CN-121981571-A - Intelligent management optimization system and method based on exploration big data

CN121981571ACN 121981571 ACN121981571 ACN 121981571ACN-121981571-A

Abstract

The invention discloses an intelligent management optimization system and method based on exploration big data, the system builds a data base of 'lake and warehouse integration', and innovatively introduces a dynamic business object model to organize multi-source heterogeneous data. The core is that a cloud primary scheduling engine drives a plurality of agents (Multi-agents) to work cooperatively, so that automatic decomposition and execution of scientific research tasks are realized. In the method, a model training paradigm of federal learning and meta learning is adopted, the model generalization capability under a small sample scene is improved while the data privacy is ensured, and a multi-objective optimization model of 'risk-cost-benefit' is constructed for decision making. The invention realizes the integrated closed loop of exploration data management, knowledge discovery and decision optimization, and remarkably improves the intelligent level and comprehensive benefit of exploration and development.

Inventors

PENG JING
PENG HAILONG
MA YONGXIN
GONG YUNLEI
XIE MINGTAO

Assignees

中海石油(中国)有限公司湛江分公司

Dates

Publication Date: 20260505
Application Date: 20260119

Claims (10)

1. An intelligent management optimization system based on exploration big data is characterized by comprising, The lake and warehouse integrated data base adopts an open table format to uniformly manage structured and unstructured data, and models and associates core business entities based on a dynamic business object model; The cloud primary intelligent scheduling engine is used for calculating the elastic expansion of resources and the task arrangement and coordination of multiple agents based on a container arrangement technology; the model factory integrating federal learning and meta learning is used for cooperatively training a global model on the premise that data cannot go out of a domain and rapidly adapting to a new exploration block; And the multi-objective intelligent decision center integrates multi-dimensional indexes of risk, cost and benefit and provides a pareto optimal solution set as decision support.
2. The intelligent management optimization system based on exploration big data according to claim 1, wherein the dynamic business object model describes business entities through an extensible Schema definition language, and utilizes association relations among storage entities of a graph database to support response of business requirements and evolution of data relations.
3. The intelligent management optimization system based on exploration big data according to claim 1 or 2, wherein the multi-Agent comprises, but is not limited to, a data quality sensing Agent, a geophysical prospecting explanation special Agent, a logging well explanation special Agent, a risk early warning Agent and a parameter optimization Agent, and each Agent performs asynchronous communication and cooperation through a publish-subscribe mechanism.
4. The intelligent management optimization system based on exploration big data according to claim 3, wherein the cloud native intelligent scheduling engine is constructed based on Kubernetes and KubeFlow technology stacks and achieves cooperation of multiple agents through event-driven dynamic workflow.
5. The intelligent management optimization system based on exploration big data of claim 4, wherein the cloud native intelligent scheduling engine comprises: The workflow composer automatically generates a directed acyclic graph according to the task type, and defines the execution sequence and the dependency relationship of each Agent; The resource scheduler monitors the cluster load in real time, dynamically allocates CPU/GPU resources according to the workflow progress, and realizes the flexible extension and contraction of computing resources; the fault-tolerant controller realizes automatic migration and task recovery of the fault node through heartbeat detection and state snapshot, and ensures high availability of the system.
6. The intelligent management optimization system based on exploration big data according to claim 4, wherein the engine adopts a publish-subscribe communication mode, and issues an event when the data quality sensing Agent finds new data ready, so as to trigger the subsequent risk early warning Agent to start working.
7. An intelligent management optimization method based on exploration big data is characterized by utilizing the intelligent management optimization system based on exploration big data as set forth in any one of claims 1 to 6, comprising the following steps, S1, constructing an enterprise-level data asset view based on a lake and warehouse integrated data base and a dynamic business object model; S2, receiving user tasks and automatically decomposing the user tasks through a cloud primary intelligent scheduling engine, and scheduling corresponding multi-agent cooperation to be completed; s3, generating an AI model suitable for a specific scene by utilizing a model factory integrating federal learning and meta learning, and deploying the AI model to corresponding agents; s4, introducing a multi-objective optimization algorithm, and constructing a risk-cost-benefit balance model; And S5, displaying a multi-objective optimization result through an interactive visual interface, and assisting a decision maker in scheme selection and decision.
8. The intelligent management optimization method based on exploration big data according to claim 7, wherein the step S3 comprises the following steps, S31, each participant locally trains a model by using private data; S32, only uploading model parameters to an aggregation server; And S33, introducing a meta learner, extracting meta knowledge from the model training process of a plurality of exploration blocks, and rapidly generating a high-performance initial model with only a small amount of data in a new block.
9. The intelligent management optimization method based on exploration big data according to claim 8, wherein in the step S4, the multi-objective optimization algorithm adopts a rapid non-dominant ordering genetic algorithm with elite strategy, and an objective function of the multi-objective optimization algorithm simultaneously minimizes operation risk and cost and maximizes predicted recovery ratio.
10. The intelligent management optimization method based on exploration big data according to claim 7, further comprising feedback reinforcement closed loop, wherein actual decision effect is compared with a prediction index, and difference data are used for triggering retraining of a model factory and fine adjustment of a business object model, so that continuous iterative optimization of a system is achieved.

