Dynamic resource allocation method in edge computing environment

Abstract

The invention relates to the technical field of edge computing, and discloses a dynamic resource allocation method in an edge computing environment. The method comprises the steps of collecting a node historical data set, training a load prediction model based on the data set, adjusting a resource allocation strategy by using a stochastic optimization algorithm, and optimizing a task unloading and resource sharing strategy in combination with a game theory model. Resource scheduling optimization in a high-load scene is realized, and task real-time performance and system efficiency are ensured. The invention further provides a dynamic resource allocation system in the edge computing environment. The system comprises a resource state model generation module, a load prediction module, an optimization control module, a game decision module, a node cooperation module and a stability analysis module. By adopting Markov process model modeling, LSTM neural network load prediction, game theory optimization resource scheduling and Lyapunov stability theory, load change prediction, resource allocation optimization and task unloading are realized, stable operation of the system under high load is ensured, resource utilization efficiency is improved, and real-time task requirements are met.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and specifically to a dynamic resource allocation method in an edge computing environment. Background Art

[0002] With the rapid development of the Internet of Things and big data, edge computing, as an emerging computing architecture, has gradually become an important means of data processing. By sinking computing tasks to edge nodes close to data sources, edge computing can significantly reduce data transmission delays and improve system response speed, which is especially 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 problem that needs to be solved urgently.

[0003] Most current resource scheduling technologies rely on static resource allocation methods and are difficult to adapt to load changes in real time. This method usually fails to take into account load fluctuations between nodes and cannot predict future load changes, resulting in uneven or delayed resource allocation and affecting the overall efficiency of the system. Especially when load changes are more drastic, the resources of the node may not be able to meet the needs of high-priority tasks in a timely manner, leading to task timeouts or waste of resources.

[0004] In addition, most of the existing load prediction methods rely on simple linear regression models or static rules, which cannot effectively process complex time series data; especially for the prediction of high-load scenarios, traditional methods have large errors; due to the failure to accurately predict future loads, the system can only rely on experience for resource scheduling, which not only affects the timeliness of tasks, but also causes performance bottlenecks in the system; moreover, existing scheduling algorithms usually do not consider the flexibility of task offloading and node collaboration, resulting in the system being unable to effectively share tasks when node resources are tight, causing problems such as uneven load and performance waste.

[0005] Existing technologies cannot effectively respond to the dynamic resource allocation needs in edge computing environments, especially in scenarios with load fluctuations and high task real-time requirements; traditional resource allocation methods fail to fully utilize prediction models and flexible scheduling strategies, and lack real-time adjustment and optimization mechanisms; therefore, the present invention proposes a dynamic resource allocation method in an edge computing environment to address the shortcomings of the existing technology. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a dynamic resource allocation method in an edge computing environment, which solves the problem that traditional static resource allocation cannot cope with load fluctuations and uneven resource scheduling and inefficiency in scenarios with high task real-time requirements.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A dynamic resource allocation method in an edge computing environment, comprising the following steps: S1. Collect the historical data set of edge computing nodes, where the historical data set includes data on the computing resource usage, storage utilization rate, and bandwidth occupancy of the nodes; S2. Based on the historical data set, predict the future load change trend of edge nodes by training a load prediction model, and identify possible high-load scenarios; S3. Based on the load prediction model, use a stochastic optimization algorithm to adjust the resource allocation strategy of each edge node to ensure that resources are preferentially allocated to tasks with the highest real-time requirements in high-load scenarios; S4. Based on the resource allocation strategy, use a game theory model to determine the task offloading and resource sharing strategies among nodes based on the current resource load status of the nodes, thereby achieving the optimization of the resource scheduling of the entire system.

[0008] Preferably, the historical data set further includes the computing resource usage data, storage occupancy data, and bandwidth usage data of edge computing nodes, and is organized in the form of a time series.

[0009] Preferably, the load prediction model is trained using an LSTM neural network model based on the historical data set, and predicts the future load change trend of edge nodes.

[0010] Preferably, the stochastic optimization algorithm optimizes the resource allocation strategy of edge nodes by minimizing the weighted sum between task delay and maximizing resource utility, and the optimization objective is to achieve optimal resource scheduling within a given time.

