CN-122021858-A - Knowledge graph reasoning method, system and equipment for intelligent decision of oil and gas field enterprise management

CN122021858ACN 122021858 ACN122021858 ACN 122021858ACN-122021858-A

Abstract

The invention discloses a knowledge graph reasoning method and a knowledge graph reasoning system for an oil and gas field enterprise management intelligent decision, which comprise the following steps of constructing a knowledge graph reasoning problem concept and a framework, automatically obtaining knowledge data in the oil and gas field, extracting information from the knowledge data, obtaining entities in the knowledge data and relations among the entities, constructing a triplet data set, constructing a knowledge graph in the oil and gas field decision field through the triplet data set, generating a relational reasoning rule base and a rule confidence probability based on a knowledge reasoning model and a method of neural logic programming, generating a knowledge reasoning model based on a cognitive graph reasoning framework, cross iterating new knowledge inclusion and a knowledge graph reasoning process based on a cognitive graph reasoning process, integrating a prediction result of the neural logic programming and the cognitive graph reasoning, and constructing a knowledge graph reasoning model for the oil and gas field intelligent decision. The invention supplements and enhances the structure of the neural logic programming reasoning through the cognitive map reasoning, and realizes the efficient reasoning and prediction in the intelligent decision-making process of the oil-gas field.

Inventors

JIANG BIN
ZHANG LI
ZHANG LIYUN
PEI SENQI
CHEN JUN
WANG XINWEI

Assignees

中国石油天然气股份有限公司

Dates

Publication Date: 20260512
Application Date: 20241112

Claims (10)

