain grouping results; an edge optimization module, which is used to determine multiple edge computing nodes according to the grouping results; a graph model establishment module, which is used to establish a weighted graph model according to the computing requirement consistency parameters and grouping results of any two unmanned retail stores, wherein the weighted graph model includes demand nodes representing unmanned retail stores and nodes representing edge nodes. The computing node of the computing node, the demand node is connected to at least one node through an edge, and the static weight of the edge is determined based on the transmission delay between the unmanned retail store and the edge computing node and the computing demand consistency parameter of any two unmanned retail stores; a demand acquisition module is used to obtain the computing demand of multiple unmanned retail stores at multiple consecutive current time points; a demand prediction module is used to predict the computing demand of multiple unmanned retail stores at multiple consecutive future time points based on the computing demand consistency parameters of any two unmanned retail stores and the computing demand of multiple unmanned retail stores at multiple consecutive current time points; a collaborative scheduling module is used to determine the dynamic weight of the edge based on the computing demand of multiple unmanned retail stores at multiple consecutive future time points, update the weighted graph model, and determine the optimal edge computing scheduling plan based on the updated weighted graph model.

[0015] Compared with the prior art, the collaborative scheduling method and system for edge computing provided by the present invention have at least the following beneficial effects:

[0016] 1. By calculating the historical computing needs of unmanned retail stores and grouping the stores accordingly, it is possible to more accurately match edge computing resources with the actual needs of the stores. This avoids over-allocation or under-allocation of resources and improves the efficiency of resource use. By establishing a weighted graph model and considering the transmission delay between the store and the edge computing node and the consistency parameters of the computing needs between the stores, this method can formulate a more reasonable scheduling strategy. This helps to reduce data transmission delays, improve system response speed, and thus improve user experience. It can predict future computing needs based on the historical and current computing needs of the store. This enables the system to schedule resources in advance to ensure that sufficient computing resources are available during peak demand, thereby avoiding service interruptions or performance degradation caused by insufficient resources. The method of dynamically updating the weighted graph model can flexibly adjust the scheduling strategy according to actual conditions. This helps to adapt to changing business needs and market environments and improve flexibility and adaptability. It can reduce data transmission delays and improve system response speed, so it can provide users with a smoother and more efficient service experience. This is of great significance to improving user satisfaction and loyalty.

[0017] 2. By calculating the mean and fluctuation parameters of the computing demand of each unmanned retail store in each valid historical time period, as well as their mean consistency and fluctuation consistency parameters with other stores, the similarity of computing demand between stores can be more finely characterized. Grouping stores based on these parameters can ensure the accuracy of grouping, thereby more accurately matching edge computing resources, improving the efficiency and accuracy of resource allocation, and avoiding excessive computing load on one or several edge computing nodes in a certain period of time.

[0018] 3. By predicting the computing needs of unmanned retail stores at multiple consecutive future time points, resource scheduling can be planned in advance to avoid resource shortages during peak demand or waste of resources during low demand. Combined with the weighted graph model and the dynamic weight update mechanism, the optimal edge computing node can be quickly determined, thereby improving scheduling efficiency. The introduction of dynamic weights can flexibly adjust the scheduling strategy according to the actual demand changes of unmanned retail stores. When the demand of a store increases, resources can be quickly dispatched to the edge computing nodes near the store to meet its needs. The computing needs of multiple unmanned retail stores at multiple consecutive future time points, as well as the configuration and computing load of edge computing nodes, are considered, so that a more reasonable resource allocation plan can be formulated. By optimizing resource allocation, it is possible to ensure that each store has sufficient computing resources while avoiding resource waste. Through prediction and scheduling, it is possible to respond to possible resource shortages or demand peaks in advance, thereby avoiding edge computing crashes or performance degradation. This improved stability helps to maintain the normal operation of unmanned retail stores and improve consumer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] 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:

[0020] Figure 1 It is a flowchart of a collaborative scheduling method for edge computing according to some embodiments of this specification;

[0021] Figure 2 is a schematic diagram of a process of grouping multiple unmanned retail stores according to some embodiments of this specification;

[0022] Figure 3 It is a schematic diagram of a process for determining an optimal edge computing scheduling solution according to some embodiments of this specification;

[0023] Figure 4 is a schematic diagram of a weighted graph model according to some embodiments of this specification;

[0024] Figure 5 It is a module diagram of a collaborative scheduling system for edge computing according to some embodiments of this specification. DETAILED DESCRIPTION

[0025] 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.

[0026] Figure 1 is a flow chart of a collaborative scheduling method for edge computing according to some embodiments of this specification, such as Figure 1 As shown, the collaborative scheduling method for edge computing may include the following steps.

[0027] Step 110, obtaining historical computing requirements of multiple unmanned retail stores.

[0028] Specifically, the historical computing requirements of the unmanned retail store include computing requirements of the unmanned retail store in multiple historical time periods.

[0029] Step 120: Calculate computing demand consistency parameters of any two unmanned retail stores based on historical computing demands of the plurality of unmanned retail stores.

