Edge cluster resource monitoring prediction method, system and device based on cloud edge collaboration and medium

Abstract

The invention discloses an edge cluster resource monitoring prediction method, system and device based on cloud edge collaboration and a medium, and relates to the technical field of power grid resource scheduling, and the method comprises the steps: collecting operation data used for predicting an edge cluster through an edge cluster platform, and carrying out the data preprocessing; based on timestamps, physical equipment configuration and historical node operation data, feature extraction operation is carried out on the preprocessed data, and a data set is constructed; constructing an SVR model in the cloud center, training the SVR model by using the data set, and predicting the operation condition of the edge cluster; the cloud center carries out model training on the data received from each node device by using a litsea rotundifolia optimization algorithm, carries out global optimization, and finds optimal model parameter configuration; and predicting the operation state of the edge cluster through optimal model parameter configuration, and formulating a scheduling strategy according to a prediction result. According to the method, the problems of edge computing power bottleneck, data link delay and insufficient global collaboration in a traditional cloud edge scheme are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid resource scheduling, and in particular to an edge cluster resource monitoring and prediction method, system, device and medium based on cloud-edge collaboration. BACKGROUND

[0002] With the development of smart grids, the management and optimization of power systems increasingly rely on advanced data analysis and machine learning techniques. Current solutions typically utilize edge computing to enable local processing and fast response of data, reducing latency and improving efficiency. By deploying node devices at various nodes of the power grid, a variety of data including power production, consumption, environmental conditions, and historical load can be collected in real time, and local computing capabilities can be utilized to perform preliminary analysis with low latency. To further improve data processing efficiency, existing methods often use models such as support vector regression (SVR) and random forest for operational state prediction, and combine a federated learning framework to implement model training, in order to alleviate the contradiction between data privacy and communication overhead.

[0003] However, the existing technology has significant limitations in dealing with complex power grid system scenarios. First, in the face of complex and dynamically changing system environments, traditional model training methods may not provide sufficient accuracy and real-time performance, especially when considering the influence of multiple factors such as weather conditions and system load on system balance. Second, while federated learning can protect privacy, its decentralized parameter update mechanism is prone to local optima, especially in multi-edge cluster collaboration scenarios, lacking a cloud-based global perspective on resource allocation strategies, leading to uneven load distribution among clusters. Finally, existing methods lack consideration of system safety constraints, and the parameter optimization process is time-consuming, making it difficult to meet real-time scheduling requirements in high-load scenarios.

[0004] Therefore, there is an urgent need for an edge cluster resource monitoring and prediction method based on cloud-edge collaboration to improve prediction accuracy and system stability. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] Therefore, the problem to be solved by the present application is that in the face of complex and dynamically changing system environments, traditional model training methods may not provide sufficient accuracy and real-time performance, especially when considering the influence of multiple factors on system balance.

[0007] To solve the above technical problems, the application provides the following technical scheme: an edge cluster resource monitoring and prediction method based on cloud edge collaboration, which comprises collecting running data of the edge cluster for prediction through an edge cluster platform and performing data preprocessing; performing feature extraction operation on the preprocessed data based on a time stamp, physical device configuration and historical node running data to construct a data set; constructing a support vector regression (SVR) model in a cloud center to predict the running condition of the edge cluster; training the SVR model by using the data set collected by each node device to generate a model parameter set of the SVR model; performing global optimization on the model parameter set by using a golden jackal optimization algorithm in the cloud center to find optimal model parameter configuration of the SVR model and generate an SVR model with optimal model parameter configuration; and predicting the running state of the edge cluster by using the SVR model with optimal model parameter configuration according to the real-time collected data and formulating a scheduling strategy according to the prediction result.

[0008] As a preferred scheme of the edge cluster resource monitoring and prediction method based on cloud edge collaboration, the running data of the edge cluster for prediction comprises service tasks, network environment data, edge cluster running state data, historical node running data and historical load data.