[0011] Preferably, the optimization objective of the stochastic optimization algorithm is to minimize the delay of system tasks and maximize the utilization efficiency of resources, and the optimization problem is: ; where, is a weight factor, representing the priority coefficient of task delay, and is used to adjust the trade-off between delay and resource utility; is a weight factor, representing the priority coefficient of resource utility, and is used to adjust the trade-off between resource utility and task delay; is the number of nodes in the edge computing system; represents the edge node at time the task delay; represents the edge node at time the resource utility; is a time variable, representing the time dimension of node load and task processing.

[0012] Preferably, the game theory model evaluates the contributions of task offloading and resource sharing between nodes by calculating the Shapley value, ensuring fairness and optimality in resource sharing among nodes.

[0013] Preferably, the optimization of resource scheduling includes comprehensively evaluating the resource utilization efficiency, task priorities, and latencies of all nodes to optimize the overall resource utilization rate and task execution performance of the system.

[0014] Preferably, the stochastic optimization algorithm combines the load prediction results and the current system state to dynamically adjust the resource allocation strategy of edge nodes.

[0015] Preferably, the game theory model further considers the computing capabilities, resource idle conditions, and task urgencies among nodes to optimize task offloading and resource sharing decisions.

[0016] The present invention also provides a dynamic resource allocation system in an edge computing environment, including: A resource status model generation module for collecting historical data based on the resource load status of edge computing nodes and constructing a status model of the system, and using a Markov process model to model the changes in resource status; A load prediction module for training a load prediction model based on a historical data set and predicting the future load change trend; An optimization control module for adjusting the resource allocation strategy of each edge node through a stochastic optimization algorithm based on the load prediction model and resource scheduling objectives; A game decision module for determining the task offloading and resource sharing strategies between nodes based on the game theory model according to the resource load status of nodes, thereby achieving the optimization of resource scheduling for the entire system; A node cooperation module for triggering a task offloading mechanism when a node is overloaded and determining the target node for task offloading based on the Shapley value algorithm of game theory; A stability analysis module for analyzing and optimizing the resource scheduling strategy based on the Lyapunov stability theory to ensure the stable operation of the system in a high-load environment and meet the real-time requirements of tasks.

[0017] The present invention provides a dynamic resource allocation method in an edge computing environment. It has the following beneficial effects: 1. The present invention adopts a resource status modeling technology based on a Markov process model, achieving the effect of efficiently capturing the dynamic changes in the resource load of edge computing nodes; compared with the traditional static load model in the prior art, the present invention flexibly models the load status of nodes by using a Markov process model, solving the deficiency that traditional methods are difficult to cope with dynamically changing loads; this method can more accurately predict the future load changes of nodes, providing a more accurate decision-making basis for subsequent optimization control.

[0018] 2. The present invention uses an LSTM neural network for load prediction, achieving a high-accuracy load trend prediction effect. Compared with the simple linear prediction method in the prior art, by introducing the LSTM network in deep learning, the present invention can better handle the long-term dependence of time series data, avoiding the prediction errors of traditional methods when the load fluctuates greatly. This enables the system to more accurately predict the node resource requirements when the load suddenly changes or varies, thereby optimizing task scheduling.

[0019] 3. The present invention optimizes resource scheduling and task offloading strategies through game theory, achieving a fair and efficient resource allocation effect. Different from the single resource allocation algorithm in the prior art, the present invention combines the Shapley value algorithm in game theory to ensure the fairness of task offloading and resource sharing decisions, avoiding the situation of excessive resource concentration or uneven distribution. This approach not only improves the utilization rate of resources but also reduces the risk of load imbalance between nodes.

[0020] 4. The present invention optimizes the resource scheduling strategy using Lyapunov stability theory, achieving the effect of stable system operation in a high-load environment. Compared with the solutions in the prior art that do not consider system stability, by introducing Lyapunov stability theory, the present invention ensures that in a high-load situation, the system can maintain stable resource scheduling and handle load fluctuations. This optimization effectively avoids the risks of system crashes or task delays, ensuring that the real-time task requirements in the edge computing environment are met. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the flowchart of the method of the present invention; Figure 2 is the system architecture diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] Please refer to Figure 1 , the embodiments of the present invention provide a dynamic resource allocation method in an edge computing environment, including the following steps: S1. Collect the historical data set of edge computing nodes, where the historical data set includes data on the computing resource usage, storage utilization, and bandwidth occupancy of the nodes; In the present invention, step S1 is the first step for dynamic resource allocation in the edge computing environment. Its main task is to collect historical data of each edge computing node; this data is crucial for subsequent load prediction and optimization control; by analyzing the historical data set, the system can understand the resource load status, task execution situation, and network bandwidth usage of each node; accurate historical data provides the necessary information for decisions such as subsequent load prediction, resource optimization, and task scheduling.

