Edge computing platform computing power distribution scheduling method and system

Abstract

The invention belongs to the technical field of computing power scheduling, and particularly relates to a computing power distribution scheduling method and system for an edge computing platform, and the method comprises the steps: generating a business pool based on a historical resource request task; constructing a business computing power demand prediction model based on XGBoost, adding the influence factors into the prediction model, generating resource request business prediction output of each node device, updating the business pool according to the prediction output, evaluating the computing power scale required by the resource request business in the updated business pool, and obtaining the computing power demand; according to the computing power demand obtained by each node device, taking energy consumption minimization and response time minimization as targets, carrying out multi-target solution based on an improved firefly algorithm, and obtaining a computing power distribution scheme which enables the energy consumption and the response time to reach the optimal; and allocating schedulable computing power resources in the edge computing platform based on the computing power allocation scheme. According to the method, overload of part of the data centers can be avoided, meanwhile, the overall calculation efficiency is improved, and calculation power resources are reasonably utilized.

Description

Technical Field

[0001] The present invention belongs to the technical field of computing power scheduling, and specifically relates to a computing power allocation and scheduling method and system for an edge computing platform. Background Art

[0002] With the continuous development of new-generation information technologies such as 5G, artificial intelligence, and big data, all walks of life have put forward more urgent demands for computing power and networks. As a new type of information infrastructure, computing power network is essentially a computing power resource service. Enterprise customers or individual users not only need networks and clouds, but also need to flexibly dispatch computing tasks to appropriate places. They need to allocate computing resources, storage resources, and network resources on demand and flexibly dispatch them between clouds, networks, and edges according to business needs.

[0003] Edge computing is a distributed computing architecture that moves data processing and analysis tasks from centralized data centers to the edge of the network, close to the data source. This architecture emerged to cope with the growing amount of data and application scenarios with increasingly high real-time requirements, such as the Internet of Things (IoT), self-driving cars, smart factories, etc. The core advantage of the edge computing platform is its ability to provide low latency, high bandwidth and real-time data processing capabilities; in the edge computing distributed computing network, the load imbalance of nodes may lead to resource waste and excessive energy consumption.

[0004] For example, some data centers may be overloaded with computing tasks, while other data centers may be relatively idle. These underutilized data centers still require a lot of energy to maintain, resulting in serious waste of resources. Currently, research on computing resource scheduling is still in its infancy. Traditional computing resource scheduling and allocation methods are usually static, which cannot adapt well to dynamic changes in tasks and network load conditions. This can easily lead to low computing resource utilization and poor accuracy in matching computing resources. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a computing power allocation and scheduling method and system for an edge computing platform, aiming to solve the technical problem that the computing power allocation and scheduling method in the prior art cannot adapt well to the dynamic changes of tasks and network load conditions, and easily causes low computing power resource utilization and poor accuracy in matching computing power resources.

[0006] To achieve the above object, the present invention provides the following technical solutions.

[0007] In one embodiment of the present invention, a method for allocating and scheduling computing power of an edge computing platform is provided, the method comprising the following steps:

[0008] Obtain the historical resource request services and influencing factors of each node device on the edge computing platform, and generate a service pool based on the historical resource request tasks;

[0009] Build a business computing power demand prediction model based on XGBoost, add influencing factors to the prediction model, generate resource request business prediction output for each node device, update the business pool according to the prediction output, evaluate the computing power scale required for resource request business in the updated business pool, and obtain the computing power demand;

[0010] According to the computing power requirements of each node device, with the goal of minimizing energy consumption and shortening response time, a multi-objective solution is performed based on the improved firefly algorithm to obtain a computing power allocation plan that achieves the best energy consumption and response time.

[0011] Based on the computing power allocation plan, the schedulable computing power resources in the edge computing platform are allocated.

[0012] In an optional implementation, the step of obtaining historical resource request services and influencing factors of each node device of the edge computing platform and generating a service pool based on the historical resource request tasks includes:

[0013] Collect historical resource usage data from each edge node, including CPU usage, memory usage, network bandwidth, and storage usage;

[0014] Collect factors that affect resource requests, including resource request time period, user type, number of users, and device type;

[0015] Integrate the collected data to generate a business pool of historical resource request tasks;

[0016] Feature extraction is performed on the data in the business pool to obtain time series data for training the business computing power demand prediction model based on XGBoost.

