Network intent processing method, computer device and readable medium

Abstract

Provided in the present disclosure is a service activation method for an intent-based network. The method comprises: on the basis of network topology information, impact factors, attribute information of a service to be activated, and a pre-trained AIGC model, determining weight vectors of the impact factors for said service; and on the basis of the network topology information, the attribute information of said service, the weight vectors of the impact factors for said service, and the model, determining a network-wide service activation configuration total score and path configuration information of said service, wherein the network-wide service activation configuration total score is a reward of a first trajectory generated after executing a first action for intent orchestration, and is used for performing first fine-tuning processing on the model, so as to obtain the path configuration information of said service having the highest network-wide service activation configuration total score, and the first trajectory is generated after executing the first action for intent orchestration, and is used for realizing an intent orchestration process for service activation.

Description

Network intent processing method, computer device and readable medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This patent application claims priority to Chinese patent application 202311777216.4 filed with the State Intellectual Property Office of China on December 21, 2023, and the disclosure of this Chinese patent application is incorporated herein by reference in its entirety. Technical Field

[0003] The present disclosure relates to the field of self-intelligent optical network technology, and in particular to a network intent processing method, a computer device, and a computer-readable medium. Background Art

[0004] The global communications industry is moving from the interconnected and cloud eras to the intelligent era. Faced with market competition, industry-wide digital transformation, and increasing network complexity, which present challenges in network planning, optimization, operations and maintenance, reliability, agility, and OPEX reduction, the evolution of optical networks towards intelligence is inevitable. The rise and development of AI (artificial intelligence) software and hardware technologies has provided a solid foundation and primary means for achieving this goal.

[0005] Since the release of ChatGPT (Chat Generative Pre-trained Transformer), a large-scale AI-generated content (AIGC) model, in November 2022, the field of AI applications has ushered in a major technological revolution. As an AIGC model, ChatGPT continuously improves its contextual semantic understanding and interactive capabilities through continuous training on massive amounts of data, demonstrating unlimited potential in numerous application scenarios. The resulting surge in research and application has also propelled the entire AI industry forward.

[0006] Leveraging the advantages of AIGC generative large-scale modeling technology and employing it to enhance the core intelligent capabilities of the Autonomous Optical Transport Network (AOTN) is a key issue currently under scrutiny and needs to be addressed in the industry. The key challenges facing autonomous optical networks include leveraging AIGC generative large-scale modeling technology to enhance the core intelligent capabilities of AOTNs (Autonomous Optical Transport Networks), achieving intelligent management, a superior user experience, and flexible and open intelligence to meet customer needs for improved user experience and reduced operating costs. Currently, the industry is focusing on the application of AIGC large-scale models in network communications, primarily in areas such as intelligent fault diagnosis, network planning, commissioning, and optimization solution design and implementation. In particular, applying AIGC large-scale modeling technology to enhance the intelligent capabilities of intent-based provisioning of OTN network services, a key application scenario for autonomous optical networks, has become an emerging technology hotspot. Summary of the Invention

[0007] In response to the above-mentioned deficiencies in the prior art, the present disclosure provides a network intent processing method, a computer device, and a computer-readable medium.

[0008] In a first aspect, an embodiment of the present disclosure provides a network intent processing method, comprising: determining a weight vector of each influencing factor of the service to be activated based on network topology information, influencing factors, attribute information of the service to be activated, and a pre-trained model; wherein the model is an artificial intelligence generated content (AIGC) model; and determining a total score for the entire network service activation configuration and path configuration information of the service to be activated based on the network topology information, the attribute information of the service to be activated, the weight vector of each influencing factor of the service to be activated, and the model, wherein the total score for the entire network service activation configuration is a reward of a first trajectory, calculated based on a pre-generated prompt template, and used to perform a first fine-tuning process on the model to obtain the path configuration information of the service to be activated with the highest total score for the entire network service activation configuration, and wherein the first trajectory is generated after executing a first action for intent orchestration, and is used to implement the intent orchestration process for service activation.

[0009] On the other hand, an embodiment of the present disclosure also provides a computer device, comprising: one or more processors; and a storage device on which one or more programs are stored, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the intention network service activation method as described above.

[0010] On the other hand, an embodiment of the present disclosure further provides a computer-readable medium having a computer program stored thereon, wherein when the program is executed, the method for provisioning an intentional network service as described above is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] FIG1 is a schematic diagram of a process for activating an intent-based network service according to an embodiment of the present disclosure;

[0012] FIG2 is a schematic diagram showing the principle of activation of an intention-based network service according to an embodiment of the present disclosure;

[0013] FIG3 is a schematic diagram of the design principle of the prompt template provided by an embodiment of the present disclosure;

[0014] FIG4 is a schematic diagram of a process for calculating a total score for network-wide service provisioning configuration in a scenario of provisioning a single service provided by an embodiment of the present disclosure;

[0015] FIG5 is a schematic diagram of a network topology for a single service activation scenario and a multi-service concurrent activation scenario provided by a specific example of the present disclosure;

[0016] FIG6 is a schematic diagram of a process for calculating a total score for network-wide service provisioning configuration in a scenario where multiple concurrent services are provisioned, according to an embodiment of the present disclosure;

[0017] FIG7 is a schematic diagram showing the principle of performing a first fine-tuning process on a model according to an embodiment of the present disclosure;

[0018] FIG8 is a schematic diagram of a first fine-tuning process of the AIGC model using the Policy-Gradient algorithm provided in an embodiment of the present disclosure;

[0019] FIG9 is a schematic diagram of an intention network service activation process provided by a specific example of the present disclosure;

[0020] FIG10 is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0021] Example embodiments will be described more fully hereinafter with reference to the accompanying drawings, but the example embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the scope of this disclosure to those skilled in the art.