[0024] In this embodiment, the historical data set of the edge computing node includes three main resource usage data: computing resource usage data, storage utilization data, and bandwidth occupancy data; specifically, these data items cover the following aspects: Computing resource usage data: It includes the CPU and GPU usage rates of each node, indicating the actual usage of the computing power of the node within a certain period of time; Through these data, it can be understood whether the node is in an overloaded state or whether there is sufficient computing resource to process the current task; Storage utilization data: Records the disk storage usage of each node, including the occupancy rate and free space of the storage device; Storing data is crucial for determining whether the node can store new tasks or data; Bandwidth occupancy data: Represents the network bandwidth usage of the node, mainly recording the bandwidth requirements of the node when transmitting data; The effective use of bandwidth can affect task offloading and data synchronization between nodes, so it is particularly important to monitor the network bandwidth usage.

[0025] Generally, these data will be organized in the form of a time series and collected at a certain time step; Time series data can reflect the historical load changes of the node, and through the analysis of these data, the system can identify the rules of load changes and the performance bottlenecks of the node; In a possible implementation, data collection is performed through real-time monitoring tools; These tools can automatically collect the status data of the nodes during the entire system operation, including information such as the CPU load, memory usage, storage occupancy, and bandwidth occupancy of each node; The data collected through these tools can be directly uploaded to the data storage system to ensure the reliability and consistency of the data.

[0026] First, for each edge computing node, the system will conduct a comprehensive monitoring of it; Specifically, the resource data of the edge node will be regularly sampled through an integrated system monitoring tool; The monitoring data includes: information such as the CPU load, memory usage rate, storage space utilization rate, and bandwidth occupancy of the node; The monitoring tool can reflect the status of the node in real time to ensure that when the load changes, the system can timely capture the resource consumption situation.

[0027] As an option, the time interval for data collection can be adjusted according to actual requirements; for example, in scenarios with large load fluctuations, the system may choose a shorter sampling time interval (such as once per second); while in cases where the load changes relatively smoothly, the sampling time interval can be appropriately increased (such as once per minute); this flexible sampling strategy helps the system make precise adjustments according to the actual load situation.

[0028] Specifically, within each time step, the system will record the resource usage data of each node and store this data in the historical dataset; in this way, the system can establish a historical load model for each node, providing the input required for subsequent prediction tasks.

[0029] To ensure the quality and accuracy of the data, data preprocessing is an important part of this embodiment; since during long-term operation, the monitoring data may contain noise or outliers, the data must be cleaned and processed; generally, preprocessing includes steps such as data denoising, outlier detection, and missing value filling.

[0030] For example, using smoothing techniques to remove sudden fluctuations in historical data, or filling missing data points through interpolation; through these techniques, the system can ensure that the historical data used for subsequent load prediction is of high quality and reliable.

[0031] The collected historical dataset provides the basis for the subsequent load prediction model; the load prediction model will be trained based on these historical datasets to predict the load situation of edge computing nodes in the future for a period of time; in this embodiment, the load prediction model uses an LSTM neural network (Long Short-Term Memory network), which can effectively capture long-term dependencies in time series data and is particularly suitable for tasks such as load prediction.

[0032] Specifically, by training the LSTM neural network model, the system can learn trends such as the periodic changes in node load and sudden increases in load; the LSTM model is trained using historical load data to obtain a predictor that can predict future load changes.

[0033] For the training of the load prediction model, assume the historical load dataset is , representing the load of node at time , and the training formula of the LSTM neural network is: ; Among them, is the input dataset of node at time (including historical load data, resource usage, etc.), are the parameters (such as weights and biases) of the LSTM model, is the output function of the LSTM network, which maps the input data to the future load prediction results of the nodes.

[0034] All historical data will be stored in the central data storage system; as an option, a distributed storage scheme can be adopted to store and manage this data, ensuring the reliability and high availability of the data in a multi-node environment; to ensure the security and consistency of the data, the system can use distributed database technologies (such as Hadoop, Cassandra, etc.) for storage and achieve efficient query and access to the data.

[0035] In a possible implementation, by real-time updating and backing up the historical data, the system can ensure the persistence and reliability of data storage; the system will also perform appropriate compression and storage optimization according to the usage frequency of the data to reduce the occupancy of storage space.

