Resource scheduling method and device, equipment, storage medium and program product
By employing multi-model collaboration and dynamic scheduling mechanisms, combined with reinforcement learning and coordinated communication, the flexibility and efficiency issues of edge computing resource scheduling are resolved, thereby improving the system's service quality and user experience.
Patent Information
- Application Number
- CN202511585990.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-31
- Publication Date
- 2026-02-10
AI Technical Summary
Traditional edge computing resource scheduling technologies are ill-suited to different tasks and their processing metrics, resulting in a decline in overall service quality and a poor user experience.
A multi-model collaborative approach is adopted, combining a pre-trained large language model with a lightweight inference model, and integrating dynamic perception and adaptive scheduling mechanisms. Through reinforcement learning and a coordination communication module, flexible resource scheduling and efficient processing are achieved.
It improves the flexibility and efficiency of resource scheduling, ensures overall service quality, and enhances user experience, especially in terms of system stability when handling unstructured data and migrating real-time tasks.
Smart Images
Figure CN121501484A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of edge computing, and particularly relates to a resource scheduling method and device, equipment, a storage medium and a program product. BACKGROUND
[0002] With the rapid development of Internet of Things and mobile communication technology, edge computing has become one of the key technologies to improve data processing speed and reduce network delay. Edge computing deploys computing resources on edge nodes, so that data can be processed close to the data source, thereby significantly improving response speed and service quality. However, how to efficiently manage and schedule these distributed resources to ensure that they can meet the changing application requirements has become an important research topic.
[0003] Traditional edge computing resource scheduling techniques mostly rely on a single type of model or static capability percentage for resource allocation, which is difficult to adapt to different tasks and their processing indicators, thereby affecting the overall service quality and reducing user experience. SUMMARY
[0004] The present application aims to at least solve one of the technical problems in the related art. To this end, the present application provides a resource scheduling method, device, equipment, storage medium and program product to solve the problem that traditional edge computing resource scheduling techniques are difficult to adapt to different tasks and their processing indicators, thereby ensuring overall service quality and improving user experience.
[0005] According to the resource scheduling method of the first aspect of the present application, the method comprises: determining at least one target model from more than one type of artificial intelligence model based on task information of a to-be-processed task; wherein different types of artificial intelligence models are used to process tasks with different types and indicators; determining a resource scheduling strategy for the to-be-processed task based on each type of target model; determining a target node from each edge computing node based on the resource scheduling strategy, real-time state information of at least one edge computing node and predicted state information of each edge computing node; wherein the predicted state information of the edge computing node is obtained based on historical state information; issuing the to-be-processed task to the target node.
[0006] According to one embodiment of the present application, the method of determining at least one target model from more than one type of artificial intelligence model based on task information of a to-be-processed task comprises: determining at least one target model from more than one type of artificial intelligence model based on task type and task indicator in the task information of the to-be-processed task through a gating network.
[0007] According to an embodiment of the present application, the target node is determined from the edge computing nodes based on the resource scheduling strategy, real-time state information of at least one edge computing node, and predicted state information of each edge computing node. The learning result is obtained by reinforcement learning based on the resource scheduling strategy, real-time state information of at least one edge computing node, and predicted state information of each edge computing node. The target node is determined from the edge computing nodes based on the learning result.
[0008] According to an embodiment of the present application, the edge computing node comprises a coordination communication module; and the edge computing nodes perform cross-node cooperation through the coordination communication module.
[0009] According to an embodiment of the present application, the edge computing node comprises a preset model, and the preset model is a neural network model for processing unstructured data.
[0010] According to an embodiment of the present application, the unstructured data comprises at least one of image data and video stream data.
[0011] The resource scheduling device according to the second aspect of the present application comprises: The first determination module is configured to determine at least one target model from more than one type of artificial intelligence model based on task information of a to-be-processed task; different types of artificial intelligence models are used to process tasks with different types and indexes. The second determination module is configured to determine a resource scheduling strategy of the to-be-processed task based on each type of target model. The third determination module is configured to determine a target node from the edge computing nodes based on the resource scheduling strategy, real-time state information of at least one edge computing node, and predicted state information of each edge computing node; the predicted state information of the edge computing node is obtained based on historical state information. The scheduling module is configured to assign the to-be-processed task to the target node.
