An intelligent service orchestration method and related apparatus

Abstract

The application discloses a kind of intelligent service arrangement method and related device, it is related to telecommunication carrier network, value-added service, cloud computing field, method includes: by obtaining customer service demand information and carrying out semantic understanding and intention analysis to convert into service delivery target, get rid of the dependence on formatted input and artificial decomposition order, solve the problem of cross-domain cooperation difficulty;By generating service target arrangement strategy including action sequence and tool calling strategy based on service delivery target, the dynamic intelligent generation of business process strategy is realized, and the defect of rigidification of business process strategy is overcome;By monitoring and capturing abnormal information in real time during the operation of service and network delivery process and updating the service target arrangement strategy using the abnormal information, the process can automatically respond to exceptions and dynamically adjust the execution plan during execution, improving the fault tolerance and delivery success rate of the system.

Description

Technical Field

[0001] This application relates to the fields of telecommunications operator networks, value-added services, and cloud computing, and in particular to an intelligent service orchestration method and related apparatus. Background Technology

[0002] In the fields of telecom operator networks, value-added services, and cloud computing, traditional networks (such as SPN (Slicing Packet Network), PTN (Packet Transport Network), SDH (Synchronous Digital Hierarchy), MPLS (Multiprotocol Label Switching)) and emerging technologies (such as SDN (Software Defined Networking), NFV (Network Functions Virtualization), cloud-native technologies, and 5G (5th Generation Mobile Communication Technology) private networks) have long coexisted, forming a complex heterogeneous environment lacking a unified architecture, with stacked generations of network technologies, internal fragmentation, and extremely difficult management and maintenance. This heterogeneity has led to the formation of "chimney-like" islands for various specialized networks such as transmission, IP (Internet Protocol), access, and wireless, each with its own independent management system, configuration interface, and maintenance process.

[0003] With the deepening of digital transformation, enterprise users' demands for network services have evolved from traditional "one-size-fits-all" leased line services to differentiated services that are generated on demand, scalable elastically, and of guaranteed quality. This places extremely high demands on the agility and intelligence of network delivery. A typical end-to-end business delivery process begins with the receipt of a customer service order and ends with the successful activation and verification of the service's availability on the network. Its core components include order parsing, resource discovery, cross-domain collaborative scheduling, and configuration distribution. When a cross-domain end-to-end service needs to be activated, manual intervention is required to perform multiple order decompositions, resource queries, instruction conversions, and collaborative scheduling across different systems. This has become the biggest bottleneck in agile business delivery.

[0004] However, under the existing technological framework, the automation level of the above processes is extremely low. For example, when it is necessary to open a "dedicated cloud line" from a customer's site to a cloud data center, this service typically needs to cross multiple specialized network domains, including access networks, transmission networks, cloud private networks, and even cloud intranets. To achieve this service, it is not only necessary to establish IP connectivity between the networks in each domain, but also to configure specific bandwidth, speed, and other parameters within each domain. However, since each specialized network (such as transmission, IP, and access) has its own independent network management system, and the networks between the systems are isolated from each other, both manual configuration through sequential login to each network management system and the mechanical execution of instructions by a simple automated system between systems present significant challenges. The entire process heavily relies on experts with complex network knowledge and rich experience to manually decompose orders, query cross-system resources, convert heterogeneous instructions, and coordinate scheduling. If the equipment configuration at any stage is incorrect, the entire end-to-end service will not function properly. This high dependence on expert experience and the technical complexity of cross-domain collaboration constitute the biggest bottleneck for agile service delivery at present.

[0005] To address the challenge of rapid network delivery, existing technologies primarily include the following solutions:

[0006] Automation based on traditional network management and pre-integrated workflows involves hard-coding fixed processes into the system, with each specialized network management system operating independently. Its drawbacks include extremely poor adaptability, requiring extensive code modifications for process changes, and the inability to achieve cross-domain (e.g., transmission, IP, access domains) automatic collaboration due to the isolation and heterogeneity between these specialized network management systems, heavily relying on manual intervention.

[0007] Automation based on external automation tools or orchestrators: This involves introducing a top-level orchestrator to connect the underlying systems. Its drawbacks include the orchestrator itself becoming a complex system, the codebase expanding rapidly with business growth, and the failure to address the rigidity and heterogeneity issues of the underlying systems themselves.

[0008] Model-driven and SDN / NFV-based cloud automation: Automation is driven by abstracting business models. Its drawbacks include high implementation costs, the need for complete overhaul of existing networks and equipment, and the risk of vendor lock-in in multi-vendor environments.

[0009] AI-based (Artificial Intelligence) intelligent closed-loop automation introduces AI models to assist in process operation. Its drawbacks include heavy reliance on high-quality, massive amounts of data and powerful computing capabilities, and the fact that AI decisions are often "black boxes," making them difficult to understand and maintain.

[0010] In summary, existing technologies generally suffer from rigid business process strategies, difficulties in cross-domain collaboration, poor fault tolerance, and over-reliance on human experience when handling business and network delivery processes. Summary of the Invention

[0011] In view of the above problems, this application provides an intelligent business orchestration method and related apparatus to solve the problems of rigid business process strategies, difficulties in cross-domain collaboration, poor fault tolerance, and over-reliance on human experience in the prior art. The specific solution is as follows:

[0012] The first aspect of this application provides an intelligent service orchestration method, including:

[0013] Obtain information about customer business needs;

[0014] The customer business needs information is semantically understood and intent-based, and then transformed into business delivery objectives.

[0015] Based on the aforementioned business delivery objectives, a business objective orchestration strategy, including action sequences and tool invocation strategies, is generated to achieve these objectives.

[0016] According to the business objective orchestration strategy, the business and network delivery process is run to complete the delivery of services and networks; during the operation of the business and network delivery process, abnormal information inside the system is monitored and captured in real time, and the abnormal information is used to update the business objective orchestration strategy.

[0017] A second aspect of this application provides an intelligent service orchestration system, comprising:

[0018] The situational awareness module is used to acquire information about customer business needs;

[0019] The intent recognition module is used to perform semantic understanding and intent analysis on the collected customer business demand information, and to transform the customer business demand information into business delivery targets.

[0020] The strategy planning module is used to generate a business objective orchestration strategy, which includes action sequences and tool invocation strategies, to achieve the business delivery objective based on the business delivery objective.

[0021] The target execution module is used to run the service and network delivery process according to the service target orchestration strategy to complete the delivery of services and network.

[0022] The self-awareness module is used to monitor and capture abnormal information within the system in real time during the operation of the business and network delivery process, and to feed the abnormal information back to the strategy planning module.

[0023] The strategy planning module is also used to update the current business objective orchestration strategy based on the abnormal information after receiving the abnormal information.

[0024] A third aspect of this application provides a computer program product, including a computer program that, when run, enables the implementation of the intelligent business orchestration method of the first aspect or any implementation thereof.

[0025] The fourth aspect of this application provides a computer storage medium storing a computer program thereon, characterized in that the program, when executed by a processor, implements the intelligent service orchestration method of the first aspect or any implementation thereof.