[0030] Specifically include:

[0031] For each historical time period, the validity parameter of the historical time period is calculated according to the computing requirements of each unmanned retail store in the historical time period;

[0032] Filter valid historical time periods from multiple historical time periods according to the validity parameters of each historical time period;

[0033] For each unmanned retail store, the mean value of the computing demand and the computing demand fluctuation parameter of the unmanned retail store in each valid historical time period are calculated according to the computing demand of the unmanned retail store in each valid historical time period. For example, the computing demand of the unmanned retail store at multiple historical time points in the valid historical time period may be averaged as the mean value of the computing demand of the unmanned retail store in the valid historical time period, and the standard deviation of the computing demand of the unmanned retail store at multiple historical time points in the valid historical time period may be calculated as the computing demand fluctuation parameter of the unmanned retail store in the valid historical time period;

[0034] For any two unmanned retail stores, the mean consistency parameter of the two unmanned retail stores is calculated based on the mean calculated demand of the two unmanned retail stores in each valid historical time period. The fluctuation consistency parameter of the two unmanned retail stores is calculated based on the calculated demand fluctuation parameter of the two unmanned retail stores in each valid historical time period. The calculated demand consistency parameter of the two unmanned retail stores is calculated based on the mean consistency parameter and fluctuation consistency parameter of the two unmanned retail stores.

[0035] Specifically, the validity parameter of the historical time period can be calculated according to the following formula:

[0036]

[0037] in, is the validity parameter of the historical time period, is the computing demand of the i-th unmanned retail store at the t-th historical time point in the historical time period, is the total number of historical time points included in the historical time period, is the total number of unmanned retail stores.

[0038] The historical time period in which the validity parameter is greater than the validity parameter threshold may be taken as the valid historical time period.

[0039] The computing demand consistency parameter of two unmanned retail stores can be calculated according to the following formula:

[0040]

[0041] in, is the computing demand consistency parameter between the i-th unmanned retail store and the j-th unmanned retail store, and is the weight, and greater than 0, , is the mean consistency parameter between the i-th unmanned retail store and the j-th unmanned retail store, is the fluctuation consistency parameter between the i-th unmanned retail store and the j-th unmanned retail store, is the average value of the demand for the i-th unmanned retail store in the n-th valid historical period, is the average value of the demand for the j-th unmanned retail store in the n-th valid historical period, is the total number of valid historical time periods, is the calculation demand fluctuation parameter of the i-th unmanned retail store in the n-th valid historical time period, Calculate the demand fluctuation parameters for the j-th unmanned retail store in the n-th valid historical time period.

[0042] Step 130: grouping the multiple unmanned retail stores according to the calculation requirement consistency parameters of any two unmanned retail stores to obtain a grouping result.

[0043] Figure 2 is a schematic diagram of a process of grouping multiple unmanned retail stores according to some embodiments of this specification, such as Figure 2 As shown, step 130 specifically includes:

[0044] S11. For each unmanned retail store, determine the basic computing demand of the unmanned retail store according to the historical computing demand of the unmanned retail store. For example, the historical computing demand of the unmanned retail store may be averaged to serve as the basic computing demand of the unmanned retail store.

[0045] S12, initializing the number of groups M according to the computing requirement consistency parameters of any two unmanned retail stores;

[0046] S13. According to the number of groups M, M unmanned retail stores are selected from the plurality of unmanned retail stores as the group center unmanned retail stores;

[0047] S14. For each unmanned retail store, sort the multiple unmanned retail stores according to the computing requirement consistency parameter between the unmanned retail store and each group center unmanned retail store to generate a first sorting result. For example, the computing requirement consistency parameter between the unmanned retail store and each group center unmanned retail store may be averaged as the first sorting value of the unmanned retail store, and the multiple unmanned retail stores may be sorted from small to large according to the first sorting value to generate a first sorting result.

[0048] S15. Determine the unmanned retail store to be currently grouped according to the first ranking result. For example, determine the unmanned retail store with the smallest first ranking value among the ungrouped unmanned retail stores as the unmanned retail store to be currently grouped.

[0049] S16, according to the computing requirement consistency parameters between the unmanned retail stores to be grouped currently and each group center unmanned retail store, the group computing requirement of the unmanned retail store group corresponding to each group center unmanned retail store, and the group computing requirement maximum threshold, determine the group center unmanned retail store that matches the unmanned retail stores to be grouped currently, update the group computing requirement of the unmanned retail store group corresponding to the matched group center unmanned retail store according to the basic computing requirement of the unmanned retail store, and execute S17;

[0050] S17, determining whether all unmanned retail stores have completed the allocation, if so, executing S18, if not, executing S15;

[0051] S18, judging whether each unmanned retail store group meets the grouping requirement according to the computing requirement consistency parameters of any two unmanned retail stores included in the unmanned retail store group, and if so, obtaining the grouping result, and if not, executing S19, for example, the computing requirement consistency parameters of any two unmanned retail stores included in the unmanned retail store group may be averaged to obtain the computing requirement consistency parameter mean corresponding to the unmanned retail store group, and the grouping requirement may be that the computing requirement consistency parameter mean is less than the computing requirement consistency parameter mean threshold;