[0009] As a preferred scheme of the edge cluster resource monitoring and prediction method based on cloud edge collaboration, the finding of the optimal model parameter configuration of the SVR model comprises generating an initial population according to the edge cluster running state data and device node real-time load rate; the initialization of the parameters of the golden jackal optimization algorithm comprises population size, maximum iteration number and initial energy factor; a multi-objective fitness function is constructed based on the prediction error of the data set collected by each node device and the system stability requirement of the SVR model to calculate the fitness of each individual; the female and male jackals are determined according to the fitness, the current energy is calculated and the energy factor is updated; the prey and leader position are updated and the population is updated according to the current energy; and the optimal model parameter configuration is output until the maximum iteration number is reached.

[0010] As a preferred scheme of the edge cluster resource monitoring and prediction method based on cloud edge collaboration, the model parameter set of the SVR model is wherein n is the number of node devices, C i is a penalty coefficient, ∈ i is an insensitive area, γ i is a kernel function parameter; the population size is dynamically calculated based on the real-time load rate of the device node, and the population size calculation formula is:

[0011]

[0012] wherein L is the real-time load rate, L maxk is an attenuation coefficient; and the parameter disturbance range Δmax is determined according to a ratio of L to L max ; an initial population containing N individuals is generated, wherein the first n individuals inherit the local SVR model parameters of each node device, and the model parameters of the subsequent individuals are generated by a population calculation formula, the population calculation formula being:

[0013]

[0014] ∈∈[0.01×P base , 0.1×P rated ]

[0015]

[0016] wherein, is the model parameter of the subsequent individual; is a parameter vector of the ith n device selected from the parameter set of the n node devices in a loop; in is the index i of the subsequent individual modulo the number of devices n; × (-1, 1) is a uniformly distributed random vector; C, ∈, γ are constraint parameter ranges; V min is a minimum voltage safety value; V max is a maximum voltage safety value; P base is a reference power; P rated is a device rated power; σ min is a minimum standard deviation of the kernel function parameter set; σ max is a maximum standard deviation of the kernel function parameter set.

[0017] As a preferred scheme of the edge cluster resource monitoring and prediction method based on cloud-edge collaboration, the determination manner of the parameter disturbance range Δmax is: when L < 0.8L max , Δmax = 0.1 × (2-sin(π×L / L max )); when L≥0.8L max ,

[0018] As a preferred scheme of the edge cluster resource monitoring and prediction method based on cloud-edge collaboration, the calculation of the fitness of each individual includes: calculating the prediction error mean of the SVR model based on the data set collected by each node device; generating a load rate curve according to the node load change rate within a preset time window; generating a safety penalty term based on the amplitude of the node load rate deviating from the safety interval and the proportion of the line load rate exceeding the limit; and performing weighted summation on the prediction error mean, the load rate fluctuation penalty term and the safety penalty term to generate the fitness value.

[0019] As a preferred scheme of the edge cluster resource monitoring and prediction method based on cloud edge collaboration provided in the application, the formula of the current energy is expressed as,

[0020]

[0021] E t = E t-1 + (E t-1 - E t-2 ) / 2 (t) E t is an individual energy value of the tth iteration; is a current parameter vector; is an optimal parameter vector of the previous iteration; is a current load rate proportion.

[0022] To solve the above technical problems, the application provides the following technical scheme: an edge cluster resource monitoring and prediction system based on cloud edge collaboration, comprising a data collection module, a feature extraction module, a model construction module, a model parameter set generation module, a parameter optimization module and a resource scheduling module; the data collection module collects edge cluster operation data for prediction and performs data preprocessing; the feature extraction module performs feature extraction operations on the preprocessed data based on timestamps, physical device configurations and historical node operation data to construct a data set; the model construction module constructs an SVR model in a cloud center to predict edge cluster operation; the model parameter set generation module trains the SVR model using the data set collected by each node device to generate a model parameter set of the SVR model; the parameter optimization module performs global optimization on the model parameter set using the cloud center to find optimal model parameter configurations of the SVR model and generate an SVR model with optimal model parameter configurations; and the resource scheduling module predicts edge cluster operation states through the SVR model with optimal model parameter configurations according to real-time collected data and formulates a scheduling strategy according to the prediction results.

