Artificial intelligence-driven edge computing resource dynamic allocation method

Abstract

The invention discloses an artificial intelligence-driven edge computing resource dynamic allocation method, and relates to the technical field of edge computing. The method comprises the following steps: firstly, based on periodically acquired multi-dimensional operation data of each node device, an edge computing network and multi-dimensional index parameters for evaluating an edge computing resource allocation result; predicting and obtaining multi-dimensional operation data of each node device in the current next period based on an artificial intelligence algorithm, and establishing an edge computing resource dynamic allocation strategy and a mapping model of the multi-dimensional operation data and multi-dimensional index parameters; then optimizing an edge computing resource dynamic allocation strategy oriented to the edge computing network and to be executed in the current next cycle based on a self-adaptive hummingbird algorithm to obtain the edge computing resource dynamic allocation strategy and an optimal search result for minimizing a fitness function; and finally carrying out edge computing resource dynamic allocation on the edge computing network according to the optimal search result, so that the overall efficiency of the edge computing network can be ensured.

Description

Technical Field

[0001] The present invention belongs to the field of edge computing technology, and specifically relates to an artificial intelligence-driven dynamic allocation method for edge computing resources. Background Art

[0002] With the rapid development of the Internet of Things and big data technologies, edge computing, as an emerging computing architecture, has gradually become a key means of data processing. Edge computing involves moving computing tasks to edge nodes close to the data source to significantly reduce data transmission latency and improve system response speed. This is particularly important in scenarios with high real-time requirements, such as autonomous driving and industrial automation. However, with the increase in computing tasks and the number of nodes, how to efficiently manage and allocate the resources of edge computing nodes has become a pressing issue.

[0003] Currently, existing edge computing resource scheduling solutions primarily rely on static resource allocation methods, which struggle to adapt to load changes in real time. These methods also fail to account for load fluctuations between nodes and cannot predict future load changes, leading to flaws such as uneven resource allocation and delays, which in turn impact the overall efficiency of the system. Furthermore, existing scheduling solutions fail to fully utilize predictive models and flexible scheduling strategies, lacking real-time adjustment and optimization mechanisms. Summary of the Invention

[0004] The purpose of the present invention is to provide an artificial intelligence-driven edge computing resource dynamic allocation method, device, computer equipment, computer-readable storage medium and computer program product to solve the problems of uneven resource allocation or delay caused by static allocation in existing edge computing resource scheduling schemes, affecting the overall efficiency of the system and lacking real-time adjustment and optimization mechanisms.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, a method for dynamically allocating edge computing resources driven by artificial intelligence is provided, which is executed by a central device for communicating with each node device in an edge computing network, including:

[0007] Periodically collecting multidimensional operating data of each node device and multidimensional indicator parameters of the edge computing network for evaluating edge computing resource allocation results;

[0008] Establishing a time series data prediction model based on a first artificial intelligence algorithm based on the multi-dimensional operating data of each node device collected in the most recent multiple consecutive cycles, and applying the time series data prediction model to predict the multi-dimensional operating data of each node device in the next cycle;

[0009] Establishing a mapping model between the edge computing resource dynamic allocation strategy and the multidimensional operating data and the multidimensional indicator parameters based on a second artificial intelligence algorithm according to the edge computing resource dynamic allocation strategy historically executed for the edge computing network and the multidimensional operating data and the multidimensional indicator parameters of each node device collected during the corresponding strategy execution cycle;

[0010] Based on the adaptive hummingbird algorithm, the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be executed is optimized to obtain the edge computing resource dynamic allocation strategy and the best search result for minimizing the fitness function, wherein the independent variables of the fitness function include the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be selected and the multidimensional operating data of each node device in the current next cycle imported into the mapping model. Multidimensional indicator parameters;

[0011] Dynamically allocate edge computing resources to the edge computing network according to the best search result.

[0012] Based on the above invention, a new solution for optimizing and adjusting the dynamic allocation strategy of edge computing resources based on an artificial intelligence algorithm and an adaptive hummingbird algorithm is provided. That is, first, based on the multidimensional operating data of each node device and the edge computing network collected periodically and multidimensional indicator parameters for evaluating the edge computing resource allocation results, the multidimensional operating data of each node device in the current next cycle is predicted based on the artificial intelligence algorithm, and a mapping model between the edge computing resource dynamic allocation strategy and the multidimensional operating data and the multidimensional indicator parameters is established. Then, based on the adaptive hummingbird algorithm, the dynamic allocation strategy of edge computing resources to be executed for the edge computing network and the current next cycle is optimized to obtain the best search result for the dynamic allocation strategy of edge computing resources and for minimizing the fitness function. Finally, the edge computing resources are dynamically allocated to the edge computing network based on the best search result. In this way, the edge computing network can adapt to the real-time load changes, load fluctuations between nodes, and future load changes of the edge computing network, and to a certain extent eliminate defects such as uneven resource allocation or delay. In addition, by optimizing the current dynamic allocation strategy of edge computing resources to be executed based on the prediction model, the mapping model, and the adaptive hummingbird algorithm, a real-time adjustment and optimization mechanism is provided, thereby ensuring the overall efficiency of the edge computing network and facilitating practical application and promotion.

[0013] In one possible design, the multidimensional operation data contains the computing resource utilization, storage resource utilization and bandwidth resource occupancy of the corresponding node devices, and the multidimensional indicator parameters include the average response delay of all computing requests and the standard deviation of the computing resource utilization, storage resource utilization and bandwidth resource occupancy of all node devices.

[0014] In a possible design, the calculation formula of the fitness function F is expressed as follows:

[0015] F=η t ×S t +η d ×S d +η u ×S u +η o ×S o

[0016] Where S t represents the average response delay in the multidimensional indicator parameters obtained by importing the dynamic allocation strategy of edge computing resources for the edge computing network and the current next cycle to be selected and the multidimensional operation data of each node device in the current next cycle into the mapping model, S d Indicates the standard deviation of computing resource utilization in this multidimensional indicator parameter, S u Indicates the standard deviation of storage resource utilization in this multi-dimensional indicator parameter, S o Indicates the standard deviation of bandwidth resource utilization in the multi-dimensional indicator parameter, η t ,η d ,η u and η o Respectively represent the preset weight coefficients.

[0017] In one possible design, the first artificial intelligence algorithm adopts a machine learning algorithm based on a long short-term memory network, a Bi-LSTM network, an Attention-LSTM network or a Prophet model, and the second artificial intelligence algorithm adopts a machine learning algorithm based on a support vector machine, a K-nearest neighbor method, a stochastic gradient descent method, a multi-layer perceptron, a decision tree, a back-propagation neural network or a radial basis function network.