[0022] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0023] The terms used herein are used only to describe specific embodiments and are not intended to limit the present disclosure. As used herein, the singular forms "a," "an," and "the" are also intended to include the plural forms, unless the context clearly indicates otherwise. It will also be understood that when the terms "comprising" and / or "made of" are used in this specification, the presence of the features, wholes, steps, operations, elements, and / or components is specified, but the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or groups thereof is not excluded.

[0024] The embodiments described herein may be described with reference to plan views and / or cross-sectional views, with the aid of idealized schematic diagrams of the present disclosure. Thus, the example illustrations may be modified based on manufacturing techniques and / or tolerances. Therefore, the embodiments are not limited to the embodiments shown in the accompanying drawings, but include modifications of the configurations formed based on the manufacturing process. Therefore, the regions illustrated in the accompanying drawings are schematic in nature, and the shapes of the regions shown in the drawings illustrate specific shapes of the regions of the elements, but are not intended to be limiting.

[0025] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted as having an idealized or overly formal meaning unless expressly defined as such herein.

[0026] With the development of the digital economy, customers are increasingly demanding services over OTN networks. Although OTN networks have sufficient comprehensive service carrying capabilities, they face the following bottlenecks in service provisioning and intelligent operation that need to be improved: Service provisioning involves multiple steps and requires a large amount of input data, requiring professionals familiar with OTN technology to complete service provisioning, making it difficult for customers to activate services according to their needs; service provisioning takes a long time (on the order of days), which falls short of the fast provisioning expected by customers; and the network's low level of intelligence means that it takes a long time to analyze and evaluate network resources, impacting service provisioning time.

[0027] To address these bottlenecks, OTN network management and control systems must gradually incorporate intelligent features. This requires a transition to an intent-driven, simplified optical network architecture called IBON (Intent-Based Optical Network), leveraging network digital twin simulation and analysis technologies and integrating SDN (Software Defined Network) and AI. This architecture will drive OTN network services toward a self-intelligent ecosystem spanning the entire service lifecycle. IBON, through its awareness of OTN network services, enables intent-based analysis of user service operational expectations and goals, ensuring intelligent and precise understanding of user intent input. Intelligent technologies are then used to orchestrate, optimize, simulate, verify, and provision IBON services, ensuring precise alignment of network resource allocation with service operational intent and application scenarios.

[0028] The basic process of IBON intention activation includes four steps:

[0029] (1) Intent input (i.e., description of service activation intention): Parameter input is simplified to reduce technical complexity and replaced with the descriptive language expected by customers; an intent completion mechanism such as data mining, collection, and KG (Knowledge Graph) reasoning is established.

[0030] (2) Intent analysis: Based on the intent input, the optimization strategy for meeting the SLA (Service Level Agreement) and KPI (Key Performance Indicator) for launching the service is analyzed, including the target, priority, and weight.

[0031] (3) Intent-based orchestration: Based on the SLA KPI optimization strategy, multi-factor routing technology is used to orchestrate, simulate optical and electrical cross-layer resources, and configure performance indicators to meet user service usage intent requirements.

[0032] (4) Issuance and verification: Automated activation, issuance, and verification of intent-based business configurations.

[0033] Intent parsing and intent orchestration are the two most important key steps in the basic IBON intent activation process. This disclosed embodiment proposes a service activation solution based on these two steps, using AIGC large-scale model technology within the intelligent OTN (AN OTN) architecture. It should be noted that this disclosed embodiment uses the OTN network as an example, but the solution can be applied to PTN (Packet Transport Network), POTN (Packet Optical Transport Network), SPN (Secret Private Network), IP network, and other fields.

[0034] FIG1 is a schematic diagram of a process for activating an intent-based network service according to an embodiment of the present disclosure, and FIG2 is a schematic diagram of a principle for activating an intent-based network service according to an embodiment of the present disclosure.

[0035] As shown in Figure 2, the embodiment of the present disclosure arranges the intent-based network service provisioning process in a sequential order, focusing on two key steps: intent parsing and intent orchestration. The same AIGC model instance (Transformer A) is used to implement the functions of these two steps, namely, perform intent parsing and intent orchestration function reasoning, thereby obtaining the final OTN network service provisioning resource orchestration plan (i.e., routing plan).

[0036] Before implementing these two steps, Transformer A needs to be prompted and fine-tuned. As shown in Figure 2, the two steps of Transformer A implementing intent parsing and intent orchestration are designed as two independent DRL (Deep Reinforcement Learning) trajectory instances: DRL Trajectory1 for intent parsing and DRL Trajectory2 for intent orchestration. Both DRL Trajectory instances consist of two states and a single action. In DRL Trajectory1, Transformer A implements the intent parsing reasoning process, executing action a0 from state S0 to state S1. At this point, Transformer A's model parameters are the action policy model parameters for executing action a0. In DRL Trajectory2, Transformer A implements the intent orchestration reasoning process, executing action a1 from state S1 to state S2. At this point, Transformer A's model parameters are the action policy model parameters for executing action a1.

[0037] 1 and 2 , the intention-based network service activation method includes steps S11 and S12 .

[0038] In step S11, the weight vector of each influencing factor of the service to be activated is determined according to the network topology information, the influencing factors, the attribute information of the service to be activated and a pre-trained model, wherein the model is an AIGC model.