[0036] S2. Based on the historical data set, by training the load prediction model, predict the future load change trend of the edge nodes and identify possible high-load scenarios; In the edge computing environment, one of the key tasks of resource scheduling is to predict the change trend of node load in advance, especially in high-load scenarios, to ensure that the system can respond in a timely manner and adjust resource allocation; to achieve this goal, step S2 uses the historical data set to predict the future load change of the edge nodes by training the load prediction model; the core of this step is to model the time series data through the LSTM neural network model, accurately predict the change trend of future load, and identify possible high-load scenarios.

[0037] In this embodiment, based on the historical data set collected in step S1, we predict the node load by training the load prediction model; the historical data set includes information such as the computing resource usage, storage occupancy, and bandwidth occupancy of each node at different time points; this data is time series data and has time dependence, that is, the current load situation is closely related to the past load status; to fully exploit this time dependence, the LSTM neural network (long short-term memory network) is selected as the load prediction model.

[0038] Generally, LSTM is a neural network model particularly suitable for processing time series data. It avoids the problem of long-term dependence by introducing a gating mechanism and can effectively capture long-term and short-term dependence relationships; Therefore, LSTM is used in the present invention to capture the time patterns of node load for accurate prediction of future load.

[0039] Specifically, before training the LSTM model, historical data needs to be preprocessed; historical data usually contains various resource usage data of nodes, and these data will first be standardized (such as normalization or Z-score normalization) so that the LSTM model can be trained more efficiently.

[0040] To train the LSTM model, the system divides the historical dataset into multiple time periods, usually using the sliding window method, and uses a fixed-size time window (such as the past 7 days, the past 30 hours, etc.) for prediction; for example, when predicting the load of a node at a future time, the system will use the historical data within the window (i.e., the load data of the past several times) to train the model.

[0041] As an option, to improve the accuracy of load prediction, the system can select time windows of different lengths, and the specific window length can be adjusted according to the load volatility of the node; if the node load fluctuates greatly, a shorter window (such as the hourly level) is selected; if the node load changes smoothly, a longer time window (such as the daily level) can be selected.

[0042] In this embodiment, the input layer of the LSTM network receives time series data, and the output layer predicts future load values; assuming that we want to predict the load of a node at time , then the training process of the LSTM model is as follows: ; ; where, represents the load prediction of node at time ; is the historical load data of node at the past time, which is input into the LSTM network; is the prediction function of the LSTM network, indicating the predicted value obtained by the model through the training process; are the parameters of the LSTM model, including weights and bias terms, which are continuously adjusted during the training process through the backpropagation algorithm to minimize the prediction error.

[0043] As an option, the model uses the mean squared error (MSE) as the loss function for optimization during training. The calculation method of this loss function is: ; where, is the loss value, is the number of training samples, is the node The true load value; is the load value predicted by the LSTM model; The training process adjusts the model parameters by minimizing this loss function.

[0044] After training is completed, the LSTM model is used to predict the load of the node for a period of time in the future; Specifically, when the system needs to predict the load of the node at a future time the model inputs the historical data within the most recent period of time and outputs the predicted load value; This process relies on the load change pattern learned by the LSTM model during the training phase.

[0045] The prediction result can help the system identify future high-load scenarios; Generally, the predicted value is compared with a preset high-load threshold; When the predicted load value exceeds this threshold, the system marks it as a possible high-load scenario.

[0046] In one possible implementation, the threshold for high-load scenarios can be dynamically adjusted according to the resource type of the node, the nature of the task, and the real-time performance requirements of the system; For example, for storage-intensive tasks, the system may set the threshold according to the storage load; For compute-intensive tasks, it will be set according to the usage of the CPU and GPU.

[0047] As an option, the system calibrates high-load scenarios by defining a load threshold; Specifically, if the load prediction value of the node exceeds this threshold, the system will consider that the node has entered a high-load state; In a high-load state, the resources of the node may be insufficient to meet the task requirements, so the system needs to perform resource scheduling in advance.

[0048] For example, if the node the predicted load values of the computing resources (CPU, GPU) and bandwidth exceed the set thresholds respectively, the system will consider offloading some tasks to nodes with lower load, or allocating more resources to this node.

[0049] In some embodiments, in order to further improve the accuracy of load prediction, the system can also retrain the LSTM model regularly; As time goes by, the workload pattern of the node may change, especially in the case of node addition, task load change, or system expansion, the original load prediction model may no longer adapt to the new load pattern; Therefore, regularly retraining the load prediction model can ensure that the system still maintains good prediction ability in a changing environment.