[0018] In one possible design, based on the multi-dimensional operating data of each node device collected in the most recent multiple consecutive cycles, a time series data prediction model is established based on a first artificial intelligence algorithm, and the time series data prediction model is applied to predict the multi-dimensional operating data of each node device in the next cycle, including:

[0019] For a certain dimension of operation data of a certain node device in the edge computing network, corresponding sample data is extracted in the following manner: multidimensional operation data collected in m consecutive cycles and of each node device are extracted from the multidimensional operation data collected in the current most recent multiple consecutive cycles and of each node device, and the extraction result is used as a prediction model input item; and the certain dimension of operation data collected in the m+1th cycle and of the certain node device are also extracted from the multidimensional operation data collected in the current most recent multiple consecutive cycles and of each node device, and the new extraction result is used as a prediction model output item, and finally the prediction model input item and the prediction model output item are used as sample data, wherein m represents a positive integer greater than or equal to 2 and less than M, M represents the total number of cycles of the current most recent multiple consecutive cycles, and the m+1th cycle refers to the next cycle after the m consecutive cycles;

[0020] Applying all the sample data to perform calibration and verification modeling on a machine learning model based on the first artificial intelligence algorithm, and obtaining a time series data prediction model corresponding to the certain node device and the certain dimension of operation data;

[0021] Importing the multi-dimensional operation data of each node device collected in the most recent m consecutive cycles into the time series data prediction model, and outputting the certain dimensional operation data of the node device predicted to be collected in the next cycle;

[0022] Calculate, based on the edge computing resource dynamic allocation strategy historically executed by the edge computing network in the most recent multiple consecutive cycles and the certain dimension operation data of the certain node device collected in the most recent multiple consecutive cycles, a correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data;

[0023] The following formula is used to update and calculate the certain dimension operation data of the certain node device in the current next period prediction collection:

[0024]

[0025] Where, Indicates the certain dimension operation data of the certain node device that is predicted and collected in the current next cycle before updating, It indicates the updated running data of a certain dimension of a certain node device that is predicted to be collected in the next period of the current period.

[0026] In one possible design, based on the edge computing resource dynamic allocation strategy historically executed by the edge computing network in the most recent multiple consecutive cycles and the certain dimension operation data of the certain node device collected in the most recent multiple consecutive cycles, the correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data is calculated, including:

[0027] The certain dimension operation data of the certain node device collected in the most recent multiple consecutive cycles is used as the first time series data S1, and the normal distribution KS check is performed on the first time series data S1 to calculate a first check statistic p value p1;

[0028] Converting the edge computing resource dynamic allocation strategy for the edge computing network executed in the current multiple consecutive cycles into a multi-dimensional timing strategy code parameter in an encoding manner;

[0029] For each dimension of the multidimensional timing strategy code parameter, the corresponding timing strategy code parameter is used as the second timing data S2 and the normal distribution KS check is performed to calculate the corresponding second check statistic p value p2. Then, if the first check statistic p value p1 and the second check statistic p value p2 are both greater than the preset threshold, the corresponding correlation coefficient r is calculated according to the following formula: q :

[0030]

[0031] In the formula, i represents a positive integer, x i Indicates the i-th value in S1, y i represents the i-th value in S2, represents the mean value of S1, represents the mean value of S2;

[0032] The correlation coefficient r corresponding to the timing strategy code parameters of each dimension q The average value is used as the correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data.

[0033] In one possible design, the edge computing resource dynamic allocation strategy for the edge computing network and the next cycle to be executed is optimized based on the adaptive hummingbird algorithm to obtain the best search result for the edge computing resource dynamic allocation strategy and for minimizing the fitness function, including the following steps S401 to S407:

[0034] S401. Initialize the optimization algorithm parameters of the adaptive hummingbird algorithm and include the population size N and the maximum number of iterations T, and randomly generate a set of parameters to be optimized and a corresponding initial search value array for each hummingbird individual in the hummingbird population including N hummingbird individuals, and then execute step S402, wherein the set of parameters to be optimized includes multidimensional parameters obtained by encoding and having a mapping relationship with the edge computing network and the current next cycle and the current edge computing resource dynamic allocation strategy to be executed, and the initial search value array x 0 It is expressed as follows:

[0035]

[0036] Wherein, k represents a positive integer less than or equal to K, K represents the total number of parameters in the parameter set to be optimized, represents the initial search value corresponding to the kth parameter in the parameter set to be optimized, u c,k represents the upper limit of the parameter search space corresponding to the kth parameter, l c,k represents the lower limit of the parameter search space corresponding to the kth parameter, and rand(0,1) represents a pure decimal random generator function;

[0037] S402. For each hummingbird individual, the corresponding initial search value array is mapped into a dynamic allocation strategy for edge computing resources to be selected for the edge computing network and the current next cycle. The dynamic allocation strategy for edge computing resources and the multidimensional operating data of each node device in the current next cycle are then imported into the mapping model to obtain a first multidimensional indicator parameter as an output. The first multidimensional indicator parameter is then imported into the fitness function as an independent variable to obtain a corresponding first fitness. Finally, step S403 is executed.

[0038] S403. The initial search value array corresponding to the minimum first fitness is used as the current optimal search value array, and the current iteration number t=1 is initialized and set, and then step S404 is executed;

[0039] S404. For each hummingbird individual, update the corresponding current search value array according to the current number of iterations, the current optimal search value array and the corresponding pre-update search value array, and then execute step S405, wherein the current search value array x t It is expressed as follows:

[0040]

[0041] Where, Represents the current search value in the array x t and the search value corresponding to the k-th parameter, represents the search value in the pre-update search value array and corresponding to the kth parameter, represents the search value corresponding to the kth parameter in the current optimal search value array, AF() represents an adaptive adjustment factor calculation function, sig() represents an adaptive Sigmoid function, Ca(0,1) represents a random variable that satisfies a standard Cauchy distribution, and Ga(0,1) represents a random variable that satisfies a standard Gaussian distribution;

[0042] S405. For each hummingbird individual, the corresponding current search value array is mapped into a dynamic allocation strategy for edge computing resources to be selected for the edge computing network and the current next cycle. The dynamic allocation strategy for edge computing resources and the multidimensional operating data of each node device in the current next cycle are then imported into the mapping model to obtain a second multidimensional indicator parameter as an output. The second multidimensional indicator parameter is then imported into the fitness function as an independent variable to obtain a corresponding second fitness. Finally, step S406 is executed.

[0043] S406. Determine whether the minimum second fitness is lower than the fitness corresponding to the current optimal search value array. If so, update the current optimal search value array to the current search value array corresponding to the minimum second fitness, and then execute step S407. Otherwise, execute step S407 directly.

[0044] S407. Determine whether the current number of iterations t reaches the maximum number of iterations T. If so, use the current optimal search value array as the best search result obtained by optimizing the dynamic allocation strategy of edge computing resources for the edge computing network and the next cycle to be executed and used to minimize the fitness function. Otherwise, add 1 to the current number of iterations t, and then return to execute step S404.

[0045] In one possible design, the edge computing resource dynamic allocation strategy includes increasing / decreasing the maximum number of computing request responses of peripheral nodes, raising / lowering the minimum level of computing request responses of peripheral nodes and / or adjusting the computing request forwarding direction between peripheral nodes.