[0023] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the edge cluster resource monitoring and prediction method based on cloud edge collaboration when executing the computer program.

[0024] A computer readable storage medium stores a computer program, and the computer program implements the steps of the edge cluster resource monitoring and prediction method based on cloud edge collaboration when executed by a processor.

[0025] The application has the advantages that: the application collects data at the edge cluster node and optimizes model parameters in the cloud through the collaborative architecture of edge lightweight collection-cloud centralized optimization, and completely concentrates model training, parameter optimization and prediction scheduling in the cloud, so that the edge side only bears data collection and lightweight preprocessing, and solves the problems of edge computing power bottleneck, data link delay and insufficient global cooperation in the traditional cloud edge scheme. The golden jackal optimization algorithm is used to dynamically generate an initial population, and the disturbance range is adaptively adjusted based on the real-time load rate of the system, and the global exploration and local development are considered. The prediction error, load rate fluctuation and safety penalty term are dynamically weighted and optimized through a multi-objective fitness function, the prediction accuracy and system safety are cooperatively optimized, the search intensity is dynamically adjusted combined with the energy driving mechanism, and the optimization response time is shortened in the high load scene. A safe, efficient and adaptive resource scheduling paradigm is provided for the smart grid, and the stability of the power grid system operation in the high proportion of new energy access scene is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating laborious work.

[0027] Figure 1 The flowchart of the edge cluster resource monitoring and prediction method based on cloud edge cooperation in embodiment 1.

[0028] Figure 2 The flowchart of the golden jackal optimization algorithm of the edge cluster resource monitoring and prediction method based on cloud edge cooperation in embodiment 1. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0030] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0031] Embodiment 1, refer to Figure 1 and Figure 2 , the first embodiment of the present application provides an edge cluster resource monitoring and prediction method based on cloud edge cooperation, which includes, as shown in Figure 1

[0032] ​S1, the edge cluster platform collects data for predicting edge cluster running data and performs data preprocessing;

[0033] In some embodiments, the data for predicting edge cluster running data includes service tasks, network environment data, edge cluster running state data, historical node running data, and historical load data.

[0034] Service tasks include but are not limited to services and processes within container nodes.

[0035] Network environment data includes but is not limited to memory, disk, network, bandwidth, load rate, disk I / O rate, CPU temperature.

[0036] Running state data includes but is not limited to node running state and service running state.

[0037] Historical node running data is an important part of analyzing and predicting system balance, which provides running state information of device nodes and their components in the past period of time, including but not limited to running records, fault reports, maintenance logs, etc. in the past period of time, which is very important for analyzing and predicting future load trends.

[0038] Historical load data refers to the changes in device node load rate in the past period of time.

[0039] After collecting the above data, a series of preprocessing steps are needed to ensure the quality and applicability of the data. Data preprocessing includes data cleaning, missing value processing, outlier processing, duplicate value processing, and data standardization.

[0040] Specifically, data cleaning: identify and correct errors or inaccurate parts in the data set, such as correcting obvious input errors or logical contradictions. Missing value processing: for missing values in the data set, different strategies can be used to fill them, such as using mean, median filling, or applying interpolation methods to estimate missing values. Outlier processing: detect and process outliers in the data, which may be caused by measurement errors or other abnormal factors. Statistical methods (such as standard deviation method) can be used to identify outliers and decide whether to remove or correct them. Duplicate value processing: check if there are duplicate records in the data set, and choose to keep unique records or merge duplicate items according to specific circumstances. Data standardization: in order to improve the effect of model training, it is usually necessary to standardize or normalize the data, so that different features have the same scale. Common methods include min-max scaling, Z-score standardization, etc.