[0050] As an option, to adapt to the characteristics of different node loads, the system can also establish a separate load prediction model for each node; in this way, the LSTM model can better capture the load patterns of each node and improve the prediction accuracy.

[0051] S3. Based on the load prediction model, use a stochastic optimization algorithm to adjust the resource allocation strategy of each edge node to ensure that resources are preferentially allocated to the tasks with the highest real-time requirements in high-load scenarios; In the foregoing step S2, the load prediction model of the edge computing node predicts the future load change trend of each node based on the historical data set and identifies possible high-load scenarios; through this step, the system can obtain effective information about the future node load to provide guidance for subsequent resource scheduling; the goal of step S3 is to use the load prediction results and the stochastic optimization algorithm to dynamically adjust the resource allocation strategy of each node to ensure that the system can preferentially allocate resources to tasks with relatively high real-time requirements in high-load scenarios and maximize the resource utility of the system.

[0052] In this embodiment, the stochastic optimization algorithm will adjust the resource allocation strategy of each edge node in an iterative manner based on the load prediction data obtained in the foregoing step S2; specifically, the system will optimize the resource allocation plan according to the load prediction results of each node and the real-time requirements of the tasks to ensure that high-priority tasks can obtain computing resources and bandwidth in a timely manner and optimize the overall resource utilization efficiency.

[0053] In this step, the optimization goal of the stochastic optimization algorithm is to minimize the weighted sum between task delay and maximize resource utility; specifically, the optimization objective function is: ; where is the weight factor, representing the priority coefficient of task delay, used to adjust the trade-off between delay and resource utility; is the weight factor, representing the priority coefficient of resource utility, used to adjust the trade-off between resource utility and task delay; is the number of nodes in the edge computing system; represents the edge node at time the task delay; represents the edge node at time the resource utility; is the time variable, representing the time dimension of node load and task processing.

[0054] Specifically, the goal of the optimization algorithm is to minimize the task delay as much as possible, while maximizing the resource utility ; By adjusting the weight factor and , the system can flexibly adjust the priority according to different application scenarios; in high-load scenarios, the weight of may be increased to prioritize optimizing the latency of tasks; while in low-load situations, the system may be more concerned with maximizing resource utilization and increase the weight of .

[0055] In this embodiment, the implementation of the random optimization algorithm adopts an iterative optimization method; in each iteration, the system dynamically adjusts the resource allocation strategy according to the current node load situation and task requirements; the optimization process includes the following steps: Initial allocation: Based on the initial load prediction data, the system allocates a certain amount of computing resources, storage resources, and bandwidth resources to each edge node; the initial resource allocation can be carried out through simple rules, such as performing basic resource allocation according to the current load situation and task priority of the node.

[0056] Target calculation: In each iteration, the system calculates the task latency and resource utility of each node under the current resource allocation, and calculates the value of the optimization objective function; the value of the objective function is the evaluation index of the current resource allocation strategy.

[0057] Adjustment strategy: According to the calculation result of the optimization objective function, the system adjusts the resource allocation strategy of each node; through the iterative process of the algorithm, the resource allocation strategy will gradually approach the optimal strategy until the optimization objective function converges.

[0058] Convergence determination: The algorithm calculates the change of the optimization objective function after each iteration; if the change of the objective function is less than the preset threshold in consecutive multiple iterations, it is considered that the optimization process converges, stops the optimization, and outputs the final resource allocation strategy.

[0059] Generally, the optimization process will perform multiple iterations to find the optimal resource allocation scheme; to improve the calculation efficiency, parallel computing or distributed optimization methods can be adopted to handle the resource scheduling problem of multiple nodes.

[0060] Specifically, the optimization algorithm adjusts the priority according to the real-time requirements of each task and the resource load situation of the node; the task priority is divided according to the real-time requirements of the task, and tasks with higher real-time requirements will be given priority to obtain computing resources and bandwidth; for example, in the autonomous driving scenario, the real-time requirements of safety-related tasks are usually high, so these tasks will be scheduled first.

[0061] As an option, the task priorities of the nodes can be further optimized through a game theory model; the game theory model takes into account the competition relationships among tasks and determines the decisions on task offloading and resource sharing by calculating the contribution of each task to the system resources; such decisions can help the system perform resource scheduling more intelligently, reduce resource waste, and ensure that the real-time requirements of critical tasks are met.

