Efficient intelligent business arrangement method for computing power network

Abstract

The invention relates to the technical field of big data and cloud computing, and particularly provides a computing power network efficient service intelligent arrangement method, which comprises the steps of normalizing modal service features, fusing and collecting multi-type heterogeneous features of services, performing normalization processing on the multi-type heterogeneous features, and obtaining the multi-type heterogeneous features of the services. Carrying out weighted fusion on the normalized features through a composite feature fusion method to obtain comprehensive service feature representation; and intelligent service type discrimination based on integrated learning: analyzing the comprehensive service feature representation by using an integrated learning model to realize accurate discrimination of the service type, and realizing accurate discrimination of the service type under constraint conditions of end-to-end time delay of a service function chain, CPU resources of a terminal computing power node, bandwidth resources of a physical link and the like. And the end-to-end time delay and deployment cost of the service function chain are jointly optimized, and deterministic service quality guarantee is provided for the service. According to the method, the deep space-time evolution rule contained in the service sequence can be accurately mined, and the service flow prediction precision can be improved.

Description

Technical Field

[0001] This invention relates to the technical field of big data and cloud computing, and in particular to a method for efficient intelligent orchestration of services in a computing power network. Background Technology

[0002] With the development of 5G, artificial intelligence, and edge computing technologies, the demand for AI and multimedia rendering applications in vertical industries is growing rapidly, leading to an explosive growth in the number of connections and data volume. China Mobile IoT, focusing on high-quality development and product capability enhancement, concentrates on four aspects: connectivity, entry points, online presence, and ecosystem, aiming to achieve a net increase of 250 million IoT connections by 2023. Based on its connection scale, they are extending downwards to secure entry points, expanding upwards to platforms and applications, and building an industrial ecosystem, deeply cultivating three major business areas: video IoT, smart IoT, and industrial IoT, achieving a full-industry chain layout for IoT. The evolution of terminal connectivity brings massive amounts of data to the network edge, accelerating the diffusion of computing power to the network edge, giving rise to the concept of computing power networks. Computing power networks aim to achieve the interconnection and coordinated scheduling of distributed computing nodes through optimized network architecture and protocols, thereby achieving efficient utilization of network and computing resources. At the national level, the construction of new computing power facilities and national hub nodes for a nationwide integrated computing power network has become an important development direction.

[0003] Intelligent service orchestration at the terminal is an effective measure to optimize the utilization of computing network resources and reduce the construction and maintenance costs of computing networks. It can support autonomous, continuous, and real-time flexible management and rational allocation of computing, storage, and network resources to meet the differentiated computing power needs of diverse services. Unified representation modeling and accurate measurement of diverse services can achieve accurate perception of computing network resources guided by diverse service needs. However, due to the diversity of service types and the dynamic evolution of communication and computing resources, it is difficult to obtain accurate service representation models, which in turn affects the ability to perceive computing network resources and intelligent service orchestration.

[0004] Network Functions Virtualization (NFV) / Software-Defined Networking (SDN) offers flexible task deployment and migration capabilities, addressing the intelligent orchestration of computing network services. However, the diverse service attributes of edge-cloud collaborative computing networks impose varying demands on the scheduling of computing and network resources. Existing methods fail to provide deterministic communication and computing resource guarantees when handling time-sensitive services requiring high bandwidth and computing power, such as video and image streams in vehicular communications and telemedicine. Furthermore, the dynamic arrival of service requests and the random, dynamic changes in the network environment lead to issues like service request backlog, service timeouts, and network instability. Moreover, balancing the deployment costs of NFV / Software-Defined Networking with end-to-end Quality of Service (QoS) guarantees is challenging. Therefore, diverse service requirements place higher demands on the intelligent orchestration of services in edge-cloud collaborative computing networks.

[0005] Existing service representation models neglect the diverse attributes and differences between attributes of services, making it difficult to accurately perceive and represent diverse service needs. This, in turn, affects the efficient resource utilization, service quality assurance, and multi-scenario adaptability of computing networks. Furthermore, current service orchestration methods ignore issues such as service request backlog, service timeouts, and uneven network resource allocation caused by the dynamic arrival of service requests. They also fail to consider jointly optimizing QoS and SFC deployment costs to provide deterministic service quality, thus limiting the efficiency and resource utilization of computing network resource scheduling.

[0006] To address the shortcomings of existing technologies, this invention provides a method for efficient intelligent service orchestration in computing power networks. Summary of the Invention