[0041] S2, based on timestamp, physical device configuration and historical node running data, feature extraction is performed on the preprocessed data to construct a data set;

[0042] Specifically, first, the timestamp data is used to generate sliding window statistical features (such as the average load rate in the past 1 hour, the standard deviation of the 3-hour fluctuation) and to encode the time period type and holiday flag; second, the device type is independently encoded, the rated capacity is normalized, and the number of adjacent nodes and the average voltage difference are derived based on the topological relationship; further, time series statistics (24-hour maximum load rate, failure frequency of similar devices) and environment-related features (sliding correlation coefficient of temperature and humidity and load rate) are extracted from historical operation data, and spatial correlation indicators such as the average voltage difference of adjacent nodes and the load rate ratio of similar devices are fused; finally, a structured dataset containing time attributes, device static parameters, dynamic operating states, and environmental sensitive factors is generated, and the target variable is defined as the load rate prediction value and the voltage out-of-limit classification label of the next period. This multi-dimensional feature fusion method enhances the spatio-temporal correlation and device heterogeneity representation ability of the dataset, providing high information density input for the cloud SVR model, effectively supporting accurate prediction and safe scheduling in complex power grid system scenarios.

[0043] S3, constructing an SVR model in a cloud center to predict the running condition of an edge cluster; an SVR (Support Vector Regression) model is established in the cloud center to predict the running condition of the edge cluster.

[0044] S4, training the SVR model using the dataset collected by each node device to generate a model parameter set of the SVR model.

[0045] In some embodiments, the SVR model is trained according to the dataset collected by each node device to generate a model parameter set of the SVR model, which can be achieved by the following steps:

[0046] S41, selecting features and target variables: according to the dataset extracted and processed in step S2, the input features and target variables for training the SVR model are determined.

[0047] S42, data segmentation: the dataset is divided into a training set and a test set, and the commonly used ratio is that 70% of the data is used for training and 30% of the data is used for testing, to ensure that the model not only performs well on the training data but also has good generalization ability on unseen data.

[0048] S43, parameter initialization: some key parameters of the SVR model are initialized, including but not limited to the penalty coefficient C, the insensitive region ∈, and the kernel function parameter γ. The selection of these parameters is crucial to the performance of the model. The initial values can be set according to experience or optimized by methods such as grid search (Grid Search), random search (Random Search), etc.

[0049] S44, model training: the data in the training set is used to train the SVR model.

[0050] S45. Model Evaluation: Evaluate the performance of the trained SVR model on the test set. Common evaluation metrics include mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). If the model performs poorly, adjust the model parameters, or even reconsider the feature engineering approach, and then train the model again.

[0051] S46. Using the data set collected by each node device, repeat steps S41 to S45 to train the SVR model respectively and generate a model parameter set of the SVR model.

[0052] S5. The cloud center uses the golden jackal optimization algorithm to globally optimize the model parameter set, find the optimal model parameter configuration of the SVR model, and generate the SVR model with the optimal model parameter configuration; Figure 2 As shown, specifically including:

[0053] S51. Generate an initial population based on the real-time load rate of the edge cluster operation data method.

[0054] In some embodiments, S51 specifically includes:

[0055] S511, use the data set collected by each node device to train the SVR model, the SVR model parameter set Among them, n is the number of node devices, Ci is the penalty coefficient, ∈i is the insensitive area, and γi is the kernel function parameter; that is, each node device sends its trained SVR model parameter set to the cloud. These parameters include the penalty coefficient C i , insensitive area∈ i and kernel function parameter γ i wait.

[0056] S512. Dynamically calculate the population size based on the real-time load rate of the system. The population size calculation formula is:

[0057]

[0058] Among them, L is the real-time load rate, L max is the maximum safe load rate, and k is the attenuation coefficient.