[0026] In summary, the intelligent business orchestration method and related apparatus provided in this application, by acquiring customer business demand information and performing semantic understanding and intent analysis to transform it into business delivery targets, can eliminate the reliance on formatted, standardized input and manual order breakdown, significantly reducing the dependence on expert experience and solving the problem of cross-domain collaboration difficulties. By generating a business target orchestration strategy containing action sequences and tool invocation strategies based on the business delivery targets, dynamic and intelligent generation of business process strategies is achieved, rather than pre-defined rigid definitions, thus overcoming the defects of rigid business process strategies. Furthermore, by monitoring and capturing internal system anomaly information in real time during the operation of business and network delivery processes, and feeding back this anomaly information to update the business target orchestration strategy, a closed-loop self-healing capability is formed, enabling the process to automatically respond to anomalies and dynamically adjust the execution plan during execution, greatly improving the system's fault tolerance and delivery success rate. Therefore, this invention effectively solves the problems of rigid business process strategies, difficulties in cross-domain collaboration, poor fault tolerance, and excessive reliance on human experience in the prior art, achieving efficient, flexible, intelligent, and robust business and network delivery. Attached Figure Description

[0027] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0028] Figure 1 This is a flowchart illustrating an intelligent service orchestration method provided in an embodiment of this application;

[0029] Figure 2 This is a schematic diagram of the process for obtaining customer business requirement information provided in an embodiment of this application;

[0030] Figure 3 This is a flowchart illustrating the process of semantic understanding and intent analysis of customer business needs information, and transforming the customer business needs information into business delivery targets, as provided in the embodiments of this application.

[0031] Figure 4This is a schematic diagram of the strategy planning and orchestration algorithm provided in the embodiments of this application;

[0032] Figure 5 This is another flowchart illustrating the intelligent service orchestration method provided in this application embodiment;

[0033] Figure 6 This is a flowchart illustrating the task execution process in the intelligent service orchestration method provided in this application embodiment;

[0034] Figure 7 This is a schematic diagram of the structure of an intelligent service orchestration system provided in an embodiment of this application. Detailed Implementation

[0035] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0036] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0037] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements but may include other elements not explicitly listed or inherent to such processes, methods, systems, products, or apparatus.

[0038] In the embodiments of this application, the terms and English abbreviations are exemplary examples given for ease of description and should not be construed as limiting the application in any way. This application does not preclude the possibility of defining other terms that can achieve the same or similar functions in existing or future agreements.

[0039] Figure 1 This is a flowchart illustrating an intelligent service orchestration method provided in an embodiment of this application, including steps 101 to 105, which are described in detail below.

[0040] 101. Obtain customer business needs information;

[0041] As the initial access point for intelligent business orchestration, it supports the collection of customer business requirements from multiple channels and in multiple formats. It can receive requirement data from various front-end channels such as customer service systems, customer self-service platforms, and business acceptance systems, covering various forms such as structured work orders, unstructured natural language text, and speech-to-text.

[0042] The acquired customer business requirements information represents the customer's needs for various network services. It may include core information such as service type, network performance requirements, and service scenarios, serving as the foundational data source for subsequent semantic understanding, policy orchestration, and service delivery.

[0043] For example, business requirement information could be a customer requesting the activation of a dedicated cloud connection for their business.

[0044] 102. Perform semantic understanding and intent analysis on the customer's business needs information, and transform the customer's business needs information into business delivery objectives;

[0045] We conduct multi-dimensional and comprehensive semantic analysis and intent mining based on the collected original customer business needs information.

[0046] Based on domain-specific natural language processing technology and a pre-defined business knowledge system, it can identify key information such as business objects, performance parameters, service scope, and delivery scenarios contained in customer business requirements information. This enables the identification of the true business demands beneath the surface-level descriptions of customers, the elimination of invalid and redundant information, and the transformation of vague, fragmented, and unstructured customer needs into clear, quantifiable, and standardized business delivery goals that can be identified and executed by the system.

[0047] The service delivery objectives clearly define the functions, performance indicators, and delivery standards that the network services need to achieve, providing a precise basis for subsequent strategy generation.

[0048] For example, after semantic understanding and intent analysis of customer business needs information to open a dedicated enterprise cloud line, a clear business delivery goal is obtained. This business delivery goal is to complete the dedicated line interconnection between the enterprise site and the cloud data center, and meet the network delivery standards of specified bandwidth, latency, and reliability.

[0049] 103. Based on the business delivery objective, generate a business objective orchestration strategy that includes action sequences and tool invocation strategies to achieve the objective;

[0050] Driven by standardized business delivery goals, and combined with preset business rules, network resource constraints, and cross-domain collaboration specifications, a business goal orchestration strategy is dynamically generated.

[0051] This business objective orchestration strategy can adaptively adjust according to differentiated business delivery objectives without relying on manually preset fixed scripts. The core of this business objective orchestration strategy includes two core elements: the standardized action execution sequence required to achieve the business delivery objectives, and the system, device, and interface tool calling strategies corresponding to each execution action, so as to realize the full-process execution plan for defining business delivery.

[0052] For example, to meet the business delivery goals of enterprise cloud access dedicated lines, generate an action sequence that includes resource verification, tunnel configuration, route distribution, cloud-network integration, service activation, and connectivity verification. At the same time, match the tool call strategies of the corresponding network management and cloud platform interfaces to form a complete orchestration scheme.

[0053] 104. Based on the orchestration strategy for this business objective, run the business and network delivery processes to complete the delivery of services and networks;

[0054] The generated business objective orchestration strategy can be parsed by the execution engine, and the entire business process can be automatically driven according to the action sequence, execution logic, and tool call rules defined by the strategy.

[0055] This end-to-end service can include operations such as network resource application, cross-domain device configuration, cloud network parameter synchronization, service activation, and performance verification. The execution of this service does not require manual intervention and automatically completes the configuration, activation, and commissioning of network services, ultimately delivering the network services required by the customer and ensuring that the actual delivery effect fully matches the preset service delivery goals.

[0056] 105. During the operation of this business and network delivery process, monitor and capture abnormal information within the system in real time, and use this abnormal information to update the business objective orchestration strategy.

[0057] Throughout the entire lifecycle of fully automated business and network delivery processes, we continuously monitor the operational status of each execution stage, tool call results, device feedback data, and process progress in real time, and capture various internal system anomalies such as process execution timeouts, interface call failures, insufficient resources, and configuration distribution errors.

[0058] The captured structured anomaly information is fed back to the strategy generation stage in real time, triggering the strategy adaptive update mechanism to correct, adjust or replan the original business objective orchestration strategy, adapt to the current anomaly scenario, ensure that the delivery process can continue to run with fault tolerance, and form a closed-loop adaptive optimization mechanism.

[0059] For example, when an interface call timeout exception occurs during the configuration distribution process, the exception is captured in real time, and the orchestration strategy is automatically updated using the exception to adjust the tool calling method or execution order and restart the process.

[0060] In this embodiment, by acquiring customer business requirements information and performing semantic understanding and intent analysis to transform it into business delivery goals, the reliance on formatted, standardized input and manual order breakdown can be eliminated, significantly reducing the dependence on expert experience and solving the problem of difficult cross-domain collaboration. By generating a business goal orchestration strategy containing action sequences and tool invocation strategies based on the business delivery goals, dynamic and intelligent generation of business process strategies is achieved, rather than pre-defined rigid definitions, thus overcoming the defects of rigid business process strategies. Furthermore, by monitoring and capturing internal system anomalies in real time during the operation of business and network delivery processes, and using this anomaly information to update the business goal orchestration strategy, a closed-loop self-healing capability is formed, enabling the process to automatically respond to anomalies and dynamically adjust the execution plan during execution, greatly improving the system's fault tolerance and delivery success rate. Therefore, this invention effectively solves the problems of rigid business process strategies, difficulties in cross-domain collaboration, poor fault tolerance, and excessive reliance on human experience in the prior art, achieving efficient, flexible, intelligent, and robust business and network delivery.

[0061] Figure 2 This is a flowchart illustrating the process of obtaining customer business needs information provided in this application embodiment, including steps 201 to 203, which are described in detail below.

[0062] 201. Collect the customer's business requirements information through at least one of the following methods: text input, voice input, or application programming interface;

[0063] The system implementing this intelligent business orchestration method can be pre-configured with multi-channel compatible access capabilities, supporting the collection of customer business requirement information through at least one of text input, voice input, or API (Application Programming Interface), and can adapt to access requirements from different scenarios and sources.