1. A knowledge graph reasoning method oriented to intelligent decision making of oil and gas field enterprise management is characterized by comprising the following steps: s1, constructing a knowledge graph reasoning problem concept and a framework; S2, knowledge data in the oil and gas field is automatically acquired, information extraction is carried out on the knowledge data, the entities in the knowledge data and the relation between the entities are acquired, a triplet data set is constructed, and a knowledge graph in the decision-making field of the oil and gas field is constructed through the triplet data set; s3, generating a relation reasoning rule base and a rule confidence probability based on a knowledge reasoning model and a method of neural logic programming; s4, a knowledge reasoning model based on a cognitive map reasoning framework carries out cross iteration on new knowledge to be incorporated into a cognitive map reasoning process; s5, integrating the prediction results of the neural logic programming and the cognitive map reasoning, and constructing a knowledge graph reasoning model oriented to the intelligent decision of the oil and gas field.
2. The method of claim 1, wherein in S1: Constructing a knowledge graph reasoning problem concept and framework, wherein a knowledge graph G constructed based on a triplet data set is expressed as G= { E, R, T }; Wherein e= { E i ,1≤i≤n e } is expressed as a set of entities in the knowledge graph, E i is the ith entity in the knowledge graph, and n e is the number of entities in the knowledge graph; R= { R i ,1≤i≤n r } is expressed as a set of relationship types in the knowledge graph, R i is expressed as the ith relationship type in the knowledge graph, and n r is expressed as the number of relationship types in the knowledge graph; T= { T i ,1≤i≤n t } is expressed as a triplet set in the knowledge graph, T i is expressed as the ith triplet in the knowledge graph, and n t is expressed as the number of triples in the knowledge graph; triplet t is denoted as t= (e 0 ,r,e 1 ),e 0 is the head entity in the triplet, e 1 is the tail entity, and r is the relationship between the two entities.
3. A method according to claim 1, wherein in step S2: Automatically acquiring domain knowledge, constructing a domain knowledge graph, constructing an information extraction model based on natural language processing based on a transducer model, extracting entities in unstructured data and relations among the entities through the constructed information extraction model, constructing a triplet data set, constructing an oil-gas field knowledge graph through the triplet, analyzing a decision target in natural language expression through a natural language processing technology and the information extraction model, and constructing a query mode applicable to knowledge graph reasoning.
4. The method of claim 1, wherein S2 comprises the steps of: Step S201, knowledge data in the oil and gas field enterprise management intelligent decision field is automatically acquired, an oil and gas field intelligent decision search keyword is constructed, an automatic data crawling program is designed, and related knowledge data is searched and crawled from multi-source data; Step S202, preprocessing the acquired data, further processing unstructured data based on an information extraction model and a method, extracting the entity in knowledge data and the relation between the entities to construct a triplet data set, specifically, Analyzing the acquired HTML or XML webpage data based on BeautifulSoup library of Python programming language to acquire text data of webpage sources and the like, further matching and filtering the text data through regular expressions and public stop word lists, finally manually marking the entity and relation of partial data, training the marked data by adopting a pre-trained BERT model containing 768 hidden layers based on Chinese, constructing a deep learning model for extracting the entity and relation oriented to intelligent decision of the oil-gas field, and finally processing unstructured data processed through the trained BERT model to obtain a triplet dataset containing the entity and relation; Step S203, constructing a visual knowledge graph based on entity and relation triplet data sets, namely, compiling an automatic data input program by using Python language, inputting the constructed triplet data sets into Neo4j software, constructing a knowledge graph visual model of the triplet data sets, and observing and finding related entity relations and knowledge connection structures in the constructed visual knowledge graph; Step S204, based on the constructed entity and relationship extraction model, unstructured data processing is performed on the target decision problem, and the extraction target query is q= (e 0 ,r,?),e 0 is the head entity of the query, r is the relationship of the query,.
5. The method of claim 1, wherein in S3: Constructing a knowledge reasoning method based on neural logic programming, and generating a reasoning rule base based on domain knowledge and a corresponding confidence probability score; The specific steps include defining a query q as: wherein M is the index of the neural logic rule, c is the rule weight, b is the relation ordering in the query, and M is the calculation mode of the corresponding relation; The tail entities defining the query are: Wherein x is the head entity of the query; defining the accuracy score s of the query tail entity in the reasoning result as follows: iteratively performing fitting calculation on the neurologic rule chain by using a recurrent neural network model: Wherein: α t ＝softmax(Wh t +β) β t ＝softmax([h 0 ,…,h t-1 ] T h t )
6. The method according to claim 1, wherein in the step S4: A knowledge graph reasoning method of the cognitive map is constructed, Defining a cognitive map reasoning process, wherein the answer entity score of the query is as follows: where alpha is the one-hot encoding vector of the querying entity, A vector representing an edge adjacent to entity e i , Representing potential representations of entity e i and relationship r, ω being the computational weight, f being the activation weight; The calculation method of X [ e ] and a t is as follows: wherein the GRU is a gated recurrent unit neural network.
7. The method according to claim 1, wherein the step S5 comprises the following specific steps: S501, weighting probability scores corresponding to answer entities obtained by knowledge reasoning to obtain: where rank 1 and rank 2 represent answer ordering for neuro-logic programming and cognitive map reasoning, w 1 and w 2 are integration weights; S502, training a constructed model by using the acquired data, selecting proper data for a triplet data set after constructing a knowledge graph reasoning method oriented to intelligent decision of an oil-gas field, dividing the data into a training set and a verification set, iteratively inputting the training set data into the constructed model, and adopting a random gradient descent method to adjust parameters such as learning rate, batch training size, iteration times and the like of a long-short-period memory neural network and a gate-controlled loop neural network in the model to search optimal parameters; S503, constructing training optimization indexes of a knowledge graph reasoning model facing intelligent decision of the oil and gas field, constructing model evaluation indexes in model training, and enabling the model to update parameters along target iteration, wherein the method comprises the following steps: s5031, constructing an evaluation index of a knowledge graph reasoning model: the bits@K represents the proportion of the number of triples with the correct answer ranking smaller than K in all predicted triples to the predicted answer ranking in the reasoning prediction of knowledge reasoning; S5032, processing triples containing the same head entity and relation according to the traversal triples data set based on the real scene that the same query has a plurality of correct answers, and expanding tail entities into tail entity sets; S5033, based on the constructed model evaluation index, traversing and inquiring the tail entity set during model training, calculating the average value of the model evaluation index for each entity in the tail entity set, calculating a loss function in the neural network according to the average value, and updating the model parameters based on a random gradient descent optimization algorithm.
8. A knowledge-graph reasoning system oriented to intelligent decision making of oil and gas field enterprise management, which is applied to the method of any one of claims 1-6, and is characterized by comprising a data acquisition module, an information extraction and graph construction module and a reasoning prediction module; the data acquisition module collects large-scale unstructured knowledge data related to the field according to decision targets and query problems; The information extraction and map construction module identifies and extracts the corresponding relation of the entities in the data according to the acquired data, and constructs a triplet data set and a corresponding knowledge map; the reasoning prediction module is used for constructing a knowledge graph reasoning method oriented to intelligent decision of the oil-gas field, and obtaining a final answer and a decision support suggestion according to decision targets and query questions.
9. The system of claim 7, wherein the data acquisition module automatically searches for domain knowledge data and extracts a relationship triplet in the knowledge data to construct a knowledge graph, and constructs a knowledge graph inference model for intelligent decision-making of the oil-gas field to realize inference prediction.
10. A computer device comprising a processor and a memory, the memory having stored thereon a computer program, characterized in that the processor implements the method of any of claims 1-6 when executing the computer program.