[0007] To achieve the above objectives, the present invention adopts the following technical solution: One aspect of the present invention provides an efficient intelligent service orchestration method for computing power networks, comprising the following steps: The high-precision service representation modeling steps involve characterizing and predicting the future traffic of diverse services within a computing network, including: Multimodal service feature normalization and fusion acquisition of multi-type heterogeneous features of services: normalize the multi-type heterogeneous features, and then perform weighted fusion of the normalized features through a composite feature fusion method to obtain a comprehensive service feature representation; Intelligent business type identification based on ensemble learning: The ensemble learning model is used to analyze the comprehensive business feature representation to achieve accurate identification of business types. Non-uniform interval traffic prediction based on time-aware long short-term memory network: In view of the non-uniform interval time sequence characteristics of service traffic, a service traffic prediction model is constructed by using a time-aware long short-term memory network. The long short-term memory network obtains short-term memory and long-term memory through subspace decomposition, and introduces a time discount factor into the short-term memory to fuse time interval information to predict the future traffic of the multi-service. The deterministic computing network resource orchestration steps, under the constraints of end-to-end latency of the service function chain, CPU resources of terminal computing nodes, and physical link bandwidth resources, jointly optimize the end-to-end latency and deployment cost of the service function chain to provide deterministic quality of service assurance for services, including: The abstraction and modeling of computing power networks abstracts the physical network of the terminal layer into an undirected graph, defines variables and parameters that characterize network state, service requirements and resource constraints, and establishes a deterministic computing network resource orchestration model with the objective function of minimizing SFC end-to-end latency and deployment cost. The parallel multi-agent TD3 algorithm based on the hierarchical Critic network architecture is used to solve the problem. The parallel multi-agent dual-delay deep deterministic policy gradient algorithm based on the hierarchical Critic network architecture is also used. In this algorithm, the global Critic network is updated according to the global network state, and guides and allocates resources to the lower-level Critic networks to minimize the total network cost. The global cost information is fed back to all lower-level Critic networks as a penalty factor, realizing multi-agent collaborative decision-making and completing the deterministic orchestration of computing network resources.

[0008] In one optional implementation, the multi-type heterogeneous features include at least service-inherent attribute features, network status features, historical traffic features, and terminal device features; the composite feature fusion method includes dynamic weighted fusion based on attention mechanism or feature fusion based on adaptive weight learning; The ensemble learning model employs a random forest algorithm based on a voting combination strategy, or a hybrid ensemble model combining gradient boosting decision trees and support vector machines. The Long Short-Term Memory (LSTM) network autoencoder inputs the learned temporal feature representations into the multilayer perceptron output to obtain the predicted traffic for subsequent multi-service traffic; the predicted traffic results are used to guide subsequent resource pre-allocation and scheduling decisions.

[0009] In one optional implementation, the nodes of the undirected graph represent terminal computing power nodes or edge / cloud computing power nodes with computing capabilities, and the edges represent physical links; the variables and parameters also include deterministic bandwidth, deterministic latency jitter limit, energy consumption, service fairness indicators, and operator revenue.

[0010] In one optional implementation, the constraints of the deterministic computing network resource orchestration model further include: link bandwidth constraints of SFC, node CPU computing power constraints, service function chain sequence constraints, and meeting the deterministic latency jitter requirements of the service.

[0011] In one optional implementation, in the hierarchical Critic network architecture, the lower-level Critic network is responsible for local resource orchestration decisions within its jurisdiction and feeds back the local decision results to the global Critic network; the global Critic network integrates all local decision results and the global network status, performs global cost evaluation and optimization, and provides global guidance information to the lower-level Critic networks, forming a closed-loop feedback mechanism.

[0012] In one optional implementation, in the parallel multi-agent TD3 algorithm, each agent corresponds to one or a group of service flows or network regions. Agents communicate indirectly by sharing some state information or through a global Critic network to achieve collaborative resource orchestration in a distributed environment. The deterministic network resource orchestration step also includes a dynamic adjustment mechanism for the orchestration results. When a drastic change in network status or the access of new high-priority services is detected, the parallel multi-agent TD3 algorithm of the hierarchical Critic network architecture is triggered to perform rapid re-optimization.

[0013] In one optional implementation, the high-precision business characterization modeling step further includes evaluating the importance of business features and dynamically adjusting the weight allocation during feature fusion based on the evaluation results, thereby further improving the accuracy of business type discrimination and traffic prediction.

[0014] Another aspect of the present invention provides a high-efficiency intelligent service orchestration system for computing power networks, comprising: The business feature perception and normalization module is used to collect various types of heterogeneous features of the business and normalize them. The composite feature fusion module is used to perform weighted fusion of normalized multi-type heterogeneous features using a composite feature fusion method to obtain a comprehensive business feature representation. The business type discrimination module is used to analyze the comprehensive business feature representation using an ensemble learning model to achieve accurate discrimination of business types; The service traffic prediction module is used to construct a service traffic prediction model using a time-aware long short-term memory network to predict the future traffic of the aforementioned diverse services. The network abstraction and model building module is used to abstract the terminal layer physical network into an undirected graph, define variables and parameters that characterize network state, service requirements and resource constraints, and establish a deterministic computing network resource orchestration model. A multi-agent resource orchestration optimization module is used to solve the deterministic computing network resource orchestration model using a parallel multi-agent dual-delay deep deterministic policy gradient algorithm with a hierarchical Critic network architecture, thereby achieving deterministic orchestration of computing network resources; and The decision execution and feedback module is used to execute resource orchestration decisions and monitor network status and service operation, providing feedback for dynamic adjustments.

[0015] In another aspect, the present invention provides an electronic device comprising: At least one memory stores computer-executable instructions non-transiently; At least one processor, configured to run the computer-executable instructions, The computer-executable instructions are executed by the processor to implement the above-described method for efficient intelligent service orchestration in a computing power network.

[0016] In another aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by at least one processor, implement the above-described method for efficient intelligent service orchestration in a computing power network.