[0017] In an optional implementation, the step of building a business computing power demand prediction model based on XGBoost and adding influencing factors to the prediction model includes:

[0018] Get the time series data in the business pool, select m features and n samples of computing power resource occupancy, and build a regression model for business computing power requests under multi-dimensional resources. The feature values ​​of the regression model data include the number of CPU cores, GPU percentage occupancy, memory capacity requirements, and video memory percentage occupancy; the target value y of the regression model data i is the business execution time of the i-th sample;

[0019] The output of the business computing power demand prediction model based on XGBoost is expressed as:

[0020]

[0021] In formula (1), represents the predicted value after the tth iteration, that is, the business execution time predicted by the model based on the computing power resources; k represents the number of decision trees obtained after the iteration, and f represents the output function of each decision tree;

[0022] The objective function of the business computing power demand prediction model based on XGBoost is expressed as:

[0023]

[0024] In formula (2), l represents the loss function for measuring the model fit; Ω is the regularization representation used to measure the model complexity;

[0025] Map the influencing factor into an item in the loss function l:

[0026]

[0027] In formula (3), represents the original loss function; α represents the weight parameter, which is used to control the importance of the influencing factors in the loss function; h(X factors ) represents the mapping function, which is used to transform the influencing factor X factors Mapped to a term in the loss function l.

[0028] In an optional implementation, the step of updating the service pool according to the prediction output, evaluating the computing power scale required for the resource request service in the updated service pool, and obtaining the computing power demand includes:

[0029] According to the predicted output, new services corresponding to the predicted output are added to the initial service pool, and unfinished services in the initial service pool are retained to generate an updated service pool;

[0030] Sort the businesses in the update business pool, evaluate the required computing power scale for each business in the current business pool, and obtain the computing power demand based on the sorting results through a nonlinear activation function.

[0031] In an optional implementation, the step of obtaining computing power requirements according to each node device, taking minimizing energy consumption and minimizing response time as the goal, performing multi-objective solving based on an improved firefly algorithm, and obtaining a computing power allocation solution that optimizes both energy consumption and response time includes:

[0032] Set the computing power allocation, business computing power demand, energy consumption, response time parameters, and initialize the Firefly algorithm parameters;

[0033] The initial population is randomly generated and the firefly sequence is initialized using piecewise linear chaotic mapping;

[0034] The brightness and attractiveness of fireflies are calculated according to equations (4) and (5), where the brightness I of fireflies gradually decreases with the increase of distance r and the absorption of the medium, expressed as:

[0035]

[0036] In formula (4), I0 represents the brightness of the autofluorescence, which is the maximum brightness; I0 is positively correlated with the objective function value corresponding to the location of the firefly, γ represents the absorption factor, r ij represents the distance between fireflies i and j; the attraction related to the absorption factor and distance is expressed as:

[0037]

[0038] In formula (5), β0 represents the maximum attraction;

[0039] Calculate the step size factor and update the firefly position;

[0040] Sorting and crossover mutation operations are performed on fireflies according to the sorting-based firefly update strategy;

[0041] When the number of iterations reaches the maximum value, the iteration stops and the optimal computing power allocation plan is output.

[0042] In an optional implementation, the step of calculating the step size factor and updating the firefly position includes:

[0043] Furthermore, the step factor of the embodiment of the present invention follows a semi-Gaussian distribution with the number of iterations, and the step factor is expressed as:

[0044]

[0045] In formula (6), η represents the number of iterations, η max represents the maximum number of iterations; σ represents a value that is one-fourth of the maximum number of iterations; the range of the step factor ω is [ω min ,ω max ], this scheme can improve the dynamic adjustment of global and local search capabilities and improve the efficiency of the algorithm;

[0046] In updating the position of fireflies, the position update of the i-th firefly attracted by the j-th firefly is expressed as:

[0047] z i+1 =z i +β·(z j -z i )+ω·(rand-ζ) (7)

[0048] In formula (7), z j and z i They represent the spatial positions of fireflies j and i respectively; ω represents the step factor, rand is a random number that follows a uniform distribution between [0,1]; β represents the attractiveness of the firefly.