[0059] For example, system parameters: maximum safe load rate L max =1000MW, current real-time load rate L = 750MW (i.e. L = 0.75L max ); attenuation coefficient k = 3, number of node devices n = 3; minimum voltage safety value V min =0.95pu; Maximum voltage safety value V max =1.05pu.

[0060] The node device has 3, using the data collected by the node device, the SVR model parameters obtained are: C1=100, C2=500, C3=200; ∈1=0.5, ∈2=0.3, ∈3=0.4; γ1=0.1, γ2=0.2, γ3=0.15.

[0061] Calculate population size

[0062] The initial population contains 4 individuals, the first 3 inherit the node device parameters, and the 4th is generated by disturbance.

[0063] S513, according to the ratio of L and L max , the parameter disturbance range Δmax includes:

[0064] When L<0.8L max , Δmax=0.1×(2-sin(π×L / L max )).

[0065] When L≥0.8L max ,

[0066] For example, L=0.75Lmax, L<0.8Lmax, so Δmax=0.1×(2-sin(π×0.75))≈0.1293.

[0067] S514, generate an initial population containing N individuals, the first n individuals inherit the local SVR model parameters of each node device, and the model parameters of the subsequent individuals are generated by population calculation formula, the population calculation formula is:

[0068]

[0069] ∈∈[0.01×P base ,0.1×P rated ]

[0070]

[0071] Where, is the model parameter of the subsequent individual; is the parameter vector of the i-th n device selected from the parameter set of n node devices; in is the index i of the subsequent individual modulo the number of devices n; u(-1,1) is a uniformly distributed random vector; C, ∈, γ are constraint parameter ranges; V min is the minimum voltage safety value; V max is the maximum voltage safety value; P base is the reference power; P rated is the rated power of the device; σmin is the minimum standard deviation of the kernel function parameter set; σ max is the maximum standard deviation of the kernel function parameter set.

[0072] For example, the initial population contains 4 individuals, the first 3 inherit the node device parameters, and the 4th is generated by disturbance. That is, the population size N = 4, the device number n = 3, and i mod n represents the modulo operation of the individual index i on the device number n, the purpose is to cyclically allocate node parameters, when n = 3:

[0073] i = 1 → 1 mod 3 = 1 → use the parameters w1 of device 1.

[0074] i = 2 → 2 mod 3 = 2 → use the parameters w2 of device 2.

[0075] i = 3 → 3 mod 3 = 0 → use the parameters w3 of device 3 (the actual implementation may be offset as w0).

[0076] i = 4 → 4 mod 3 = 1 → use the parameters of device 1 again.

[0077] Then, the 4th individual is generated based on the disturbance of the parameters of device 1:

[0078]

[0079] Assume that the random vector U (-1, 1) = [0.5, -0.3, 0.2]:

[0080]

[0081] Back-propagated parameters: C4 = 102.1293 ≈ 134.9, ∈4 = 0.4806 MW, γ4 = 0.1003.

[0082] S52, initialize the parameters of the golden eagle optimization algorithm: population size, maximum number of iterations, and initial energy factor golden eagle optimization algorithm.

[0083] S53, calculate the fitness of each individual based on the prediction error of the SVR model on the training data set and the system stability requirement.

[0084] Further, S53 specifically includes:

[0085] S531, calculate the mean prediction error of the SVR model based on the training data set.

[0086] S532, generate a load rate curve according to the node load change rate within the preset time window.

[0087] S533, generate a safety penalty term based on the magnitude of the node load rate deviating from the safety interval and the proportion of line load rate exceeding the limit.

[0088] S534, the mean prediction error, the load rate fluctuation penalty term and the safety penalty term are weighted and summed to generate the fitness value.

[0089] In some embodiments, the multi-objective fitness function is:

[0090] F = MSE local + λ × P 波 + S

[0091] Wherein, MSE local is the mean prediction error of the SVR model; P 波 is the load rate fluctuation; λ is the load rate fluctuation penalty coefficient; S is the safety penalty term.