[0046] In a second aspect, an artificial intelligence-driven edge computing resource dynamic allocation device is provided, which is arranged in a central device for communicating with each node device in the edge computing network, and includes a data cycle acquisition unit, an operation data prediction unit, a mapping model establishment unit, an allocation strategy optimization unit, and a resource allocation scheduling unit;

[0047] The data periodic collection unit is used to periodically collect multidimensional operating data of each node device and multidimensional indicator parameters of the edge computing network for evaluating edge computing resource allocation results;

[0048] The operation data prediction unit is communicatively connected to the data cycle acquisition unit, and is configured to establish a time series data prediction model based on a first artificial intelligence algorithm based on the multidimensional operation data of each node device collected in the most recent multiple consecutive cycles, and to apply the time series data prediction model to predict the multidimensional operation data of each node device in the next cycle;

[0049] The mapping model establishment unit is communicatively connected to the data period collection unit and is used to establish, based on a second artificial intelligence algorithm, a mapping model of the edge computing resource dynamic allocation strategy and the multidimensional operation data and the multidimensional indicator parameters according to the edge computing resource dynamic allocation strategy historically executed for the edge computing network and the multidimensional operation data and the multidimensional indicator parameters of each node device collected during the corresponding strategy execution cycle;

[0050] The allocation strategy optimization unit is communicatively connected to the operation data prediction unit and the mapping model establishment unit, respectively, and is used to optimize the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be executed based on the adaptive hummingbird algorithm, and obtain the edge computing resource dynamic allocation strategy and the best search result for minimizing the fitness function, wherein the independent variables of the fitness function include the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be selected and the multidimensional operation data of each node device in the current next cycle imported into the mapping model. Multidimensional index parameters;

[0051] The resource allocation scheduling unit is communicatively connected to the allocation strategy optimization unit and is used to dynamically allocate edge computing resources to the edge computing network according to the best search result.

[0052] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the edge computing resource dynamic allocation method as described in the first aspect or any possible design of the first aspect.

[0053] In a fourth aspect, the present invention provides a computer-readable storage medium having instructions stored thereon. When the instructions are run on a computer, the method for dynamically allocating edge computing resources as described in the first aspect or any possible design of the first aspect is executed.

[0054] In a fifth aspect, the present invention provides a computer program product, comprising a computer program or instructions, which, when executed by a computer, implements the method for dynamic allocation of edge computing resources as described in the first aspect or any possible design of the first aspect.

[0055] Beneficial effects of the above scheme:

[0056] (1) The present invention creatively provides a new solution for optimizing and adjusting the dynamic allocation strategy of edge computing resources based on artificial intelligence algorithms and adaptive hummingbird algorithms, namely, first based on the multidimensional operating data of each node device and the edge computing network collected periodically and multidimensional indicator parameters for evaluating the edge computing resource allocation results, the multidimensional operating data of each node device in the current next cycle is predicted based on the artificial intelligence algorithm, and a mapping model of the edge computing resource dynamic allocation strategy and the multidimensional operating data and the multidimensional indicator parameters is established, and then based on the adaptive hummingbird algorithm, the edge computing resource dynamic allocation strategy for the edge computing network and the current next cycle to be executed is optimized to obtain the best search result for the edge computing resource dynamic allocation strategy and for minimizing the fitness function, and finally, the edge computing resource is dynamically allocated to the edge computing network according to the best search result, so that the real-time load changes of the edge computing network, the load fluctuations between nodes and the future load changes can be adaptively adjusted, and defects such as uneven resource allocation or delay can be eliminated to a certain extent, and the current dynamic allocation strategy of the edge computing resources to be executed is optimized based on the prediction model, the mapping model and the adaptive hummingbird algorithm, thereby providing a real-time adjustment and optimization mechanism, thereby ensuring the overall efficiency of the edge computing network, and facilitating practical application and promotion;

[0057] (2) By calculating the correlation coefficient between the allocation strategy and the node equipment and operating data, and using the correlation coefficient as the influencing factor for the correction and update of the time series prediction results, the adverse effects of the allocation strategy execution on the operating data of the current node equipment can be eliminated to a certain extent, ensuring the accuracy of the load prediction of each edge node under the assumption that the strategy is not executed, thereby ensuring the reliability of subsequent strategy determination and execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0059] Figure 1A flowchart of the artificial intelligence-driven dynamic allocation method for edge computing resources provided in an embodiment of the present application.

[0060] Figure 2 A schematic diagram of the structure of the artificial intelligence-driven edge computing resource dynamic allocation device provided in an embodiment of the present application.

[0061] Figure 3 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly introduced below in conjunction with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structures of the drawings is only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these embodiments without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0063] It should be understood that although the terms first, second, etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are merely used to distinguish one object from another. For example, a first object can be referred to as a second object, and similarly, a second object can be referred to as a first object without departing from the scope of the exemplary embodiments of the present invention.

[0064] It should be understood that the term "and / or" that may appear in this document is merely a description of the association relationship between associated objects, indicating that there may be three relationships. For example, A and / or B can indicate three situations: A exists alone, B exists alone, or A and B exist at the same time. For another example, A, B and / or C can indicate the existence of any one of A, B and C or any combination of them. The term " / and" that may appear in this document describes another type of association object relationship, indicating that there may be two relationships. For example, A / and B can indicate two situations: A exists alone or A and B exist at the same time. In addition, the character " / " that may appear in this document generally indicates that the previous and next associated objects are in an "or" relationship.

[0065] Example

[0066] like Figure 1 As shown, the method for dynamically allocating edge computing resources provided by the first aspect of this embodiment and driven by artificial intelligence can be, but is not limited to, executed by a central device having certain computing resources and used to communicate and connect each node device in the edge computing network. Figure 1As shown, the edge computing resource dynamic allocation method may include, but is not limited to, the following steps S1 to S5.

[0067] S1. Periodically collect multi-dimensional operating data of each node device and multi-dimensional indicator parameters of the edge computing network for evaluating edge computing resource allocation results.

[0068] In the step S1, the node device is an edge node in the edge computing network that can handle sinking computing tasks, and the multi-dimensional operation data and the multi-dimensional indicator parameters can be obtained based on the conventional collection of existing monitoring tools. Specifically, the multi-dimensional operation data includes but is not limited to the computing resource utilization rate, storage resource utilization rate and bandwidth resource occupancy rate of the corresponding node device; through these data, the operation status of each node device can be grasped in real time. Specifically, the multi-dimensional indicator parameters include but are not limited to the average response delay of all computing requests and the standard deviation of computing resource utilization rate, storage resource utilization rate and bandwidth resource occupancy rate of all node devices; through these data, the excellence of the edge computing resource dynamic allocation strategy corresponding to the edge computing resource allocation result can be evaluated. In addition, the specific collection period can be pre-set according to actual needs, for example, the multi-dimensional operation data and the multi-dimensional indicator parameters can be collected every 10 minutes.

[0069] S2. Based on the multi-dimensional operating data of each node device collected in the most recent multiple consecutive cycles, a time series data prediction model is established based on the first artificial intelligence algorithm, and the time series data prediction model is applied to predict the multi-dimensional operating data of each node device in the next cycle.