[0049] In an optional implementation, the accuracy evaluation index of the business computing power demand prediction model is the mean absolute error and the weighted mean absolute error percentage, where:

[0050] The mean absolute error of the embodiment of the present invention is expressed as:

[0051]

[0052] The weighted mean absolute error of the embodiment of the present invention is expressed as:

[0053]

[0054] In formula (8)-formula (9), m represents the number of evaluation samples; y′ represents the predicted result of computing resource demand; and y represents the actual result of computing resource demand.

[0055] In another embodiment of the present invention, a computing power allocation and scheduling system for an edge computing platform is also provided, and the allocation and scheduling system includes:

[0056] A service pool generation unit is used to obtain historical resource request services and influencing factors of each node device of the edge computing platform, and generate a service pool based on historical resource request tasks;

[0057] The computing power demand evaluation unit is used to build a business computing power demand prediction model based on XGBoost, add influencing factors to the prediction model, generate resource request business prediction outputs for each node device, update the business pool according to the prediction output, evaluate the computing power scale required for the resource request business in the updated business pool, and obtain the computing power demand;

[0058] The allocation solution unit is used to obtain the computing power requirements according to each node device, and to minimize energy consumption and response time. Based on the improved firefly algorithm, multi-objective solutions are performed to obtain a computing power allocation plan that optimizes both energy consumption and response time.

[0059] The computing power resource allocation unit is used to allocate the schedulable computing power resources in the edge computing platform based on the computing power allocation plan.

[0060] Compared with the prior art, the edge computing platform computing power allocation and scheduling method and system of the present invention have the following beneficial effects:

[0061] First, the present invention constructs a business computing power demand prediction model based on XGBoost, adds influencing factors to the prediction model, generates resource request business prediction outputs for each node device, and updates the business pool according to the prediction output. The historical computing power usage of the target node device and the computing power demand in the future can be predicted, and then the computing power scale required for the resource request business in the updated business pool can be evaluated. Based on the evaluated computing power scale, the computing power demand of each node device is further obtained, providing a basis for computing power allocation and scheduling;

[0062] Secondly, the present invention obtains the computing power demand according to each node device, takes minimizing energy consumption and shortening response time as the goal, performs multi-objective solving based on the improved firefly algorithm, and obtains a computing power allocation plan that achieves the best energy consumption and response time. Based on the obtained optimal computing power allocation plan, the computing power resources are allocated and scheduled, which can effectively optimize the load balancing between data centers and migrate computing power resources from nodes with lighter loads to nodes with heavier loads according to the real-time load conditions.

[0063] In summary, the edge computing platform computing power allocation and scheduling method and system of the present invention can avoid overloading of some data centers, while improving the overall computing efficiency and making reasonable use of computing power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0065] Figure 1 This is a flow chart of an implementation method of edge computing platform computing power allocation and scheduling according to the present invention;

[0066] Figure 2 It is a sub-flow chart of the edge computing platform computing power allocation and scheduling method of the present invention;

[0067] Figure 3 Another sub-flow chart of the edge computing platform computing power allocation and scheduling method of the present invention;

[0068] Figure 4 This is a structural block diagram of a computing power allocation and scheduling system for an edge computing platform of the present invention;

[0069] Figure 5 The present invention provides a structural block diagram of a computer device. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0071] In order to make the objectives, technical solutions and advantages of the present application more clear, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0072] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims.

[0073] Currently, edge computing is a distributed computing architecture that transfers data processing and analysis tasks from centralized data centers to the edge of the network, that is, close to the data source. This architecture emerged to cope with the growing amount of data and application scenarios with increasingly high real-time requirements, such as the Internet of Things (IoT), self-driving cars, smart factories, etc. The core advantage of the edge computing platform is its ability to provide low latency, high bandwidth and real-time data processing capabilities; in the edge computing distributed computing network, the load imbalance of nodes may lead to resource waste and excessive energy consumption.

[0074] Some data centers may carry too many computing tasks, while other data centers may be relatively idle. These underutilized data centers still require a lot of energy to maintain, which is a serious waste of resources. At present, the research on computing resource scheduling is still in its infancy. Traditional computing resource scheduling and allocation methods are usually static, which cannot adapt well to the dynamic changes of tasks and network load conditions. This easily leads to problems such as low computing resource utilization and poor accuracy in matching computing resources.