[0039] In DRL Trajectory1, the input parameters of the Transformer A model that executes action a0 include: network topology information t0, multiple influencing factors related to the service t1, and attribute information x of one or more services to be activated. i ,i∈[1,svr num ], svr num , svr num The Transformer A model is pre-trained, and its output parameters include: the weight vector y of each influencing factor of the service to be launched i ,i∈[1,svr num ], svr num is the number of services to be opened. That is, each service to be opened has a corresponding weight vector y i , a weight vector y of the service to be opened i Including the weights of various influencing factors of this business.

[0040] Action a0 is a second action for intent resolution. After executing action a0, a second trajectory is generated. The second trajectory can realize the intent resolution process of service activation.

[0041] In some embodiments, in DRL Trajectory1, the output parameters of the Transformer A model may also include the overall customer satisfaction score y0 of the services to be launched. It should be noted that the overall customer satisfaction score y0 of the services to be launched is the customer satisfaction of all services to be launched, which is the reward of the second trajectory generated by executing action a0 and is used to perform a second fine-tuning process on the Transformer A model to obtain the weight vector y of each influencing factor of the services to be launched. i This is consistent with the overall customer satisfaction score y0 for the services to be launched.

[0042] By continuously improving the overall customer satisfaction score y0 of the services to be launched and using the back gradient propagation algorithm, the action a0 policy model in DRL Trajectory1 can be trained and tuned, thereby achieving fine-tuning of the Transformer A model during the intent parsing process.

[0043] In step S12, the total score for the entire network service provisioning configuration and the path configuration information for the service to be provisioned are determined based on the network topology information, attribute information of the service to be provisioned, the weight vectors of the influencing factors of the service to be provisioned, and the model. The total score for the entire network service provisioning configuration is the reward of the first trajectory, calculated based on a pre-generated prompt template, and used to perform a first fine-tuning process on the model to obtain the path configuration information for the service to be provisioned with the highest total score for the entire network service provisioning configuration. The first trajectory is generated after executing the first action for intent orchestration and is used to implement the intent orchestration process for service provisioning.

[0044] In DRL Trajectory2, the input parameters of the Transformer A model that executes action a1 include: network topology information t0, attribute information x of one or more services to be activated i , the weight vector y of each influencing factor of the service to be launched i The output parameters obtained by Transformer A model inference include at least: the total score x′0 of the entire network service activation configuration and the path configuration information x′ of the service to be activated i It should be noted that in DRL Trajectory2, the Transformer A model can also output network topology update information t'0.

[0045] Action a1 is the first action used for intent orchestration. Executing action a1 generates the first trajectory, which implements the intent orchestration process for service provisioning. The total score x′0 for the entire network service provisioning configuration generated by the first trajectory is the reward for executing action a1. By continuously improving the total score x′0 for the entire network service provisioning configuration and using the back gradient propagation algorithm, the action a1 policy model in DRL Trajectory 2 can be trained and optimized, thereby fine-tuning the Transformer A model during the intent orchestration process.

[0046] The network intent processing method provided by the disclosed embodiment determines the weight vector of each influencing factor of the service to be activated based on network topology information, influencing factors, attribute information of the service to be activated, and a pre-trained AIGC model; and determines the total score of the entire network service activation configuration and the path configuration information of the service to be activated based on the network topology information, attribute information of the service to be activated, the weight vector of each influencing factor of the service to be activated, and the model; the total score of the entire network service activation configuration is the reward of the first trajectory generated after executing the first action for intent orchestration, calculated based on a pre-generated prompt template, and used to perform a first fine-tuning process on the model to obtain the path configuration information of the service to be activated with the highest total score of the entire network service activation configuration. By leveraging the technical advantages of the AIGC large model, the disclosed embodiment can solve problems such as long network service activation time, high technical complexity of activation operation, inability to accurately understand user intentions and expectations for activating and using services, and inability to obtain the optimal solution for concurrent multi-service resource orchestration and configuration based on user activation intentions. At the same time, with the help of the large-scale data analysis capabilities of the big model, the optimal solution for resource orchestration and configuration for batch multi-business intentions can be obtained efficiently and accurately.

[0047] The disclosed embodiments propose a solution for implementing OTN network service provisioning using AIGC generative technology. This solution employs a combination of natural language, Chinese characters, and mathematical symbols to construct and form a self-contained "self-intelligence language" for the field of intelligent OTN provisioning and operation and maintenance. Through pre-training and prompt templates, AIGC uses "self-intelligence language" to describe model inputs and outputs, enabling it to semantically understand model inputs and infer expected model outputs.

[0048] The disclosed embodiment provides specific definitions of the input and output parameters of the Transformer A model, an AIGC model example for implementing intention-based network activation, in the form of "self-intelligent language".

[0049] 1. Definition of the input and output parameters of the Transformer A model that implements the intent parsing function.

[0050] In some embodiments, the network topology information t0 may include node information t 0_NodeLst and link information t 0_LnkLst , node information t 0_NodeLst Includes at least one of the following: node type, cross-connect capacity, occupied cross-connect capacity, number of transmission directions, and number of wavelengths in each transmission direction. Link information t 0_LnkLst Include at least one of the following: link type, identification, optical power, optical attenuation, optical signal-to-noise ratio, number of wavelengths of the link; identification of each wavelength, transmission rate, channel resource occupancy information, power, bit error rate, optical signal-to-noise ratio, optical path data unit multiplexing relationship, modulation mode, baud rate, spectrum efficiency, center frequency, and spectrum width.