[0070] In step S2, the artificial intelligence algorithm is a core artificial intelligence algorithm that specifically studies how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their own performance. It is the fundamental way to make computers intelligent. In order to achieve the purpose of time series data prediction, specifically, the first artificial intelligence algorithm can be, but is not limited to, a machine learning algorithm based on a long short-term memory network, a Bi-LSTM (Bi-directional Long Short-Term Memory) network, an Attention-LSTM (long short-term memory based on an attention mechanism) network, or a Prophet model (which is a time series prediction algorithm open sourced by Facebook, suitable for data with seasonal, trend, and holiday effects; its core idea is to decompose the time series into trend terms, seasonal terms, and holiday terms, and fit and predict them through an additive model, which is easy to use, robust, and highly interpretable), etc., so as to quickly and accurately find the time series patterns in the data. Therefore, based on a certain amount of sample data, the time series data prediction model that has passed the verification can be trained through conventional calibration and verification modeling methods (the specific process includes the calibration process and the verification process of the model, that is, first by comparing the model simulation results with the measured data, and then adjusting the model parameters according to the comparison results, so that the simulation results are consistent with the actual process). In order to ensure the accuracy of the prediction results, preferably, based on the multi-dimensional operating data of each node device collected in the most recent multiple consecutive cycles, a time series data prediction model is established based on the first artificial intelligence algorithm, and the time series data prediction model is applied to predict the multi-dimensional operating data of each node device in the next cycle, including but not limited to the following steps S21 to S24.

[0071] S21. For a certain dimension of operation data of a certain node device in the edge computing network, the corresponding sample data is extracted in the following manner: the multidimensional operation data collected in m consecutive cycles and of each node device are extracted from the multidimensional operation data collected in the current most recent multiple consecutive cycles and of each node device, and the extraction result is used as the input item of the prediction model; and the certain dimension of operation data collected in the m+1th cycle and of the certain node device are also extracted from the multidimensional operation data collected in the current most recent multiple consecutive cycles and of each node device, and the new extraction result is used as the output item of the prediction model; finally, the prediction model input item and the prediction model output item are used as sample data, wherein m represents a positive integer greater than or equal to 2 and less than M, M represents the total number of cycles of the current most recent multiple consecutive cycles, and the m+1th cycle refers to the next cycle after the m consecutive cycles.

[0072] In step S21, for example, if the edge computing network includes node device A, node device B, node device C and node device D, and the collection cycle is 10 minutes, then for the computing resource utilization rate of node device A, the computing resource utilization rate, storage resource utilization rate and bandwidth resource occupancy rate of node device A, node device B, node device C and node device D collected from 10:01 to 12:00 (i.e., m=12 consecutive cycles) can be extracted from the multi-dimensional operation data of each node device collected in the most recent 1000 consecutive cycles (i.e., M=1000) as a prediction model input item, and the computing resource utilization rate of node device A collected from 12:01 to 12:10 can also be extracted from the multi-dimensional operation data of each node device collected in the most recent 1000 consecutive cycles as a prediction model output item. Finally, the prediction model input item and the prediction model output item are used as sample data corresponding to the node device A and the computing resource utilization rate.

[0073] S22. Apply all the sample data to calibrate and verify the machine learning model based on the first artificial intelligence algorithm to obtain a time series data prediction model corresponding to the certain node device and the certain dimension of operation data.

[0074] In step S22, all the sample data will be divided into a training set, a test set and a validation set, and then these sample sets will be used to calibrate and validate the modeling of the machine learning model based on the first artificial intelligence algorithm, and finally the conventional training will obtain the time series data prediction model with the ability to perform time series data prediction on the certain dimension operation data of the certain node device. Considering that the Attention-LSTM network is an improved network that adds a layer of attention mechanism to the existing long short-term memory network, and the added attention mechanism is used to allow the long short-term memory network to dynamically focus on part of the information during training, so as to be able to capture more important information and improve the performance of the algorithm, this embodiment preferably uses the Attention-LSTM network to train the time series data prediction model. In addition, the parameters of the time series data prediction model can be, but are not limited to, searched using a grid search algorithm or a Bayesian search algorithm.

[0075] S23. Import the multi-dimensional operation data of each node device collected in the latest m consecutive cycles into the time series data prediction model, and output the certain dimensional operation data of the node device predicted to be collected in the next cycle.

[0076] S24. Based on the edge computing resource dynamic allocation strategy historically executed for the edge computing network in the current most recent multiple consecutive cycles and the certain dimension operation data collected in the current most recent multiple consecutive cycles and of the certain node device, calculate the correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data.

[0077] In step S24, specifically, the dynamic allocation strategy of edge computing resources includes but is not limited to increasing / decreasing the maximum number of computing request responses of peripheral nodes (for example, setting the maximum number of computing request responses of a certain edge node device to 100. If a new computing request is received from an adjacent data source when the number of computing requests currently being responded to by the certain edge node device has reached 100, the new computing request needs to be transferred to other edge node devices for response processing), increasing / decreasing the minimum level of computing request responses of peripheral nodes (for example, setting the minimum level of computing request responses of a certain edge node device to medium. If the certain edge node device receives a new computing request from an adjacent data source that is low-level, the new computing request needs to be transferred to other edge node devices for response processing) and / or adjusting the computing request transfer direction between peripheral nodes (for example, if a new computing request from a data source adjacent to edge node device A needs to be transferred to edge node device B, the computing request transfer direction between edge node device A and edge node device B can be set to forward, otherwise it can be configured to reverse), etc. Based on the specific content of the aforementioned allocation strategy, it can be known that the execution of the edge computing resource dynamic allocation strategy will have more or less adverse effects on the certain dimension operation data of the current certain node device, so it is necessary to calculate the correlation coefficient γ as the size of the impact factor and use it for subsequent correction updates. Specifically, based on the edge computing resource dynamic allocation strategy executed historically for the edge computing network in the current recent multiple consecutive cycles and the certain dimension operation data of the certain node device collected in the current recent multiple consecutive cycles, the correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data is calculated, including but not limited to the following steps S241 to S243.

[0078] S241. Take the certain dimension operation data of the certain node device collected in the most recent multiple consecutive cycles as the first time series data S1, and perform normal distribution KS check on the first time series data S1 to calculate the first check statistic p value p1.

[0079] In step S241, the normal distribution KS test is the Kolmogorov-Smirnov test, which is an existing statistical test method used to determine whether the observed values ​​of a set of data conform to a specific distribution. Therefore, it can be conventionally used to obtain the first check statistic p value p1 and the subsequent second check statistic p value p2.

[0080] S242. Convert the edge computing resource dynamic allocation strategy for the edge computing network executed in the current recent multiple consecutive cycles into a multi-dimensional timing strategy code parameter using an encoding method.

[0081] In the step S242, the specific technical means of the encoding method is the existing conventional method. For example, for the maximum number of calculation request responses, the corresponding numerical value can be directly used as a one-dimensional strategy code parameter value; for the lowest level of calculation request responses, the corresponding high-level code can be 1, the corresponding intermediate code can be 2, and the corresponding low-level code can be 3, thereby obtaining another dimension of strategy code parameter value; for the calculation request forwarding direction, the corresponding forward code can be 1, and the corresponding reverse code can be -1, thereby obtaining another dimension of strategy code parameter value; finally, the strategy code parameter values ​​of all dimensions are arranged in chronological order to obtain the multi-dimensional timing strategy code parameter.

[0082] S243. For each dimension of the multidimensional timing strategy code parameter, the corresponding timing strategy code parameter is used as the second timing data S2 and the normal distribution KS check is performed to calculate the corresponding second check statistic p value p2. Then, if the first check statistic p value p1 and the second check statistic p value p2 are both greater than the preset threshold, the corresponding correlation coefficient r is calculated according to the following formula: q :

[0083]

[0084] In the formula, i represents a positive integer, x i Indicates the i-th value in S1, y i represents the i-th value in S2, represents the mean value of S1, represents the mean of S2.