[0075] To solve the above problems, the edge computing platform computing power allocation and scheduling method and system provided by the present invention obtain the historical resource request business and influencing factors of each node device of the edge computing platform, and generate a business pool based on the historical resource request tasks; construct a business computing power demand prediction model based on XGBoost, add the influencing factors to the prediction model, generate the resource request business prediction output of each node device, update the business pool according to the prediction output, evaluate the computing power scale required for the resource request business in the updated business pool, and obtain the computing power demand; obtain the computing power demand according to each node device, take minimizing energy consumption and shortening response time as the goals, perform multi-objective solving based on the improved firefly algorithm, and obtain a computing power allocation plan that optimizes both energy consumption and response time; based on the computing power allocation plan, allocate the schedulable computing power resources in the edge computing platform.

[0076] Among them, the edge computing platform computing power allocation and scheduling method and system of the present invention can effectively optimize the load balancing between data centers. According to the real-time load conditions, computing power resources can be migrated from nodes with lighter loads to nodes with heavier loads, which can avoid overloading of some data centers and improve the overall computing efficiency, so that computing power resources can be reasonably utilized.

[0077] The following is a detailed description of the specific implementation of the edge computing platform computing power allocation and scheduling method and system of the present invention in conjunction with specific embodiments.

[0078] like Figure 1 As shown, in one embodiment of the present invention, a method for allocating and scheduling computing power of an edge computing platform includes the following steps:

[0079] Step S101: Obtain historical resource request services and influencing factors of each node device of the edge computing platform, and generate a service pool based on the historical resource request tasks;

[0080] In a preferred embodiment of the present invention, historical computing resource usage data is collected from each edge node, including data such as CPU usage, memory usage, network bandwidth, and storage usage; at the same time, influencing factors affecting resource usage requests are also collected, including resource request time period, user type, number of users, and device type, which will affect the computing resource requirements of user-side services to a certain extent; further, the present invention also integrates the collected data to generate a business pool for historical resource request tasks, and the historical resource requests contained in the business pool include computing resource usage data executed by the user side under the influencing factors, reflecting the computing resource usage requirements of the user side; the present invention also performs feature extraction on the data in the business pool to obtain time series data for training a business computing power demand prediction model based on XGBoost.

[0081] Step S102: construct a business computing power demand prediction model based on XGBoost, add influencing factors to the prediction model, generate resource request business prediction outputs for each node device, update the business pool according to the prediction output, evaluate the computing power scale required for the resource request business in the updated business pool, and obtain the computing power demand;

[0082] In a preferred embodiment of the present invention, the business computing power demand prediction model of XGBoost is used to predict the computing power resource capacity required by the user side to execute its business under specific influencing factors in the future;

[0083] By adding influencing factors to the prediction model and adjusting the weights of each influencing factor, we can better weigh the changes in the computing resource demand caused by the influencing factors and make the prediction results more accurate.

[0084] In an optional implementation, the step of building a business computing power demand prediction model based on XGBoost and adding influencing factors to the prediction model includes:

[0085] Get the time series data in the business pool, select m features and n samples of computing power resource occupancy, and build a regression model for business computing power requests under multi-dimensional resources. The feature values ​​of the regression model data include the number of CPU cores, GPU percentage occupancy, memory capacity requirements, and video memory percentage occupancy; the target value y of the regression model data i is the business execution time of the i-th sample;

[0086] In a preferred embodiment of the present invention, the output of the business computing power demand prediction model based on XGBoost is expressed as:

[0087]

[0088] In formula (1), represents the predicted value after the tth iteration, that is, the business execution time predicted by the model based on the computing power resources; k represents the number of decision trees obtained after the iteration, and f represents the output function of each decision tree;

[0089] In a preferred embodiment of the present invention, the objective function of the business computing power demand prediction model based on XGBoost is expressed as:

[0090]

[0091] In formula (2), l represents the loss function for measuring the model fit; Ω is the regularization representation used to measure the model complexity;

[0092] Map the influencing factor into an item in the loss function l:

[0093]

[0094] In formula (3), represents the original loss function; α represents the weight parameter, which is used to control the importance of the influencing factors in the loss function; h(X factors ) represents the mapping function, which is used to transform the influencing factor X factors Mapped to a term in the loss function l.