[0064] Text input can include forms, text work orders, and natural language messages; voice input can convert requirements through speech-to-text technology and assist in error correction and verification; the API interface can connect to various business systems to automatically push structured order data. It can passively receive external calls or proactively detect changes in data from external systems, without relying on fixed standardized interface protocols. It is compatible with various heterogeneous information, including structured, semi-structured, and unstructured data, covering multiple channels and formats for collecting original customer requirements, achieving comprehensive and complete collection of customer business needs.

[0065] 202. Obtain the vector corresponding to the customer's business requirements information;

[0066] By using a built-in pre-trained semantic embedding encoding model, various unstructured, semi-structured, and structured customer business requirement information can be uniformly vectorized, eliminating data differences caused by different input formats and different expression methods, and transforming the original requirement text into a high-dimensional, uniform-dimensional semantic requirement vector.

[0067] This demand vector can accurately represent the core semantic features of customer business needs, providing a calculable and comparable standardized data foundation for subsequent template matching, semantic retrieval, and intent analysis.

[0068] 203. Using predefined business templates, extract key information from the vector to form a perception result set.

[0069] This preset service template, denoted as T, is a standardized feature template pre-configured and adapted to various telecommunications network services. This predefined service template can integrate service information from one or more external customer systems, covering customer system data from different channels and with different permissions within the operator, supplementing the service attribute information scattered across various systems, and ensuring the completeness of requirement collection. The preset service template includes template information and attribute information. The template information describes the service overview corresponding to the template, while the attribute information carries various core parameters corresponding to the service. This attribute information can include n types of attributes such as customer name, service type, bandwidth, rate, access address, and service level.

[0070] This preset service template can cover all types of network services, including leased lines, network slicing, and cloud-network interconnection.

[0071] The preset business template is denoted as T, and the corresponding predefined business template containing attributes is denoted as T=[T1,T2,…,Tn], where n is an integer greater than 2. The collected multi-source heterogeneous customer business requirement information is denoted as M, and the m requirement data is denoted as M=[M1,M2,…,Mm], where m is an integer greater than 2.

[0072] After the demand vector transformation is completed, the demand vector set M is compared and matched with the preset business template set T dimension by dimension based on the intelligent matching algorithm of predefined business templates. Valid demand features that meet the similarity threshold are selected, and key feature information such as business type, bandwidth, access point, service level, and latency requirements are extracted. Invalid and redundant features are eliminated, and finally a structured perception result set R with unified format, complete parameters, and accurate semantics is formed.

[0073] Among them, the perception result set R is a valid feature subset of the original requirement information M, which filters out invalid, redundant, and non-business-related miscellaneous information, and retains only the valid business requirement information that matches the standard business template.

[0074] Determining the perception result set using a predefined business template intelligent matching algorithm may include the following steps:

[0075] Step 1: Calculate the external demand information vector M i The similarity S with the predefined business template vector Tj ij The calculation uses the following formula:

[0076] (1)

[0077] Among them, S ij M represents the similarity between the external demand information vector and the predefined business template vector; i This indicates the existence of i demand vectors, T j This indicates that there are j predefined business templates, where i and j are integers with values ​​greater than 1.

[0078] Step 2: Obtain the preset similarity threshold θ. The preset similarity threshold can be 0.9, and this threshold can be customized and adjusted according to business complexity and delivery accuracy requirements.

[0079] Step 3: Filter all that satisfy S ij The matching pairs ≥ θ form a standardized perception result set R, which is expressed as:

[0080] R={(T x :M y )|T x ∈[T1,T j ],M y ∈[M1,M i ],(∀x∈[1,i],∀y∈[1,j],S xy ≥θ)}(2)

[0081] Among them, T x M represents the x-th predefined business template vector; y T represents the y-th requirement vector; T1 represents the 1st predefined business template vector; T j M1 indicates the existence of j predefined business templates; M1 indicates the existence of the first requirement vector; M i This indicates the existence of i demand vectors; S xy This indicates the similarity between the external demand information vector and the predefined business template vector.

[0082] The final generated perception result set R serves as standardized and highly accurate input data for subsequent intent recognition.

[0083] For example, based on a preset business template, key information such as business type, bandwidth specifications, access area, and cloud connection node are extracted from the customer demand vector to generate a structured perception result set and complete the standardized preprocessing of the demand.

[0084] In this embodiment, customer business requirement information is collected through multiple channels, heterogeneous requirement information is uniformly vectorized, and key features are extracted using a preset business template and a dedicated matching algorithm to generate a perception result set. This fully realizes the intelligent and standardized collection and preprocessing of customer business requirement information. This approach effectively solves the problems of messy formatting, low efficiency of manual input, and inaccurate feature extraction in traditional requirement collection methods. At the same time, it overcomes the technical limitations of traditional interface receiving modes, such as passivity, protocol restrictions, and inability to process unstructured information. Relying on the template fusion capabilities of multiple systems, it achieves unified and standardized conversion of heterogeneous requirements, laying a solid data foundation for subsequent accurate semantic understanding and intelligent orchestration.

[0085] Figure 3 This is a flowchart illustrating the process of semantic understanding and intent analysis of customer business needs information, and transforming the customer business needs information into business delivery targets, provided by an embodiment of this application. It includes steps 301 to 302, which are described in detail below.

[0086] 301. Search the customer's business requirements information in a pre-built knowledge base to find one or more document fragments, which are more relevant to the customer's business requirements information than the remaining document fragments in the knowledge base.

[0087] This knowledge base is a business and network delivery knowledge base. It is a set of domain-specific knowledge that is pre-built by the system to support intent recognition. This knowledge base comprehensively includes network business knowledge of all professional telecommunications sectors, covering business knowledge, activation specifications, equipment configuration rules, interface call standards, SLA (Service Level Agreement) service specifications, etc. of various subnets such as transmission networks, IP networks, access networks, wireless networks, and cloud private networks.

[0088] The materials used to build the knowledge base include, but are not limited to: telecommunications service activation specifications and SLA documents, network equipment configuration manuals, API (Application Programming Interface) manuals for various professional network management systems (NMS) / EMS, and historical successful delivery case work orders. The collected raw knowledge data is cleaned, deduplicated, and segmented into independent, searchable knowledge fragments stored in semantic vector form. These fragments are then uniformly organized and stored as document blocks, with each document block corresponding to a complete piece of business knowledge content.

[0089] For example, the knowledge base can store various document blocks related to service and network delivery, such as the MPL leased line service activation process, 5G network slicing configuration rules, cloud-network interconnection service level protocol, and router QoS (Quality of Service) configuration command manual.

[0090] During the retrieval process in the knowledge base, the customer business requirement information (the structured perception result set R in this embodiment) after situational awareness processing is mapped into semantic vectors. Approximate retrieval is carried out in the knowledge base. By calculating vector similarity and using the ANN (Approximate Nearest Neighbor) search algorithm, the semantic relevance between the requirement vector and the knowledge fragment vectors in the knowledge base can be accurately calculated, and the set of n (n is an integer greater than 1) optimal document fragments with the highest relevance can be selected.

[0091] This approach abandons the traditional keyword matching mode and, based on global semantic matching capabilities, can effectively adapt to customers' needs scenarios that are colloquial, non-standard, and have different expressions, thereby improving the robustness and adaptability of intent recognition.

[0092] The specific algorithm logic of the network approximate target recognition technology in this application is as follows:

[0093] The system maps customer requirement text and knowledge base document text into a unified semantic vector E; it also presets a dedicated similarity metric function, Similarity, to quantify the semantic matching degree between requirements and knowledge fragments; and it retrieves and filters the optimal set of relevant document blocks.

[0094] The screening process can be calculated using the following formula:

[0095] (3)

[0096] Where Cs represents the document fragments selected from the database, E(R) represents the semantic vector obtained by mapping customer business requirements information, and E(Ki) represents the text of the i-th document in knowledge base K.