[0012] The electronic device according to the third aspect of the present application comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the resource scheduling method according to any of the above embodiments when executing the computer program.
[0013] The storage medium according to the fourth aspect of the present application is a non-transitory computer readable storage medium, and a computer program is stored on the storage medium, and the computer program is executed by a processor to implement the resource scheduling method according to any of the above embodiments.
[0014] The computer program product provided in the fifth aspect of the embodiments of the present application comprises a computer program, and the computer program is executed by a processor to implement the resource scheduling method according to any one of the above.
[0015] The one or more technical solutions described above in the embodiments of the present application have at least the following technical effects: Since at least one target model is determined from more than one type of artificial intelligence model based on the task information of the to-be-processed task, and based on the target models of different types and indicators for processing tasks with different types and indicators, the to-be-processed task with different types and indicators can be satisfied, and thus the resource scheduling strategy of the to-be-processed task can be accurately determined. Furthermore, based on the resource scheduling strategy, the real-time state information and the predicted state information of each edge computing node, the target node can be accurately determined from each edge computing node. Since the real-time state and the predicted future state of each edge computing node are considered when determining the target node, the rationality of resource scheduling can be effectively improved, and the to-be-processed task is then assigned to the target node, which can serve as the optimal edge computing node to process the to-be-processed task, so as to guarantee the overall service quality and improve the user experience.
[0016] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS
[0017] In order to more clearly illustrate the technical solutions of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.
[0018] Figure 1 is a flowchart of the resource scheduling method provided by the embodiments of the present application.
[0019] Figure 2 is a structural diagram of the edge cloud computing resource intelligent scheduling system based on multi-model cooperation provided by the embodiments of the present application.
[0020] Figure 3 is a schematic diagram of the edge side optimization module in the edge cloud computing resource intelligent scheduling system based on multi-model cooperation provided by the embodiments of the present application.
[0021] Figure 4 is a structural diagram of the electronic device provided by the present application. DETAILED DESCRIPTION
[0022] In order to make the purposes, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.
[0023] It should be noted that although the existing edge computing resource scheduling scheme has made certain progress in some aspects, there are still many deficiencies in actual application. These problems mainly manifest in single model limitations, static scheduling mechanisms, insufficient real-time decision support, and lack of support for edge side, etc.: 1. Single model limitations and static scheduling mechanisms. Most existing systems rely on a single type of model or percentage of capabilities for resource allocation, lack flexibility, and are difficult to dynamically adjust according to real-time running state information, affecting overall service quality and user experience. Most scheduling systems use static allocation strategies and fail to fully utilize real-time running state information, resulting in low efficiency when processing complex or large-scale tasks, relying on pre-set capability allocation, and failing to fully consider the low latency and high bandwidth requirements in edge computing environments.
[0024] 2. Insufficient real-time decision support. Existing scheduling methods usually focus on static text analysis and lack sufficient real-time decision support capabilities in dynamic environments. When there is a need for quick response to changing workloads, these methods may not meet the demand, especially in edge computing environments, which require higher real-time and adaptability.
[0025] 3. Lack of edge side implementation details. Existing technologies fail to fully utilize the low latency and high bandwidth advantages of edge computing environments. In distributed deployment situations, the coordination consistency between multiple agents is also a problem that has not been completely solved. Edge devices have limited resources, and developing small models suitable for edge device operation to improve efficiency is a major challenge.
[0026] 4. Limited unstructured data processing capabilities. Existing technologies mainly focus on document processing tasks and have limited processing capabilities for unstructured data such as images and video streams, limiting their applicability in more extensive application scenarios and failing to solve the coordination consistency problem between multiple agents in distributed deployment.
[0027] In order to overcome the deficiencies of the above-mentioned existing technologies, the present application proposes a resource scheduling method, device, equipment, storage medium and program product, aiming to realize more intelligent, flexible and efficient computing resource scheduling by integrating different types of artificial intelligence models and fully utilizing the advantages of edge computing.
[0028] 1. Multi-model collaborative operation. Combining various types of artificial intelligence models (such as pre-trained large language models and lightweight inference models) to handle different types of tasks and their processing metrics. The models can complement each other, improving the system's response speed and accuracy while reducing computation time and resource consumption.