[0092] The mean prediction error of the SVR model is calculated by the formula:

[0093]

[0094] Wherein, n is the number of node devices; m i is the number of validation set samples of the i-th node device; y ij is the true value of the j-th sample of the i-th node device; is the predicted value of the j-th sample of the i-th node device.

[0095] The load fluctuation is calculated by the formula:

[0096]

[0097] Wherein, P t is the current system load rate; P t-1 is the system load rate at the previous moment; T is the monitoring time window;

[0098] The safety penalty term is calculated by the formula:

[0099]

[0100] Wherein, p i is the load value of the i-th node; P max is the maximum allowed load value; P min is the minimum allowed load value; L is the current load rate; L max is the current maximum allowed load rate.

[0101] The fitness function F is defined as the weighted sum of three parts: the local prediction error MSElocal, the load rate fluctuation Pwave and the safety penalty term S. This multi-objective optimization method aims to balance the prediction accuracy of the model and the safety and stability of the system operation.

[0102] S54, according to the fitness, determine the male and female dholes, calculate the current energy and update the energy factor.

[0103] The energy calculation formula is:

[0104]

[0105] where E (t) is the energy value of the individual at the tth iteration; is the current parameter vector; is the optimal parameter vector of the previous iteration; is the current load rate proportion.

[0106] First, all individuals (i.e., a specific set of model parameter configurations) are sorted in ascending order of their fitness F. The lower the fitness, the better the performance of the model corresponding to the individual, the more accurate the prediction of system balance, and the more in line with safety and stability requirements.

[0107] According to certain rules, a portion of the top-ranked individuals are designated as "female jackals", and these individuals are considered to be high-quality solutions in the current population. The remaining individuals are designated as "male jackals". These individuals may perform slightly worse, but still have the opportunity to improve themselves through interaction with other individuals.

[0108] According to the roles of the individuals (female or male jackals) and their performance, the energy factor is updated.

[0109] S55, according to the current energy, update the prey and leader positions and update the population.

[0110] In the golden jackal optimization algorithm, the prey represents a potential better solution or target position, and the leader is the best individual in the group, i.e., the individual with the lowest fitness (best performance).

[0111] The prey position is updated based on energy, and the energy of each individual determines whether it can approach the prey. High-energy individuals are more likely to successfully capture the prey (find better solutions), while low-energy individuals may need to reposition to find new prey. The update formula can be expressed as:

[0112]

[0113] where X i is the current position of individual i; P best is the position of the prey; L best is the current position of the leader; r1 and r2 are two random numbers between 0 and 1, used to introduce randomness and simulate the uncertainty in nature.

[0114] The leader position is updated based on energy, and if an individual shows higher fitness (lower fitness value) after an iteration, it has the potential to become the new leader. The new leader position can be determined based on the best position of all individuals, ensuring that the entire population moves towards the optimal solution.

[0115] The new position of each individual is calculated according to the above formula

[0116] S56, repeat S53-S55 until the maximum number of iterations is reached, and output the optimal model parameter configuration.

[0117] In the above steps S51-S52, each node device uploads the collected data to the cloud, and the cloud uses the SVR model parameter set obtained by training the data of each node device. The cloud dynamically calculates the population size N according to the real-time load rate of the system, and initializes the related parameters of the golden wolf optimization algorithm. In steps S511-S514, assuming that the current load rate of the system is 70% (L=0.7L max ), the initial population size N is calculated according to the formula. An initial population containing N individuals is generated, the first n individuals directly use the SVR model parameter set, and the remaining individuals are generated by random disturbance. In step S53, a multi-objective fitness function is constructed to calculate the fitness of each individual. In steps S54-S55, the energy factor and position are updated according to the fitness, and the optimal solution is found by simulating the predatory behavior in nature. In step S56, after multiple iterations, a set of globally optimal model parameter configurations is finally found.