[0097] For example, if a customer needs to activate a dedicated cloud line, the most relevant document fragments (Cs) can be retrieved from the knowledge base, such as the cloud dedicated line service activation process, cloud gateway interconnection configuration specifications, and enterprise dedicated line performance indicator definitions.

[0098] 302. Input the customer's business requirements information and the document fragment into the large language model to generate the business delivery target.

[0099] The system takes customer business requirements information, highly relevant document fragments retrieved, and preset task instructions as input to construct prompts, which are then fed into the domain LLM (Large Language Model).

[0100] For example, the constructed prompt template could be: Based on the customer's original requirements: {customer requirement text}, and relevant business knowledge: {retrieved document fragments}, please generate a clear, structured, and executable business delivery target, which should include key attributes such as: business type, source and destination endpoints, bandwidth, latency, and reliability level.

[0101] The large language model performs comprehensive understanding, reasoning, and induction based on the input content, thereby transforming fuzzy and unstructured natural language requirements into clear, structured, and actionable business objectives. This process does not rely on manually configured fixed rules and achieves knowledge-enhanced semantic understanding and reasoning.

[0102] This application can employ a dedicated search-generation algorithm to drive large-scale model inference and accurately output standardized delivery targets. This dedicated search-generation algorithm can be represented by the following formula:

[0103] (4)

[0104] Where A represents the structured business delivery target; Cs is the document fragments selected from the database; and R represents the structured perception result set R. This represents the t-th item of the structured business delivery objective; This represents all preceding content before the t-th item of the structured business delivery objective; This means that, given all the known structured business delivery objectives generated before the tth term, the content of the tth term is generated.

[0105] The large language model combines the input customer business requirements information with highly relevant document fragments to perform deep semantic reasoning and content summarization, and outputs a structured business delivery target with a unified format that can be directly called by subsequent functional modules.

[0106] The business delivery objective is a clear and actionable task description that accurately corresponds to the customer's real business needs. It can include quantifiable and actionable content such as business type, professional subnets involved in network delivery, equipment type, network performance indicators, service scope, resource constraints, and delivery standards. The output format supports structured natural language or directly parsable data formats, which can be directly used as the basis for generating business objective orchestration strategies.

[0107] The data format can be JSON.

[0108] For example, given a customer's vague requirement to open a dedicated enterprise cloud line, after knowledge base retrieval and large-scale model inference, the final standardized business delivery objective is output as: "Business Delivery Objective: Establish an MPLS dedicated line connection for customer A from site 'Site_A' to cloud service provider 'Cloud_C' Beijing region 'VPC_X'. Requirements: bandwidth 1Gbps (Gigabits Per Second), end-to-end latency ≤50ms, availability ≥99.99%, requiring coordination with the access network, metropolitan area transmission network, and cloud private network to complete equipment configuration, IP connectivity, and service activation." This objective clearly defines the business type, involved network domain, equipment type, quantified performance indicators, and collaborative delivery scope, fully adapting to subsequent business orchestration processes.

[0109] In this embodiment, highly relevant document fragments are retrieved from a pre-built, comprehensive professional network knowledge base based on customer business requirements. Leveraging the powerful processing capabilities of a large-scale model, the requirements and document fragments are input into a large language model for deep reasoning, ultimately outputting standardized and structured business delivery targets. Unlike the rudimentary methods of traditional fixed-rule matching and keyword recognition, this approach, through knowledge augmentation and large-scale model reasoning, eliminates the need for customers to input standardized format requirements. It automatically parses colloquial and ambiguous requirements, accurately outputting delivery targets encompassing the entire network dimension. This overcomes the shortcomings of traditional technologies, such as poor accuracy in intent recognition, weak generalization ability, and reliance on human expert experience, providing precise core input for subsequent dynamic and intelligent business orchestration.

[0110] In one possible implementation, a business objective orchestration strategy, comprising action sequences and tool invocation strategies, is generated based on the business delivery objective to achieve that objective, including:

[0111] Based on the business delivery goal, a pre-built action model and tool model are invoked. Through a preset strategy planning and orchestration algorithm, the corresponding process execution logic is selected and combined from the action model, and the corresponding operation interface is matched from the tool model. This generates the process, steps, execution relationships between steps, and tool invocation strategy from the beginning to the end of the delivery process, so as to obtain the orchestration strategy for the business goal. Among them, the action model defines the process execution logic, and the tool model defines the operation interface of the external system or device to be invoked.

[0112] This pre-built action model can refer to a data structure or knowledge base that abstracts and defines the execution logic of a business process. Its purpose is to shield the specific technical details of the underlying network domain and standardize complex business processes into reusable logical units. This network domain can be a transport, IP, cloud, or other domains.

[0113] The action model includes the start node, end node, sequential execution logic, parallel execution logic, conditional branching logic, loop execution logic, and exception rollback logic of the process.

[0114] This tool model refers to a set of definitions that encapsulate the operation interfaces of external systems or devices to be called. Its source can be a digital mapping of various network management systems, cloud platform API (Application Programming Interface) documents, or device command-line manuals. The tool model specifically defines the tool object and its corresponding operation interface.

[0115] For example, the tool object can be a transmission network management system, IP router, cloud resource pool proxy gateway, etc., and the operation interface can adopt REST (Representational State Transfer) API address, CLI (Command-Line Interface) command template, SNMP (Simple Network Management Protocol) OID (Object Identifier), etc.

[0116] By invoking action models and tool models, the identified business delivery goals can be broken down into logical requirements and execution carrier requirements, providing basic data support for subsequent strategy generation.

[0117] For example, when the service delivery target requires configuring the access network first, then the transmission network, and rolling back if it fails, the action model defines the corresponding sequential execution and abnormal rollback logic templates.

[0118] This preset strategy planning and orchestration algorithm can be set in the strategy planning module. The input of the strategy planning module is the business delivery goal and the action model and tool model to be invoked, and the output is a preliminary business orchestration scheme. The function of this preset strategy planning and orchestration algorithm is to autonomously search for the optimal action-tool combination path under the goal constraints.

[0119] The pre-defined strategy planning and orchestration algorithm first selects process execution logic segments from the action model that match the performance indicators (such as bandwidth and latency) and business type in the business delivery objectives. For example, for leased line services with high reliability requirements, it automatically selects and combines logic segments that include dual-route protection and automatic failover. Based on the execution environment required by the selected logic segments, it matches specific operation interfaces from the tool model. In this process, the logical framework provided by the action model and the interface capabilities provided by the tool model work together. The action model determines the objectives and the order of execution operations, while the tool model provides the specific means of execution. Together, they achieve automatic mapping from abstract business objectives to specific executable instruction sets.

[0120] Figure 4 This is a schematic diagram of the strategy planning and orchestration algorithm provided in an embodiment of this application. The algorithm takes business delivery goals and customer business requirements as inputs and also invokes a pre-built knowledge base, action model, and tool model. The implementation of this algorithm includes the following steps:

[0121] 401. Large Language Model Analysis Process and Steps;

[0122] By searching the database for multiple document fragments that are highly relevant to the customer's business needs, and inputting these document fragments and the business needs information into the large language model, the business delivery target is obtained.

[0123] The large language model, based on the business delivery goals, invokes the action model to analyze the process and steps.

[0124] 402. Tool invocation strategy within the large language model analysis stage;

[0125] The large language model calls the tool model to implement the tool calling strategy within the analysis process.

[0126] 403. Integrate processes, steps, and tool invocation strategies to generate business objective orchestration strategies;

[0127] 404, Vector verification and adjustment of the generated process, steps, and tool calling strategies;

[0128] The adjusted results are fed back to step 401, and the large language model continues to analyze the results in conjunction with the adjusted results.

[0129] 405. Output business objective orchestration strategy.