[0095] like Figure 2 As shown, in an optional implementation, the step of updating the service pool according to the prediction output, evaluating the computing power scale required for the resource request service in the updated service pool, and obtaining the computing power demand includes:

[0096] Step S201: according to the predicted output, a new service corresponding to the predicted output is added to the initial service pool, and unfinished services in the initial service pool are retained to generate an updated service pool;

[0097] Among them, the unfinished services in the initial service pool still need to be completed in the next period of time. Therefore, they need to be included in the service pool to obtain the actual computing power resource demand of each user-side service in the service pool in the next period of time.

[0098] Step S202: Sort the services in the updated service pool, evaluate the required computing power scale for each service in the current service pool, and obtain the computing power requirement based on the sorting result through a nonlinear activation function.

[0099] Therefore, the present invention constructs a business computing power demand prediction model based on XGBoost, adds influencing factors to the prediction model, generates a resource request business prediction output for each node device, and updates the business pool according to the prediction output. The historical computing power usage of the target node device and the computing power demand in the future can be predicted, and then the computing power scale required for the resource request business in the updated business pool can be evaluated. Based on the evaluated computing power scale, the computing power demand of each node device is further obtained, providing a basis for computing power allocation and scheduling.

[0100] For further information, please refer to Figure 1 , the edge computing platform computing power allocation and scheduling method of the present invention also includes:

[0101] Step S103: Obtain computing power requirements according to each node device, and take minimizing energy consumption and minimizing response time as the goals, perform multi-objective solving based on the improved firefly algorithm, and obtain a computing power allocation plan that optimizes both energy consumption and response time;

[0102] Step S104: Based on the computing power allocation plan, allocate the schedulable computing power resources in the edge computing platform.

[0103] In a preferred embodiment of the present invention, Figure 3 As shown, the steps of obtaining the computing power demand according to each node device, taking the minimization of energy consumption and the shortest response time as the goal, performing multi-objective solving based on the improved firefly algorithm, and obtaining the computing power allocation scheme that achieves the optimal energy consumption and response time specifically include:

[0104] Step S301: Set the computing power allocation, business computing power demand, energy consumption, and response time parameters, and initialize the firefly algorithm parameters;

[0105] Step S302: randomly generate an initial population and use piecewise linear chaotic mapping to initialize the firefly sequence;

[0106] In a preferred embodiment of the present invention, due to the uneven distribution of the initial positions of randomly generated fireflies during the initialization of the firefly algorithm, the algorithm performs poorly in optimization and converges slowly. Based on this, the present invention uses a piecewise linear chaotic map to initialize the firefly sequence, wherein the chaotic operator has the characteristics of non-repetitive ergodicity, randomness, and regularity, so that the algorithm maintains population diversity and improves the global search performance of the algorithm.

[0107] Further, immediately after step S302, the embodiment of the present invention further includes:

[0108] Step S303: Calculate the brightness and attractiveness of fireflies according to equations (4) and (5), wherein the brightness I of fireflies gradually decreases with the increase of distance r and the absorption of the medium, expressed as:

[0109]

[0110] In formula (4), I0 represents the brightness of the autofluorescence, which is the maximum brightness; I0 is positively correlated with the objective function value corresponding to the location of the firefly, γ represents the absorption factor, r ij represents the distance between fireflies i and j; the attraction related to the absorption factor and distance is expressed as:

[0111]

[0112] In formula (5), β0 represents the maximum attraction;

[0113] Step S304: Calculate the step size factor and update the firefly position;

[0114] Step S305: performing sorting and crossover mutation operations on the fireflies according to the sort-based firefly update strategy;

[0115] Step S306: When the number of iterations reaches the maximum value, the iteration is stopped and the optimal computing power allocation plan is output.