[0118] S6, according to the real-time collected data, through the SVR model with the optimal model parameter configuration, the running state of the edge cluster is predicted, and a scheduling strategy is formulated according to the prediction result.

[0119] The extracted feature vectors are predicted using the SVR model with the optimal model parameter configuration, and the running state of the node device in the future period of time is obtained. These prediction results usually include estimated values of future power demand.

[0120] Based on the prediction results, the edge cluster system can formulate corresponding scheduling strategies. For example:

[0121] If it is predicted that the load rate will exceed the limit soon, the operation and maintenance personnel are notified in advance to reduce related non-critical operations to reduce the load. Through the steps of the present application, the node device can ensure stable operation of the system while realizing efficient resource management and scheduling optimization. Not only does it improve the efficiency and response speed of system operation, but also enhances the flexibility and adaptability of the system, better coping with various complex and variable operating scenarios.

[0122] Embodiment 2, which is different from the first embodiment, is a cloud edge collaboration based edge cluster resource monitoring and prediction system, comprising a data collection module, a feature extraction module, a model construction module, a model parameter set generation module, a parameter optimization module and a resource scheduling module; the data collection module collects edge cluster running data for prediction and performs data preprocessing; the feature extraction module is used for performing feature extraction operation on the preprocessed data based on timestamp, physical device configuration and historical node running data to construct a data set; the model construction module is used for constructing an SVR model in a cloud center to predict edge cluster running conditions; the model parameter set generation module is used for training the SVR model by using the data set collected by each node device to generate a model parameter set of the SVR model; the parameter optimization module is used for globally optimizing the model parameter set by the cloud center by using the golden jackal optimization algorithm to find the optimal model parameter configuration of the SVR model to generate the SVR model with the optimal model parameter configuration; and the resource scheduling module is used for predicting the edge cluster running state by the SVR model with the optimal model parameter configuration according to the real-time collected data and formulating a scheduling strategy according to the prediction result.

[0123] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0124] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logical functions, and can be specifically embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch the instructions from an instruction execution system, apparatus or device and execute the instructions. For the purpose of this specification, the "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in connection with an instruction execution system, apparatus or device or in conjunction with these instruction execution systems, apparatus or devices.

[0125] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer readable medium can be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, as necessary, and stored in a computer memory.

[0126] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any forms of hardware, or combinations of hardware and software can be used, such as: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and so on.

[0127] It should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced, without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for monitoring and predicting edge cluster resources based on cloud-edge collaboration, characterized by: include, Collect and pre-process the edge cluster operation data used to predict the edge cluster through the edge cluster platform; Based on timestamps, physical device configurations, and historical node operation data, feature extraction is performed on the pre-processed data to construct a dataset. Build a support vector regression (SVR) model in the cloud center to predict the operation status of the edge cluster; The SVR model is trained using the data set collected by each node device to generate a model parameter set of the SVR model; The cloud center uses the golden jackal optimization algorithm to globally optimize the model parameter set, find the optimal model parameter configuration of the SVR model, and generate the SVR model with the optimal model parameter configuration; Based on the data collected in real time, the SVR model with optimal model parameters is used to predict the operating status of the edge cluster and formulate a scheduling strategy based on the prediction results.

2. The edge cluster resource monitoring and prediction method based on cloud-edge collaboration according to claim 1 is characterized by: The operation data used to predict the edge cluster includes service tasks, network environment data, edge cluster operation status data, historical node operation data and historical load data.

3. The edge cluster resource monitoring and prediction method based on cloud-edge collaboration according to claim 2 is characterized by: The finding of the optimal model parameter configuration of the SVR model includes generating an initial population based on the edge cluster operation status data and the real-time load rate of the device node; Initializing the parameters of the golden jackal optimization algorithm include population size, maximum number of iterations and initial energy factor; A multi-objective fitness function is constructed based on the prediction error of the data set collected by the SVR model at each node device and the system stability requirements, and the fitness of each individual is calculated; According to the fitness, determine the male and female jackals, calculate the current energy and update the energy factor; Based on the current energy, the prey and leader positions are updated and the population is updated; Until the maximum number of iterations is reached, the optimal model parameter configuration is output.