[0029] 2. Dynamic Sensing and Adaptive Scheduling. An adaptive learning algorithm is introduced to predict future load conditions based on historical data and adjust resource allocation strategies accordingly. A real-time sensing and dynamic task migration mechanism is implemented to ensure that when a node becomes overloaded, some tasks can be quickly transferred to other idle nodes to continue execution, guaranteeing service continuity and efficiency.
[0030] 3. Enhance edge support. Focus on local processing capabilities on edge computing nodes to reduce latency issues caused by reliance on remote servers. Through a distributed architecture design, each edge device can operate independently and support cross-device collaboration, enhancing the system's robustness and reliability.
[0031] 4. Efficiently process unstructured data. Develop lightweight inference models suitable for edge computing environments, not limited to text processing, but also capable of efficiently processing unstructured data such as images and video streams. Ensure efficient agent collaboration on resource-constrained edge computing nodes, providing stable and reliable quality of service.
[0032] It should be noted that all actions involving the acquisition of signals, information, or data in this application are carried out in compliance with the relevant data protection laws and regulations of the locality and with authorization from the owner of the relevant device.
[0033] Figure 1 This is a flowchart illustrating the resource scheduling method provided in an embodiment of this application, as shown below. Figure 1 As shown, the resource scheduling method includes: Step 110: Based on the task information of the task to be processed, determine at least one target model from one or more artificial intelligence models; wherein, different types of artificial intelligence models are used to process tasks with different types and indicators.
[0034] Step 120: Based on various target models, determine the resource scheduling strategy for the tasks to be processed.
[0035] Step 130: Based on the resource scheduling strategy, the real-time status information of at least one edge computing node and the predicted status information of each edge computing node, determine the target node from each edge computing node; wherein, the predicted status information of the edge computing node is obtained based on historical status information.
[0036] Step 140: Distribute the tasks to be processed to the target node.
[0037] It should be noted that the execution entity of the resource scheduling method provided in this application embodiment can be an edge cloud computing resource intelligent scheduling system based on multi-model collaboration (hereinafter referred to as the system). The system can be deployed on a remote server.
[0038] Figure 2 This is a schematic diagram of the structure of the intelligent scheduling system for edge cloud computing resources based on multi-model collaboration provided in this application embodiment, as shown below. Figure 2 As shown, the edge cloud computing resource intelligent scheduling system based on multi-model collaboration in this application may include a multi-model collaboration module, a dynamic perception and adaptive scheduling module, an edge-side optimization module, an unstructured data processing module, and an intelligent operation and maintenance module.
[0039] Specifically, the multi-model collaboration module combines a pre-trained Large Language Model (LLM) with a lightweight inference model to handle different types of tasks and their processing metrics; the dynamic perception and adaptive scheduling module introduces an adaptive learning algorithm to predict future load based on historical data and adjust resource allocation strategies accordingly; the edge-side optimization module focuses on the local processing capabilities of edge computing nodes, reducing latency issues caused by reliance on remote servers; the unstructured data processing module develops a lightweight inference model suitable for edge computing environments to efficiently process unstructured data such as images and video streams; and the intelligent operation and maintenance module uses intelligent agents to perform fully automated core network fault analysis and processing, automatically advancing the processing flow and scheduling each processing module. These modules work together to ensure the system's efficiency and flexibility, solving problems in existing technologies such as the limitations of single models, static scheduling mechanisms, insufficient real-time decision support, and lack of edge-side support. The lightweight inference model can be a model obtained by quantizing, pruning, distilling, low-rank / sparse, parameter sharing, and lightweight architecture based on a pre-trained large language model.
[0040] Specifically, the system in this application can connect to multiple edge computing nodes. These edge computing nodes can be, for example, smart gateways, edge servers, and multi-access edge computing (MEC) servers.
[0041] It should be further noted that the system of this application can be equipped with multiple types of artificial intelligence models. These may include, but are not limited to, large language models and lightweight inference models, to cope with different types of tasks and their computational metrics.
[0042] Therefore, when the system receives a task, it can treat it as a task to be processed. For example, a task to be processed can be a client request (such as a video stream, an API call, etc.).