Description

Intelligent management optimization system and method based on exploration big data Technical Field The invention belongs to the technical field of oil and gas exploration and development information, and particularly relates to an intelligent management optimization system and method based on exploration big data. Background Currently, the oil and gas exploration industry is working on digital conversion, but still faces many challenges. On the data level, although the data lake concept is widely accepted, static, stiff data models are difficult to adapt to rapidly changing business requirements, resulting in high data governance costs. On the technical level, a single AI model does not perform well when facing small sample scenes such as new areas, new layers and the like, and the 'model island' formed by each professional software prevents the collaborative optimization across disciplines. At the decision level, the existing system is optimized from a single dimension (such as lowest risk), and multiple objectives of cost reduction, synergy and security of enterprises are difficult to meet. In practical applications, some companies have progressed in intelligent seismic interpretation, but their models have severely relied on large amounts of annotation data. In new area exploration, obtaining sufficiently high quality samples is expensive and time consuming. Meanwhile, the DELFI cognitive exploration and development environment of part of companies realizes multidisciplinary collaboration, but the core is centralized data and model management, and the bottlenecks of data privacy and network transmission exist for companies with a plurality of independent operation bases or related to sensitive data. Therefore, a new intelligent system which can realize global optimization and respect data privacy and adapt to agile changes of business is urgently needed. Disclosure of Invention The invention aims to provide an intelligent management optimization system and method based on exploration big data, the data model of the method is flexible, the generalization capability of the AI model is strong in a small sample scene, and multi-objective collaborative optimization can be performed. In order to solve the technical problems, the technical scheme adopted by the invention is that the intelligent management optimization system based on the exploration big data comprises, The lake and warehouse integrated data base adopts an open table format to uniformly manage structured and unstructured data, and models and associates core business entities based on a dynamic business object model; The cloud primary intelligent scheduling engine is used for calculating the elastic expansion of resources and the task arrangement and coordination of multiple agents based on a container arrangement technology; the model factory integrating federal learning and meta learning is used for cooperatively training a global model on the premise that data cannot go out of a domain and rapidly adapting to a new exploration block; And the multi-objective intelligent decision center integrates multi-dimensional indexes of risk, cost and benefit and provides a pareto optimal solution set as decision support. Furthermore, the dynamic business object model describes business entities through an extensible Schema definition language, stores association relations among the entities by using a graph database, and supports the response of business requirements and the evolution of data relations. Further, the multi-Agent includes, but is not limited to, a data quality sensing Agent, a geophysical prospecting interpretation special Agent, a logging interpretation special Agent, a risk early warning Agent and a parameter optimization Agent, and each Agent performs asynchronous communication and cooperation through a publish-subscribe mechanism. Further, the cloud native intelligent scheduling engine is constructed based on Kubernetes and KubeFlow technology stacks, and realizes the cooperation of multiple agents through event-driven dynamic workflow. Further, the cloud native intelligent scheduling engine includes: The workflow composer automatically generates a directed acyclic graph according to the task type, and defines the execution sequence and the dependency relationship of each Agent; The resource scheduler monitors the cluster load in real time, dynamically allocates CPU/GPU resources according to the workflow progress, and realizes the flexible extension and contraction of computing resources; The fault-tolerant controller realizes automatic migration and task recovery of the fault node through heartbeat detection and state snapshot, and ensures high availability of the system; Furthermore, the engine adopts a publish-subscribe communication mode, and issues an event when the data quality sensing Agent discovers that new data is ready, so as to trigger the subsequent risk early warning Agent to start working. Further, the intelligent manag