CN-122022941-A - Service processing system based on heterogeneous computing power

CN122022941ACN 122022941 ACN122022941 ACN 122022941ACN-122022941-A

Abstract

The invention provides a business processing system based on heterogeneous computing power, which comprises a computing power scheduling component, a knowledge engineering component, a large model component and an intelligent decision component, wherein the computing power scheduling component builds a uniform resource model, maps heterogeneous computing power resources into standard computing units and distributes corresponding standard computing units for computing tasks, the knowledge engineering component builds a multilingual knowledge graph, extracts incremental knowledge from business interaction data and updates the incremental knowledge to the multilingual knowledge graph, the large model component trains the universal large model according to corpus data in a target field to generate a vertical field large model, the intelligent decision component receives business instructions, invokes the vertical field large model to combine the multilingual knowledge graph to perform reasoning to generate an executable decision flow, and invokes an external business system to perform specific operation according to the executable decision flow and returns execution result feedback data. The invention improves the utilization rate of the whole computing power and the cross-language business processing capacity.

Inventors

Geng Xiaoqiao
CUI HONGZHI
Shen linjiang
QIU SHUQING

Assignees

浪潮通信信息系统有限公司

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (10)

1. The service processing system based on heterogeneous computing power is characterized by comprising a computing power scheduling component, a knowledge engineering component, a large model component and an intelligent decision component, wherein: The computing power scheduling component is used for constructing a uniform resource model, mapping heterogeneous computing power resources into standard computing units under the uniform resource model, and distributing the corresponding standard computing units for computing tasks in the knowledge engineering component, the large model component and the intelligent decision component; The knowledge engineering component is used for constructing a multilingual knowledge graph, extracting incremental knowledge from service interaction data and updating the incremental knowledge to the multilingual knowledge graph; The large model component is used for training the general large model according to the corpus data of the target field to generate a vertical field large model; the intelligent decision component is used for receiving service instructions, calling the vertical domain big model to perform reasoning by combining the multilingual knowledge graph so as to generate an executable decision flow, calling an external service system to perform specific operation according to the executable decision flow, and returning feedback data of an execution result to the knowledge engineering component and the big model component.
2. The heterogeneous computing force based business processing system of claim 1, wherein the computing force scheduling component comprises a heterogeneous computing force provisioning engine to: configuring standard resource access interface, loading computing power adapting plug-in to map different computing resource parameters into standard computing units in the uniform resource model, and And detecting the resource idle state and the network connection state of the edge equipment through the lightweight agent deployed on the edge equipment, and registering the edge equipment computing power meeting the preset condition as an edge computing service node in the uniform resource model.
3. The heterogeneous computing power based business processing system of claim 1, wherein the computing power scheduling component further comprises an intelligent computing power scheduling engine for: receiving a task to be processed, and extracting task feature data of the task to be processed, wherein the task feature data comprises a computing density parameter and a data compliance label; acquiring computing unit characteristic data of a currently available standard computing unit, wherein the computing unit characteristic data comprises resource state data, network connection data, energy consumption cost data and a data attribution identifier; and sending the task to be processed to a matched target available standard computing unit according to the task characteristic data and the computing unit characteristic data.
4. The heterogeneous computing power based business processing system of claim 1, wherein the knowledge engineering component comprises a global data flywheel engine for: Acquiring implicit service data, and carrying out stream training on the corresponding vertical field large model according to the implicit service data; and acquiring explicit service data, performing privacy field filtering and multilingual semantic alignment processing on the explicit service data, and performing off-line full-scale fine adjustment on the vertical domain large model according to the processed explicit service data.
5. The heterogeneous computing power based business processing system of claim 1, wherein the knowledge engineering component further comprises a knowledge engineering engine for: Calling a translation management system to translate source language data into a text of a target language, writing the text back to a corresponding language branch, identifying a changed target text, and performing incremental semantic alignment on the target text; extracting a triplet in the text according to the entity recognition model, and storing the triplet to the multilingual knowledge graph, wherein the triplet comprises an entity, a relationship and an attribute, and And when the vertical field large model receives input inquiry, searching associated nodes and document fragments matched with the input inquiry in the multilingual knowledge graph, and combining the associated nodes and the document fragments into prompt word context to input the vertical field large model.
6. The heterogeneous computing power-based business processing system of claim 1, wherein the large model component comprises a corpus construction module, a directional fine tuning module, a word segmentation optimization module and a reinforcement learning module, wherein: the corpus construction module is used for constructing corpus data in the target field; the directional fine tuning module is used for carrying out parameter fine tuning on the pre-trained general large model according to the corpus data of the target field to generate a vertical field large model; The word segmentation optimization module is used for carrying out word segmentation on the corpus data in the target field based on a text segmentation rule; the reinforcement learning module is used for acquiring a reward signal according to feedback data in a service scene and carrying out iterative optimization on the vertical field large model based on the reward signal.
7. The heterogeneous computing power based business processing system of claim 1, wherein the intelligent decision component comprises an agent infrastructure module comprising a perception unit, a decision unit, an execution unit, and a learning unit, wherein: The sensing unit is used for receiving and processing multi-mode data from the target field and generating an environmental state representation; the decision unit is used for generating an executable decision flow corresponding to the service instruction according to the environmental state representation; the execution unit is used for analyzing the executable decision flow, calling an external service system to execute specific operation through a predefined interface protocol, and monitoring the execution state; The learning unit is used for collecting feedback data of the external service system and updating decision parameters of the decision unit according to the feedback data.
8. The heterogeneous computing power based business processing system of claim 7, wherein the intelligent decision component further comprises a tool chain module for: establishing a standard communication protocol between the agent and the external business system and maintaining a tool registry to manage call interfaces and parameter configuration of the external tool, and And establishing an isolated execution environment, pre-running the specific operation in the isolated execution environment before the execution unit calls an external service system, and performing risk verification and blocking on the specific operation based on a security policy.
9. The heterogeneous computing power based business processing system of claim 1, wherein the intelligent decision component further comprises an agent operation and maintenance module and an agent assessment module, wherein: the intelligent agent operation and maintenance module is used for providing a visual arrangement interface to configure workflow logic of the intelligent agent and carrying out deployment management and operation log analysis; the agent evaluation module is used for constructing a multi-dimensional evaluation system and generating optimization suggestion data based on an evaluation result corresponding to the multi-dimensional evaluation system.
10. The heterogeneous computing force based business processing system of any of claims 1 to 9, wherein the system further comprises an application execution component comprising: The supply chain management agent is used for generating an inventory optimization strategy based on historical sales data and market trend data and calling a rule knowledge base to execute compliance verification operation; The intelligent customer service intelligent body is used for identifying the consultation intention of the multilingual user according to the multilingual knowledge graph and outputting response data of corresponding languages; the intelligent logistics intelligent body is used for determining an optimal transportation path according to real-time state data of a logistics server and adjusting the optimal transportation path according to real-time road condition data; The intelligent recommendation agent is used for generating a commodity recommendation strategy according to the matching degree of the user preference characteristics and the commodity characteristics and recommending commodities according to the commodity recommendation strategy.