[0051] Therefore, network topology information t0 may include, but is not limited to, the number and distribution of OTN nodes, the number and distribution of OTN links, the resource scheduling structure characteristics of each node, and the resource occupancy information of each link. The resource scheduling structure characteristics of a node may include, but are not limited to, whether optical-electrical hybrid scheduling is used; the cross-connect capacity and bandwidth resource occupancy information of the relevant OXC (Optical Cross-Connect) and DXC (Digital Cross Connect); the number of transmission directions, the number of wavelengths in each direction, and the single-wavelength transmission rate. Link resource occupancy information may include, but is not limited to, the distribution and number of OCH (Optical Channel Layer) channels, the distribution and number of ODU (Optical Channel Data Unit) time slots created on each OCH channel, and the distribution, number, and occupied bandwidth of OTN services in each ODU time slot.

[0052] The information description of the current OTN network topology node i is defined as follows:

[0053] t0i_NodeType: The type of node i. The data type is an enumeration type. Currently, the values ​​are OXC, DXC, and OXC&DXC mixed scheduling.

[0054] t0i_OptCrssCptcy: Optical cross-connect capacity of node i. The data type is an integer and the unit is Tbit / s. For DXC type nodes, this element has a value of 0.

[0055] Occupied optical cross-connect capacity of node i. The data type is an integer and the unit is Tbit / s. For DXC type nodes, this element has a value of 0.

[0056] The electrical cross-connect capacity of node i. The data type is real number and the unit is Tbit / s. For OXC type nodes, this element is 0.

[0057] Occupied electrical cross-connect capacity of node i. The data type is real number and the unit is Tbit / s.

[0058] The number of transmission directions of node i, the data type is integer.

[0059] The number of wavelengths in transmission direction 1 of node i. The data type is an integer.

[0060] The information description of the current OTN network topology link j is defined as follows:

[0061] Wavelength 1 refers to a wavelength of link j.

[0062] In some embodiments, the influencing factor t1 includes at least one of the following: delay factor, bandwidth utilization factor, energy consumption factor, cost factor, security factor, computing power factor, and hop factor. 11 ,t 12 ,t 13 ,t 14 ,t 15 ,t 16 ,t 17 ,…,t 1k ,…,t 1fn ];

[0063] t 11 is the delay factor, t 12 is the bandwidth utilization factor, t 13 is the energy consumption factor, t 14 is the cost factor, t 15 is the safety factor, t 16 is the computing power factor, t 17 is the hop factor.

[0064] Indicates the number of influencing factors considered in OTN service provisioning and optimized path calculation in the current OTN network topology.

[0065] In some embodiments, the attribute information x of the service to be activated i Includes at least one of the following: source and sink node information, service level agreement grade information, application scenario type information, occupied bandwidth information, and routing constraint information.

[0066] Attribute information of the service to be activated x i,i∈[1,svr num ], svr num is the number of services to be opened, and the attribute information of the services to be opened x i is defined as follows:

[0067] The weight vector y of each influencing factor of the service to be launched i ,i∈[1,svr num ], is the weight of each influencing factor of the i-th service to be opened, and is a decimal between 0 and 1.

[0068] 2. Definition of the input and output parameters of the Transformer A model that implements the intent orchestration function.

[0069] The input parameters of the Transformer A model that uses the large model to implement intent orchestration (i.e., action a1) reasoning include the network topology information t0 and the attribute information x of the service to be activated in the input parameters of the Transformer A model that implements the intent parsing step function. i , and the weight vector y of each influencing factor of the service to be launched in the output parameters of the Transformer A model that implements the intent parsing step function i The output parameters of the Transformer A model that uses a large model to implement intention orchestration (i.e., action a1) reasoning can include network topology update information t'0, the total score of the entire network service activation configuration x'0, and the path configuration information x' of the service to be activated. i .

[0070] The network topology update information t'0 is the updated OTN network topology information after the large model is used to orchestrate and configure network resources for OTN services according to user intent (ie, action a1 is executed). The data structure of this parameter is the same as that of the network topology information t0.

[0071] The total score x′0 for the entire network service provisioning configuration is the evaluation score of the solution that uses the large model to orchestrate the entire network OTN service and configure network resources (i.e., execute action a1) based on the user intention.

[0072] Path configuration information x′ of the service to be activated i ,i∈[1,svr num ] is the attribute information of the service after the impact factor path calculation and configuration are implemented for the i-th OTN service, including the configuration values ​​of each impact factor on the service path.

[0073] In some embodiments, the path configuration information x′ iIt includes at least one of the following: source and sink node information, service identification information, service level agreement grade information, total score of activation configuration, attribute information corresponding to impact factors, and attribute information corresponding to topology information.

[0074] In the embodiment of the present disclosure, the prompt template can wake up and guide the model. FIG3 is a schematic diagram of the design principle of the prompt template provided in the embodiment of the present disclosure.

[0075] As shown in Figure 3, the prompt templates include a single-service alternative path screening prompt template, a single-service optimized routing prompt template, a multi-service concurrent alternative path screening prompt template, and a multi-service concurrent optimized routing prompt template. The guidance capability of the single-service optimized routing prompt template is generated based on the guidance capability of the single-service alternative path screening prompt template, the guidance capability of the multi-service concurrent alternative path screening prompt template is generated based on the guidance capability of the single-service optimized routing prompt template, and the guidance capability of the multi-service concurrent optimized routing prompt template is generated based on the guidance capability of the multi-service concurrent alternative path screening prompt template.

[0076] The disclosed embodiment proposes a hierarchical prompt scheme design in the form of "self-intelligent language" to guide the Transformer A model instance to have the ability to reason and realize the intention orchestration of the OTN network. Taking into account factors such as the development of AIGC model capabilities in the current and future period, the mathematical derivation and function definition involved in the disclosed embodiment are limited to addition and multiplication operations, and do not involve other complex mathematical derivations and calculations. The prompt template has the ability to guide the Transformer A model instance to have the ability to discover several alternative paths that meet the constraints and the ability to select the optimal path among the alternative paths. It does not prompt and guide the Transformer A model instance to have the path optimization algorithm capability.