[0052] S19. According to the computing requirement consistency parameters of any two unmanned retail stores included in the group of unmanned retail stores that do not meet the grouping requirements, a new group center unmanned retail store is added from the group of unmanned retail stores that do not meet the grouping requirements, and S14 is executed. For example, for each unmanned retail store that is not a center unmanned retail store included in the group of unmanned retail stores that do not meet the grouping requirements, the standard deviation of the computing requirement consistency parameters of other unmanned retail stores and the unmanned retail store is calculated as the first center value of the unmanned retail store, and the unmanned retail store with the largest first center value is selected as the newly added group center unmanned retail store.

[0053] Specifically, a multiple regression model can be established, wherein the independent variables of the multiple regression model can include the mean of the calculation demand consistency parameter, the standard deviation of the calculation demand consistency parameter, and the median of the calculation demand consistency parameter, and the dependent variable of the multiple regression model can be the number of groups M. Collect multiple groups of sample data to solve the multiple regression model. According to the calculation demand consistency parameters of any two unmanned retail stores, determine the mean of the calculation demand consistency parameter, the standard deviation of the calculation demand consistency parameter, and the median of the calculation demand consistency parameter, substitute the mean of the calculation demand consistency parameter, the standard deviation of the calculation demand consistency parameter, and the median of the calculation demand consistency parameter into the solved multiple regression model to obtain the initialized number of groups M.

[0054] For each unmanned retail store, the standard deviation of the calculation demand consistency parameters of other unmanned retail stores and the unmanned retail store is calculated as the second central value of the unmanned retail store.

[0055] The unmanned retail stores are sorted from large to small according to the second center value, and the top M unmanned retail stores are selected as the group center unmanned retail stores.

[0056] In some embodiments, according to the computing requirement consistency parameter between the unmanned retail store currently to be grouped and each group center unmanned retail store, the group computing requirement of the unmanned retail store group corresponding to each group center unmanned retail store, and the group computing requirement maximum threshold, determining the group center unmanned retail store that matches the unmanned retail store currently to be grouped includes:

[0057] S161, sorting the plurality of group center unmanned retail stores according to the computing requirement consistency parameters between the current unmanned retail store to be grouped and each group center unmanned retail store, and generating a second sorting result, wherein the group center unmanned retail store with a smaller computing requirement consistency parameter with the current unmanned retail store to be grouped in the second sorting result is sorted higher;

[0058] S162: Determine the current group center unmanned retail store according to the second sorting result, for example, the group center unmanned retail store with the smallest calculation requirement consistency parameter among the unmatched unmanned retail stores to be grouped is used as the current group center unmanned retail store;

[0059] S163, judging whether the current group center unmanned retail store is the group center unmanned retail store that matches the current unmanned retail store to be grouped according to the basic computing demand of the unmanned retail store, the group computing demand of the unmanned retail store group corresponding to the current group center unmanned retail store, and the maximum threshold of the group computing demand, if so, determining the group center unmanned retail store that matches the current unmanned retail store to be grouped, and executing S164, if not, executing S162, for example, if the sum of the basic computing demand of the unmanned retail store and the group computing demand of the unmanned retail store group corresponding to the current group center unmanned retail store is less than the maximum threshold of the group computing demand, then determining that the current group center unmanned retail store is the group center unmanned retail store that matches the current unmanned retail store to be grouped;

[0060] S164, determining whether all the group center unmanned retail stores have completed matching, if so, executing S165;

[0061] S165. Update the group computing requirements of the unmanned retail store group corresponding to each group center unmanned retail store based on the basic computing requirements of the unmanned retail stores, and use the group center unmanned retail store of the unmanned retail store group with the smallest difference between the updated group computing requirements and the maximum threshold of the group computing requirements as the group center unmanned retail store that matches the unmanned retail stores to be grouped.

[0062] Step 140: determine multiple edge computing nodes according to the grouping result.

[0063] Specifically include:

[0064] For each unmanned retail store group, the location and configuration of the edge computing node corresponding to the unmanned retail store group are determined based on the location of each unmanned retail store included in the unmanned retail store group, the computing requirement consistency parameters of any two unmanned retail stores, and the basic computing requirements of each unmanned retail store.

[0065] Specifically, the location and configuration of the edge computing nodes corresponding to the unmanned retail store group can be determined by the edge computing optimization model according to the location of each unmanned retail store included in the unmanned retail store group, the computing requirement consistency parameters of any two unmanned retail stores, and the basic computing requirements of each unmanned retail store. The edge computing optimization model can be a convolutional neural network model.

[0066] Step 150, establish a weighted graph model based on the computing demand consistency parameters and grouping results of any two unmanned retail stores, wherein the weighted graph model includes demand nodes representing unmanned retail stores and computing nodes representing edge computing nodes, the demand nodes are connected to at least one node through an edge, and the static weight of the edge is determined based on the transmission delay between the unmanned retail store and the edge computing node and the computing demand consistency parameters of any two unmanned retail stores.