[0085] In the step S242 , the preset threshold may be 0.05, for example.

[0086] S244. The correlation coefficient r corresponding to the timing strategy code parameters of each dimension q The average value is used as the correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data.

[0087] S25. Update and calculate the certain dimension operation data of the certain node device in the current next cycle prediction collection according to the following formula:

[0088]

[0089] Where, Indicates the certain dimension operation data of the certain node device that is predicted and collected in the current next cycle before updating, It indicates the updated running data of a certain dimension of a certain node device that is predicted to be collected in the next period of the current period.

[0090] Therefore, based on the above steps S21 to S25, by calculating the correlation coefficient between the allocation strategy and the node device and the operating data, and using the correlation coefficient as the influencing factor size for the correction and update of the timing prediction results, the adverse effects of the execution of the allocation strategy on the operating data of the current node device can be eliminated to a certain extent, ensuring the accuracy of the load prediction of each edge node under the assumption that the strategy is not executed, and thus ensuring the reliability of subsequent strategy determination and execution.

[0091] S3. Based on the edge computing resource dynamic allocation strategy historically executed for the edge computing network and the multidimensional operation data and the multidimensional indicator parameters of each node device collected during the corresponding strategy execution cycle, a mapping model of the edge computing resource dynamic allocation strategy and the multidimensional operation data and multidimensional indicator parameters is established based on a second artificial intelligence algorithm.

[0092] In step S3, the specific construction process of the mapping model is to apply multiple groups of edge computing resource dynamic allocation strategies for the historical execution of the edge computing network and the multidimensional operation data of the node devices collected in the corresponding strategy execution cycle based on the second artificial intelligence algorithm, and the multidimensional indicator parameters collected in the corresponding strategy execution cycle (that is, each group of data is used as a sample data, in which the edge computing resource dynamic allocation strategy for the historical execution of the edge computing network and the multidimensional operation data of the node devices collected in the corresponding strategy execution cycle are used as mapping model input items, and the multidimensional indicator parameters collected in the corresponding strategy execution cycle are used as mapping model output items), and trained through conventional calibration modeling method (the specific process includes the calibration process and verification process of the model, that is, first comparing the model simulation results with the measured data, and then adjusting the model parameters according to the comparison results to make the simulation results consistent with the actual process); in the calibration verification modeling process of the mapping model, the model parameters can be specifically tuned by a tree-structured Bayesian optimization algorithm. In addition, the second artificial intelligence algorithm specifically but not limited to adopts a machine learning algorithm based on support vector machine, K nearest neighbor method, stochastic gradient descent method, multi-layer perceptron, decision tree, back propagation neural network or radial basis function network, and the edge computing resource dynamic allocation strategy needs to be converted into multi-dimensional strategy code parameters through encoding when applied.

[0093] S4. Based on the adaptive hummingbird algorithm, the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be executed is optimized to obtain the edge computing resource dynamic allocation strategy and the best search result for minimizing the fitness function, wherein the independent variables of the fitness function include the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be selected and the multidimensional indicator parameters obtained by importing the multidimensional operation data of each node device in the current next cycle into the mapping model.

[0094] In step S4, the adaptive hummingbird algorithm is a bio-inspired optimization algorithm based on the intelligent behavior of hummingbirds, which aims to solve optimization problems; it is inspired by the foraging behavior and group characteristics of hummingbirds in nature, and gradually optimizes the solution by simulating the random flight, food source tracking and detection, and information exchange of hummingbirds. Therefore, it can be modified and applied in this embodiment to achieve the purpose of optimizing the edge computing network and the edge computing resource dynamic allocation strategy to be executed in the next week. The fitness function is used to comprehensively consider multiple key indicators for evaluating the edge computing resource allocation results, and integrate them into a comprehensive evaluation index through weighted summation or other combination methods, so as to measure the pros and cons of the edge computing resource dynamic allocation strategy represented by each hummingbird individual (that is, the lower the comprehensive evaluation index value, the better the edge computing resource dynamic allocation strategy, and vice versa). Specifically, the calculation formula of the fitness function F is expressed as follows:

[0095] F=η t ×S t +η d ×S d +η u ×S u +η o ×S o

[0096] Where S t represents the average response delay in the multidimensional indicator parameters obtained by importing the dynamic allocation strategy of edge computing resources for the edge computing network and the current next cycle to be selected and the multidimensional operation data of each node device in the current next cycle into the mapping model, S d Indicates the standard deviation of computing resource utilization in this multidimensional indicator parameter, S u Indicates the standard deviation of storage resource utilization in this multi-dimensional indicator parameter, S o Indicates the standard deviation of bandwidth resource utilization in the multi-dimensional indicator parameter, η t ,η d ,η u and η o Respectively represent the preset weight coefficients. The aforementioned weight coefficient η t ,η d ,η u and η o To balance the importance of different objectives, they first need to normalize parameters according to the dimension, and can obtain parameters that meet the requirements by testing different weight combinations in simulation. Then, after deployment, they can be dynamically adjusted in real time according to the on-site situation. For example, when the average response delay increases, the weight coefficient η can be temporarily increased. t ; The aforementioned weight coefficient η t ,ηd ,η u and η o The sum of does not need to be 1 and can all be positive. Thus, based on the aforementioned fitness function F, complex allocation result evaluations can be converted into a single numerical value, facilitating comparison and decision-making in subsequent optimization steps. Furthermore, to ensure consistency in the dimensions of the independent variables in the fitness function F and thereby improve the accuracy of the fitness calculation results, the aforementioned average response delay, standard deviation of computing resource utilization, standard deviation of storage resource utilization, and standard deviation of bandwidth resource utilization are preferably normalized.

[0097] In step S4, during the operation of the adaptive hummingbird algorithm, it is necessary to adjust and control the Cauchy and Gaussian fusion mutation operators to perturb the hummingbird individuals, thereby iteratively updating the population position. In order to balance global search and local development capabilities during the operation of the algorithm and to prevent the algorithm from falling into a local optimum too early, preferably, based on the adaptive hummingbird algorithm, the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be executed is optimized to obtain the optimal search result for the edge computing resource dynamic allocation strategy and for minimizing the fitness function, including but not limited to the following steps S401 to S407.

[0098] S401. Initialize the optimization algorithm parameters of the adaptive hummingbird algorithm and include the population size N and the maximum number of iterations T, and randomly generate a set of parameters to be optimized and a corresponding initial search value array for each hummingbird individual in the hummingbird population including N hummingbird individuals, and then execute step S402, wherein the set of parameters to be optimized includes multidimensional parameters obtained by encoding and having a mapping relationship with the edge computing network and the current next cycle and the current edge computing resource dynamic allocation strategy to be executed, and the initial search value array x 0 It is expressed as follows:

[0099]

[0100] Wherein, k represents a positive integer less than or equal to K, K represents the total number of parameters in the parameter set to be optimized, represents the initial search value corresponding to the kth parameter in the parameter set to be optimized, u c,k represents the upper limit of the parameter search space corresponding to the kth parameter, l c,k represents the lower limit of the parameter search space corresponding to the kth parameter, and rand(0,1) represents a pure decimal random generation function.