[0077] As shown in FIG3 , the features of the hierarchical prompt solution design that enables the AIGC model (here, the AIGC model example is Transformer A) to implement OTN network intent orchestration are described as follows:

[0078] (1) From top to bottom, design a multi-level inverted pyramid prompt architecture mechanism from shallow to deep, from easy to difficult, and from simple to complex.

[0079] (2) The prompt template of the previous layer can awaken the large model to learn easier knowledge and reasoning abilities related to specific scene areas.

[0080] (3) The next layer of prompt templates defaults to and inherits the knowledge and reasoning ability obtained by the previous layer prompt template to awaken the large model, as what should be known and what should be done, providing the premise and foundation for the prompt awakening large model to learn the knowledge of this layer.

[0081] (4) Progressively, from shallow to deep, after completing the prompt template awakening process at all levels, the large model will have complex knowledge and reasoning capabilities related to specific scene areas.

[0082] Figure 4 is a schematic diagram of a process for calculating a total score for network-wide service provisioning configuration in a scenario of provisioning a single service according to an embodiment of the present disclosure. As shown in Figure 4 , the steps of calculating the total score for network-wide service provisioning configuration according to a prompt template include S21 and S22 .

[0083] In step S21, when the service to be activated is a single service, the single service optimization routing prompt template is used to wake up and guide the model so as to determine various first candidate paths that meet the preset first constraint condition according to the network topology information and the source and sink nodes of the service to be activated.

[0084] The first constraint condition includes but is not limited to at least one of the following: nodes that must be avoided (i.e., must-avoid nodes), bandwidth, latency, security indicators, and energy consumption.

[0085] In step S22, the single-service optimized routing prompt template is used to wake up and guide the model so as to calculate the activation configuration score of each first candidate path according to the evaluation function and weight corresponding to each influencing factor.

[0086] Each impact factor corresponds to an evaluation function and a weight. For a first alternative path, a weighted average method is adopted to calculate the opening configuration score of the first alternative path according to the evaluation function and weight corresponding to each impact factor.

[0087] In the case that the service to be opened is a single service, the path configuration information x of the service to be opened i The configuration information of the first candidate path corresponding to the highest score among the activation configuration scores of the first candidate paths is configured.

[0088] 5 , a single service to be activated with source and sink nodes A and B is used as an example for description.

[0089] In step A, a prompt template for screening a single service activation and optimizing an alternative path that meets the first constraint condition wakes up and guides the Transformer A model.

[0090] For the OTN network shown in Figure 5, the OMC (Operation and Maintenance Center) network management issues a single service with source and sink AB. The constraint satisfaction of each first candidate path of this service is shown in Table 1:

[0091] Table 1

[0092] Based on the first constraint, the first candidate paths A—D—B, A—B, and A—E—B meet the routing constraints for a single OTN service with source and sink AB and can be used as alternative paths for resource orchestration of this OTN service. The first candidate path A—C—E—B does not meet the routing constraints for this service and therefore cannot be used as an alternative path for resource orchestration of this service. Therefore, based on the current network topology shown in Figure 5, for the provisioning requirement of a single OTN service with source and sink AB, a search is conducted to identify alternative paths for this service and select the first candidate paths A—D—B, A—B, and A—E—B that meet the first constraint.

[0093] In step B, after Transformer A is awakened and guided through step A above, it has the ability to "screen single services that meet the constraints and optimize alternative paths". The prompt template for optimizing the routing of a single service awakens and guides the Transformer A model and determines the path configuration information x for the service to be opened. i .

[0094] The comprehensive evaluation objective function for the k-th service routing is defined as follows:

[0095] Among them, ω L is the path delay weight, which can be set to 0.45; ω E is the path energy consumption weight, which can be set to 0.2; ω S is the path security weight, which can be set to 0.15; ω C is the path cost weight, which can be set to 0.15; ω H is the path hop weight, which can be set to 0.05. eval-L (R k ) is the delay evaluation function of the kth service alternative path, f eval-E (R k ) is the energy consumption evaluation function of the kth service alternative path, f eval-S (R k ) is the security evaluation function of the kth business alternative path, f eval-C (R k ) is the cost evaluation function of the k-th business alternative path, f eval-H (R k ) is the hop count evaluation function of the kth service alternative path, f eval-L (R k ),f eval-E (R k ),f eval-S (R k ),f eval-C (R k),f eval-H (R k ) is a positive number in the range [0,1].

[0096] According to the above objective optimization function, the three first alternative paths are evaluated respectively, and the evaluation results and decision conclusions are shown in Table 2:

[0097] Table 2

[0098] Figure 6 is a schematic diagram of a process for calculating a total score for network-wide service provisioning configuration in a scenario where multiple concurrent services are provisioned according to an embodiment of the present disclosure. As shown in Figure 6 , the step of calculating the total score for network-wide service provisioning configuration according to a prompt template includes steps S31 and S32.

[0099] In step S31, when the services to be activated are multiple concurrent services, a multi-service concurrent alternative path screening prompt template is used to wake up and guide the model to determine a multi-service concurrent routing solution in which each service to be activated meets the preset second constraint condition.

[0100] Each service to be activated has one or more second alternative paths. The second alternative paths of each service to be activated are permuted and combined to obtain multiple multi-service concurrent routing schemes, each of which includes the second alternative paths of all services to be activated. The second alternative paths of each service to be activated are screened based on a preset second constraint condition to obtain one or more multi-service concurrent routing schemes in which each second alternative path satisfies the second constraint condition.