[0067] Specifically include:

[0068] For each unmanned retail store, determine the candidate edge computing nodes of the unmanned retail store based on the transmission delay between the unmanned retail store and each edge computing node, and calculate the static weight of the edge connecting the demand node representing the unmanned retail store and the computing node representing the candidate edge computing node based on the transmission delay between the unmanned retail store and the candidate edge computing node and the computing demand consistency parameter of each unmanned retail store included in the unmanned retail store group corresponding to the unmanned retail store and the candidate edge computing node. For example, an edge computing node with a transmission delay less than a transmission delay threshold may be used as a candidate edge computing node of the unmanned retail store;

[0069] A weighted graph model is established based on the candidate edge computing nodes of each unmanned retail store and the static weights of the edges connecting the demand nodes representing the unmanned retail store and the computing nodes representing the candidate edge computing nodes.

[0070] Specifically, the static weight of the edge connecting the demand node representing the unmanned retail store and the computing node representing the candidate edge computing node can be calculated according to the following formula:

[0071]

[0072] in, is the static weight of the edge connecting the demand node representing the i-th unmanned retail store and the computing node representing the k-th candidate edge computing node of the i-th unmanned retail store, is the importance parameter of the edge connecting the demand node representing the i-th unmanned retail store and the computing node representing the k-th candidate edge computing node of the i-th unmanned retail store, is the importance parameter of the edge connecting the demand node representing the i-th unmanned retail store and the computing node representing the j-th candidate edge computing node of the i-th unmanned retail store, K is the total number of candidate edge computing nodes of the i-th unmanned retail store, and is the weight, and greater than 0, , and is the normalization parameter, and greater than 0, is the transmission delay between the i-th unmanned retail store and the k-th candidate edge computing node of the i-th unmanned retail store, is the computing requirement consistency parameter of the mth unmanned retail store included in the unmanned retail store group corresponding to the kth candidate edge computing node of the i-th unmanned retail store, M is the total number of unmanned retail stores included in the unmanned retail store group corresponding to the kth candidate edge computing node of the i-th unmanned retail store, The calculation method of The calculation method of is similar and will not be repeated here.

[0073] For example only, Figure 4 is a schematic diagram of a weighted graph model according to some embodiments of this specification, such as Figure 4 As shown, there are six unmanned retail stores (A1, A2, A3, A4, A5, A6) and three edge computing nodes (B1, B2, B3), among which the candidate edge computing nodes of the unmanned retail store A1 are B1 and B2, and the static weights of their corresponding edges are 0.4 and 0.6 respectively; the candidate edge computing nodes of the unmanned retail store A2 are B1, B2, B3, and the static weights of their corresponding edges are 0.3, 0.5, and 0.2 respectively; the candidate edge computing nodes of the unmanned retail store A3 are B1 and B3, and the static weights of their corresponding edges are 0.2 and 0.8 respectively; the candidate edge computing nodes of the unmanned retail store A4 are B1 and B3, and the static weights of their corresponding edges are 0.3 and 0.7 respectively; the candidate edge computing nodes of the unmanned retail store A5 are B1 and B2, and the static weights of their corresponding edges are 0.2 and 0.8 respectively; the candidate edge computing nodes of the unmanned retail store A6 are B2 and B3, and the static weights of their corresponding edges are 0.5 and 0.5 respectively.

[0074] Step 160, obtaining computing requirements of multiple unmanned retail stores at multiple consecutive current time points.

[0075] Step 170 , predicting the computing requirements of the plurality of unmanned retail stores at a plurality of consecutive future time points based on the computing requirement consistency parameters of any two unmanned retail stores and the computing requirements of the plurality of unmanned retail stores at a plurality of consecutive current time points.

[0076] Specifically include:

[0077] For each unmanned retail store, a reference unmanned retail store is determined based on the computing demand consistency parameters of any two unmanned retail stores, and the initial computing demand of the unmanned retail store at multiple consecutive future time points is predicted based on the computing demand of the unmanned retail store at multiple consecutive current time points through the demand prediction model, and the initial computing demand of the unmanned retail store at multiple consecutive future time points is corrected based on the initial computing demand of the reference unmanned retail store at multiple consecutive future time points through the demand correction model, so as to predict the computing demand of the unmanned retail store at multiple consecutive future time points. For example, two unmanned retail stores whose computing demand consistency parameters are greater than the computing demand consistency parameter threshold are reference unmanned retail stores for each other, and the demand prediction model and the demand correction model can both be long short-term memory network models.

[0078] Step 180, according to the computing requirements of multiple unmanned retail stores at multiple consecutive future time points, determine the dynamic weights of the edges, update the weighted graph model, and determine the optimal edge computing scheduling solution based on the updated weighted graph model.