[0017] Effects of the invention: This paper addresses the problem of low resource utilization in the collaborative processing of massive heterogeneous data computing tasks by focusing on business representation modeling and resource allocation optimization based on multi-attribute joint perception. First, a composite feature fusion method is designed to fuse multiple features and combined with a classification algorithm to achieve differentiated business type discrimination. Then, a long short-term memory (LSTM) autoencoder network is used to construct a business traffic prediction model. The LSM autoencoder possesses fine-grained deep perception capabilities when facing non-uniformly spaced evolving business demand sequences, accurately uncovering the deep spatiotemporal evolution patterns inherent in the business sequences and improving the accuracy of business traffic prediction. Finally, by combining business discrimination and traffic prediction methods, a business representation model is obtained, enabling precise perception by the computing network.

[0018] From the perspective of communication, computing, and storage resources required for edge-cloud collaborative computing networks, this paper introduces deterministic network technology to model the service orchestration problem of edge-cloud collaborative computing networks. Addressing the challenges of optimizing service orchestration solutions, such as excessively large state space, high-dimensional action space, and unknown state transition probabilities, a dual-delay deep deterministic resource orchestration algorithm is proposed. The TD3 algorithm is introduced, employing a multi-agent parallel approach to process massive service computing requests, reducing agent training time and action space. The TD3 algorithm is suitable for high-dimensional continuous action space scenarios, featuring strong parallel processing capabilities and fast convergence speed, making it well-suited for solving resource orchestration problems involving massive, diverse, and high-concurrency computing services. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an efficient intelligent service orchestration method for computing power networks provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of the deterministic computing network resource orchestration provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of the high-precision business representation modeling process provided in Embodiment 2 of the present invention; Figure 4 This is a framework diagram of an efficient intelligent service orchestration system for computing power networks provided in Embodiment 3 of the present invention; Figure 5 This is a block diagram of the electronic device provided in Embodiment 4 of the present invention; Figure 6 This is a block diagram of a computer-readable storage medium provided in Embodiment 4 of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Hereinafter, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0022] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integral part; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. Furthermore, unless otherwise explicitly specified and limited, the term "coupling" should be interpreted broadly. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components; it can also be understood as an electrical connection between different components in a circuit structure through physical lines capable of transmitting electrical signals, such as copper foil or wires on a printed circuit board (PCB), to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in a non-contact manner, such as an electrical connection between two components using capacitive coupling to transmit electrical signals.

[0023] In this embodiment of the invention, directional terms such as "up," "down," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and can change accordingly depending on the orientation of the components in the accompanying drawings.

[0024] Example 1: like Figure 1 As shown, this embodiment of the invention provides a method for efficient intelligent service orchestration in a computing power network, comprising the following steps: S1: High-precision service representation modeling steps, which provide unified and accurate feature descriptions and future traffic predictions for diverse services in the computing network, including: Multimodal service feature normalization and fusion acquisition of multi-type heterogeneous features of services: normalize the multi-type heterogeneous features, and then perform weighted fusion of the normalized features through a composite feature fusion method to obtain a comprehensive service feature representation; Intelligent business type identification based on ensemble learning: The ensemble learning model is used to analyze the comprehensive business feature representation to achieve accurate identification of business types. Non-uniform interval traffic prediction based on time-aware long short-term memory network: In view of the non-uniform interval time sequence characteristics of service traffic, a service traffic prediction model is constructed by using a time-aware long short-term memory network. The long short-term memory network obtains short-term memory and long-term memory through subspace decomposition, and introduces a time discount factor into the short-term memory to fuse time interval information to predict the future traffic of the multi-service. S2: Deterministic computing network resource orchestration steps, under the constraints of end-to-end latency of the service function chain, CPU resources of terminal computing nodes, and physical link bandwidth resources, jointly optimize the end-to-end latency and deployment cost of the service function chain to provide deterministic quality of service assurance for services, including: The abstraction and modeling of computing power networks abstracts the physical network of the terminal layer into an undirected graph, defines variables and parameters that characterize network state, service requirements and resource constraints, and establishes a deterministic computing network resource orchestration model with the objective function of minimizing SFC end-to-end latency and deployment cost. The parallel multi-agent TD3 algorithm based on the hierarchical Critic network architecture is used to solve the problem. The parallel multi-agent dual-delay deep deterministic policy gradient algorithm based on the hierarchical Critic network architecture is also used. In this algorithm, the global Critic network is updated according to the global network state, and guides and allocates resources to the lower-level Critic networks to minimize the total network cost. The global cost information is fed back to all lower-level Critic networks as a penalty factor, realizing multi-agent collaborative decision-making and completing the deterministic orchestration of computing network resources.

[0025] The above embodiments include high-precision service representation modeling and deterministic computing network resource orchestration. Considering the low resource utilization in collaborative processing of massive heterogeneous data computing tasks, a composite feature fusion method combined with a classification algorithm is designed. Simultaneously, a long short-term memory network autoencoder is used to construct a service traffic prediction model. Combining service discrimination and traffic prediction methods improves the accuracy of service traffic prediction and achieves precise perception of the computing network. Furthermore, addressing the service orchestration problem of edge-cloud collaborative computing networks, a dual-delay deep deterministic resource orchestration algorithm is proposed. The TD3 algorithm is used to achieve parallel processing by multiple agents, thus providing a novel and efficient solution for the resource orchestration problem of massive, diverse, and high-concurrency computing services.