[0101] The second constraint condition includes but is not limited to at least one of the following: nodes that must be avoided (i.e., must-avoid nodes), bandwidth, latency, security indicators, energy consumption, and links that must be passed through (i.e., must-pass links).

[0102] In step S32, the multi-service concurrent optimization routing prompt template is used to wake up and guide the model so as to calculate the total score of the activation configuration of each multi-service concurrent routing solution according to the evaluation function and weight corresponding to each influencing factor.

[0103] In the case that there are multiple multi-service concurrent routing schemes in which each second alternative path satisfies the second constraint condition, the multi-service concurrent optimization routing prompt template is used to wake up and guide the model, and the total activation configuration score of each multi-service concurrent routing scheme is calculated respectively. The total activation configuration score of each multi-service concurrent routing scheme is the sum of the activation configuration scores of the second alternative paths of each service to be activated in the multi-service concurrent routing scheme. Therefore, for a multi-service concurrent routing scheme, the activation configuration score of the second alternative path of each service to be activated in the multi-service concurrent routing scheme is calculated respectively, and the activation configuration scores of the second alternative paths of each service to be activated in the multi-service concurrent routing scheme are summed to obtain the total activation configuration score of the multi-service concurrent routing scheme.

[0104] In the case that the services to be opened are multiple concurrent services, the path configuration information x of the services to be opened i For the activation of each multi-service concurrent routing scheme, configuration information of the second candidate path of each service to be activated in the multi-service concurrent routing scheme corresponding to the highest score in the total score is configured.

[0105] 5 , an explanation will be given below using three concurrent services to be activated whose source and sink nodes are AB, BC, and DC as an example.

[0106] In step C, after Transformer A is awakened and guided in step B above and has the capability of "optimizing routing for a single OTN service," the Prompt template for screening multiple services that meet the second constraint and optimizing alternative paths awakens and guides the Transformer A model to determine a multiple service concurrent routing solution in which each service to be activated meets the preset second constraint.

[0107] The OMC network management system issues activation for three concurrent services, whose sources and sinks are AB, BC, and DE, respectively. The routing schemes for the three concurrent services, their respective second alternative paths, and the satisfaction of the second constraint conditions are as follows. Multi-service concurrent routing scheme 1, its second alternative path, and the satisfaction of the second constraint conditions are shown in Table 3:

[0108] Table 3

[0109] Table 4 shows the multi-service concurrent routing solution 2 and the satisfaction of its second alternative path and second constraint conditions:

[0110] Table 4

[0111] Table 5 shows the multi-service concurrent routing solution 3 and the satisfaction of its second alternative path and second constraint conditions:

[0112] Table 5

[0113] Multi-service concurrent routing solutions 1 and 3 satisfy the second constraint requirements for each service routing and can be used as alternative solutions for orchestrating resources for these multiple concurrent services. Multi-service concurrent routing solution 2 fails to satisfy the second constraint requirements for each service routing and therefore cannot be used as an alternative solution for this scenario. Therefore, based on the current network topology shown in Figure 5, for the provisioning requirements of three concurrent services with source and sink nodes AB, BC, and DC, we select multi-service concurrent routing solutions 1 and 3 that satisfy the preset second constraint.

[0114] In step D, after Transformer A is awakened and guided through step C above, and has the ability to "screen multiple services that meet the constraints and optimize the alternative paths", the prompt template for multi-service activation and optimized routing decision-making wakes up and guides the Transformer A model, calculates the total activation configuration score of each multi-service concurrent routing solution, and determines the configuration information x of the second alternative path of each service to be activated in the multi-service concurrent routing solution corresponding to the highest score. i .

[0115] The comprehensive evaluation objective function for k concurrent service routing decisions is defined as follows:

[0116] Among them, ω L is the path delay weight, which can be set to 0.45; ω E is the path energy consumption weight, which can be set to 0.2; ω S is the path security weight, which can be set to 0.15; ω C is the path cost weight, which can be set to 0.15; ω H is the path hop weight, which can be set to 0.05. eval-L (R k ) is the delay evaluation function of the kth service alternative path, f eval-E (R k ) is the energy consumption evaluation function of the kth service alternative path, f eval-S (R k ) is the security evaluation function of the kth business alternative path, f eval-C (R k ) is the cost evaluation function of the k-th business alternative path, f eval-H (R k ) is the hop count evaluation function of the kth service alternative path, f eval-L (R k ),f eval-E (R k ),f eval-S (R k ),f eval-C(R k ),f eval-H (R k ) is a positive number in the range [0,1].

[0117] Based on the above objective optimization function, the two multi-service concurrent routing solutions are evaluated respectively. The evaluation results and decision conclusions are shown in Table 6:

[0118] Table 6

[0119] Figure 7 is a schematic diagram illustrating the principle of performing a first fine-tuning process on a model according to an embodiment of the present disclosure, and Figure 8 is a schematic diagram illustrating performing a first fine-tuning process on an AIGC model using the Policy-Gradient algorithm according to an embodiment of the present disclosure. The following describes the process of performing the first fine-tuning process on the model in detail, combining Figures 7 and 8.

[0120] The entire process of obtaining a network path orchestration solution is considered as a single-step DRL algorithm reasoning process. The AIGC large model (including the Adaptor) is the action strategy provider and action executor of the single-step DRL algorithm from state S1 to state S2. The θ parameter in the Adaptor, i.e., the action strategy π, can be iteratively tuned through the Policy-Gradient algorithm. θ (a1, s1), thereby obtaining the network path orchestration solution with the highest total score for provisioning configuration.