[0079] Figure 3 is a flow chart of determining an optimal edge computing scheduling solution according to some embodiments of this specification, such as Figure 3 As shown, step 180 specifically includes:

[0080] S21. Determine the unmanned retail store to be scheduled according to the computing requirements of multiple unmanned retail stores at multiple consecutive future time points and the configuration of the edge computing nodes corresponding to the unmanned retail store group. For example, for each unmanned retail store group, the computing requirements of each unmanned retail store included in the unmanned retail store group at multiple consecutive future time points may be summed to obtain the group computing requirements of the unmanned retail store group at multiple consecutive future time points, and determine whether there is a future time point at which the group computing requirement is greater than the group computing requirement threshold. If so, the unmanned retail store included in the unmanned retail store group with the smallest average computing requirement at multiple consecutive future time points is used as the unmanned retail store to be scheduled;

[0081] S22, sorting the unmanned retail stores to be scheduled according to the computing demands of the unmanned retail stores to be scheduled at multiple consecutive future time points to generate a third sorting result, for example, sorting the unmanned retail stores to be scheduled from large to small according to the average of the computing demands of the unmanned retail stores to be scheduled at multiple consecutive future time points to generate a third sorting result;

[0082] S23. Determine the computing loads of the plurality of edge computing nodes at the plurality of consecutive future time points according to the computing demands of the other unmanned retail stores at the plurality of consecutive future time points. Specifically, the computing load of the edge computing node at the future time point may be the sum of the computing demands of each corresponding other unmanned retail store at the future time point.

[0083] S24. Determine the current unmanned retail store to be scheduled according to the third sorting result. For example, the unmanned retail store to be scheduled that is not scheduled and has the highest third sorting result may be used as the current unmanned retail store to be scheduled.

[0084] S25, determining the dynamic weights of the edges connecting the demand nodes representing the current unmanned retail store to be scheduled and the candidate edge computing nodes representing the current unmanned retail store to be scheduled according to the computing loads of the multiple edge computing nodes at multiple consecutive future time points, and updating the weighted graph model;

[0085] S26. Determine the optimal edge computing node of the current unmanned retail store to be scheduled according to the static weights and dynamic weights of the edges of the computing nodes representing the candidate edge computing nodes of the current unmanned retail store to be scheduled, for example, by weighted summing the static weights and dynamic weights of the edges to obtain the comprehensive weights of the edges, and use the edge computing node connected to the edge with the largest comprehensive weight as the optimal edge computing node of the current unmanned retail store to be scheduled;

[0086] S27. According to the computing requirements of the current unmanned retail store to be scheduled at multiple consecutive future time points, update the computing load of the optimal edge computing node of the current unmanned retail store to be scheduled at multiple consecutive future time points. Specifically, the computing load of the updated edge computing node at the future time point may be the sum of the computing requirements of each other corresponding unmanned retail store at the future time point and the computing requirements of the current unmanned retail store to be scheduled at the future time point.

[0087] S28: Determine whether all unmanned retail stores to be scheduled have completed scheduling. If so, generate an optimal edge computing scheduling plan. If not, execute S24.

[0088] In some embodiments, determining the dynamic weight of the edge connecting the demand node representing the current unmanned retail store to be scheduled and the candidate edge computing node representing the current unmanned retail store to be scheduled according to the computing loads of the plurality of edge computing nodes at a plurality of consecutive future time points includes:

[0089] According to the computing loads of multiple edge computing nodes at multiple consecutive future time points and the computing requirements of the current unmanned retail store to be scheduled at multiple consecutive future time points, determine the valid candidate edge computing nodes and invalid candidate edge computing nodes of the current unmanned retail store to be scheduled, and assign the dynamic weight of the invalid candidate edge computing nodes to be 0. For example, the computing requirements of the current unmanned retail store to be scheduled at multiple consecutive future time points and the computing requirements of the candidate edge computing nodes of the current unmanned retail store to be scheduled at multiple consecutive future time points can be summed, and the candidate edge computing nodes whose computing requirements at the future time points after the summation are still less than the computing requirement threshold are used as valid candidate edge computing nodes of the current unmanned retail store to be scheduled, otherwise they are used as invalid candidate edge computing nodes;

[0090] For the valid candidate edge computing nodes of the current unmanned retail store to be scheduled, the dynamic weight of the edge connecting the demand node representing the current unmanned retail store to be scheduled and the computing node representing the candidate edge computing node of the current unmanned retail store to be scheduled is determined according to the computing load of the valid candidate edge computing nodes of the current unmanned retail store to be scheduled at multiple consecutive future time points.