[0101] In step S401, the initial search value array maps the initial allocation strategy of edge computing resources facing the edge computing network and the current next cycle and currently to be executed. The upper / lower limit of the parameter search space corresponding to the kth parameter can be conventionally determined based on the coding range corresponding to the kth parameter. For example, for the parameter of the lowest level of computing request response, the corresponding coding range is 1 to 3 and is an integer, then the upper limit of the parameter search space corresponding to the parameter is 3, and the lower limit is 1.

[0102] S402. For each hummingbird individual, the corresponding initial search value array is mapped to the edge computing resource dynamic allocation strategy for the edge computing network and the current next cycle to be selected, and then the edge computing resource dynamic allocation strategy and the multidimensional operation data of each node device in the current next cycle are imported into the mapping model, and the first multidimensional indicator parameter is output, and the first multidimensional indicator parameter is imported into the fitness function as an independent variable to obtain the corresponding first fitness, and finally step S403 is executed.

[0103] In step S402, when mapping the initial search value array to the edge computing resource dynamic allocation strategy to be selected for the edge computing network and the current next cycle, it is necessary to consider the rounding requirements of certain parameters. For example, for the parameter of the minimum level of computing request response, rounding processing needs to be performed during the mapping process.

[0104] S403. Use the initial search value array corresponding to the minimum first fitness as the current optimal search value array, initialize and set the current iteration number t=1, and then execute step S404.

[0105] S404. For each hummingbird individual, update the corresponding current search value array according to the current number of iterations, the current optimal search value array and the corresponding pre-update search value array, and then execute step S405, wherein the current search value array x t It is expressed as follows:

[0106]

[0107] Where, Represents the current search value in the array x t and the search value corresponding to the k-th parameter, represents the search value in the pre-update search value array and corresponding to the kth parameter, represents the search value corresponding to the k-th parameter in the current optimal search value array, AF() represents the adaptive adjustment factor calculation function, sig() represents the adaptive Sigmoid function, Ca(0,1) represents a random variable that satisfies the standard Cauchy distribution, and Ga(0,1) represents a random variable that satisfies the standard Gaussian distribution.

[0108] In step S404, the adaptive adjustment factor calculation function AF() is used to adjust and control the Cauchy and Gaussian fusion mutation operators to perturb the hummingbird individuals; since the standard sigmoid function may have the following deficiencies when balancing exploration and development: (1) insufficient attenuation speed in the initial stage, resulting in low global search efficiency; (2) insufficient local development capabilities in the later stage, and easy to fall into suboptimal solutions, by using the adaptive Sigmoid function to adjust the parameters, it can be more in line with the dynamic needs of edge computing resource allocation, and then the adaptive adjustment factor can be dynamically adjusted according to the current number of iterations, so as to achieve the purpose of performing global search in the early stage of the algorithm and focusing on local development in the later stage.

[0109] S405. For each hummingbird individual, the corresponding current search value array is mapped to a dynamic allocation strategy for edge computing resources to be selected for the edge computing network and the current next cycle, and then the dynamic allocation strategy for edge computing resources and the multidimensional operation data of each node device in the current next cycle are imported into the mapping model, and a second multidimensional indicator parameter is output, and the second multidimensional indicator parameter is imported into the fitness function as an independent variable to obtain the corresponding second fitness, and finally step S406 is executed.

[0110] In the step S405, the specific details are derived from the aforementioned step S402 and will not be repeated here.

[0111] S406. Determine whether the minimum second fitness is lower than the fitness corresponding to the current optimal search value array. If so, update the current optimal search value array to the current search value array corresponding to the minimum second fitness, and then execute step S407. Otherwise, directly execute step S407.

[0112] S407. Determine whether the current number of iterations t reaches the maximum number of iterations T. If so, use the current optimal search value array as the best search result obtained by optimizing the dynamic allocation strategy of edge computing resources for the edge computing network and the next cycle to be executed and used to minimize the fitness function. Otherwise, add 1 to the current number of iterations t, and then return to execute step S404.

[0113] S5. Dynamically allocate edge computing resources to the edge computing network according to the best search result.

[0114] In the step S5, the best search result is specifically mapped to the edge computing network and the dynamic allocation strategy of edge computing resources to be executed in the next week (for specific details, please refer to the derivation of the aforementioned step S402 and will not be repeated here), and then the computing resources of the edge computing network are dynamically scheduled according to the dynamic allocation strategy of edge computing resources. In this way, it can adapt to the real-time load changes of the edge computing network, the load fluctuations between nodes and future load changes, and to a certain extent eliminate defects such as uneven resource allocation or delays. It also optimizes the dynamic allocation strategy of edge computing resources to be executed based on the prediction model, mapping model and adaptive hummingbird algorithm, providing a real-time adjustment and optimization mechanism, thereby ensuring the overall efficiency of the edge computing network and facilitating practical application and promotion.

[0115] Therefore, based on the edge computing resource dynamic allocation method described in the above steps S1 to S5, a new solution is provided for optimizing and adjusting the edge computing resource dynamic allocation strategy based on artificial intelligence algorithm and adaptive hummingbird algorithm, that is, first based on the multi-dimensional operation data of each node device and the edge computing network collected periodically and the multi-dimensional indicator parameters used to evaluate the edge computing resource allocation results, based on the artificial intelligence algorithm, the multi-dimensional operation data of each node device in the current next cycle is predicted, and a mapping model of the edge computing resource dynamic allocation strategy and the multi-dimensional operation data and the multi-dimensional indicator parameters is established, and then based on the adaptive hummingbird algorithm, the edge computing network and the current next cycle to be executed are optimized. The edge computing resource dynamic allocation strategy is optimized to obtain the best search result for minimizing the fitness function. Finally, the edge computing resources are dynamically allocated to the edge computing network based on the best search result. This can adapt to the real-time load changes of the edge computing network, the load fluctuations between nodes, and future load changes, and to a certain extent eliminate defects such as uneven resource allocation or delays. It also provides a real-time adjustment and optimization mechanism by optimizing the current edge computing resource dynamic allocation strategy to be executed based on the prediction model, mapping model, and adaptive hummingbird algorithm, thereby ensuring the overall efficiency of the edge computing network and facilitating practical application and promotion.

[0116] like Figure 2 As shown, the second aspect of this embodiment provides a virtual device for implementing the edge computing resource dynamic allocation method described in the first aspect, which is arranged in a central device for respectively communicating with each node device in the edge computing network, and includes a data cycle acquisition unit, an operation data prediction unit, a mapping model establishment unit, an allocation strategy optimization unit, and a resource allocation scheduling unit;

[0117] The data periodic collection unit is used to periodically collect multidimensional operating data of each node device and multidimensional indicator parameters of the edge computing network for evaluating edge computing resource allocation results;

[0118] The operation data prediction unit is communicatively connected to the data cycle acquisition unit, and is configured to establish a time series data prediction model based on a first artificial intelligence algorithm based on the multidimensional operation data of each node device collected in the most recent multiple consecutive cycles, and to apply the time series data prediction model to predict the multidimensional operation data of each node device in the next cycle;

[0119] The mapping model establishment unit is communicatively connected to the data period collection unit and is used to establish, based on a second artificial intelligence algorithm, a mapping model of the edge computing resource dynamic allocation strategy and the multidimensional operation data and the multidimensional indicator parameters according to the edge computing resource dynamic allocation strategy historically executed for the edge computing network and the multidimensional operation data and the multidimensional indicator parameters of each node device collected during the corresponding strategy execution cycle;

[0120] The allocation strategy optimization unit is communicatively connected to the operation data prediction unit and the mapping model establishment unit, respectively, and is used to optimize the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be executed based on the adaptive hummingbird algorithm, and obtain the edge computing resource dynamic allocation strategy and the best search result for minimizing the fitness function, wherein the independent variables of the fitness function include the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be selected and the multidimensional operation data of each node device in the current next cycle imported into the mapping model. Multidimensional index parameters;

[0121] The resource allocation scheduling unit is communicatively connected to the allocation strategy optimization unit and is used to dynamically allocate edge computing resources to the edge computing network according to the best search result.