[0026] This includes high-precision service representation modeling, designing a unified service representation method for joint perception of multi-dimensional terminal attributes, and a service traffic prediction method for perception of non-uniform interval temporal features. It also involves constructing a unified representation model for diverse services, providing a basis for accurate perception of service characteristics and for rational scheduling and on-demand allocation of computing resources. The main steps are: Step 1: In the computing network, normalize the various characteristics of the services to make them have the same scale.

[0027] Step 2: Derive the fused features by taking a weighted sum of multiple features, thereby obtaining a more comprehensive and representative feature representation.

[0028] Step 3: Combining the voting combination strategy of ensemble learning, the random forest algorithm is used to accurately identify the business type of the extracted high-dimensional features.

[0029] Step 4: Integrate a time-aware Long Short-Term Memory (LSTM) network to extract and encode features from non-equidistant sequences. This network performs subspace decomposition to obtain short-term and long-term memory separately. A time discount factor is then added to the short-term memory to fuse time interval information. The learned representation is then passed through the output MLP function to obtain the predicted traffic for subsequent multi-service traffic.

[0030] This includes deterministic network resource orchestration, which, under the constraints of end-to-end latency of SFC (Special Function Center) and CPU resources and physical link bandwidth resources of terminal computing nodes, jointly optimizes the end-to-end latency and deployment cost of SFC, thereby providing deterministic quality of service (QoS) guarantees for latency-sensitive services. Therefore, the steps for establishing the deterministic network resource orchestration model are as follows: Step 1: The terminal layer physical network is abstracted as an undirected graph, where nodes represent terminal computing power nodes and edges represent physical links.

[0031] Step 2: A series of variables and constraints are defined, such as deterministic bandwidth, deterministic latency, SFC deployment cost, energy consumption, service fairness, and operator revenue.

[0032] Step 3: A parallel multi-agent TD3 algorithm with a hierarchical Critic network architecture is proposed to solve the resource orchestration optimization problem. The global Critic network is updated based on the global network and allocates resources to lower-level Critic networks to minimize the total network cost. This cost is then fed back to all lower-level Critic networks as a penalty factor.

[0033] Example 2: like Figure 2 , Figure 3 As shown, based on Embodiment 1, the steps provided in this embodiment of the invention include the following: the multi-type heterogeneous features include at least service-inherent attribute features, network status features, historical traffic features, and terminal device features; the composite feature fusion method includes dynamic weighted fusion based on attention mechanism or feature fusion based on adaptive weight learning. The ensemble learning model employs a random forest algorithm based on a voting combination strategy, or a hybrid ensemble model combining gradient boosting decision trees and support vector machines. The Long Short-Term Memory (LSTM) network autoencoder inputs the learned temporal feature representations into the multilayer perceptron output to obtain the predicted traffic for subsequent multi-service traffic; the predicted traffic results are used to guide subsequent resource pre-allocation and scheduling decisions.

[0034] Specifically, the nodes of the undirected graph represent terminal computing power nodes or edge / cloud computing power nodes with computing capabilities, and the edges represent physical links; the variables and parameters also include deterministic bandwidth, deterministic latency jitter limit, energy consumption, service fairness indicators, and operator revenue.

[0035] Specifically, the constraints of the deterministic computing network resource orchestration model also include: link bandwidth constraints of SFC, node CPU computing power constraints, service function chain order constraints, and meeting the deterministic latency jitter requirements of services.

[0036] Specifically, in the hierarchical Critic network architecture, the lower-level Critic network is responsible for local resource orchestration decisions within its jurisdiction and feeds back the local decision results to the global Critic network; the global Critic network integrates all local decision results and the global network status, performs global cost evaluation and optimization, and provides global guidance information to the lower-level Critic networks, forming a closed-loop feedback mechanism.

[0037] Specifically, in the parallel multi-agent TD3 algorithm, each agent corresponds to one or a group of service flows or network regions. Agents communicate indirectly by sharing some state information or through the global Critic network to achieve collaborative resource orchestration in a distributed environment. The deterministic network resource orchestration step also includes a dynamic adjustment mechanism for the orchestration results. When a drastic change in network status or the access of new high-priority services is detected, the parallel multi-agent TD3 algorithm of the hierarchical Critic network architecture is triggered to perform rapid re-optimization.

[0038] Specifically, the high-precision business characterization modeling step also includes assessing the importance of business features and dynamically adjusting the weight allocation during feature fusion based on the assessment results, thereby further improving the accuracy of business type discrimination and traffic prediction.

[0039] In the above embodiments, to address the problem that indiscriminate resource allocation for diverse business needs cannot guarantee service quality and leads to resource waste, a multi-attribute-aware business representation model is established. This involves researching multi-feature fusion methods for business identification and non-uniform time-series feature-aware business traffic prediction methods to construct a unified representation model for diverse businesses. This enables accurate perception of business characteristics and provides a decision-making basis for the rational scheduling and on-demand allocation of computing resources. Furthermore, to address the issues of diverse computing network business types, dynamic arrival of service requests, and time-varying network topology leading to service request backlog, service timeouts, and uneven network resource allocation, deterministic network technology is introduced to construct a service orchestration model. Under the constraints of SFC end-to-end latency and terminal computing node CPU resources and physical link bandwidth resources, this model jointly optimizes SFC end-to-end latency and SFC deployment costs, thereby providing deterministic service quality assurance for latency-sensitive services. The specific implementation process of this solution consists of the following two steps.