[0091] Specifically, the dynamic weight of the edge connecting the demand node representing the current unmanned retail store to be scheduled and the computing node representing the candidate edge computing node of the current unmanned retail store to be scheduled can be calculated according to the following formula:

[0092]

[0093] in, is the dynamic weight of the edge connecting the demand node representing the current unmanned retail store to be scheduled (i.e., the i-th unmanned retail store) and the computing node representing the f-th valid candidate edge computing node of the current unmanned retail store to be scheduled, is the computing load ratio of the fth valid candidate edge computing node of the current unmanned retail store to be scheduled, is the computing load ratio of the e-th valid candidate edge computing node of the current unmanned retail store to be scheduled, F is the total number of valid candidate edge computing nodes of the current unmanned retail store to be scheduled, is the average of the computing load of the fth valid candidate edge computing node of the current unmanned retail store to be scheduled at multiple consecutive future time points, is the average of the computing loads of the e-th valid candidate edge computing node of the current unmanned retail store to be scheduled at multiple consecutive future time points.

[0094] Figure 5is a schematic diagram of a module of a collaborative scheduling system for edge computing according to some embodiments of this specification, such as Figure 5 As shown, the collaborative scheduling system for edge computing may include a data acquisition module, a data analysis module, a store grouping module, an edge optimization module, a graph model building module, a demand acquisition module, a demand forecasting module and a collaborative scheduling module.

[0095] A data acquisition module, used to obtain historical computing requirements of multiple unmanned retail stores;

[0096] A data analysis module, used to calculate the computing demand consistency parameters of any two unmanned retail stores based on the historical computing demands of multiple unmanned retail stores;

[0097] A store grouping module is used to group multiple unmanned retail stores according to the calculation requirement consistency parameters of any two unmanned retail stores to obtain grouping results;

[0098] An edge optimization module, used to determine multiple edge computing nodes according to the grouping results;

[0099] A graph model building module, used to build a weighted graph model according to the computing demand consistency parameters and grouping results of any two unmanned retail stores, wherein the weighted graph model includes a demand node representing the unmanned retail store and a computing node representing the edge computing node, the demand node is connected to at least one node through an edge, and the static weight of the edge is determined based on the transmission delay between the unmanned retail store and the edge computing node and the computing demand consistency parameters of any two unmanned retail stores;

[0100] A demand acquisition module, used to obtain the computing demands of multiple unmanned retail stores at multiple consecutive current time points;

[0101] A demand prediction module, used to predict the computing demands of multiple unmanned retail stores at multiple consecutive future time points based on the computing demand consistency parameters of any two unmanned retail stores and the computing demands of multiple unmanned retail stores at multiple consecutive current time points;

[0102] The collaborative scheduling module is used to determine the dynamic weights of edges and update the weighted graph model according to the computing needs of multiple unmanned retail stores at multiple consecutive future time points, and determine the optimal edge computing scheduling solution based on the updated weighted graph model.

[0103] The collaborative scheduling system for edge computing can be used to execute the collaborative scheduling method for edge computing, which will not be repeated here.

[0104] 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. A collaborative scheduling method for edge computing, characterized in that: include: Obtain historical computing requirements for multiple unmanned retail stores; Calculate the computing demand consistency parameters of any two unmanned retail stores based on the historical computing demands of multiple unmanned retail stores; According to the computing requirement consistency parameters of any two unmanned retail stores, multiple unmanned retail stores are grouped to obtain grouping results; According to the grouping results, multiple edge computing nodes are determined; A weighted graph model is established according to the computing demand consistency parameters and grouping results of any two unmanned retail stores, wherein the weighted graph model includes a demand node representing the unmanned retail store and a computing node representing the edge computing node, the demand node is connected to at least one node through an edge, and the static weight of the edge is determined based on the transmission delay between the unmanned retail store and the edge computing node and the computing demand consistency parameters of any two unmanned retail stores; Obtain computing requirements of multiple unmanned retail stores at multiple consecutive current time points; Predicting the computing demands of the multiple unmanned retail stores at multiple consecutive future time points based on the computing demand consistency parameters of any two unmanned retail stores and the computing demands of the multiple unmanned retail stores at multiple consecutive current time points; According to the computing needs of multiple unmanned retail stores at multiple consecutive future time points, the dynamic weights of the edges are determined, the weighted graph model is updated, and the optimal edge computing scheduling solution is determined based on the updated weighted graph model; The historical computing requirements of unmanned retail stores include the computing requirements of unmanned retail stores in multiple historical time periods; Based on the historical computing requirements of multiple unmanned retail stores, the computing requirement consistency parameters of any two unmanned retail stores are calculated, including: For each historical time period, calculating a validity parameter of the historical time period according to the computing requirements of each unmanned retail store in the historical time period; Filter valid historical time periods from multiple historical time periods according to the validity parameters of each historical time period; For each unmanned retail store, based on the computing demand of the unmanned retail store in each valid historical time period, calculate the average computing demand and computing demand fluctuation parameter of the unmanned retail store in each valid historical time period; For any two unmanned retail stores, the mean consistency parameter of the two unmanned retail stores is calculated based on the mean calculated demand of the two unmanned retail stores in each valid historical time period. The fluctuation consistency parameter of the two unmanned retail stores is calculated based on the calculated demand fluctuation parameter of the two unmanned retail stores in each valid historical time period. The calculated demand consistency parameter of the two unmanned retail stores is calculated based on the mean consistency parameter and fluctuation consistency parameter of the two unmanned retail stores.

