Advantageous cloud workload prediction algorithm based on deep learning

Abstract

The invention discloses an accompanying cloud workload prediction algorithm based on deep learning, and relates to the technical field of network resource prediction. The non-ground mobile network is specifically a flexible and efficient communication and calculation network constructed by an unmanned aerial vehicle carrying a server and a 5G mobile base station in a maneuvering process, and can meet the communication and calculation requirements of resource-limited Internet of Things equipment in a presence scene; according to the adjoint cloud workload prediction algorithm based on deep learning, both computing resources and network resources are provided through a non-ground mobile network, and the algorithm is mainly applied to providing resource guarantee for service requirements of highly-imminent scenes. The scene generally has no preset infrastructure, and the nature, the social and the information environment show stronger uncertainty. The intelligent service also embodies the obvious characteristic of rapid change of time, space and resource demands.

Description

Technical Field

[0001] The present invention relates to the technical field of network resource prediction, and in particular to a companion cloud workload prediction algorithm based on deep learning. Background Art

[0002] Due to the high variability and multi-dimensionality of heterogeneous service types and dynamic workloads, accurate estimation of resource usage is a complex and challenging problem. In the past few years, resource usage prediction has attracted extensive attention.

[0003] In large-scale distributed systems, failure to accurately estimate resource usage may lead to resource waste and increased operating costs, or insufficient resource allocation, affecting system performance and service quality, or even causing system crashes. For example, a video streaming service failed to accurately estimate resource requirements, causing severe freezes when a popular series was released, significantly reducing user experience.

[0004] Therefore, how to effectively solve the problem of accurately estimating resource usage caused by the high variability and multidimensionality of heterogeneous service types and dynamic workloads has become an important issue that needs to be solved in the current technology field. Summary of the invention

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a companion cloud workload prediction algorithm based on deep learning, comprising: The non-ground mobile network module is a flexible and efficient communication and computing network built by drones equipped with servers and 5G mobile base stations during maneuvers. It can meet the communication and computing needs of resource-constrained IoT devices in emergency scenarios.

[0006] Non-terrestrial mobile networks provide both computing resources and network resources. Their main application direction is to provide resource guarantees for service needs in highly temporary scenarios. Such scenarios usually have no preset infrastructure, and the natural, social and information environments present stronger uncertainties. Intelligent services also show the characteristics of rapid changes in time, space and resource requirements. Due to the flexibility of their structure, the mobility of their nodes and the diversity of resources they can carry, non-terrestrial mobile networks are a good supplement to terrestrial networks, while also adapting to the resource guarantee requirements of temporary scenarios.

[0007] Local prediction model, which is specifically a method for predicting the resource margin of a single node in a non-terrestrial mobile network. This method focuses on using the local historical data of the node, mainly the resource status within a specific time window in the past. By deeply analyzing these data, the prediction model can capture the changing rules and trends of resource usage, and estimate the resource margin of the node in the future.

[0008] Predicting available resources is crucial for non-terrestrial mobile networks to achieve efficient allocation of network and computing resources. It is the basis for maintaining stable operation of non-terrestrial mobile networks. By continuously predicting available resources, we can have a basic understanding of the future trends of network status and node performance. Such predictions can help achieve intelligent configuration of network and computing resources, allowing non-terrestrial mobile networks to automatically adjust bandwidth and allocate computing resources based on prediction results and service requirements. Doing so can prevent potential performance issues, reduce the possibility of service interruptions, and ensure that the system can maintain stable operation when facing different workloads and network conditions.

[0009] The available resource collaborative prediction module uses a federated learning framework to make full use of data distributed on different nodes. In this process, the model of each node is trained based on the local data set, and then the model's parameter updates are aggregated to the global model through an encrypted communication protocol. In this way, the global model can integrate knowledge and information from multiple nodes, thereby improving the accuracy and robustness of the prediction.