[0043] The tasks to be processed may include task information such as task type and task metrics.
[0044] The specific task types can include text processing, image recognition, and video stream parsing, etc.
[0045] Task metrics are the requirements for handling a task, which may include latency, error, and security, among other things.
[0046] Furthermore, the system can obtain task information by parsing the task to be processed. It can then further analyze the task information using the Mixture of Experts (MoE) system within the multi-model collaboration module, thereby determining which type or multiple types of artificial intelligence models should handle the task. Each of these determined artificial intelligence models is then used as the target model.
[0047] Furthermore, the system can analyze the tasks to be processed individually or collaboratively based on the identified target models, thereby determining the resource scheduling strategy for the tasks. For example, the resource scheduling strategy may include information such as the resources required to process the tasks.
[0048] It should be noted that the system in this application can monitor the status information of each edge computing node in real time through a dynamic perception and adaptive scheduling module. Specifically, it can acquire the real-time status of each edge computing node and predict its status information for a specified future time period based on its historical status information. The status of the edge computing node can include available resources, specifically the available resources of the Central Processing Unit (CPU), Graphics Processing Unit (GPU), and memory.
[0049] The process of predicting the state information of the corresponding edge computing node in a future specified time period based on the historical state information of each edge computing node can be implemented by a traditional neural network model, such as the traditional Long Short-Term Memory (LSTM) network model. This application does not make any specific limitations.
[0050] In other words, the system can monitor the real-time status information and predicted status information of each edge computing node in real time through the dynamic perception and adaptive scheduling module.
[0051] Furthermore, the system can use the reinforcement learning framework in the dynamic perception and adaptive scheduling module to combine the resource scheduling strategy of the task to be processed, the real-time status information of each edge computing node and the predicted status information of each edge computing node to evaluate the reward of each edge computing node when executing the task to be processed under different resource allocations, and then determine one or more edge computing nodes as target nodes based on the reward.
[0052] After the target node is determined, the system can send the tasks to be processed to the target node.
[0053] Once the target node receives the task to be processed, it can use its resources to process the task and finally return the processing result to the initiator of the task.
[0054] According to the resource scheduling method of this application embodiment, at least one target model is determined from more than one type of artificial intelligence models based on the task information of the task to be processed. Based on the target models used to process tasks with different types and indicators, tasks with different types and indicators can be satisfied, thereby accurately determining the resource scheduling strategy for the task to be processed. Furthermore, based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node, the target node can be accurately determined from each edge computing node. Since the real-time status and predicted future status of each edge computing node are considered when determining the target node, the rationality of resource scheduling can be effectively improved. Then, the task to be processed is sent to the target node, and the target node can be used as the optimal edge computing node to process the task to be processed. Therefore, the overall service quality can be guaranteed, which helps to improve the user experience.
[0055] In one embodiment, based on task information of the task to be processed, at least one target model is determined from one or more types of artificial intelligence models, including: Based on the task type and task indicators in the task information of the task to be processed, at least one target model is determined from one or more artificial intelligence models through a gating network.
[0056] Specifically, the hybrid expert system of the multi-model collaborative module in this application is equipped with a gating network, so the task type and task indicators in the task information can be used as input through the gating network. And determine which AI models the input should be assigned to.
[0057] Specifically, the processing method of a hybrid expert system can be as follows: ; in, This is the final output. It's a gating function that determines the input. Should be assigned to the The probability of an artificial intelligence model. Indicates the first An artificial intelligence model processes the input. The result.
[0058] That is, the gating network first analyzes And calculate a weight for each type of artificial intelligence model. This weight represents "making the artificial intelligence model..." To handle The probability or confidence level of "".
[0059] Therefore, one can choose the AI model with the highest weight or multiple AI models with relatively high weight as the target model.
[0060] This application combines a pre-trained large language model with a lightweight inference model through a multi-model collaborative module, which can handle tasks with different types and processing requirements, effectively improve the flexibility and efficiency of edge computing resource scheduling, ensure overall service quality, and help improve user experience.
[0061] In one embodiment, determining the target node from the edge computing nodes based on a resource scheduling strategy, real-time status information of at least one edge computing node, and predicted status information of each edge computing node includes: Reinforcement learning is performed based on resource scheduling strategies, real-time status information of at least one edge computing node, and predicted status information of each edge computing node to obtain learning results. Based on the learning results, the target node is determined from each edge computing node.