[0040] High-precision business representation modeling: The diverse business needs within an edge-cloud collaborative computing network inherently lead to differentiated requirements for computing and network resources, which can be categorized into latency-sensitive and compute-intensive types. Accurate business identification can be achieved by combining multiple attributes, including the inherent characteristics of the business, the required computing resources, and the required network resources. The resource record required for the computing network to complete one service is shown in the following formula:

[0041] In the formula, This indicates the business's own attributes, including the business origin, unique identifier, and data representation, etc. The computing resources required for business operations include the type, size, location, and cache location of the computing resources used. This represents the network resources required for the service, including the quality of service requirements for computing power transmission, transmission bandwidth, transmission latency, packet loss rate, etc. Furthermore, to characterize the statistical properties of identical service requests, the time characteristics of the service are defined. This includes arrival time, waiting time, and service time.

[0042] To achieve accurate identification of diverse business types, we first normalize the features of various business types, and then design a composite feature fusion method to integrate multi-dimensional business features. The fused features are derived by taking the weighted sum of multiple features in various ways.

[0043] Furthermore, for high-dimensional features containing rich information, a voting combination strategy based on ensemble learning is used, and the random forest algorithm is employed to accurately identify the business type of the extracted high-dimensional features. These features are categorized into four types: latency-sensitive and computationally intensive businesses, latency-sensitive and non-computationally intensive businesses, latency-tolerant and computationally intensive businesses, and latency-tolerant and non-computationally intensive businesses. This facilitates subsequent traffic prediction and on-demand resource allocation for different types of businesses.

[0044] By deeply mining service arrival patterns from service traffic data, this project aims to predict diverse service traffic in advance, enabling resource reservation for the computing network and ensuring subsequent network resource scheduling and coordination. Periodic services exhibit strong temporal evolution patterns and arrival statistics. This project proposes a service traffic demand prediction method that is aware of non-equidistant time-series features. It combines a time-aware Long-Short Term Memory (LSTM) network to extract and encode fine-grained features from non-equidistant sequences. An LSTM network autoencoder is then used to mine deep features from the service traffic sequence and learn an effective and unique representation of the sequence. Specifically, the LSTM network structure first undergoes subspace decomposition to obtain short-term and long-term memory separately, and a time discount factor is added to the short-term memory to fuse time interval information, as shown in the following equation:

[0045] The adjusted short-term memory and long-term memory are combined to obtain memory cells. After the subspace decomposition stage of the Long Short-Term Memory network, the standard gating architecture of LSTM is used, which includes the forget gate, input gate, output gate, candidate memory, current memory, and current hidden state. The obtained feature representation is shown in the following formula:

[0046] Hidden states of Long Short-Term Memory (LSTM) network encoders The initial hidden state and the memory content of the decoder are respectively used as the unit memory. When the reconstruction error Er in equation (16) is minimized, the long short-term memory network encoder is applied to the original sequence to obtain the learned representation of the business traffic features, that is, the hidden state of the encoder at the end of the sequence.

[0047]

[0048] Finally, the learned representation is passed through the output MLP function to obtain the predicted traffic for subsequent diversified services.

[0049] Therefore, the steps for high-precision service representation modeling in computing power networks are as follows: Step 1: In the computing network, services are divided into two categories: latency-sensitive and computationally intensive, taking into account various attributes such as service-specific properties, computing resource requirements, and network resource requirements. To achieve accurate service identification, the various features of the services are normalized to ensure they have the same scale.

[0050] Step 2: A composite feature fusion method was designed, which derives the fused feature by taking a weighted sum of multiple features. This step aims to combine different types of features to obtain a more comprehensive and representative feature representation.

[0051] Step 3: Combining the voting strategy of ensemble learning, the random forest algorithm is used to accurately identify the business type of the extracted high-dimensional features. The business types are divided into four categories: latency-sensitive and computationally intensive businesses, latency-sensitive and non-computationally intensive businesses, latency-tolerant and computationally intensive businesses, and latency-tolerant and non-computationally intensive businesses.

[0052] Step 4: A service traffic demand prediction method based on non-uniform time-series features is proposed. This method combines a time-aware Long Short-Term Memory (LSTM) network to extract and encode features from non-uniformly spaced sequences. The LSTM network structure first performs subspace decomposition to obtain short-term and long-term memory separately. Then, a time discount factor is added to the short-term memory to fuse time interval information. Finally, the learned representation is passed through the output MLP function to obtain the predicted traffic for subsequent multi-service traffic.

[0053] Deterministic computing network resource orchestration: The terminal layer physical network is abstracted as an undirected graph. , This refers to a physical node, or terminal computing node, which provides the CPU resources for the instantiation of VNFs, and each terminal computing node can instantiate multiple VNFs. Represents the set of physical links. Each underlying terminal computing node CPU capacity is Connect adjacent terminal computing nodes and The physical link bandwidth capacity is .