[0116] In a preferred embodiment of the present invention, in step S304 of the embodiment of the present invention, the step of calculating the step factor and updating the firefly position includes:

[0117] The step factor follows a semi-Gaussian distribution with the number of iterations, and the step factor is expressed as:

[0118]

[0119] In formula (6), η represents the number of iterations, η max represents the maximum number of iterations; σ represents a value that is one-fourth of the maximum number of iterations; the range of the step factor ω is [ω min ,ω max ], this scheme can improve the dynamic adjustment of global and local search capabilities and improve the efficiency of the algorithm;

[0120] In a preferred embodiment of the present invention, a quadratic Bezier curve is selected as a fitting curve of the step factor ω, and the selection of parameters is further optimized;

[0121] As a preferred method, the number of iterations t is taken as the horizontal coordinate, the parameter φ is taken as the vertical coordinate, the number of firefly populations is N, and the starting point and the end point are respectively expressed as P0(0, φ0) and P2(N, 0);

[0122] Among them, the control point is set to P1(μ, v), and the parametric equation of the Bezier curve is expressed as:

[0123]

[0124] In formula (10), φ(t) represents a decreasing function, and the value range of the control point P1(μ, v) is determined by the monotonicity of the function; μ represents the horizontal coordinate of the control point; and v represents the vertical coordinate of the control point.

[0125] Among them, in a preferred embodiment of the present invention, in the initial iteration stage of the firefly algorithm, the search range of the firefly individual can be expanded by increasing the step size factor ω, which helps to improve the global exploration performance.

[0126] In a preferred embodiment of the present invention, in updating the position of fireflies, the position update of the i-th firefly attracted and moved by the j-th firefly is expressed as:

[0127] z i+1 =z i +β·(z j -z i )+ω·(rand-ζ) (7)

[0128] In formula (7), z j and z iThey represent the spatial positions of fireflies j and i respectively; ω represents the step factor; rand is a random number with uniform distribution between [0,1]; β represents the attractiveness of fireflies.

[0129] Therefore, the present invention obtains the computing power demand according to each node device, takes minimizing energy consumption and shortening response time as the goal, performs multi-objective solving based on the improved firefly algorithm, and obtains a computing power allocation plan that achieves the best energy consumption and response time. Based on the obtained optimal computing power allocation plan, the computing power resources are allocated and scheduled.

[0130] To sum up, the present invention can effectively optimize the load balancing between data centers. According to the real-time load conditions, computing power resources can be migrated from nodes with lighter loads to nodes with heavier loads, which can avoid overloading of some data centers. At the same time, it can improve the overall computing efficiency and make rational use of computing power resources.

[0131] In a preferred embodiment of the present invention, the accuracy evaluation index of the business computing power demand prediction model is the mean absolute error and the weighted mean absolute error percentage, where:

[0132] The mean absolute error of the present invention is expressed as:

[0133]

[0134] The weighted mean absolute error of the present invention is expressed as:

[0135]

[0136] In formula (8)-formula (9), m represents the number of evaluation samples; y′ represents the predicted result of computing resource demand; and y represents the actual result of computing resource demand.

[0137] like Figure 4 As shown, in another embodiment of the present invention, a computing power allocation and scheduling system for an edge computing platform is also provided, and the allocation and scheduling system 400 includes:

[0138] The service pool generation unit 401 is used to obtain the historical resource request services and influencing factors of each node device of the edge computing platform, and generate a service pool based on the historical resource request tasks;

[0139] The computing power demand evaluation unit 402 is used to build a business computing power demand prediction model based on XGBoost, add influencing factors to the prediction model, generate resource request business prediction outputs for each node device, update the business pool according to the prediction output, evaluate the computing power scale required for the resource request business in the updated business pool, and obtain the computing power demand;

[0140] The allocation solution solving unit 403 is used to obtain the computing power demand according to each node device, and to perform multi-objective solving based on the improved firefly algorithm with the goal of minimizing energy consumption and minimizing response time, so as to obtain the computing power allocation solution that optimizes both energy consumption and response time;

[0141] The computing power resource allocation unit 404 is used to allocate the schedulable computing power resources in the edge computing platform based on the computing power allocation plan.