[0130] In one possible implementation, the system receives the business delivery target and, in conjunction with the action model library and tool model library, dynamically matches and combines them through an internal orchestration algorithm to ultimately generate a strategy scheme containing a complete execution path.

[0131] For example, if the logic segment involves the activation of a transmission circuit, the algorithm will search all available transmission network management system interfaces in the tool model and match the optimal operation interface based on the current network resource status (such as port idle status).

[0132] A business objective orchestration strategy is a structured data object or collection of scripts that describes the entire lifecycle execution plan from business order receipt to network service activation.

[0133] From the beginning to the end of the delivery process, including the process steps, the execution relationships between the steps, and the tool invocation strategy, the process is the overall timeline of business delivery; the steps are the specific work stages divided in the process, such as resource verification, configuration distribution, business testing, etc.; the execution relationships between the steps clarify the dependencies or concurrency between the steps, such as step B can only start after step A is successful, or step C and step D can be executed in parallel.

[0134] The tool invocation strategy specifies the target interface in the target tool model to be invoked in each stage, as well as the specific format and values ​​of the input parameters. For example, the generated orchestration strategy specifies: the first stage calls the resource query interface to verify port availability; the second stage calls the IP network management configuration interface and the transmission network management circuit creation interface in parallel; and the third stage calls the end-to-end test interface to verify connectivity.

[0135] By solidifying discrete action logic and specific tool interfaces into an ordered orchestration strategy, complex operations that originally required manual cross-system coordination are transformed into standardized instruction flows that can be executed automatically by machines. This solves the problem of collaboration difficulties caused by system fragmentation in the traditional model and improves the efficiency and accuracy of business delivery.

[0136] In this embodiment, a decoupled action model and tool model are pre-built, and a preset strategy planning and orchestration algorithm is used to combine the action model and tool model, realizing hierarchical management of business logic and execution carrier. The action model defines the process execution logic, so that the same logic can be adapted to different professional networks, avoiding the tedious process of rewriting fixed scripts for each new business; the tool model defines the operation interface of the external system or device to be called, supporting plug-and-play for new devices or systems without modifying the core orchestration algorithm. Moreover, the preset strategy planning algorithm automatically searches and combines the optimal execution path according to the real-time business delivery target, which not only breaks down the silo barriers between different professional networks and realizes cross-domain end-to-end automatic collaboration, but also significantly reduces the strategy development cycle and maintenance cost, enabling efficient and accurate generation of executable business target orchestration strategies when facing diverse and rapidly changing customer needs.

[0137] Figure 5This is another flowchart of the intelligent service orchestration method provided in this application embodiment, including steps 501 to 502. After generating the target orchestration strategy, the orchestration strategy is modified. Between steps 103 and 104, these steps are described in detail below.

[0138] 501. Compare the generated orchestration strategy for this business objective with historical order data for similarity.

[0139] Comparing the generated business objective orchestration strategy with historical order data can be achieved by calculating the similarity between the newly generated strategy and past successful strategies stored in the historical order database. The similarity can be determined using a vector space model, mapping the strategy and order data to vector spaces and calculating Euclidean distance or cosine similarity. Alternatively, a semantic matching algorithm can be used to calculate the semantic similarity between the strategy and order data, and this semantic similarity can be used as the overall similarity between the two strategies.

[0140] This similarity comparison process not only compares the similarity of the overall process architecture, but also focuses on comparing whether the tool invocation strategies of key links are consistent, such as whether they all call the same SDN controller interface or configure the same bandwidth guarantee parameters. Through this multi-dimensional similarity comparison, the differences between the currently generated strategy and historically mature strategies that have been verified in the live network can be quantified, thereby identifying atypical combinations or unverified novel paths that may exist in the current strategy, providing data support for subsequent strategy risk assessment.

[0141] For example, suppose the currently generated business objective orchestration strategy is strategy P_new, and its vector representation is V_new; the historical order database contains N (N is an integer greater than 1) historical order data P_hist_1 to P_hist_N for the same type of dedicated line opening service, and their vector representations are V_hist_1 to V_hist_N respectively. The similarity score S_i between V_new and each V_hist_i is calculated one by one using the cosine similarity formula Cosine(V_new, V_hist_i) or the reciprocal of Euclidean distance.

[0142] 502. When the similarity is lower than the similarity threshold, the orchestration strategy for the business objective shall be modified based on the comparison results.

[0143] The similarity between each historical order data and the business target orchestration strategy is sorted, and the one with the highest similarity is selected as the closest historical order data. The orchestration strategy corresponding to the closest historical order data is the one that is closest to the target business orchestration strategy.

[0144] The similarity of the closest historical order data is compared with a threshold. If the similarity is lower than the threshold, the currently generated business objective orchestration strategy is deemed to have a high execution risk or lack sufficient historical experience to support it. Accordingly, the business objective orchestration strategy needs to be revised.

[0145] The similarity threshold can be dynamically set according to the complexity of the business scenario, the fault tolerance requirements, and the richness of historical data. For example, it can be set to 0.85 in core network slicing services and 0.75 in ordinary broadband access services. Once the similarity is determined to be below the threshold, the generated business objective orchestration strategy is corrected. This correction mechanism is based on the part with the greatest difference in the comparison results, and refers to the historical order data with the highest similarity to make targeted adjustments to the current business objective orchestration strategy.

[0146] The correction process includes analyzing the specific differences between the current strategy and the best historical strategy in terms of action sequence, execution logic, or tool selection. If the difference lies in a certain step calling a tool interface that was not frequently used in historical success cases, that tool can be replaced with a high-frequency and reliable tool from the corresponding step in the historical strategy. If the difference lies in the execution order not conforming to historical practices, the order of steps can be adjusted to conform to mature operating patterns.

[0147] For example, when the calculated similarity is below a similarity threshold, the historical order data with the highest similarity is selected as the best historical matching strategy. Features of this best historical matching strategy are extracted, and a comparison reveals that the business objective orchestration strategy uses a real-time bidding model in the resource reservation stage, while the best historical matching strategy uses a fixed quota model and has a very high historical execution success rate. In this case, the resource reservation action in the current strategy is modified to a fixed quota model, and the relevant parameter configurations are updated synchronously to generate a revised business objective orchestration strategy.

[0148] By making corrections based on historical experience, we can avoid execution failures caused by strategies that are too novel or deviate from the norm. We can use proven success patterns to calibrate new strategies, improve the robustness and feasibility of business objective orchestration strategies, and shorten the optimization cycle from strategy generation to successful execution.

[0149] In this embodiment, by comparing the generated business objective orchestration strategy with historical order data for similarity, and making corrections based on the comparison results when the similarity falls below a threshold, the orchestration strategy generation process is upgraded from simple logical reasoning to reasoning analogy. While the initially generated business objective orchestration strategy meets the customer's personalized needs, it may lack consideration for adaptability to the complex environment of the current network. Introducing comparison with historical order data allows for the immediate access to accumulated strategy assets, identifying risk points in the initially generated strategy that deviate from mature experience. Moreover, using highly similar historical success cases as a reference, the action sequence and tool invocation strategy of the initially generated strategy are fine-tuned, retaining not only the innovative parts that meet customer needs but also integrating historically validated stable execution logic. This ensures that the final issued business objective orchestration strategy possesses both the flexibility to respond to new requirements and the reliability proven over time, effectively reducing the probability of initial execution failure, reducing the cost of manual intervention and debugging, and promoting the continuous evolution and self-improvement of business orchestration capabilities.

[0150] In one possible implementation, the exception information includes process exception, interface call exception, execution failure, or resource scheduling exception;

[0151] This anomaly information is a real-time capture of abnormal internal system signals throughout the entire lifecycle of the operational business and network delivery processes. It can include process anomalies, interface call anomalies, execution failures, or resource scheduling anomalies, covering various fault scenarios from the logical layer to the physical resource layer. This enables comprehensive monitoring of any deviations from expectations during the business delivery process, providing complete input data for subsequent root cause analysis.