[0062] In a possible implementation, under high load conditions, the system determines whether certain tasks need to be offloaded from nodes with heavy loads to other nodes based on the results of the load prediction model and the optimization algorithm; the task offloading process is optimized according to the Shapley value in the game theory model to ensure that the task offloading decision can balance the loads among nodes to the greatest extent and improve the resource utilization efficiency of the system.

[0063] For example, if the load of a certain node exceeds a predetermined threshold, the system determines through the game theory model whether a part of the tasks of this node can be offloaded to other nodes with lighter loads; the offloaded tasks are allocated to the target nodes according to their priorities and resource requirements; the offloading decision calculates the Shapley value to ensure the fairness and optimality of resource sharing and avoid a certain node undertaking too many tasks.

[0064] In some embodiments, the system can dynamically adjust the parameters in the optimization algorithm according to different application scenarios, such as adjusting the weight factor and values to adapt to different load conditions; for example, when processing computationally intensive tasks, the system may reduce the weight of latency and pay more attention to maximizing resource utility; while for tasks that require low latency, the system will increase the weight of latency to ensure that the tasks can be completed as soon as possible.

[0065] In a possible implementation, different resource scheduling priorities can also be set for different nodes according to factors such as the computing power, storage capacity, and bandwidth of the nodes; for example, under high load conditions, more resources are preferentially allocated to nodes with stronger computing power, while for nodes with weaker computing power, the resource allocation is optimized through task offloading and other means.

[0066] S4. Based on the resource allocation strategy, use the game theory model to determine the task offloading and resource sharing strategy among nodes based on the current resource load status of the nodes, so as to achieve the optimization of the resource scheduling of the entire system; In the aforementioned step S3, the system dynamically adjusts the resource allocation strategy of each edge node through a load prediction model and a stochastic optimization algorithm to ensure that resources are preferentially allocated to tasks with higher real-time requirements in high-load scenarios; through this step, the system allocates resources to each node and performs preliminary resource scheduling based on the predicted load and task priorities; however, resource allocation is not a static process but a dynamic process that requires continuous adjustment. Especially in high-load situations, the system may face resource bottlenecks; to better address this challenge, step S4 uses a game theory model to optimize the resource scheduling of the entire system through task offloading and resource sharing strategies based on the current resource load status of the nodes.

[0067] In this embodiment, the core of step S4 is to evaluate and determine the task offloading and resource sharing strategies between nodes through a game theory model based on the current resource load status of each node; the introduction of the game theory model enables the system to perform efficient resource allocation and load balancing in a multi-node environment, ensuring optimal resource scheduling in high-load scenarios and that tasks with higher real-time requirements can be preferentially processed.

[0068] Generally, the resources of each node in an edge computing system are limited, and the tasks in the system have different real-time requirements and resource demands; when the load of a certain node is too high, it may not be able to continue processing new tasks. At this time, task offloading becomes an effective means to solve this problem; the game theory model mainly determines the optimal offloading and allocation decisions by evaluating the benefits of resource sharing and task offloading between nodes in this step.

[0069] As an option, the game theory model uses the Shapley value for the optimal decision of task offloading; the Shapley value is a way of fairly allocating resources and can ensure the reasonable allocation of tasks and resources according to the contribution of each node to the overall system benefit in a multi-party game; through this method, the system can ensure the fairness of task offloading and resource allocation while maximizing the resource utilization efficiency of the entire system.

[0070] Specifically, assume there are nodes in the system, and the resource load status of node is . At a certain time point , the resource usage of node may cause the node to be overloaded or the task cannot be processed in a timely manner; to solve this problem, the game theory model evaluates the task offloading decision by calculating the Shapley value of each node.

[0071] The formula for the Shapley value is as follows: ; Among them, represents node The Shapley value in the cooperation set represents the contribution of a node to the overall system benefit; denotes the number of nodes in the cooperation set; denotes all permutations of the cooperation set i.e., all possible permutations and combinations of nodes; represents the benefit when all nodes in the cooperation set cooperate together, the total resource benefit; represents the benefit of the cooperation set after removing node i.e., the total resource benefit of the remaining nodes after removing node

[0072] By calculating the Shapley value, the system can quantify the contribution of each node to the overall system resource benefit and determine whether to offload tasks to that node based on this; in the offloading decision, the system calculates the Shapley value of node to determine whether the node should undertake additional tasks, thus achieving load balancing of the system.