Description

Service processing system based on heterogeneous computing power Technical Field The invention relates to the technical field of artificial intelligence and cloud computing, in particular to a service processing system based on heterogeneous computing power. Background Under the globalization background, some enterprises have huge market scale and potential consumer groups, and the enterprises can realize the rapid growth of the business by expanding the global market. With the popularity of large model technology and large data applications, enterprises need to continually train and deploy more complex models, which dramatically increase the demand for computing resources. Currently, some businesses are relatively weak in infrastructure in a partial area, and local computing power often cannot independently support the high computing demands of the business. Meanwhile, the existing general large model and data analysis tool have the problem of multilingual insufficient coverage and cultural context understanding deviation, and are difficult to support the effective processing and decision assistance of enterprises on multilingual business data. Disclosure of Invention The invention provides a business processing system based on heterogeneous computing power, which is used for solving the technical problems of insufficient local computing power and difficult cross-language and cross-cultural data processing of enterprises in partial areas in the prior art. The invention provides a service processing system based on heterogeneous computing power, which comprises a computing power scheduling component, a knowledge engineering component, a large model component and an intelligent decision component, wherein: The computing power scheduling component is used for constructing a uniform resource model, mapping heterogeneous computing power resources into standard computing units under the uniform resource model, and distributing the corresponding standard computing units for computing tasks in the knowledge engineering component, the large model component and the intelligent decision component; The knowledge engineering component is used for constructing a multilingual knowledge graph, extracting incremental knowledge from service interaction data and updating the incremental knowledge to the multilingual knowledge graph; The large model component is used for training the general large model according to the corpus data of the target field to generate a vertical field large model; the intelligent decision component is used for receiving service instructions, calling the vertical domain big model to perform reasoning by combining the multilingual knowledge graph so as to generate an executable decision flow, calling an external service system to perform specific operation according to the executable decision flow, and returning feedback data of an execution result to the knowledge engineering component and the big model component. According to the service processing system based on heterogeneous computing force, the computing force scheduling component comprises a heterogeneous computing force supply engine, wherein the heterogeneous computing force supply engine is used for: configuring standard resource access interface, loading computing power adapting plug-in to map different computing resource parameters into standard computing units in the uniform resource model, and And detecting the resource idle state and the network connection state of the edge equipment through the lightweight agent deployed on the edge equipment, and registering the edge equipment computing power meeting the preset condition as an edge computing service node in the uniform resource model. According to the service processing system based on heterogeneous computing power, the computing power scheduling component further comprises an intelligent computing power scheduling engine, wherein the intelligent computing power scheduling engine is used for: receiving a task to be processed, and extracting task feature data of the task to be processed, wherein the task feature data comprises a computing density parameter and a data compliance label; acquiring computing unit characteristic data of a currently available standard computing unit, wherein the computing unit characteristic data comprises resource state data, network connection data, energy consumption cost data and a data attribution identifier; and sending the task to be processed to a matched target available standard computing unit according to the task characteristic data and the computing unit characteristic data. According to the business processing system based on heterogeneous computing power provided by the invention, the knowledge engineering component comprises a global data flywheel engine, and the global data flywheel engine is used for: Acquiring implicit service data, and carrying out stream training on the corresponding vertical field large model according to the implicit