Description

Knowledge graph reasoning method, system and equipment for intelligent decision of oil and gas field enterprise management Technical Field The invention relates to the technical field of data processing, in particular to a knowledge graph reasoning method for intelligent decision making of oil and gas field enterprise management. Background The energy development is always an important guarantee for national development and social stability, and the problems of complex environment, great decision influence, limited reference cases and the like of enterprise management decisions for oil and gas fields often exist. In addition, in the field development and production enterprise management environment, the development of equipment automation and management informatization generates large amounts of structured and unstructured data. These data are often difficult to process, analyze and utilize by efficient means, thereby providing valuable empirical knowledge and reference advice for the decision making process of management development and management of oil and gas field enterprises. How to excavate and characterize knowledge data in the field of massive unstructured oil and gas fields, and support is provided for solving decision-making problems in a real environment through an effective mathematical model, so that the knowledge data is a scientific problem to be solved urgently. With the development of BP neural network and artificial intelligence technology, natural language processing models have been developed to a great extent, and great potential is presented in solving nonlinear problems and unstructured data processing. The cyclic neural network, the convolutional neural network and the models such as a transducer based on a self-attention mechanism have reached the capability of being comparable to human beings in tasks such as entity and relationship extraction, text classification, text generation and the like. With the accumulation of knowledge data and the progress of natural language processing technology, a knowledge graph has become a new method and new thought for knowledge characterization, knowledge management and knowledge prediction. More and more researches focus on how to construct a high-efficiency knowledge graph to organize and represent knowledge data, develop application and scene based on the knowledge graph, and exert the advantage that the knowledge graph provides decision support in a complex reality environment. However, most of the technologies and applications based on knowledge graph are based on storage and query, and it is difficult to solve the problem that does not exist in the past knowledge, that is, it is impossible to infer new entities and link relationships through existing entities and relationships. In addition, the current knowledge graph-based reasoning has the problems of poor reasoning performance, interpretation, insufficient reliability and the like. In particular, in the field of intelligent decision-making for oil and gas fields, there are still few large-scale knowledge maps and inference-related researches based on knowledge maps. Disclosure of Invention The invention discloses a knowledge graph reasoning method oriented to intelligent decision making of oil and gas field enterprise management, which provides a novel knowledge graph reasoning method through integrating the advantages of neuro-logic programming reasoning and cognitive map reasoning and solves the problems that the traditional reasoning method is poor in multi-knowledge reasoning performance and insufficient in interpretation ability. The application aims to provide a knowledge graph reasoning method oriented to intelligent decision of oil and gas field enterprise management, which can process massive complex unstructured knowledge data and fully mine entities and relations in experience data during prediction reasoning, and the method constructs a neuro-logic programming model and a knowledge reasoning model based on a cognitive map, the method realizes efficient entity and information extraction, constructs knowledge graph in the field of oil and gas field and intelligent decision-making problem oriented to the reality of the oil and gas field, provides reliable and interpretable knowledge reasoning and prediction results, and solves the defects of the traditional knowledge graph reasoning method. The above object of the present application is achieved by the following technical solutions: The invention provides a knowledge graph reasoning method for intelligent decision making of oil and gas field enterprise management, which comprises the following steps: s1, constructing a knowledge graph reasoning problem concept and a framework S1, knowledge data in the oil and gas field is automatically acquired, information extraction is carried out on the knowledge data, the entities in the knowledge data and the relation between the entities are acquired, a triplet data set is constructed, an