[0054] The SFC set in the network is , will the The SFC is formalized into a directed graph. , indicating the first A set of different types of VNFs on an SFC. Represents the set of virtual links on the SFC, the first The first SFC A VNF is represented as The terminal computing power node is allocated to the i-th SFC. The CPU resources of a VNF are represented as follows The physical link allocates bandwidth resources to the virtual link connecting adjacent VNFs. Define a boolean variable. When the first The first SFC A VNF is mapped to the server. hour, ,otherwise .

[0055] Remaining CPU capacity of terminal computing nodes in time slots It is represented as follows:

[0056] Remaining bandwidth resources in time slot t It is represented as follows:

[0057] Deterministic bandwidth. To guarantee deterministic bandwidth requirements for services, all bandwidth resources allocated to virtual links must be greater than the sum of the data transmission rates requested by all users served by that link, while also being less than the physical link bandwidth. (19) Deterministic latency, when calculating SFC latency, mainly considers queuing latency, processing latency, and link transmission latency. SFC, making This indicates the queue length within the time slot. This represents the arrival process of the i-th SFC data packet. This project assumes that the arrival of the data packet... Obtain the parameter as Poisson distribution, data packet size Obtain the parameter as If the distribution follows an exponential distribution, then the queue update process can be represented as follows:

[0058] in, This represents the service rate of the first VNF ​​in the i-th SFC of time slot, and the... Article 1 of SFC Service rate of each VNF The amount of CPU resources allocated to a physical terminal computing node is determined by that amount. ,in This is the service rate coefficient, representing the ratio between CPU resources and service rate. Then the... The deterministic delay of each SFC must satisfy:

[0059] in, Indicates the time slot in VNF Data packet arrivals at [location] Indicates from VNF To VNF The amount of data at that location, i.e., the time slot in VNF The number of data packets at that location has been reached. Let i be the deterministic delay requirement for the i-th SFC.

[0060] SFC deployment cost mainly consists of two aspects: one is the cost of CPU resources occupied by the VNF after deployment is complete. On the other hand, after the virtual link is deployed, it occupies physical link bandwidth resources, which incurs costs. . Represented as physical link Remaining bandwidth resources Inversely proportional, that is Then in Time slot number The deployment cost of a single SFC is expressed as follows: (twenty two) The energy consumed during the computational task in time slot t is calculated as follows: (twenty three) in, For the calculated power coefficient, This refers to the CPU's computing power.

[0061] Service fairness is assessed by constructing a service fairness function using the service latency of each service chain. A higher function value indicates poorer fairness during the computing power request process. The service fairness function is defined as follows: (twenty four) To ensure deterministic latency, deterministic bandwidth, and service fairness, while minimizing SFC deployment costs and terminal energy consumption, this project proposes to define operator revenue based on constraints including the CPU resources of physical terminal computing nodes and physical link bandwidth resources. As shown in the following formula:

[0062] in, ~ Indicates the weight value, and , This represents the maximum deployment cost of SFC. This indicates the maximum energy consumption of the terminal. This is represented as the minimum value of service fairness. Represents a node The maximum throughput is determined. The operator's revenue is defined as the optimization objective, and deterministic latency, deterministic bandwidth, terminal node energy consumption, terminal computing resources, and communication resources are used as constraints to establish an optimization problem.

[0063] Considering that business requests arrive randomly and state transition probabilities are difficult to determine, iterative optimization methods cannot be used to solve the optimization problem. Furthermore, traditional model-based optimization methods often rely on idealized assumptions, which have limitations. Therefore, a deep reinforcement learning framework is deployed at the edge layer, and a parallel multi-agent twin-delayed deep deterministic (TD3) algorithm with a hierarchical Critic network architecture is proposed to optimize network resource orchestration. A two-layer Critic network is designed, including a global Critic network and subordinate Critic networks. The global Critic network updates based on the global network, allocates resources to lower-level Critic networks, determines the degree of global constraint satisfaction, minimizes the total network cost, and then uses this as a penalty factor to feed back to all lower-level Critic networks.

[0064] Therefore, the steps for deterministic network resource orchestration are as follows: Step 1: The terminal layer physical network is abstracted as an undirected graph, where nodes represent terminal computing nodes and edges represent physical links. The CPU capacity and bandwidth capacity connecting adjacent nodes of each terminal computing node are defined. The SFC set contains different service function chains, and each SFC consists of a series of VNFs and virtual links.

[0065] Step Two: A series of variables and constraints were defined, such as the remaining CPU capacity of the terminal computing nodes, the remaining bandwidth resources of the time slots, and the requirements for deterministic bandwidth and latency. The deployment cost of SFC and the energy consumption of the terminal computing nodes when performing computing tasks were defined. Service fairness and operator revenue were defined. Service fairness was measured by constructing a service fairness function; a larger function value indicates worse fairness in the service chain. Operator revenue was defined as the optimization objective, aiming to minimize the deployment cost of SFC and terminal energy consumption while ensuring deterministic latency, deterministic bandwidth, and service fairness.