[0121] 7 and 8 , the step of performing the first fine-tuning process on the model includes:

[0122] Using the DRL action optimization algorithm, the model is used to perform a preset number of service activation intention orchestrations on the same group of multiple concurrent service activation requests, and a preset number of intention orchestration results are obtained. The steps of performing service activation intention orchestration each time include: i , the weight vector y of each of the influencing factors of the service to be opened i Input the model to obtain the total score x′0 of the entire network service activation configuration and the path configuration information x′ of the service to be activated i That is to say, the preset number of times is τ N (τ N =N), using the Transformer A model to perform τ N The service activation intention is arranged, and τ is obtained. N Intention arrangement results. Calculate τ N The sum of the average value of the intention arrangement results is used to adjust the parameter θ of the adapter in the model according to the sum of the average value, and the action strategy (π θ(a1, s1)), the action strategy of the first action is used to determine the path configuration information x′ of the service to be activated with the highest total score of the entire network service activation configuration i .

[0123] according to It can be concluded that:

[0124] There is τ n =n;

[0125] Among them, R(τ n ) is the τth 0 The reward obtained by the multi-service concurrent routing solution generated by round-robin inference is equal to the total score of the activation configuration of the multi-service concurrent routing solution.

[0126] The AIGC model takes as input the current OTN network topology, the head and tail nodes and constraints of each service to be activated and optimized, and the weights of various resources and performance indicators. The model then executes the first action a1 for the orchestration intent in the form of an action policy, and obtains the corresponding routing plan and the total score x′0 for the entire network service activation configuration, thus completing a network path orchestration plan inference trajectory.

[0127] To clearly illustrate the solution of the embodiment of the present disclosure, the following describes in detail the process of opening an intentional network service through a specific example in conjunction with Figure 9. As shown in Figure 9, the process includes steps 1 to 4.

[0128] In step 1, based on the current network topology information t0, a prompt template is designed to guide the AIGC model instance Transformer A to process batch input services. Enable intent parsing reasoning. Transformer A's reasoning is action a0 of the DRL intent parsing trajectory.

[0129] In step 2, based on the overall customer satisfaction score y0 output by Transformer A and the weight vector of each influencing factor of each business The policy gradient algorithm is used to tune the action strategy for action a0, that is, to fine-tune the intention parsing and reasoning process of Transformer A.

[0130] In step 3, a hierarchical prompt template is designed based on the current network topology information t0 to guide the AIGC model instance Transformer A to process batch input services. and the weight vector of each influencing factor of each business Perform intent orchestration reasoning. Transformer A's reasoning is action a1 of the DRL intent orchestration trajectory.

[0131] In step 4, based on the total score x′0 of the entire network service activation configuration output by Transformer A and the path configuration information x′ of the service to be activated, i , the Policy Gradient algorithm is used to tune the action strategy for action a1, that is, to fine-tune the intention orchestration reasoning process of Transformer A.

[0132] Steps 1 and 2 are used to guide and fine-tune the Transformer A model, enabling it to perform intent parsing and reasoning. Steps 3 and 4 are used to guide and fine-tune the Transformer A model, enabling it to perform intent orchestration and reasoning.

[0133] An embodiment of the present disclosure further provides a computer device. As shown in FIG10 , the computer device includes at least one processor 1001 , a memory 1002 , and at least one I / O interface 1003 .

[0134] At least one program is stored in the memory 1002. When the at least one program is executed by the at least one processor, the at least one processor implements the intent network service provisioning method provided in the aforementioned embodiments.

[0135] At least one I / O interface 1003 is connected between the processor and the memory, and is configured to implement information exchange between the processor and the memory.

[0136] The processor 1001 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 1002 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read-write interface) 1003 is connected between the processor 1001 and the memory 1002, and can realize information interaction between the processor 1001 and the memory 602, including but not limited to a data bus (Bus), etc.

[0137] In some embodiments, the processor 1001 , the memory 1002 , and the I / O interface 1003 are connected to each other via a bus, and further connected to other components of the computing device.

[0138] The embodiment of the present disclosure further provides a computer-readable medium having a computer program stored thereon, wherein when the program is executed, the method for provisioning an intentional network service as described above is implemented.

[0139] It will be appreciated by those skilled in the art that all or some of the steps in the method disclosed above, and the functional modules / units in the device can be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As is well known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media generally embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0140] Example embodiments have been disclosed herein, and although specific terms are employed, they are used and should be interpreted only in a general illustrative sense and not for purposes of limitation. In some instances, it will be apparent to those skilled in the art that, unless otherwise expressly indicated, features, characteristics, and / or elements described in conjunction with a particular embodiment may be used alone or in combination with features, characteristics, and / or elements described in conjunction with other embodiments. Therefore, it will be understood by those skilled in the art that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A method for opening an intentional network service, characterized in that: include: Determine the weight vector of each influencing factor of the service to be launched according to network topology information, influencing factors, attribute information of the service to be launched and a pre-trained model; wherein the model is an artificial intelligence generated content (AIGC) model; and Determine the total score of the service provisioning configuration of the entire network and the path configuration information of the service to be provided according to the network topology information, the attribute information of the service to be provided, the weight vector of each influencing factor of the service to be provided and the model, The total score of the whole network service activation configuration is the return of the first trajectory, which is calculated according to the pre-generated prompt template and used to perform a first fine-tuning process on the model, so as to obtain the path configuration information of the service to be activated with the highest total score of the whole network service activation configuration, and The first track is generated after executing the first action for intent orchestration, and is used to implement the intent orchestration process of service activation.