[0073] In this embodiment, when the load of a certain node reaches a preset high load threshold, the system will decide whether to offload the tasks of that node to other nodes through a game theory model; the task offloading decision of the node is based on its contribution value to the system resource benefit; specifically, the system will calculate the Shapley value of each node and judge the offloading priority according to the size of the Shapley value; a node with a larger Shapley value usually indicates stronger resource processing ability, so it may be preferentially selected as the target node for task offloading.

[0074] Specifically, if the load of node exceeds the preset threshold, the system will consider offloading tasks to a node with a lighter load and a larger Shapley value; this decision takes into account not only the current load of the node but also multiple factors such as the computing power, storage capacity, and bandwidth of the node.

[0075] In some embodiments, in addition to solving the task offloading problem, the game theory model also involves resource sharing decisions; in high load scenarios, resource sharing among nodes can effectively improve resource utilization and avoid resource idleness and task delay problems; through resource sharing, multiple nodes with lighter loads can cooperate to share the tasks of overloaded nodes, thus achieving load balancing of the entire system.

[0076] As an option, the decision of resource sharing is also optimized through the Shapley value; by calculating the Shapley value, the system can determine which nodes can share resources and which nodes have a greater resource contribution and should be given priority for resource sharing; through the game theory model, the system can fairly and effectively share resources such as computing power, storage, and bandwidth among nodes to ensure the real-time completion of tasks.

[0077] For example, if the resource load of node is relatively high, while the resource load of node is relatively low, and node has strong computing power and bandwidth resources, the system may decide to offload the task to node through the game theory model, while sharing the resources of node to reduce the load on node and avoid resource bottlenecks in the system.

[0078] In a possible implementation, to improve the adaptability of the system in different load scenarios, the parameters in the game theory model can be dynamically adjusted; the adjustment of the parameters depends not only on the resource load situation of the nodes but also on factors such as the priority of the task, the resource requirements of the task, and the processing capabilities of the nodes; changes in these factors may lead to changes in the resource sharing strategy among nodes. Therefore, the system needs to adjust the parameters according to the real-time situation to achieve optimal resource scheduling.

[0079] For example, in the case of high task latency or heavy node load, the system may increase the weight of task offloading to ensure that tasks can be offloaded to nodes with lighter loads in a timely manner; while in the case of low load, the system may give priority to resource sharing to improve the overall utilization rate of resources. Please refer to Figure 2 This invention also provides a dynamic resource allocation system in an edge computing environment, including: A resource status model generation module. The task of this module is to collect historical data based on the resource load status of edge computing nodes and construct the status model of the system; the resource status model generation module uses a Markov process model to model the changes in the resource status of the system; the Markov process model is a stochastic process in which the current state of the system depends only on the previous state and is independent of the previous historical states; through the Markov process model, the system can capture the dynamic change law of the node resource load status over time and predict the possible future load conditions; this model provides basic data and theoretical support for subsequent load prediction, resource scheduling, and optimal control. Load prediction module. The load prediction module trains a load prediction model using a historical dataset. By analyzing the past load change trends, it predicts the load change situation of edge computing nodes in a future period of time. This module uses a deep learning model (such as an LSTM neural network) for training to better capture the time dependence and fluctuation patterns of the load. The prediction results will provide a basis for subsequent resource scheduling. Especially in high-load scenarios, it can help the system identify overloaded nodes in advance and optimize resource allocation to ensure the timely completion of tasks. Optimization control module. The optimization control module adjusts the resource allocation strategy of each edge node through a stochastic optimization algorithm based on the load prediction model and resource scheduling goals. The goal of this module is to minimize the task latency of the system and maximize the resource utilization efficiency by optimizing resource configuration. The optimization algorithm dynamically adjusts the resource allocation strategy according to the current load prediction results, task priorities, and resource requirements to ensure that the system can reasonably allocate computing resources, storage, and bandwidth, etc., under high load, and improve the overall performance of the system. Game decision-making module. The game decision-making module determines the task offloading and resource sharing strategies among nodes based on a game theory model and using the current resource load status of the nodes. The game decision-making module makes the optimal task offloading decision by analyzing the resource competition and cooperation relationships among nodes. This module uses algorithms such as the Shapley value in game theory to evaluate the contribution of each node in task offloading and resource sharing, ensuring that system resources are fairly and efficiently allocated among different nodes, thereby achieving the optimization of the overall system resource scheduling. Node cooperation module. The node cooperation module is mainly responsible for triggering the task offloading mechanism when a node is overloaded. This module uses the Shapley value algorithm in game theory to determine the target node for task offloading. The Shapley value algorithm helps the system evaluate the contribution of each node to the overall system resource benefit, ensuring that tasks are offloaded to nodes with lighter loads and stronger computing capabilities. Through the intelligent task offloading mechanism, the system can optimize resource allocation, avoid node overload, and ensure that high-priority tasks can be processed in a timely manner. Stability analysis module. The stability analysis module analyzes and optimizes the resource scheduling strategy through Lyapunov stability theory to ensure the stable operation of the system in a high-load environment and meet the real-time requirements of tasks. Lyapunov stability theory is a mathematical method used to analyze the stability of the system and the balance of system states. The role of this module is to ensure that the system can adapt to load fluctuations and avoid the occurrence of over-concentration of resources or resource bottlenecks by optimizing the resource scheduling strategy. By adjusting resource allocation in real time, the stability analysis module can ensure the stability of the system during drastic load changes, thereby avoiding task latency or loss and ensuring efficient operation.