[0066] Step 3: A parallel multi-agent TD3 algorithm with a hierarchical Critic network architecture is proposed to solve the resource orchestration optimization problem. The global Critic network is updated based on the global network and allocates resources to lower-level Critic networks to minimize the total network cost. This cost is then fed back to all lower-level Critic networks as a penalty factor.

[0067] In summary, this invention addresses the low resource utilization problem encountered in the collaborative processing of massive heterogeneous data computing tasks. It designs a composite feature fusion method to fuse multiple features and combines it with a classification algorithm to achieve differentiated business type identification. Furthermore, it employs a long short-term memory network autoencoder to construct a business traffic prediction model, accurately uncovering the deep spatiotemporal evolution patterns inherent in business sequences, improving business traffic prediction accuracy, and achieving precise perception of the computing power network. To address the challenges of optimizing business orchestration in edge-cloud collaborative computing power networks, such as excessively large state space, high action space dimensionality, and unknown state transition probabilities, this invention proposes the TD3 algorithm for deep deterministic resource orchestration with dual delays, solving the resource orchestration problem for massive, diverse, and high-concurrency computing services. Thus, through the design of both business representation and deterministic computing network resource orchestration, a highly efficient and powerful intelligent business orchestration model is constructed.

[0068] Example 3 like Figure 4 As shown, based on Embodiment 1, this embodiment of the invention provides a high-efficiency intelligent service orchestration system for computing power networks, comprising: Business Feature Awareness and Normalization Module: Used to collect various types of heterogeneous features of the business and normalize them; The composite feature fusion module is used to perform weighted fusion of normalized multi-type heterogeneous features using a composite feature fusion method to obtain a comprehensive business feature representation. The business type discrimination module is used to analyze the comprehensive business feature representation using an ensemble learning model to achieve accurate discrimination of business types; The service traffic prediction module is used to construct a service traffic prediction model using a time-aware long short-term memory network to predict the future traffic of the aforementioned diverse services. The network abstraction and model building module is used to abstract the terminal layer physical network into an undirected graph, define variables and parameters that characterize network state, service requirements and resource constraints, and establish a deterministic computing network resource orchestration model. A multi-agent resource orchestration optimization module is used to solve the deterministic computing network resource orchestration model using a parallel multi-agent dual-delay deep deterministic policy gradient algorithm with a hierarchical Critic network architecture, thereby achieving deterministic orchestration of computing network resources; and The decision execution and feedback module is used to execute resource orchestration decisions and monitor network status and service operation, providing feedback for dynamic adjustments.

[0069] Example 4 Figure 5 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present invention is shown.

[0070] The electronic device may include a central processing unit / microprocessor / main control chip, etc. 4; and a storage medium 5, coupled to the central processing unit / microprocessor / main control chip, etc. 4, and storing computer-executable instructions therein for performing the steps of various methods of embodiments of the present invention when executed by the processor.

[0071] The central processing unit / microprocessor / main control chip, etc., can include, but are not limited to, one or more processors or microprocessors.

[0072] Storage medium 5 may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0073] In addition, the electronic device may also include (but is not limited to) a data bus 6, an input / output bus / external bus / device bus 7, a display 8, and input / output devices 9 (e.g., keyboard, mouse, speaker, etc.).

[0074] The central processing unit / microprocessor / main control chip, etc. 4 can communicate with external devices (8, 9, etc.) via I / O bus 7 through wired or wireless network (not shown).

[0075] The storage medium 5 may also store at least one computer-executable instruction for performing the steps of various functions and / or methods in the embodiments described herein when the central processing unit / microprocessor / main control chip, etc., 4 is running.

[0076] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0077] Figure 6 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0078] like Figure 6As shown, the non-transitory computer-readable storage medium 11 stores instructions, such as computer-readable instructions 10. When the computer-readable instructions 10 are executed by a processor, the various methods described above can be performed. The non-transitory computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-transitory non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium 11 can be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions 10 stored on the computer-readable storage medium 11, the various methods described above can be performed.

[0079] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0080] 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0082] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part 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 for executing all or part of the steps of the methods of the various embodiments of this invention through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0083] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for efficient intelligent service orchestration in a computing power network, characterized in that, Includes the following steps: The high-precision service representation modeling steps involve characterizing and predicting the future traffic of diverse services within a computing network, including: Multimodal service feature normalization and fusion acquisition of multi-type heterogeneous features of services: normalize the multi-type heterogeneous features, and then perform weighted fusion of the normalized features through a composite feature fusion method to obtain a comprehensive service feature representation; Intelligent business type identification based on ensemble learning: The ensemble learning model is used to analyze the comprehensive business feature representation to achieve accurate identification of business types. Non-uniform interval traffic prediction based on time-aware long short-term memory network: In view of the non-uniform interval time sequence characteristics of service traffic, a service traffic prediction model is constructed by using a time-aware long short-term memory network. The long short-term memory network obtains short-term memory and long-term memory through subspace decomposition, and introduces a time discount factor into the short-term memory to fuse time interval information to predict the future traffic of the multi-service. The deterministic computing network resource orchestration steps, under the constraints of end-to-end latency of the service function chain, CPU resources of terminal computing nodes, and physical link bandwidth resources, jointly optimize the end-to-end latency and deployment cost of the service function chain to provide deterministic quality of service assurance for services, including: The abstraction and modeling of computing power networks abstracts the physical network of the terminal layer into an undirected graph, defines variables and parameters that characterize network state, service requirements and resource constraints, and establishes a deterministic computing network resource orchestration model with the objective function of minimizing SFC end-to-end latency and deployment cost. The parallel multi-agent TD3 algorithm based on the hierarchical Critic network architecture is used to solve the problem. The parallel multi-agent dual-delay deep deterministic policy gradient algorithm based on the hierarchical Critic network architecture is also used. In this algorithm, the global Critic network is updated according to the global network state, and guides and allocates resources to the lower-level Critic networks to minimize the total network cost. The global cost information is fed back to all lower-level Critic networks as a penalty factor, realizing multi-agent collaborative decision-making and completing the deterministic orchestration of computing network resources.