2. The method according to claim 1, characterized in that The prompt template can wake up and guide the model, and includes a single-service alternative path screening prompt template, a single-service optimized routing prompt template, a multi-service concurrent alternative path screening prompt template, and a multi-service concurrent optimized routing prompt template. The guidance capability of the single-service optimized routing prompt template is generated based on the guidance capability of the single-service alternative path screening prompt template, the guidance capability of the multi-service concurrent alternative path screening prompt template is generated based on the guidance capability of the single-service optimized routing prompt template, and the guidance capability of the multi-service concurrent optimized routing prompt template is generated based on the guidance capability of the multi-service concurrent alternative path screening prompt template.

3. The method according to claim 2, characterized in that The step of calculating the total score of the whole network service activation configuration according to the prompt template comprises: In the case where the service to be activated is a single service, the model is awakened and guided by using the single service optimization routing prompt template, so as to determine each first candidate path that meets the preset first constraint condition according to the network topology information and the source and sink nodes of the service to be activated; and The single-service optimized routing prompt template is used to wake up and guide the model, so as to calculate the activation configuration score of each of the first candidate paths according to the evaluation function and weight corresponding to each of the influencing factors.

4. The method according to claim 3, characterized in that In the case that the service to be activated is a single service, the path configuration information of the service to be activated is the configuration information of the first candidate path corresponding to the highest score among the activation configuration scores of the first candidate paths.

5. The method according to claim 2, characterized in that The step of calculating the total score of the whole network service activation configuration according to the prompt template comprises: In the case where the service to be activated is a plurality of concurrent services, the model is awakened and guided by using the multi-service concurrent candidate path screening prompt template, so as to determine a multi-service concurrent routing scheme in which each service to be activated satisfies a preset second constraint condition, wherein each of the multi-service concurrent routing schemes includes the second candidate paths of all the services to be activated; and The multi-service concurrent optimization routing prompt template is used to wake up and guide the model so as to calculate the total score of the activation configuration of each multi-service concurrent routing scheme according to the evaluation function and weight corresponding to each influencing factor, wherein the total score of the activation configuration of each multi-service concurrent routing scheme is the sum of the activation configuration scores of the second alternative paths of each service to be activated in the multi-service concurrent routing scheme.

6. The method according to claim 5, characterized in that In the case that the service to be activated is multiple concurrent services, the path configuration information of the service to be activated is the configuration information of the second candidate path of each service to be activated in the multi-service concurrent routing scheme corresponding to the highest score in the total activation configuration score of each multi-service concurrent routing scheme.

7. The method according to claim 1, characterized in that The step of performing a first fine-tuning process on the model comprises: Adopting a deep reinforcement learning (DRL) action tuning algorithm and utilizing the model, performing a preset number of service activation intention orchestrations on the same group of multiple concurrent service activation requests, and obtaining a preset number of intention orchestration results; and Calculating the summed average of the preset number of intention orchestration results, adjusting the parameters of the adapter in the model according to the summed average, and obtaining the action strategy of the first action, wherein the action strategy of the first action is used to determine the path configuration information of the service to be activated with the highest total score of the entire network service activation configuration, Among them, each step of orchestrating the service activation intention includes: inputting the network topology information, the attribute information of the service to be activated, and the weight vector of each influencing factor of the service to be activated into the model to obtain the total score of the service activation configuration of the entire network and the path configuration information of the service to be activated.

8. The method according to claim 1, characterized in that Before determining the total score of the service provisioning configuration of the entire network and the path configuration information of the service to be provided according to the network topology information, the attribute information of the service to be provided, the weight vector of each influencing factor of the service to be provided and the model, the method further includes: determining an overall customer satisfaction score of the service to be launched according to the network topology information, the influencing factors, the attribute information of the service to be launched and the model, The overall customer satisfaction score of the service to be launched is the return of the second trajectory, which is used to perform a second fine-tuning process on the model so that the obtained weight vector of each of the influencing factors of the service to be launched is consistent with the overall customer satisfaction score of the service to be launched, and The second trajectory is generated after executing the second action for intent resolution, and is used to implement the intent resolution process for service activation.

9. The method according to any one of claims 1 to 8, characterized in that The topology information includes node information and link information, wherein the node information includes at least one of the following: node type, cross-connect capacity, occupied cross-connect capacity, number of transmission directions, and number of wavelengths in each transmission direction. The link information includes at least one of the following: the type, identification, optical power, optical attenuation, optical signal-to-noise ratio, and number of wavelengths of the link; the identification, transmission rate, channel resource occupancy information, power, bit error rate, optical signal-to-noise ratio, optical path data unit multiplexing relationship, modulation mode, baud rate, spectrum efficiency, center frequency, and spectrum width of each wavelength.

10. The method according to any one of claims 1 to 8, characterized in that The influencing factors include at least one of the following: delay factor, bandwidth utilization factor, energy consumption factor, cost factor, security factor, computing power factor, and hop factor.

11. The method according to any one of claims 1 to 8, characterized in that: The attribute information includes at least one of the following: source and destination node information, service level agreement grade information, application scenario type information, occupied bandwidth information, and routing constraint information.

12. The method according to any one of claims 1 to 8, characterized in that: The path configuration information includes at least one of the following: source and sink node information, service identification information, service level agreement grade information, activation configuration score, attribute information corresponding to the impact factor, and attribute information corresponding to the topology information.

13. A computer device comprising: one or more processors; as well as a storage device having one or more programs stored thereon, Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the intention network service provisioning method as described in any one of claims 1-12.

14. A computer readable medium having a computer program stored thereon, wherein: When the program is executed, the intention network service activation method as described in any one of claims 1 to 12 is implemented.

Images