[0152] Among them, process anomalies can be caused by the disorder of the order, missing links, or infinite loops in the execution of business orchestration logic. For example, in the cloud dedicated line activation process, the logic is interrupted because the subsequent configuration instructions are forcibly triggered due to the failure to confirm the completion of the previous link.

[0153] Interface call exceptions can occur when calling external systems, such as connection timeouts, protocol parsing errors, or authentication failures. These external systems can be transport network management systems, IP network management systems, cloud platform APIs, etc. For example, calling an SDN controller interface might return an HTTP (Hypertext Transfer Protocol) 503 Service Unavailable status.

[0154] Among them, execution failure can be an error code explicitly returned by the target device or system after a specific operation command is issued, such as a switch port configuration command being rejected.

[0155] Resource scheduling anomalies can occur when allocating network bandwidth, computing resources, or storage resources, due to resource pool exhaustion, severe fragmentation, or resource locking conflicts, leading to allocation failures. For example, when creating a 5G network slice, the underlying physical resources may be unable to meet the required latency and bandwidth specifications.

[0156] Accordingly, feeding back the anomaly information to the business objective orchestration strategy to update the business objective orchestration strategy includes: matching and analyzing the anomaly information with a preset anomaly rule base to generate a structured anomaly perception result; and using the anomaly perception result to update the business objective orchestration strategy.

[0157] This pre-built anomaly rule base is a structured collection of knowledge that includes anomaly characteristics, potential root causes, impact assessments, and remediation suggestions. This rule base not only stores simple error code mappings but also incorporates the experiential knowledge of domain experts, transforming unstructured raw error messages into machine-understandable diagnostic conclusions.

[0158] The process of generating structured anomaly perception results may include: First, extracting key feature vectors from the captured raw anomaly information, such as error type, occurrence time, involved network element ID, stack information, etc.; Second, performing similarity matching or logical reasoning between the key feature vectors and entries in the anomaly rule base to find the most matching rule item; Finally, assembling the matched rule item into a structured object containing fields such as anomaly type, root cause, affected links, and recommended handling strategies.

[0159] After obtaining the structured anomaly detection results, the business objective orchestration strategy is adjusted and updated using these results. The update may include replacing only the faulty node, skipping the faulty link, adjusting the execution order, adjusting the execution parameters, or switching to backup resources. This application does not limit the specific content of the update.

[0160] For example, when an interface call timeout is detected and accompanied by a target device CPU (Central Processing Unit) utilization exceeding 95%, the structured anomaly detection result generated by matching the entry on device overload in the anomaly rule base is as follows: {Anomaly type: Interface call anomaly, Root cause: Target network management server CPU overload, Scope of impact: Current configuration distribution stage, Recommendation: Switch to backup network management node or delay retry}.

[0161] By structurally processing the results of anomaly detection, when feedback is received, there is no need to perform semantic parsing on messy logs again. Instead, local strategy adjustments can be made directly based on clear attributions and suggestions, thereby improving the system's self-healing response speed and accuracy, and avoiding resource waste and time delays caused by blindly rerunning the entire process.

[0162] In this embodiment, by clearly defining the scope of multi-dimensional anomaly information and combining it with a preset anomaly rule base for deep matching analysis, an intelligent upgrade of anomaly handling is achieved. Anomaly information is defined to include process anomalies, interface call anomalies, execution failures, and resource scheduling anomalies, enabling keen perception of various faults during the delivery process. Based on this, the anomaly rule base transforms raw, unstructured error information into structured anomaly perception results containing root cause diagnosis and repair suggestions. This not only reduces the information parsing burden in generating business objective orchestration strategies but also provides a high-quality, reasonable, and executable decision-making basis. It can pinpoint the root cause of the problem the instant anomaly is detected and dynamically update the business objective orchestration strategy accordingly. This solves the technical problems of relying on manual log checks, delayed response, and blind strategy updates in traditional solutions, enhancing the robustness and adaptability of the business delivery process.

[0163] In one possible implementation, the process of orchestrating business and network delivery according to the business objective orchestration strategy also includes:

[0164] When the error message is received and the task is re-executed, it is determined whether the tool in the current task of the business objective orchestration strategy has been completed previously; if it has been completed, the execution of the tool is skipped.

[0165] When the business and network delivery process is interrupted due to the triggering of abnormal information, and the re-execution mechanism is initiated according to the updated business objective orchestration strategy, the status of each tool call in the current sequence of tasks to be executed is verified.

[0166] The exception information comes from real-time captured data such as process exceptions, interface call exceptions, execution failures, or resource scheduling exceptions. The exception information is used to correct the initially generated business target orchestration strategy. After obtaining the corrected strategy, the retry logic for the orchestration strategy of the business target is triggered.

[0167] The tools in the current sequence of tasks to be executed can be operation interface units defined in the tool model, including but not limited to network device configuration distribution interfaces, cloud resource allocation interfaces, and proxy gateway capability call interfaces.

[0168] Upon receiving a re-execution instruction, the process doesn't blindly restart from the task sequence's beginning. Instead, it first reads a pre-defined state memory or execution log. This state memory records the final status flags of each tool call during the previous execution, with flag types including completed, in progress, failed, or not executed. The judgment process compares the tool identifier pointed to by the current task pointer with the record in the state memory. If a tool's status flag is "completed," it's determined that the tool was successfully executed previously; if it's marked as "failed" or "not executed," it's determined that the tool needs to be re-executed. By predicting based on state memory, successful steps before the failure point can be accurately identified, providing a basis for subsequent differentiated execution decisions and avoiding resource redundancy caused by full retries.

[0169] For example, in a task sequence that includes three tool calls: creating a virtual private line, configuring bandwidth policies, and activating ports, if a timeout exception occurs during the bandwidth policy configuration step, causing the process to be interrupted, the state memory will mark the creation of the virtual private line as completed, while the subsequent two tools will be marked as not executed or failed. When an exception feedback is received and the process is re-executed, the status of the creation of the virtual private line tool will be checked first to confirm that it has been completed.

[0170] If a specific tool in the current task has already reached the completed state in a previous execution, the system actively ignores any further calls to that tool and directly moves the execution pointer to the next incomplete task in the task sequence or terminates the processing of that subtask. Skipping means not sending any operation requests corresponding to that tool to the underlying network devices, cloud platforms, or external systems, and not consuming related computing resources or network bandwidth.

[0171] When the judgment logic outputs a "yes" conclusion, it indicates that the tool in the current task has been previously executed. The execution engine removes the tool from the current execution queue, or sets a breakpoint in the execution flow control to directly jump to the first tool whose status flag is not "executed." This approach achieves idempotency control of the execution flow, meaning that regardless of how many times it is retried, successfully executed operations will not produce side effects, and there will be no conflict errors caused by duplicate resource creation.

[0172] For example, once it is confirmed that the virtual private line creation tool has been completed, the tool's calling code will be skipped directly, and the creation request will not be re-initiated. Instead, the process will start from the failed node of configuring the bandwidth policy and attempt to re-execute.

[0173] By re-executing tasks according to the business objective orchestration strategy, it determines whether the tools in the current task have already been completed previously. Combined with skipping the execution of tools that have already been completed, this allows for corrective execution only on the truly failed aspects during anomaly recovery. This shortens the anomaly recovery time window and improves the service level agreement (SLA) assurance capabilities for business delivery. Simultaneously, it avoids secondary failures caused by repeated issuance of the same configuration commands, such as device status conflicts, resource contention, or data inconsistencies, enhancing the stability and robustness of the entire business and network delivery system.