4. The edge cluster resource monitoring and prediction method based on cloud-edge collaboration according to claim 3 is characterized by: The model parameter set of the SVR model is Where n is the number of node devices, C i is the penalty coefficient, ∈ i is the insensitive area, γ i is the kernel function parameter; The population size is dynamically calculated based on the real-time load rate of the device nodes. The population size calculation formula is: Among them, L is the real-time load rate, L max is the maximum safe load rate, k is the attenuation coefficient; According to L and L max The ratio of determines the parameter disturbance range Δmax; Generate an initial population of N individuals, where the first n individuals inherit the local SVR model parameters of each node device, and the model parameters of subsequent individuals are generated by the population calculation formula, which is: ∈∈[0.01×P base ,0.1×P rated ] in, are the model parameters for subsequent individuals; is the parameter vector of the ith % nth device selected cyclically from the parameter set of n node devices; i%n is the subsequent individual index i modulo the number of devices n; u(-1,1) is a uniformly distributed random vector; C,∈,γ is the constraint parameter range; V min is the minimum voltage safety value; V max is the maximum voltage safety value; P base is the reference power; P rated is the rated power of the equipment; σ min is the minimum standard deviation of the kernel function parameter set; σ max is the maximum standard deviation of the kernel function parameter set.

5. The edge cluster resource monitoring and prediction method based on cloud-edge collaboration according to claim 4 is characterized by: The parameter disturbance range Δmax is determined as follows: When L<0.8L max When Δmax=0.1×(2-sin(π×L / L max )); When L≥0.8L max hour, 6. The edge cluster resource monitoring and prediction method based on cloud-edge collaboration according to claim 5, characterized in that: The calculating of the fitness of each individual includes calculating a prediction error mean of the SVR model based on a data set collected by each node device; Generate a load rate curve based on the node load change rate within a preset time window; Generate safety penalty items based on the extent to which the node load rate deviates from the safety interval and the proportion of line load rate exceeding the limit; The prediction error mean, the load rate fluctuation penalty term, and the safety penalty term are weightedly summed to generate a fitness value.

7. The edge cluster resource monitoring and prediction method based on cloud-edge collaboration according to claim 6 is characterized by: The calculation formula of the current energy is expressed as, Among them, E (t) is the individual energy value of the tth iteration; is the current parameter vector; is the optimal parameter vector of the previous iteration; is the current load ratio.

8. A cloud-edge collaboration-based edge cluster resource monitoring and prediction system, applying the cloud-edge collaboration-based edge cluster resource monitoring and prediction method according to any one of claims 1 to 7, characterized in that: It includes data collection module, feature extraction module, model construction module, model parameter set generation module, parameter optimization module and resource scheduling module; The data collection module collects data used to predict edge cluster operation and performs data preprocessing; The feature extraction module is used to perform feature extraction operations on the pre-processed data based on timestamps, physical device configurations and historical node operation data to construct a data set; The model building module is used to build an SVR model in the cloud center to predict the operation status of the edge cluster; The model parameter set generation module is used to train the SVR model using the data set collected by each node device to generate a model parameter set of the SVR model; The parameter optimization module is used in the cloud center to perform global optimization on the model parameter set using the Golden Jackal optimization algorithm, find the optimal model parameter configuration of the SVR model, and generate the SVR model with the optimal model parameter configuration; The resource scheduling module is used to predict the operating status of the edge cluster based on the real-time collected data and the SVR model configured with the optimal model parameters, and to formulate a scheduling strategy based on the prediction results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for monitoring and predicting edge cluster resources based on cloud-edge collaboration according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for monitoring and predicting edge cluster resources based on cloud-edge collaboration according to any one of claims 1 to 7 are implemented.

Images