2. The method for efficient intelligent service orchestration in a computing power network as described in claim 1, characterized in that, The multi-type heterogeneous features include at least service-inherent attribute features, network status features, historical traffic features, and terminal device features; the composite feature fusion method includes dynamic weighted fusion based on attention mechanism or feature fusion based on adaptive weight learning; The ensemble learning model employs a random forest algorithm based on a voting combination strategy, or a hybrid ensemble model combining gradient boosting decision trees and support vector machines. The Long Short-Term Memory (LSTM) network autoencoder inputs the learned temporal feature representations into the multilayer perceptron output to obtain the predicted traffic for subsequent multi-service traffic; the predicted traffic results are used to guide subsequent resource pre-allocation and scheduling decisions.

3. The method for efficient intelligent service orchestration in a computing power network as described in claim 1, characterized in that, The nodes of the undirected graph represent terminal computing power nodes or edge / cloud computing power nodes with computing capabilities, and the edges represent physical links; the variables and parameters also include deterministic bandwidth, deterministic latency jitter limit, energy consumption, service fairness indicators and operator revenue.

4. The method for efficient intelligent service orchestration in a computing power network as described in claim 1, characterized in that, The constraints of the deterministic computing network resource orchestration model also include: SFC link bandwidth constraints, node CPU computing power constraints, service function chain sequence constraints, and meeting the deterministic latency jitter requirements of the services.

5. The method for efficient intelligent service orchestration in a computing power network as described in claim 1, characterized in that, In the hierarchical Critic network architecture, the lower-level Critic network is responsible for local resource orchestration decisions within its jurisdiction and feeds back the local decision results to the global Critic network. The global Critic network integrates all local decision results and the global network status, performs global cost evaluation and optimization, and provides global guidance information to the lower-level Critic networks, forming a closed-loop feedback mechanism.

6. The method for efficient intelligent service orchestration in a computing power network as described in claim 1, characterized in that, In the parallel multi-agent TD3 algorithm, each agent corresponds to one or a group of service flows or network regions. Agents communicate indirectly by sharing some state information or through the global Critic network to achieve collaborative resource orchestration in a distributed environment. The deterministic network resource orchestration step also includes a dynamic adjustment mechanism for the orchestration results. When a drastic change in network status or the access of new high-priority services is detected, the parallel multi-agent TD3 algorithm of the hierarchical Critic network architecture is triggered to perform rapid re-optimization.

7. The method for efficient intelligent service orchestration in a computing power network as described in claim 1, characterized in that, The high-precision business characterization modeling step also includes evaluating the importance of business features and dynamically adjusting the weight allocation during feature fusion based on the evaluation results, thereby further improving the accuracy of business type discrimination and traffic prediction.

8. A computing power network efficient service intelligent orchestration system applying the computing power network efficient service intelligent orchestration method as described in any one of claims 1 to 7, characterized in that, include: Business Feature Awareness and Normalization Module: Used to collect various types of heterogeneous features of the business and normalize them; The composite feature fusion module is used to perform weighted fusion of normalized multi-type heterogeneous features using a composite feature fusion method to obtain a comprehensive business feature representation. The business type discrimination module is used to analyze the comprehensive business feature representation using an ensemble learning model to achieve accurate discrimination of business types; The service traffic prediction module is used to construct a service traffic prediction model using a time-aware long short-term memory network to predict the future traffic of the aforementioned diverse services. The network abstraction and model building module is used to abstract the terminal layer physical network into an undirected graph, define variables and parameters that characterize network state, service requirements and resource constraints, and establish a deterministic computing network resource orchestration model. The multi-agent resource orchestration optimization module is used to solve the deterministic computing network resource orchestration model using a parallel multi-agent dual-delay deep deterministic policy gradient algorithm with a hierarchical Critic network architecture, thereby realizing the deterministic orchestration of computing network resources. as well as The decision execution and feedback module is used to execute resource orchestration decisions and monitor network status and service operation, providing feedback for dynamic adjustments.

9. An electronic device, comprising: At least one memory stores computer-executable instructions non-transiently; At least one processor, configured to run the computer-executable instructions, The computer-executable instructions are executed by the processor at runtime, and the method for efficient intelligent orchestration of computing power network services is described in any one of claims 1-7.

10. A computer-readable storage medium, wherein, The computer-readable storage medium stores computer-executable instructions, which, when executed by at least one processor, implement the intelligent orchestration method for efficient services in a computing power network according to any one of claims 1-7.

Images