[0174] In this embodiment, by introducing a real-time judgment and skipping mechanism for the historical execution status of tools during the re-execution phase, the traditional overall rollback and retry mode is upgraded to a breakpoint-resumption intelligent recovery mode. With the help of immediate response to anomaly information and automatic filtering of completed tools, it can not only quickly recover from failures but also ensure that the delivered results are not destroyed or repeatedly consumed. Combined with an orchestration strategy based on anomaly information updates, the entire delivery process possesses dynamic adaptability and high reliability, fundamentally solving the inefficiency and resource waste caused by the complete rerun of the process due to partial failures in traditional automated workflows. Ultimately, it achieves efficient, stable, and self-healing intelligent business orchestration.

[0175] Figure 6 This is a flowchart illustrating the task execution process in the intelligent service orchestration method provided in this application embodiment, including the following steps:

[0176] 601. Execute tools sequentially according to the tool invocation business objective orchestration strategy in the process;

[0177] 602. Re-execute after checking for any abnormalities;

[0178] If no re-execution occurs, continue with error 603; if a re-execution occurs, continue with error 609.

[0179] 603. Execution tools;

[0180] Continue executing the tools in the order specified in the target business orchestration strategy.

[0181] 604. Search for tool models by tool name;

[0182] 605. Select the tool object and all operations of that tool;

[0183] 606. Select the operation for this tool according to the tool invocation strategy;

[0184] The target business orchestration strategy can specify the tool invocation strategy or preset the default invocation strategy. The tool invocation strategy is used to select the operation of the tool.

[0185] 607. Assemble the input parameters according to the operation and execute the interface corresponding to the operation;

[0186] 608. Execution complete;

[0187] The tool has finished executing.

[0188] 609. Determine if the previous execution has been completed;

[0189] "Previous execution completed" means that the tool in the current task has been previously executed. If it has been previously executed, skip the execution of the tool and proceed directly to step 610; if it has not been previously executed, proceed to step 603.

[0190] 610. Tool execution complete;

[0191] 611. Determine if all tools have completed execution;

[0192] After each tool completes its execution, determine whether all tools in the target business orchestration strategy have completed their execution. If all tools have completed their execution, the task ends; otherwise, return to step 601.

[0193] Corresponding to the aforementioned intelligent service orchestration method embodiments, this application also provides a system that applies the intelligent service orchestration method.

[0194] Figure 7 This is a schematic diagram of the structure of an intelligent service orchestration system provided in an embodiment of this application, such as... Figure 7 As shown, the system includes: a situational awareness module 701, an intent recognition module 702, a strategy planning module 703, a target execution module 704, and a self-awareness module 705;

[0195] Among them, the situational awareness module 701 is used to acquire customer business requirement information;

[0196] The situational awareness module is a logical functional unit responsible for collecting and processing customer business demand information from outside the system in real time. As the window through which the system perceives external business demands, the situational awareness module can receive input from at least one of the following methods: text input, voice input, or application programming interface (API). This application embodiment does not impose any special limitations on this.

[0197] The situational awareness module proactively senses changes in external systems and supports the input of unstructured information. It includes a data collector and a preprocessing unit. The data collector gathers order information, customer information, and network demand information from customer systems across different channels. The preprocessing unit vectorizes and stores the collected information, extracting key information using predefined business templates to form a perception result set. The module's function is to transform dispersed, multi-source external inputs into standardized cognition that the system can understand, providing an accurate data foundation for subsequent intent recognition.

[0198] Among them, the intent recognition module 702 is used to perform semantic understanding and intent analysis on the collected customer business demand information and transform it into business delivery goals;

[0199] The intent recognition module receives the perception result set output by the situational awareness module and uses natural language processing technology and large language models to perform deep semantic analysis to determine the user's true intent and generate clear business delivery goals. This intent recognition module works in conjunction with the situational awareness module, receiving its output perception result set and interacting with a pre-built knowledge base.

[0200] The intent recognition module searches for relevant document fragments in the knowledge base using the perceived results set. These fragments, along with the original requirement information, are then input into the large language model to generate specific business delivery goals, such as determining the performance indicators of the leased line, the types of subnets involved, and the types of equipment. The role of the intent recognition module in the overall technical solution is to transform vague user requirements into precise, executable goals, solving the problem of traditional methods that require standardized, formatted requirements to be input for service activation.

[0201] Among them, the strategy planning module 703 is used to generate a business objective orchestration strategy that includes action sequences and tool invocation strategies to achieve the business delivery objective based on the business delivery objective.

[0202] The strategy planning module is the core processing unit that intelligently generates processes, steps, and tool invocation strategies based on the business delivery goals generated by the intent recognition module, combined with action models and tool models, and through preset strategy planning and orchestration algorithms.

[0203] The strategy planning module is responsible for deducing the sequence of actions required to achieve the goal. The strategy planning module works closely with the intent recognition module and the goal execution module. The strategy planning module receives the business delivery goal output by the intent recognition module, calls the action model that defines the process execution logic and the tool model that defines the operation interface, and generates a complete orchestration strategy from the beginning to the end of the delivery process.

[0204] In addition, the strategy planning module has dynamic correction capabilities, which can compare the generated strategy with historical order data for similarity and make corrections when the similarity is below a threshold. Moreover, it also receives abnormal information from the self-awareness module and updates or replans the current business objective orchestration strategy accordingly to ensure the adaptability and robustness of the delivery process.

[0205] The strategy planning module may be composed of a strategy generation subunit, a history verification subunit, and a dynamic adjustment subunit, wherein the strategy generation subunit is used for initial process planning, the history verification subunit is used for comparison and correction, and the dynamic adjustment subunit is used for responding to abnormal feedback; it may also include a reasoning engine based on a large language model, which directly generates strategies based on the target and knowledge base; or it may include a hybrid architecture of rule engine and machine learning model, which handles deterministic processes and uncertain decisions respectively. This application does not impose special limitations on the specific internal architecture of the strategy planning module.

[0206] The target execution module 704 is used to run the service and network delivery process according to the service target orchestration strategy in order to complete the delivery of services and network.

[0207] The target execution module is responsible for parsing the business target orchestration strategy generated by the strategy planning module and converting it into specific executable instructions to drive the underlying network or business system to complete the actual delivery operation.

[0208] The target execution module translates abstract strategies into physical or logical resource configuration and distribution. Working in conjunction with the strategy planning module, it receives the orchestration strategy, executes each step of the process sequentially, and locates the tool model based on the tool invocation strategy, connecting to the corresponding system or network device interface for operation. During operation, the target execution module also has a fault tolerance mechanism. When it receives a re-execution command triggered by an exception, it checks whether the tool in the current task has already been completed previously. If so, it skips the execution of that tool, avoiding resource waste or state conflicts caused by repeated operations. The specific implementation of the target execution module can be a distributed task scheduling cluster or an agent program embedded in various professional network management systems; this application embodiment does not impose any special limitations on this.

[0209] Among them, the self-awareness module 705 is used to monitor and capture abnormal information inside the system in real time during the operation of the business and network delivery process, and to feed back the abnormal information to the strategy planning module.

[0210] The strategy planning module is also used to update the current business objective orchestration strategy based on the abnormal information received.

[0211] The self-awareness module is a monitoring component that monitors the internal operating status of the system in real time throughout the entire lifecycle of business and network delivery processes, capturing and structurally processing various abnormal information. This self-awareness module works in parallel with or is embedded in the target execution module, and its monitoring scope covers a variety of situations, including process anomalies, interface call anomalies, execution failures, and resource scheduling anomalies.

[0212] The self-awareness module detects deviations in process execution and matches the captured anomaly information with a pre-defined anomaly rule base to generate structured anomaly detection results. These results are then fed back to the strategy planning module, forming a closed-loop control chain. Through this coordinated approach, the self-awareness module enables the system to automatically detect internal faults and trigger strategy adjustments without manual intervention, achieving stable and rapid execution of business processes and self-healing capabilities.