[0122] The working process, working details and technical effects of the aforementioned device provided in the second aspect of this embodiment can be found in the edge computing resource dynamic allocation method described in the first aspect, and will not be repeated here.

[0123] like Figure 3As shown, the third aspect of this embodiment provides a computer device for executing the edge computing resource dynamic allocation method as described in the first aspect, including a memory, a processor and a transceiver that are sequentially connected in communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the edge computing resource dynamic allocation method as described in the first aspect or the possible design one. For example, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in first-out memory (FIFO) and / or first-in last-out memory (FILO), etc.; the processor may be, but is not limited to, a microprocessor of the STM32F105 series. In addition, the computer device may also include, but is not limited to, a power module, a display screen and other necessary components.

[0124] The working process, working details and technical effects of the aforementioned computer equipment provided in the third aspect of this embodiment can be found in the edge computing resource dynamic allocation method described in the first aspect, and will not be repeated here.

[0125] In a fourth aspect of this embodiment, a computer-readable storage medium is provided that stores instructions for the method for dynamically allocating edge computing resources as described in the first aspect, that is, the computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the method for dynamically allocating edge computing resources as described in the first aspect is executed. The computer-readable storage medium refers to a carrier for storing data, and may include, but is not limited to, a computer-readable storage medium such as a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0126] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment can be referred to the dynamic allocation method of edge computing resources described in the first aspect, and will not be repeated here.

[0127] A fifth aspect of this embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, implements the method for dynamically allocating edge computing resources as described in the first aspect. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0128] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. An artificial intelligence-driven dynamic allocation method for edge computing resources, characterized in that: Executed by a central device that communicates with each node device in the edge computing network, including: Periodically collecting multidimensional operating data of each node device and multidimensional indicator parameters of the edge computing network for evaluating edge computing resource allocation results; Establishing a time series data prediction model based on a first artificial intelligence algorithm based on the multi-dimensional operating data of each node device collected in the most recent multiple consecutive cycles, and applying the time series data prediction model to predict the multi-dimensional operating data of each node device in the next cycle; Establishing a mapping model between the edge computing resource dynamic allocation strategy and the multidimensional operating data and the multidimensional indicator parameters based on a second artificial intelligence algorithm according to the edge computing resource dynamic allocation strategy historically executed for the edge computing network and the multidimensional operating data and the multidimensional indicator parameters of each node device collected during the corresponding strategy execution cycle; Based on the adaptive hummingbird algorithm, the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be executed is optimized to obtain the edge computing resource dynamic allocation strategy and the best search result for minimizing the fitness function, wherein the independent variables of the fitness function include the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be selected and the multidimensional operating data of each node device in the current next cycle imported into the mapping model. Multidimensional indicator parameters; Dynamically allocate edge computing resources to the edge computing network according to the best search result.

2. The method for dynamic allocation of edge computing resources according to claim 1, characterized in that: The multidimensional operation data contains the computing resource utilization, storage resource utilization and bandwidth resource occupancy of the corresponding node devices, and the multidimensional indicator parameters contain the average response delay of all computing requests and the standard deviation of the computing resource utilization, storage resource utilization and bandwidth resource occupancy of all node devices.

3. The method for dynamic allocation of edge computing resources according to claim 2, characterized in that: The calculation formula of the fitness function F is as follows: F=η t ×S t +n d ×S d +n u ×S u +n o ×S o Where S t represents the average response delay in the multidimensional indicator parameters obtained by importing the dynamic allocation strategy of edge computing resources for the edge computing network and the current next cycle to be selected and the multidimensional operation data of each node device in the current next cycle into the mapping model, S d Indicates the standard deviation of computing resource utilization in this multidimensional indicator parameter, S u Indicates the standard deviation of storage resource utilization in this multi-dimensional indicator parameter, S o Indicates the standard deviation of bandwidth resource utilization in the multi-dimensional indicator parameter, η t ,η d ,η u and η o Respectively represent the preset weight coefficients.

4. The method for dynamic allocation of edge computing resources according to claim 1, characterized in that: The first artificial intelligence algorithm adopts a machine learning algorithm based on a long short-term memory network, a Bi-LSTM network, an Attention-LSTM network or a Prophet model, and the second artificial intelligence algorithm adopts a machine learning algorithm based on a support vector machine, a K-nearest neighbor method, a stochastic gradient descent method, a multi-layer perceptron, a decision tree, a back-propagation neural network or a radial basis function network.

5. The method for dynamic allocation of edge computing resources according to claim 1, characterized in that: Based on the multi-dimensional operating data of each node device collected in the most recent multiple consecutive cycles, a time series data prediction model is established based on a first artificial intelligence algorithm, and the time series data prediction model is applied to predict the multi-dimensional operating data of each node device in the next cycle, including: For a certain dimension of operation data of a certain node device in the edge computing network, corresponding sample data is extracted in the following manner: multidimensional operation data collected in m consecutive cycles and of each node device are extracted from the multidimensional operation data collected in the current most recent multiple consecutive cycles and of each node device, and the extraction result is used as a prediction model input item; and the certain dimension of operation data collected in the m+1th cycle and of the certain node device are also extracted from the multidimensional operation data collected in the current most recent multiple consecutive cycles and of each node device, and the new extraction result is used as a prediction model output item, and finally the prediction model input item and the prediction model output item are used as sample data, wherein m represents a positive integer greater than or equal to 2 and less than M, M represents the total number of cycles of the current most recent multiple consecutive cycles, and the m+1th cycle refers to the next cycle after the m consecutive cycles; Applying all the sample data to perform calibration and verification modeling on a machine learning model based on the first artificial intelligence algorithm, and obtaining a time series data prediction model corresponding to the certain node device and the certain dimension of operation data; Importing the multi-dimensional operation data of each node device collected in the most recent m consecutive cycles into the time series data prediction model, and outputting the certain dimensional operation data of the node device predicted to be collected in the next cycle; Calculate, based on the edge computing resource dynamic allocation strategy historically executed by the edge computing network in the most recent multiple consecutive cycles and the certain dimension operation data of the certain node device collected in the most recent multiple consecutive cycles, a correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data; The following formula is used to update and calculate the certain dimension operation data of the certain node device in the current next period prediction collection: Where, Indicates the certain dimension operation data of the certain node device that is predicted and collected in the current next cycle before updating, It indicates the updated running data of a certain dimension of a certain node device that is predicted to be collected in the next period of the current period.