[0010] Preferably, the non-terrestrial mobile network specifically comprises the following steps: Step 1: clustering the non-ground mobile nodes according to their location information, dividing the non-ground mobile network into multiple clusters; Step 2: Elect a master node in each cluster. The master node is responsible for the aggregation and distribution of parameters within the cluster. Step 3: Train the local model on each worker node in the cluster and send the parameters to the master node; Step 4: The master node builds a global model according to the aggregation strategy and sends the updated model to the worker node.

[0011] Preferably, the specific algorithm for clustering the location information of the non-ground mobile nodes in step 1 is: The node set of the non-terrestrial mobile network is , the location information set is , through clustering, we can get the cluster set of non-terrestrial mobile networks ,in .

[0012] Preferably, the step 2 of selecting a master node is as follows: The score is calculated based on the node's location information and current resource status, and the master node for the current cycle is elected based on the score.

[0013] Preferably, the specific algorithm for local model training in step 3 is: Indicates the first The cluster, worker nodes, its available resource data can be expressed as , a total of time interval; is the target available resource to be predicted, then the available resource prediction problem can be described as: in, represents the selected prediction model, are the corresponding model parameters.

[0014] An improved heuristic method based on the black hole phenomenon is used to train neurons in feedforward neural networks. The prediction error trend is also captured by calculating the deviation of the most recent predictions, further improving the accuracy of the prediction.

[0015] The top sparse autoencoder (TSA) is designed to efficiently extract the essential representation of the workload from the raw high-dimensional workload data. TSA and gated recurrent unit (GRU) blocks are integrated into RNN to eliminate long-term memory dependencies to achieve adaptive and accurate prediction of highly varying workloads.

[0016] The present invention provides a deep learning-based companion cloud workload prediction algorithm, which has the following beneficial effects: First, the deep learning-based companion cloud workload prediction algorithm trains neurons by using an improved heuristic method based on the black hole phenomenon in a feedforward neural network. It also captures the prediction error trend by calculating the deviation of the most recent prediction, further improving the accuracy of the prediction.

[0017] Second, the deep learning-based companion cloud workload prediction algorithm aims to effectively extract the basic representation of workload from the original high-dimensional workload data through the top sparse autoencoder (TSA). TSA and gated recurrent unit (GRU) blocks are integrated into RNN to eliminate long-term memory dependencies to achieve adaptive and accurate prediction of highly varying workloads. DETAILED DESCRIPTION

[0018] The present invention will be further described in detail below in conjunction with specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0019] In the first embodiment, the present invention provides a technical solution: a companion cloud workload prediction algorithm based on deep learning, including a non-ground mobile network module. The non-ground mobile network is specifically a flexible and efficient communication and computing network built by a drone equipped with a server and a 5G mobile base station during a maneuver, which can meet the communication and computing needs of resource-constrained IoT devices in emergency scenarios.

[0020] Non-terrestrial mobile networks provide both computing resources and network resources. Their main application direction is to provide resource guarantees for service needs in highly temporary scenarios. Such scenarios usually have no preset infrastructure, and the natural, social and information environments present stronger uncertainties. Intelligent services also show the characteristics of rapid changes in time, space and resource requirements. Due to the flexibility of their structure, the mobility of their nodes and the diversity of resources they can carry, non-terrestrial mobile networks are a good supplement to terrestrial networks, while also adapting to the resource guarantee requirements of temporary scenarios.

[0021] Local prediction model, which is specifically a method for predicting the resource margin of a single node in a non-terrestrial mobile network. This method focuses on using the local historical data of the node, mainly the resource status within a specific time window in the past. By deeply analyzing these data, the prediction model can capture the changing rules and trends of resource usage, and estimate the resource margin of the node in the future.

[0022] Predicting available resources is crucial for non-terrestrial mobile networks to achieve efficient allocation of network and computing resources. It is the basis for maintaining stable operation of non-terrestrial mobile networks. By continuously predicting available resources, we can have a basic understanding of the future trends of network status and node performance. Such predictions can help achieve intelligent configuration of network and computing resources, allowing non-terrestrial mobile networks to automatically adjust bandwidth and allocate computing resources based on prediction results and service requirements. Doing so can prevent potential performance issues, reduce the possibility of service interruptions, and ensure that the system can maintain stable operation when facing different workloads and network conditions.