[0213] By constructing an intelligent closed-loop orchestration architecture driven by both situational awareness and self-awareness, the system combines dynamic perception of external customer needs with real-time monitoring of internal operational anomalies, breaking through the limitations of traditional systems in cross-domain collaboration, rigid execution, and passive response.

[0214] In the intelligent service orchestration system of this embodiment, the situational awareness module first collects and parses external customer service requirements, forming a perception result set which is then passed to the intent recognition module. The intent recognition module uses a knowledge base and a large language model to transform the requirements into clear service delivery goals. Based on these goals, the strategy planning module generates an initial service goal orchestration strategy by combining action and tool models, and can verify and correct it using historical data. The goal execution module drives the underlying system to execute the service and network delivery process according to this strategy. During this process, the self-awareness module monitors the execution status in real time, and once an anomaly is detected, it immediately generates an anomaly perception result and feeds it back to the strategy planning module. The strategy planning module updates or re-plans the original strategy based on the anomaly information and reissues it to the goal execution module until service delivery is completed. This solves the problems of rigid business process strategies, difficulties in cross-domain collaboration, poor fault tolerance, and over-reliance on human experience in existing technologies, achieving efficient, flexible, intelligent, and robust service and network delivery.

[0215] To facilitate understanding of this solution, this application also provides an application scenario embodiment of the intelligent service orchestration method, the specific process of which is as follows:

[0216] Suppose a corporate customer requests a dedicated line via voice input, requiring 10G bandwidth and less than 20ms latency to connect data centers in Beijing and Shanghai.

[0217] The situational awareness module collects the voice information, converts it into a vector, and extracts key information such as origin: Beijing, destination: Shanghai, bandwidth: 10G, and latency: <20ms through a predefined template to form a perception result set.

[0218] The intent recognition module receives the result set, retrieves relevant cloud leased line delivery document fragments from the knowledge base, and combines them with large language model analysis to generate specific business delivery targets, including the optical transmission path, IP address planning, and equipment configuration parameters involved.

[0219] The strategy planning module receives the target, calls the action model and tool model, and automatically generates an orchestration strategy that includes resource query, path calculation, configuration distribution, business testing, etc., and distributes it after confirming that it is correct by comparing with historical data.

[0220] The target execution module executes each step sequentially, calling the corresponding network management interface to configure the device.

[0221] If, during the configuration distribution phase, the self-awareness module detects a timeout in an interface call on a router (as an exception), it immediately reports the exception to the policy planning module.

[0222] The strategy planning module determined that there was temporary network congestion based on the abnormal rule base, so it updated the strategy, added a retry mechanism and switched to the backup path, and then sent it to the target execution module for execution again, and finally successfully completed the dedicated line opening.

[0223] Because of the dual-drive mechanism consisting of a situational awareness module and a self-awareness module, it can simultaneously cope with the diversity of external demands and the uncertainty of internal operation, solving the problems of traditional systems' inability to collaborate across domains and lack of self-healing capabilities. Because the strategy planning module has the ability to dynamically adjust based on historical data verification and anomaly feedback updates, it can generate flexible and robust orchestration strategies, avoiding the rigidity and maintenance difficulties caused by hard-coded processes. Because the target execution module has task status judgment and skipping mechanisms, it can effectively avoid repeated operations in abnormal re-execution scenarios, improving delivery efficiency and system stability.

[0224] This application also provides a product including a computer program, which, when run, enables an electronic device to implement any of the intelligent service orchestration methods provided in this application.

[0225] This application also provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it enables an electronic device to implement any of the intelligent service orchestration methods provided in this application.

[0226] In the several embodiments provided in this application, 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; multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interface, or the indirect coupling or communication connection of the apparatus or unit may be electrical, mechanical, or other forms.

[0227] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0228] In summary, the above description is merely a preferred embodiment of the technical solution of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An intelligent service orchestration method, characterized in that, include: Obtain information about customer business needs; The customer business needs information is semantically understood and intent-based, and then transformed into business delivery objectives. Based on the aforementioned business delivery objectives, a business objective orchestration strategy, including action sequences and tool invocation strategies, is generated to achieve these objectives. Based on the orchestration strategy for the stated business objectives, run the service and network delivery processes to complete the delivery of services and networks; During the operation of the business and network delivery process, abnormal information within the system is monitored and captured in real time, and the abnormal information is used to update the business objective orchestration strategy.

2. The intelligent service orchestration method according to claim 1, characterized in that, The acquisition of customer business needs information includes: The customer business requirements information is collected through at least one of text input, voice input, or application programming interface. Obtain the vector corresponding to the customer's business requirements information; Using predefined business templates, key information is extracted from the vector to form a perception result set.

3. The intelligent service orchestration method according to claim 1 or 2, characterized in that, The step of performing semantic understanding and intent analysis on the customer business requirement information, and transforming the customer business requirement information into business delivery objectives, includes: The customer business requirement information is searched in a pre-built knowledge base to find one or more document fragments, wherein the relevance of the one or more document fragments to the customer business requirement information is higher than that of the remaining document fragments in the knowledge base; The customer business requirements information and the document fragment are input into the large language model to generate the business delivery target.

4. The intelligent service orchestration method according to claim 1 or 2, characterized in that, The process of generating a business objective orchestration strategy, which includes action sequences and tool invocation strategies, based on the business delivery objective, to achieve that objective includes: Based on the business delivery objective, a pre-built action model and tool model are invoked. Through a preset strategy planning and orchestration algorithm, the corresponding process execution logic is selected and combined from the action model, and the corresponding operation interface is matched from the tool model. This generates the process, steps, execution relationships between steps, and tool invocation strategy from the beginning to the end of the delivery process, in order to obtain the business objective orchestration strategy. The action model defines the process execution logic, and the tool model defines the operation interface of the external system or device to be invoked.

5. The intelligent service orchestration method according to claim 4, characterized in that, Also includes: The generated business objective orchestration strategy is compared with historical order data for similarity. When the similarity is lower than the similarity threshold, the orchestration strategy for the business objectives is modified based on the comparison results.

6. The intelligent service orchestration method according to claim 1 or 2, characterized in that, The abnormal information includes process abnormalities, interface call abnormalities, execution failures, or resource scheduling abnormalities. Feeding back the anomaly information to the generation of the business objective orchestration strategy, so as to update the business objective orchestration strategy, includes: By matching and analyzing the abnormal information with a preset abnormal rule base, a structured abnormal perception result is generated; The anomaly detection results are then used to update the business objective orchestration strategy.

7. The intelligent service orchestration method according to claim 1 or 2, characterized in that, The process of orchestrating services and network delivery according to the business objective orchestration strategy also includes: When the error message is received and the task is re-executed, it is determined whether the tool in the current task of the business objective orchestration strategy has been completed previously; and if it has been completed, the execution of the tool is skipped.

8. An intelligent business orchestration system, characterized in that, include: The situational awareness module is used to acquire information about customer business needs; The intent recognition module is used to perform semantic understanding and intent analysis on the collected customer business demand information, and to transform the customer business demand information into business delivery targets. The strategy planning module is used to generate a business objective orchestration strategy, which includes action sequences and tool invocation strategies, to achieve the business delivery objective based on the business delivery objective. The target execution module is used to run the service and network delivery process according to the service target orchestration strategy to complete the delivery of services and network. The self-awareness module is used to monitor and capture abnormal information within the system in real time during the operation of the business and network delivery process, and to feed the abnormal information back to the strategy planning module. The strategy planning module is also used to update the current business objective orchestration strategy based on the abnormal information after receiving the abnormal information.

9. A computer program product, characterized in that, Includes a computer program, which, when run, causes the intelligent service orchestration method as described in any one of claims 1 to 7 to be executed.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the intelligent service orchestration method as described in any one of claims 1 to 7.

Images