[0080] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic resource allocation method in an edge computing environment, characterized in that: The following steps are involved: S1. Collect historical data sets of edge computing nodes, where the historical data sets include computing resource usage, storage utilization, and bandwidth occupancy data of the nodes; S2. Based on historical data sets, the load prediction model is trained to predict the future load change trend of edge nodes and identify possible high-load scenarios; S3. Based on the load prediction model, a random optimization algorithm is used to adjust the resource allocation strategy of each edge node to ensure that resources are allocated preferentially to tasks with the highest real-time requirements in high-load scenarios. S4. Based on the resource allocation strategy, the game theory model is used to determine the task offloading and resource sharing strategy between nodes based on the current resource load status of the nodes, so as to achieve optimal resource scheduling of the entire system.

2. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The historical data set also includes computing resource usage data, storage occupancy data, and bandwidth usage data of edge computing nodes, and is organized in the form of a time series.

3. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The load prediction model is trained using an LSTM neural network model based on historical data sets, and predicts future load change trends of edge nodes.

4. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The random optimization algorithm optimizes the edge node resource allocation strategy by minimizing the weighted sum between task delay and maximizing resource utility, and the optimization goal is to achieve optimal resource scheduling within a given time.

5. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The optimization goal of the random optimization algorithm is to minimize the delay of system tasks and maximize the utilization efficiency of resources. The optimization problem is: ; in, is the weight factor, which represents the priority coefficient of task delay and is used to adjust the trade-off between delay and resource utility; is the weight factor, which represents the priority coefficient of resource utility and is used to adjust the trade-off between resource utility and task delay; is the number of nodes in the edge computing system; Represents an edge node In time Task delays Represents an edge node In time Resource utility at the time of is a time variable, which represents the time dimension of node load and task processing.

6. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The game theory model evaluates the contribution of task offloading and resource sharing between nodes by calculating the Shapley value, ensuring that resource sharing between nodes is fair and optimal.

7. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The resource scheduling optimization includes a comprehensive evaluation of resource utilization efficiency, task priority and delay of all nodes to optimize the overall resource utilization of the system and task execution performance.

8. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The random optimization algorithm combines the load prediction results and the current system status to dynamically adjust the resource allocation strategy of the edge nodes.

9. The method for dynamic resource allocation in an edge computing environment according to claim 1, characterized in that: The game theory model further considers the computing power, resource availability and task urgency between nodes to optimize task offloading and resource sharing decisions.

10. A dynamic resource allocation system in an edge computing environment, applied to the dynamic resource allocation method in an edge computing environment as described in any one of claims 1 to 9, characterized in that: include: The resource status model generation module is used to collect historical data and build a system status model based on the resource load status of the edge computing node, and use the Markov process model to model the resource status change; The load prediction module is used to train the load prediction model based on the historical data set and predict the future load change trend; The optimization control module is used to adjust the resource allocation strategy of each edge node through a random optimization algorithm based on the load prediction model and resource scheduling objectives; The game decision module is used to determine the task offloading and resource sharing strategies between nodes based on the game theory model and the resource load status of the nodes, thereby achieving optimal resource scheduling for the entire system; The node collaboration module is used to trigger the task offloading mechanism when the node is overloaded, and determine the target node for task offloading based on the Shapley value algorithm of game theory; The stability analysis module is used to analyze and optimize resource scheduling strategies based on Lyapunov stability theory to ensure that the system runs stably under high-load environments and meets the real-time requirements of tasks.

Images