[0062] Specifically, the dynamic perception and adaptive scheduling module of this application can incorporate a reinforcement learning framework. This framework defines the state space as the current resource allocation of the system, the action space as possible resource allocation decisions, and the reward function as designed based on task completion quality and response speed. More specifically, it includes: ; in, It is a state Take action below Expected return; Used to measure the actions performed. In state Instant rewards obtained; It is a discount factor, with a value ranging from [0, 1]. It is a state Take action below Expected return; Indicates in The next state Next, for all possible actions Iterate through the actions and select the one that yields the highest Q value.
[0063] It should be noted that the reinforcement learning process in this application may include state observation, action selection, reward feedback, and Q-value update in a sequential manner.
[0064] Therefore, by performing reinforcement learning, we can obtain the optimal reward for each edge computing node when executing the resource scheduling strategy corresponding to the task to be processed in its real-time state and predicted state, and we can further use each reward as the learning result.
[0065] Furthermore, based on the learning results, the edge computing node with the highest reward value among all edge computing nodes can be selected as the target node, or a specified number of edge computing nodes with relatively large reward values among all edge computing nodes can be selected as the target nodes as needed.
[0066] This application utilizes reinforcement learning to combine the resource scheduling strategy of the task to be processed, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node. This allows for the accurate identification of the target node from among the edge computing nodes. Since the real-time status and predicted future status of each edge computing node are considered when identifying the target node, the rationality of resource scheduling can be effectively improved. Consequently, the task to be processed is distributed to the target node, which can then serve as the optimal edge computing node for processing the task. This ensures the overall service quality and helps improve the user experience.
[0067] In one embodiment, the edge-side optimization module in this application focuses on the local processing capabilities of edge computing nodes, reducing latency issues caused by reliance on remote servers. Specifically, the edge-side optimization module designs a distributed architecture for each edge computing node, enabling each edge computing node to operate independently.
[0068] Figure 3 This is a schematic diagram of the edge-side optimization module in the edge cloud computing resource intelligent scheduling system based on multi-model collaboration provided in this application embodiment, as shown below. Figure 3 As shown, each edge computing node (e.g., edge node A and edge node B) includes a coordination communication module. Therefore, the edge computing nodes communicate with each other through the coordination communication module, thereby enabling the edge computing nodes of this application to cooperate across nodes.
[0069] Therefore, when an edge computing node becomes overloaded, some tasks can be quickly transferred to other idle edge computing nodes to continue execution, ensuring service continuity and efficiency.
[0070] It should be noted that the unstructured data processing module in this application can also be used to develop a lightweight model suitable for edge computing environments. This lightweight model can be used to efficiently process unstructured data, including but not limited to images, video streams, and audio.
[0071] Furthermore, the lightweight model can be deployed as a preset model on each edge computing node. Thus, since the edge computing nodes themselves support text processing, after deploying the preset model, each edge computing node can include text processing modules, image processing modules, and video processing modules, etc.
[0072] In one embodiment, the lightweight model employs a variant of the Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN), the Long Short-Term Memory (LSTM), for processing temporal data and image feature extraction. Its expression can be: ; in, It is a time step The hidden state, and It is a weight matrix. It is a bias term. It is the activation function, x t Indicates input, h t-1 It is the hidden state at time step t-1.
[0073] Therefore, this application utilizes a distributed architecture design to enable each edge computing node to operate independently and support cross-node collaboration. It also develops a lightweight model suitable for edge computing environments, capable of efficiently processing not only text but also unstructured data such as images and video streams. This ensures efficient agent collaboration on resource-constrained edge nodes and provides stable and reliable service quality.
[0074] In one embodiment, the intelligent operation and maintenance module in this application can use intelligent agents to perform fully automated core network fault analysis and processing, automatically advance the processing flow, and schedule each processing module. The intelligent agents communicate with each other through a message passing protocol, sharing task status and resource information, and jointly formulating a globally optimal scheduling scheme. The learning process of the intelligent agents can be implemented using the Q-learning algorithm.