6. The method for dynamic allocation of edge computing resources according to claim 5, characterized in that: According to the edge computing resource dynamic allocation strategy executed historically for the edge computing network in the most recent multiple consecutive cycles and the certain dimension operation data of the certain node device collected in the most recent multiple consecutive cycles, a correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data is calculated, including: The certain dimension operation data of the certain node device collected in the most recent multiple consecutive cycles is used as the first time series data S1, and the normal distribution KS check is performed on the first time series data S1 to calculate a first check statistic p value p1; Converting the edge computing resource dynamic allocation strategy for the edge computing network executed in the current multiple consecutive cycles into a multi-dimensional timing strategy code parameter in an encoding manner; For each dimension of the multidimensional timing strategy code parameter, the corresponding timing strategy code parameter is used as the second timing data S2 and the normal distribution KS check is performed to calculate the corresponding second check statistic p value p2. Then, if the first check statistic p value p1 and the second check statistic p value p2 are both greater than the preset threshold, the corresponding correlation coefficient r is calculated according to the following formula: q : In the formula, i represents a positive integer, x i Indicates the i-th value in S1, y i represents the i-th value in S2, represents the mean value of S1, represents the mean value of S2; The correlation coefficient r corresponding to the timing strategy code parameters of each dimension q The average value is used as the correlation coefficient γ between the edge computing resource dynamic allocation strategy and the certain node device and the certain dimension operation data.

7. The method for dynamic allocation of edge computing resources according to claim 1, characterized in that: Optimizing the edge computing resource dynamic allocation strategy for the edge computing network and the next cycle to be executed based on the adaptive Hummingbird algorithm to obtain the optimal search result for the edge computing resource dynamic allocation strategy and minimizing the fitness function includes the following steps S401 to S407: S401. Initialize the optimization algorithm parameters of the adaptive hummingbird algorithm and include the population size N and the maximum number of iterations T, and randomly generate a set of parameters to be optimized and a corresponding initial search value array for each hummingbird individual in the hummingbird population including N hummingbird individuals, and then execute step S402, wherein the set of parameters to be optimized includes multidimensional parameters obtained by encoding and having a mapping relationship with the edge computing network and the current next cycle and the current edge computing resource dynamic allocation strategy to be executed, and the initial search value array x 0 It is expressed as follows: Wherein, k represents a positive integer less than or equal to K, K represents the total number of parameters in the parameter set to be optimized, represents the initial search value corresponding to the kth parameter in the parameter set to be optimized, u c,k represents the upper limit of the parameter search space corresponding to the kth parameter, l c,k represents the lower limit of the parameter search space corresponding to the kth parameter, and rand(0,1) represents a pure decimal random generator function; S402. For each hummingbird individual, the corresponding initial search value array is mapped into a dynamic allocation strategy for edge computing resources to be selected for the edge computing network and the current next cycle. The dynamic allocation strategy for edge computing resources and the multidimensional operating data of each node device in the current next cycle are then imported into the mapping model to obtain a first multidimensional indicator parameter as an output. The first multidimensional indicator parameter is then imported into the fitness function as an independent variable to obtain a corresponding first fitness. Finally, step S403 is executed. S403. The initial search value array corresponding to the minimum first fitness is used as the current optimal search value array, and the current iteration number t=1 is initialized and set, and then step S404 is executed; S404. For each hummingbird individual, update the corresponding current search value array according to the current number of iterations, the current optimal search value array and the corresponding pre-update search value array, and then execute step S405, wherein the current search value array x t It is expressed as follows: Where, Represents the current search value in the array x t and the search value corresponding to the k-th parameter, represents the search value in the pre-update search value array and corresponding to the kth parameter, represents the search value corresponding to the kth parameter in the current optimal search value array, AF() represents an adaptive adjustment factor calculation function, sig() represents an adaptive Sigmoid function, Ca(0,1) represents a random variable that satisfies a standard Cauchy distribution, and Ga(0,1) represents a random variable that satisfies a standard Gaussian distribution; S405. For each hummingbird individual, the corresponding current search value array is mapped into a dynamic allocation strategy for edge computing resources to be selected for the edge computing network and the current next cycle. The dynamic allocation strategy for edge computing resources and the multidimensional operating data of each node device in the current next cycle are then imported into the mapping model to obtain a second multidimensional indicator parameter as an output. The second multidimensional indicator parameter is then imported into the fitness function as an independent variable to obtain a corresponding second fitness. Finally, step S406 is executed. S406. Determine whether the minimum second fitness is lower than the fitness corresponding to the current optimal search value array. If so, update the current optimal search value array to the current search value array corresponding to the minimum second fitness, and then execute step S407. Otherwise, execute step S407 directly. S407. Determine whether the current number of iterations t reaches the maximum number of iterations T. If so, use the current optimal search value array as the best search result obtained by optimizing the dynamic allocation strategy of edge computing resources for the edge computing network and the next cycle to be executed and used to minimize the fitness function. Otherwise, add 1 to the current number of iterations t, and then return to execute step S404.

8. The method for dynamic allocation of edge computing resources according to claim 1, characterized in that: The edge computing resource dynamic allocation strategy includes increasing / decreasing the maximum number of computing request responses of peripheral nodes, raising / lowering the minimum level of computing request responses of peripheral nodes and / or adjusting the computing request forwarding direction between peripheral nodes.

9. An artificial intelligence-driven edge computing resource dynamic allocation device, characterized in that: It is arranged in the central device for communicating with each node device in the edge computing network, and includes a data cycle acquisition unit, an operation data prediction unit, a mapping model establishment unit, an allocation strategy optimization unit, and a resource allocation scheduling unit; The data periodic collection unit is used to periodically collect multi-dimensional operating data of each node device and multi-dimensional indicator parameters of the edge computing network for evaluating edge computing resource allocation results; The operation data prediction unit is communicatively connected to the data cycle acquisition unit, and is configured to establish a time series data prediction model based on a first artificial intelligence algorithm based on the multidimensional operation data of each node device collected in the most recent multiple consecutive cycles, and to apply the time series data prediction model to predict the multidimensional operation data of each node device in the next cycle; The mapping model establishment unit is communicatively connected to the data period collection unit and is used to establish, based on a second artificial intelligence algorithm, a mapping model of the edge computing resource dynamic allocation strategy and the multidimensional operation data and the multidimensional indicator parameters according to the edge computing resource dynamic allocation strategy historically executed for the edge computing network and the multidimensional operation data and the multidimensional indicator parameters of each node device collected during the corresponding strategy execution cycle; The allocation strategy optimization unit is communicatively connected to the operation data prediction unit and the mapping model establishment unit, respectively, and is used to optimize the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be executed based on the adaptive hummingbird algorithm, and obtain the edge computing resource dynamic allocation strategy and the best search result for minimizing the fitness function, wherein the independent variables of the fitness function include the edge computing resource dynamic allocation strategy for the edge computing network and the current next week to be selected and the multidimensional operation data of each node device in the current next cycle imported into the mapping model. Multidimensional index parameters; The resource allocation scheduling unit is communicatively connected to the allocation strategy optimization unit and is used to dynamically allocate edge computing resources to the edge computing network according to the best search result.

10. A computer device, characterized in that: It includes a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the edge computing resource dynamic allocation method as described in any one of claims 1 to 8.

Images