[0142] like Figure 5 As shown, in yet another embodiment of the present invention, the present invention further provides a computer device;

[0143] The device 500 includes a memory 501, a processor 502, and computer-readable instructions stored in the memory 501 and executable on the processor 502. When the processor 502 executes the computer-readable instructions, the edge computing platform computing power allocation and scheduling method provided in Example 1 is implemented. The method includes the following steps:

[0144] Step S101: Obtain historical resource request services and influencing factors of each node device of the edge computing platform, and generate a service pool based on the historical resource request tasks;

[0145] Step S102: construct a business computing power demand prediction model based on XGBoost, add influencing factors to the prediction model, generate resource request business prediction outputs for each node device, update the business pool according to the prediction output, evaluate the computing power scale required for the resource request business in the updated business pool, and obtain the computing power demand;

[0146] Step S103: Obtain computing power requirements according to each node device, and take minimizing energy consumption and minimizing response time as the goals, perform multi-objective solving based on the improved firefly algorithm, and obtain a computing power allocation plan that optimizes both energy consumption and response time;

[0147] Step S104: Based on the computing power allocation plan, allocate the schedulable computing power resources in the edge computing platform.

[0148] In addition, the device 500 provided in the embodiment of the present invention may also have a communication interface 503 for receiving control instructions.

[0149] In an exemplary embodiment provided by the present disclosure, a computer-readable storage medium is also provided.

[0150] Specifically, in an exemplary embodiment of the present disclosure, the storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the edge computing platform computing power allocation and scheduling method provided in the above embodiment.

[0151] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the processes does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0152] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed over multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0153] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0154] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several requests for a computer device (which can be a personal computer, server or network device, etc., specifically a processor in a computer device) to perform some or all of the steps of the edge computing platform computing power allocation and scheduling method described in each embodiment of the present invention.

[0155] A person skilled in the art can understand that part or all of the steps in the various methods of the embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0156] The above is a detailed introduction to the edge computing platform computing power allocation and scheduling method disclosed in the embodiment of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for allocating and scheduling computing power of an edge computing platform, characterized in that: The edge computing platform computing power allocation and scheduling method includes the following steps: Obtain the historical resource request services and influencing factors of each node device on the edge computing platform, and generate a service pool based on the historical resource request tasks; Build a business computing power demand prediction model based on XGBoost, add influencing factors to the prediction model, generate resource request business prediction output for each node device, update the business pool according to the prediction output, evaluate the computing power scale required for resource request business in the updated business pool, and obtain the computing power demand; According to the computing power requirements of each node device, with the goal of minimizing energy consumption and shortening response time, a multi-objective solution is performed based on the improved firefly algorithm to obtain a computing power allocation plan that achieves the best energy consumption and response time. Based on the computing power allocation plan, the schedulable computing power resources in the edge computing platform are allocated.

2. The method for allocating and scheduling computing power of an edge computing platform according to claim 1, characterized in that: The step of obtaining historical resource request services and influencing factors of each node device of the edge computing platform and generating a service pool based on the historical resource request tasks includes: Collect historical resource usage data from each edge node, including CPU usage, memory usage, network bandwidth, and storage usage; Collect factors that affect resource requests, including resource request time period, user type, number of users, and device type; Integrate the collected data to generate a business pool of historical resource request tasks; Feature extraction is performed on the data in the business pool to obtain time series data for training the business computing power demand prediction model based on XGBoost.

3. The edge computing platform computing power allocation and scheduling method according to claim 2 is characterized in that: The steps to build a business computing power demand prediction model based on XGBoost and add influencing factors to the prediction model include: Get the time series data in the business pool, select m features and n samples of computing power resource occupancy, and build a regression model for business computing power requests under multi-dimensional resources. The feature values ​​of the regression model data include the number of CPU cores, GPU percentage occupancy, memory capacity requirements, and video memory percentage occupancy; the target value y of the regression model data i is the business execution time of the i-th sample; The output of the business computing power demand prediction model based on XGBoost is expressed as: In formula (1), represents the predicted value after the tth iteration, that is, the business execution time predicted by the model based on the computing power resources; k represents the number of decision trees obtained after the iteration, and f represents the output function of each decision tree; The objective function of the business computing power demand prediction model based on XGBoost is expressed as: In formula (2), l represents the loss function for measuring the model fit; Ω is the regularization representation used to measure the model complexity; Map the influencing factor into an item in the loss function l: In formula (3), represents the original loss function; α represents the weight parameter, which is used to control the importance of the influencing factors in the loss function; h(X factors ) represents the mapping function, which is used to transform the influencing factor X factors Mapped to a term in the loss function l.