[0023] The available resource collaborative prediction module uses a federated learning framework to make full use of data distributed on different nodes. In this process, the model of each node is trained based on the local data set, and then the model's parameter updates are aggregated to the global model through an encrypted communication protocol. In this way, the global model can integrate knowledge and information from multiple nodes, thereby improving the accuracy and robustness of the prediction.

[0024] The non-terrestrial mobile network specifically comprises the following steps: Step 1: clustering the non-ground mobile nodes according to their location information, dividing the non-ground mobile network into multiple clusters; The specific algorithm for clustering the location information of non-ground mobile nodes is: The node set of the non-terrestrial mobile network is , the location information set is , through clustering, we can get the cluster set of non-terrestrial mobile networks ,in .

[0025] Step 2: Elect a master node in each cluster. The master node is responsible for the aggregation and distribution of parameters within the cluster. The specific steps for electing a master node are: The score is calculated based on the node's location information and current resource status, and the master node for the current cycle is elected based on the score.

[0026] Step 3: Train the local model on each worker node in the cluster and send the parameters to the master node; Indicates the first The cluster, worker nodes, its available resource data can be expressed as , a total of time interval; is the target available resource to be predicted, then the available resource prediction problem can be described as: in, represents the selected prediction model, are the corresponding model parameters.

[0027] Step 4: The master node builds a global model according to the aggregation strategy and sends the updated model to the worker node.

[0028] An improved heuristic method based on the black hole phenomenon is used to train neurons in feedforward neural networks. The prediction error trend is also captured by calculating the deviation of the most recent predictions, further improving the accuracy of the prediction.

[0029] The top sparse autoencoder (TSA) is designed to efficiently extract the essential representation of the workload from the raw high-dimensional workload data. TSA and gated recurrent unit (GRU) blocks are integrated into RNN to eliminate long-term memory dependencies to achieve adaptive and accurate prediction of highly varying workloads.

[0030] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A deep learning-based companion cloud workload prediction algorithm, characterized in that: include: A non-terrestrial mobile network module, which is used for drones equipped with servers and 5G mobile base stations to build communication and computing networks during maneuvers; A local prediction model, wherein the local prediction model is specifically for predicting the resource margin of a single node in a non-terrestrial mobile network; An available resource collaborative prediction module, which makes full use of data distributed on different nodes by adopting a federated learning framework; The model of each node in the available resource collaborative prediction module is trained based on a local data set, and then the parameters of the model are updated and aggregated to the global model through an encrypted communication protocol.

2. The deep learning-based accompanying cloud workload prediction algorithm according to claim 1, characterized in that: The non-terrestrial mobile network specifically comprises the following steps: Step 1: clustering the non-ground mobile nodes according to their location information, dividing the non-ground mobile network into multiple clusters; Step 2: Elect a master node in each cluster. The master node is responsible for the aggregation and distribution of parameters within the cluster. Step 3: Train the local model on each worker node in the cluster and send the parameters to the master node; Step 4: The master node builds a global model according to the aggregation strategy and sends the updated model to the worker node.

3. The deep learning-based accompanying cloud workload prediction algorithm according to claim 2, characterized in that: The specific algorithm for clustering the location information of non-ground mobile nodes in step 1 is: The node set of the non-terrestrial mobile network is , the location information set is , through clustering, we can get the cluster set of non-terrestrial mobile networks ,in .

4. The deep learning-based companion cloud workload prediction algorithm according to claim 3, characterized in that: The specific steps of selecting a master node in step 2 are as follows: The score is calculated based on the node's location information and current resource status, and the master node for the current cycle is elected based on the score.

5. The deep learning-based accompanying cloud workload prediction algorithm according to claim 4, characterized in that: The specific algorithm for local model training in step 3 is: Indicates the first The cluster, worker nodes, its available resource data can be expressed as , a total of time interval; is the target available resource to be predicted, then the available resource prediction problem can be described as: ； in, represents the selected prediction model, are the corresponding model parameters.

Images