2. The collaborative scheduling method for edge computing according to claim 1, characterized in that: According to the computing requirement consistency parameters of any two unmanned retail stores, multiple unmanned retail stores are grouped to obtain grouping results, including: S11. For each unmanned retail store, determine the basic computing requirements of the unmanned retail store based on the historical computing requirements of the unmanned retail store; S12, initializing the number of groups M according to the computing requirement consistency parameters of any two unmanned retail stores; S13. According to the number of groups M, M unmanned retail stores are selected from the plurality of unmanned retail stores as the group center unmanned retail stores; S14. For each unmanned retail store, sort the multiple unmanned retail stores according to the computing requirement consistency parameter between the unmanned retail store and each group center unmanned retail store to generate a first sorting result; S15. Determine the unmanned retail stores to be grouped according to the first sorting result; S16, according to the computing requirement consistency parameters between the unmanned retail stores to be grouped currently and each group center unmanned retail store, the group computing requirement of the unmanned retail store group corresponding to each group center unmanned retail store, and the group computing requirement maximum threshold, determine the group center unmanned retail store that matches the unmanned retail stores to be grouped currently, update the group computing requirement of the unmanned retail store group corresponding to the matched group center unmanned retail store according to the basic computing requirement of the unmanned retail store, and execute S17; S17, determining whether all unmanned retail stores have completed the allocation, if so, executing S18, if not, executing S15; S18, judging whether each unmanned retail store group meets the grouping requirements according to the computing requirement consistency parameters of any two unmanned retail stores included in the unmanned retail store group, and if so, obtaining the grouping result, and if not, executing S19; S19. According to the computing requirement consistency parameters of any two unmanned retail stores included in the unmanned retail store group that does not meet the grouping requirements, a group center unmanned retail store is added to the unmanned retail store group that does not meet the grouping requirements, and S14 is executed.

3. The collaborative scheduling method for edge computing according to claim 2, characterized in that: According to the computing requirement consistency parameters between the unmanned retail stores to be grouped and each group center unmanned retail store, the group computing requirement of the unmanned retail store group corresponding to each group center unmanned retail store, and the maximum threshold of the group computing requirement, the group center unmanned retail store matching the unmanned retail stores to be grouped is determined, including: S161, sorting the plurality of group center unmanned retail stores according to the computing requirement consistency parameters between the current unmanned retail store to be grouped and each group center unmanned retail store, and generating a second sorting result, wherein the group center unmanned retail store with a smaller computing requirement consistency parameter with the current unmanned retail store to be grouped in the second sorting result is sorted higher; S162. Determine the current group center unmanned retail store according to the second sorting result; S163, judging whether the current group center unmanned retail store is the group center unmanned retail store that matches the current unmanned retail store to be grouped, according to the basic computing requirements of the unmanned retail store, the group computing requirements of the unmanned retail store group corresponding to the current group center unmanned retail store, and the maximum threshold of the group computing requirements; if so, determining the group center unmanned retail store that matches the current unmanned retail store to be grouped, and executing S164; if not, executing S162; S164, determining whether all the group center unmanned retail stores have completed matching, if so, executing S165; S165. Update the group computing requirements of the unmanned retail store group corresponding to each group center unmanned retail store based on the basic computing requirements of the unmanned retail stores, and use the group center unmanned retail store of the unmanned retail store group with the smallest difference between the updated group computing requirements and the maximum threshold of the group computing requirements as the group center unmanned retail store that matches the unmanned retail stores to be grouped.

4. The collaborative scheduling method for edge computing according to claim 2, characterized in that: Based on the grouping results, multiple edge computing nodes are determined, including: For each unmanned retail store group, the location and configuration of the edge computing node corresponding to the unmanned retail store group are determined based on the location of each unmanned retail store included in the unmanned retail store group, the computing requirement consistency parameters of any two unmanned retail stores, and the basic computing requirements of each unmanned retail store.

5. The collaborative scheduling method for edge computing according to claim 4, characterized in that: According to the computing demand consistency parameters and grouping results of any two unmanned retail stores, a weighted graph model is established, including: For each unmanned retail store, determine a candidate edge computing node of the unmanned retail store based on the transmission delay between the unmanned retail store and each edge computing node, and calculate a static weight of an edge connecting a demand node representing the unmanned retail store and a computing node representing the candidate edge computing node based on the transmission delay between the unmanned retail store and the candidate edge computing node and a computing demand consistency parameter of each unmanned retail store included in an unmanned retail store group corresponding to the unmanned retail store and the candidate edge computing node; The weighted graph model is established according to the candidate edge computing nodes of each unmanned retail store and the static weights of the edges connecting the demand nodes representing the unmanned retail store and the computing nodes representing the candidate edge computing nodes.