4. The edge computing platform computing power allocation and scheduling method according to claim 3 is characterized in that: The steps of updating the business pool according to the prediction output, evaluating the computing power scale required for the resource request business in the updated business pool, and obtaining the computing power demand include: According to the predicted output, new services corresponding to the predicted output are added to the initial service pool, and unfinished services in the initial service pool are retained to generate an updated service pool; Sort the businesses in the update business pool, evaluate the required computing power scale for each business in the current business pool, and obtain the computing power demand based on the sorting results through a nonlinear activation function.

5. The method for allocating and scheduling computing power of an edge computing platform according to claim 4, characterized in that: According to the computing power requirements of each node device, with the goal of minimizing energy consumption and shortening response time, multi-objective solution is performed based on the improved firefly algorithm to obtain the steps of the computing power allocation plan that achieves the optimal energy consumption and response time, including: Set the computing power allocation, business computing power demand, energy consumption, response time parameters, and initialize the Firefly algorithm parameters; The initial population is randomly generated and the firefly sequence is initialized using piecewise linear chaotic mapping; The brightness and attractiveness of fireflies are calculated according to equations (4) and (5), where the brightness I of fireflies gradually decreases with the increase of distance r and the absorption of the medium, expressed as: In formula (4), I0 represents the brightness of the autofluorescence, which is the maximum brightness; I0 is positively correlated with the objective function value corresponding to the location of the firefly, γ represents the absorption factor, r ij represents the distance between fireflies i and j; the attraction related to the absorption factor and distance is expressed as: In formula (5), β0 represents the maximum attraction; Calculate the step size factor and update the firefly position; Sorting and crossover mutation operations are performed on fireflies according to the sorting-based firefly update strategy; When the number of iterations reaches the maximum value, the iteration stops and the optimal computing power allocation plan is output.

6. The method for allocating and scheduling computing power of an edge computing platform according to claim 5, characterized in that: The steps to calculate the step factor and update the firefly position include: The step factor follows a semi-Gaussian distribution with the number of iterations, and the step factor is expressed as: In formula (6), η represents the number of iterations, η max represents the maximum number of iterations; σ represents a value that is one-fourth of the maximum number of iterations; the range of the step factor ω is [ω min ,ω max ]; In updating the position of fireflies, the position update of the i-th firefly attracted by the j-th firefly is expressed as: With i+1 =from i +β (from j -With i )+ω (rand-ζ) (7) In formula (7), z j and z i They represent the spatial positions of fireflies j and i respectively; ω represents the step size factor, rand is a random number that follows a uniform distribution between [0,1]; β represents the attractiveness of the firefly.

7. The method for allocating and scheduling computing power of an edge computing platform according to claim 6, characterized in that: The accuracy evaluation indicators of the business computing power demand prediction model are the mean absolute error and the weighted mean absolute error percentage, where: The mean absolute error is expressed as: The weighted mean absolute error is expressed as: In formula (8)-formula (9), m represents the number of evaluation samples; y′ represents the predicted result of computing resource demand; and y represents the actual result of computing resource demand.

8. A computing power allocation and scheduling system for implementing a computing power allocation and scheduling method for an edge computing platform as claimed in any one of claims 1 to 7, characterized in that: The computing power allocation and scheduling system includes: A service pool generation unit is used to obtain historical resource request services and influencing factors of each node device of the edge computing platform, and generate a service pool based on historical resource request tasks; The computing power demand evaluation unit is used to build a business computing power demand prediction model based on XGBoost, add influencing factors to the prediction model, generate resource request business prediction outputs for each node device, update the business pool according to the prediction output, evaluate the computing power scale required for the resource request business in the updated business pool, and obtain the computing power demand; The allocation solution unit is used to obtain the computing power requirements according to each node device, and to minimize energy consumption and response time. Based on the improved firefly algorithm, multi-objective solutions are performed to obtain a computing power allocation plan that optimizes both energy consumption and response time. The computing power resource allocation unit is used to allocate the schedulable computing power resources in the edge computing platform based on the computing power allocation plan.

Images