[0075] In this application, the intelligent operation and maintenance module uses intelligent agents to perform fully automated core network fault analysis and processing. It automatically advances the processing flow and schedules various processing modules, enabling it to automatically perform preliminary analysis of fault content and adopt multiple methods to resolve the fault. Simultaneously, the operation and maintenance process is solidified into templates to ensure its reliability and controllability. By continuously adding templates, new operation and maintenance needs can be met.
[0076] By learning from a large amount of existing operation and maintenance data and real-time data through intelligent agents and large language models, abnormal behavior patterns in the network can be identified, thereby more accurately detecting and diagnosing faults and improving the overall efficiency of fault management. This can improve the accuracy of fault detection and diagnosis.
[0077] Secondly, this application can take immediate action upon detecting a fault, and in some cases, achieve automatic repair by providing rapid response capabilities through a pre-trained model. This contrasts sharply with traditional fault handling methods, which rely on manual judgment and scheduling. For example, in traditional core network-based fault recovery methods, disaster recovery operations are performed manually after the cause of the fault is analyzed, increasing the fault response time. The rapid response capability of this application is crucial for ensuring network service quality and user experience.
[0078] This application utilizes intelligent agents and large models for fault handling, significantly reducing reliance on human labor and lowering operation and maintenance costs. Traditional operation and maintenance methods require real-time human monitoring of equipment, resulting in significant manpower consumption. For example, end-to-end network intent decomposition and decision-making methods assisted by pre-trained large models, despite proposing the use of large models for processing, still require substantial manual intervention. This application, however, significantly reduces manpower requirements through automation and intelligent means, saving enterprises substantial funds.
[0079] The intelligent agent in this application can adaptively learn based on historical data, continuously optimizing and improving its fault handling capabilities, thereby enhancing the robustness and intelligence of the entire network system. Furthermore, the agent can automatically learn manual handling methods, enabling fully automated fault handling when similar faults recur. While existing methods improve information retrieval efficiency, they do not involve automated learning and updating mechanisms; this application represents a further technological advancement in this area.
[0080] This application solidifies the operation and maintenance process into a professional process template, combined with a report template, to ensure the reliability and controllability of the fault handling process. By automating the updating or addition of templates, the efficiency and accuracy of fault handling can be further improved. Traditional fault handling methods lack standardized process templates, making it difficult to guarantee consistency and reliability in each process. They often rely on ad-hoc solutions, which involve significant uncertainty and risk. This invention, through template-based methods, ensures the consistency and repeatability of the operation and maintenance process, reduces the risk of human error, and improves operation and maintenance efficiency and reliability.
[0081] The resource scheduling apparatus provided in this application is described below. The resource scheduling apparatus described below can be referred to in correspondence with the resource scheduling method described above.
[0082] Furthermore, this application also provides a resource scheduling device.
[0083] The resource scheduling device includes: The first determining module is used to determine at least one target model from one or more artificial intelligence models based on the task information of the task to be processed; wherein, different types of artificial intelligence models are used to process tasks with different types and indicators. The second determining module is used to determine the resource scheduling strategy for the task to be processed based on the various target models described above. The third determining module is used to determine a target node from the edge computing nodes based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node; wherein the predicted status information of the edge computing node is obtained based on historical status information. The scheduling module is used to distribute the tasks to be processed to the target node.
[0084] The resource scheduling device of this application determines at least one target model from more than one type of artificial intelligence models by using the task information of the task to be processed. Based on the target models used to process tasks with different types and indicators, it can satisfy the task to be processed with different types and indicators, thereby accurately determining the resource scheduling strategy for the task to be processed. Furthermore, based on the resource scheduling strategy, the real-time status information of at least one edge computing node and the predicted status information of each edge computing node, it can accurately determine the target node from each edge computing node. Since the real-time status and predicted future status of each edge computing node are considered when determining the target node, the rationality of resource scheduling can be effectively improved. Then, the task to be processed is sent to the target node, and the target node can be used as the optimal edge computing node to process the task to be processed. Therefore, it can ensure the overall service quality and help improve the user experience.
[0085] In one embodiment, the first determining module is specifically used for: Based on the task type and task indicators in the task information of the task to be processed, at least one target model is determined from one or more artificial intelligence models through a gating network.