6. The collaborative scheduling method for edge computing according to any one of claims 1 to 5, characterized in that: According to the computing demand consistency parameters of any two unmanned retail stores and the computing demands of the multiple unmanned retail stores at multiple consecutive current time points, the computing demands of the multiple unmanned retail stores at multiple consecutive future time points are predicted, including: For each unmanned retail store, a reference unmanned retail store is determined based on the computing demand consistency parameters of any two unmanned retail stores, and the initial computing demand of the unmanned retail store at multiple consecutive future time points is predicted based on the computing demand of the unmanned retail store at multiple consecutive current time points through the demand prediction model. The initial computing demand of the unmanned retail store at multiple consecutive future time points is corrected based on the initial computing demand of the reference unmanned retail store at multiple consecutive future time points through the demand correction model to predict the computing demand of the unmanned retail store at multiple consecutive future time points.

7. The collaborative scheduling method for edge computing according to claim 5, characterized in that: According to the computing needs of multiple unmanned retail stores at multiple consecutive future time points, the dynamic weights of the edges are determined, the weighted graph model is updated, and based on the updated weighted graph model, the optimal edge computing scheduling solution is determined, including: S21. Determine the unmanned retail store to be scheduled according to the computing requirements of the multiple unmanned retail stores at multiple consecutive future time points and the configuration of the edge computing nodes corresponding to the unmanned retail store group; S22, sorting the unmanned retail stores to be scheduled according to the computing requirements of the unmanned retail stores to be scheduled at multiple consecutive future time points, and generating a third sorting result; S23. Determine the computing loads of multiple edge computing nodes at multiple consecutive future time points according to the computing requirements of other unmanned retail stores at multiple consecutive future time points; S24. Determine the current unmanned retail store to be scheduled according to the third sorting result; S25, determining the dynamic weights of the edges connecting the demand nodes representing the current unmanned retail store to be scheduled and the candidate edge computing nodes representing the current unmanned retail store to be scheduled according to the computing loads of the multiple edge computing nodes at multiple consecutive future time points, and updating the weighted graph model; S26. Determine the optimal edge computing node of the current unmanned retail store to be scheduled according to the static weight and dynamic weight of the edge of the computing node representing the candidate edge computing node of the current unmanned retail store to be scheduled; S27. updating the computing load of the optimal edge computing node of the current unmanned retail store to be scheduled at multiple consecutive future time points according to the computing requirements of the current unmanned retail store to be scheduled at multiple consecutive future time points; S28: Determine whether all unmanned retail stores to be scheduled have completed scheduling. If so, generate an optimal edge computing scheduling plan. If not, execute S24.

8. The collaborative scheduling method for edge computing according to claim 7, characterized in that: Determining the dynamic weight of the edge connecting the demand node representing the current unmanned retail store to be scheduled and the candidate edge computing node representing the current unmanned retail store to be scheduled according to the computing loads of the multiple edge computing nodes at multiple consecutive future time points, including: Determine valid candidate edge computing nodes and invalid candidate edge computing nodes of the current unmanned retail store to be scheduled according to the computing loads of the multiple edge computing nodes at multiple consecutive future time points and the computing requirements of the current unmanned retail store to be scheduled at multiple consecutive future time points, and assign a dynamic weight of 0 to the invalid candidate edge computing node; For the valid candidate edge computing nodes of the current unmanned retail store to be scheduled, the dynamic weight of the edge connecting the demand node representing the current unmanned retail store to be scheduled and the computing node representing the candidate edge computing node of the current unmanned retail store to be scheduled is determined according to the computing load of the valid candidate edge computing nodes of the current unmanned retail store to be scheduled at multiple consecutive future time points.

9. A collaborative scheduling system for edge computing, characterized in that: The collaborative scheduling method for edge computing applied to any one of claims 1 to 8 comprises: A data acquisition module, used to obtain historical computing requirements of multiple unmanned retail stores; A data analysis module, used to calculate the computing demand consistency parameters of any two unmanned retail stores based on the historical computing demands of multiple unmanned retail stores; A store grouping module is used to group multiple unmanned retail stores according to the calculation requirement consistency parameters of any two unmanned retail stores to obtain grouping results; An edge optimization module, used to determine multiple edge computing nodes according to the grouping results; A graph model building module, used to build a weighted graph model according to the computing demand consistency parameters and grouping results of any two unmanned retail stores, wherein the weighted graph model includes a demand node representing the unmanned retail store and a computing node representing the edge computing node, the demand node is connected to at least one node through an edge, and the static weight of the edge is determined based on the transmission delay between the unmanned retail store and the edge computing node and the computing demand consistency parameters of any two unmanned retail stores; A demand acquisition module, used to obtain the computing demands of multiple unmanned retail stores at multiple consecutive current time points; A demand prediction module, used to predict the computing demands of multiple unmanned retail stores at multiple consecutive future time points based on the computing demand consistency parameters of any two unmanned retail stores and the computing demands of multiple unmanned retail stores at multiple consecutive current time points; The collaborative scheduling module is used to determine the dynamic weights of edges and update the weighted graph model according to the computing needs of multiple unmanned retail stores at multiple consecutive future time points, and determine the optimal edge computing scheduling solution based on the updated weighted graph model.

Images