[0086] In one embodiment, the third determining module is specifically used for: Reinforcement learning is performed based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node to obtain the learning result. Based on the learning results, the target node is determined from each of the edge computing nodes.
[0087] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, communications interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute the following method: based on task information of the task to be processed, determine at least one target model from more than one class of artificial intelligence models; wherein different classes of artificial intelligence models are used to process tasks with different types and metrics. Based on the various target models, the resource scheduling strategy for the task to be processed is determined; Based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node, a target node is determined from the edge computing nodes; wherein, the predicted status information of the edge computing node is obtained based on historical status information. The task to be processed is sent to the target node.
[0088] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0089] In another aspect, embodiments of this application also provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments, such as: determining at least one type of target model from one or more types of artificial intelligence models based on task information of a task to be processed; wherein, different types of artificial intelligence models are used to process tasks with different types and indicators; Based on the various target models, the resource scheduling strategy for the task to be processed is determined; Based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node, a target node is determined from the edge computing nodes; wherein, the predicted status information of the edge computing node is obtained based on historical status information. The task to be processed is sent to the target node.
[0090] In another aspect, embodiments of this application also provide a computer program product having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to perform the methods provided in the above embodiments, such as: determining at least one type of target model from one or more types of artificial intelligence models based on task information of the task to be processed; wherein, different types of artificial intelligence models are used to process tasks with different types and indicators. Based on the various target models, the resource scheduling strategy for the task to be processed is determined; Based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node, a target node is determined from the edge computing nodes; wherein, the predicted status information of the edge computing node is obtained based on historical status information. The task to be processed is sent to the target node.
[0091] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0092] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0093] Finally, it should be noted that the above embodiments are only used to illustrate this application and are not intended to limit this application. Although this application has been described in detail with reference to the embodiments, those skilled in the art should understand that various combinations, modifications, or equivalent substitutions of the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application.
Claims
1. A resource scheduling method, characterized in that, include: Based on the task information of the task to be processed, at least one target model is determined from one or more artificial intelligence models; wherein, different types of artificial intelligence models are used to process tasks with different types and indicators. Based on the various target models, the resource scheduling strategy for the task to be processed is determined; Based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node, a target node is determined from the edge computing nodes; wherein, the predicted status information of the edge computing node is obtained based on historical status information. The task to be processed is sent to the target node.
2. The resource scheduling method according to claim 1, characterized in that, The process of determining at least one target model from more than one type of artificial intelligence models based on task information of the task to be processed includes: Based on the task type and task indicators in the task information of the task to be processed, at least one target model is determined from one or more artificial intelligence models through a gating network.
3. The resource scheduling method according to claim 1, characterized in that, The step of determining the target node from the edge computing nodes based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node includes: Reinforcement learning is performed based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node to obtain the learning result. Based on the learning results, the target node is determined from each of the edge computing nodes.
4. The resource scheduling method according to claim 1, characterized in that, The edge computing node includes a coordination communication module; the edge computing nodes collaborate across nodes through the coordination communication module.
5. The resource scheduling method according to claim 1, characterized in that, The edge computing node includes a preset model, which is a neural network model for processing unstructured data.
6. The resource scheduling method according to claim 5, characterized in that, The unstructured data includes at least one of image data and video stream data.
7. A resource scheduling device, characterized in that, include: The first determining module is used to determine at least one target model from one or more artificial intelligence models based on the task information of the task to be processed; wherein, different types of artificial intelligence models are used to process tasks with different types and indicators. The second determining module is used to determine the resource scheduling strategy for the task to be processed based on the various target models described above. The third determining module is used to determine a target node from the edge computing nodes based on the resource scheduling strategy, the real-time status information of at least one edge computing node, and the predicted status information of each edge computing node; wherein the predicted status information of the edge computing node is obtained based on historical status information. The scheduling module is used to distribute the tasks to be processed to the target node.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the resource scheduling method as described in any one of claims 1 to 6.
9. A storage medium, said storage medium being a non-transitory computer-readable storage medium, wherein a computer program is stored thereon, characterized in that, When the computer program is executed by the processor, it implements the resource scheduling method as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the resource scheduling method according to any one of claims 1 to 6.