Analysis of Intelligent Question-answer Technology for Oilfield Safety

Supervision based on Knowledge Graph

Yuan-yuan Wang

, Chao Yang

1, *

, Wei-bin Wang

, Jing-yu Zha

and Shan Huang

Technology Inspection Center of Shengli Oilfield, SINOPEC, Dongying, 257000, China

Shengli Oilfield Testing and Evaluation Research Co., Ltd., SINOPEC, Dongying, 257000, China

zhajingyu.slyt@sinopec.com, huangshan.slyt@sinopec.com

Keywords: Knowledge Graph, Safety Supervision, Intelligent Q&A.

Abstract: Safety management is a typical knowledge-intensive task. In the process of oilfield safety supervision and

management, it needs the support of a large amount of professional knowledge, which is scattered and stored

in various data. Due to large amounts of data, various types and extensive sources, it is difficult to obtain

valuable knowledge quickly and accurately by using manual methods. Therefore, with the help of relevant

tools and methods in the field of artificial intelligence, it can provide intelligent knowledge support for oilfield

safety supervision and management, thus helping to improve the safety management efficiency.

1 INTRODUCTION

Oilfield safety management is the key to ensuring

oilfield safety production. With the advent of the era

of “industry 4.0”, adopting big data intelligent

knowledge and realizing intelligent operation and

maintenance of oilfield safety management is a key

link to realize the construction of smart oilfields. The

foundation of intelligent knowledge support is to

have rich and structured available knowledge. In the

absence of sufficient knowledge, it is a practical

scheme to obtain knowledge from enormous data.

According to the DIKW hierarchical model (Figure

1) (data, information, knowledge, wisdom), the

transformation and promotion process of “data-

information-knowledge-wisdom” is a process of

continuous in-depth processing and refining of data

and knowledge, and finally promoting the

improvement of cognitive ability, problem-solving

ability and innovation ability (Paulheim 2017, Kumar

2017). In this process, relevance and understanding

are two very important factors. Only through

correlation can we promote understanding. Only

through in-depth understanding can we find new

knowledge. This process also reflects the process of

knowledge integration, accumulation and innovation.

Therefore, based on the characteristics of the oilfield

safety knowledge system, this paper analyzes the

intelligent question-answer technical framework of

oilfield safety supervision, to provide a reference for

the follow-up construction of smart oilfields.

Be advised that papers in a technically unsuitable

form will be returned for retyping. After returned the

manuscript must be appropriately modified.

Figure 1: DIKW hierarchical model and data processing

process (taking accident data as an example).

2 SAFETY KNOWLEDGE

SYSTEM

We strongly encourage authors to use this document

for the preparation of the camera-ready. Please follow

the instructions closely in order to make the volume

look as uniform as possible (Moore and Lopes 1999).

Wang, Y., Yang, C., Wang, W., Zha, J. and Huang, S.

Analysis of Intelligent Question-answer Technology for Oilﬁeld Safety Supervision based on Knowledge Graph.

DOI: 10.5220/0011174400003444

In Proceedings of the 2nd Conference on Artiﬁcial Intelligence and Healthcare (CAIH 2021), pages 64-68

ISBN: 978-989-758-594-4

Please remember that all the papers must be in

English and without orthographic errors.

Do not add any text to the headers (do not set

running heads) and footers, not even page numbers,

because text will be added electronically.

For a best viewing experience the used font must

be Times New Roman, on a Macintosh use the font

named times, except on special occasions, such as

program code (Section 2.3.7).

2.1 Features

(1) Driven by the demand for safety management

knowledge. The object of intelligent knowledge

support service is safety management. The core work

of safety management includes the identification and

analysis of safety risk factors, the formulation of

safety risk response measures and accident

emergency rescue plans, potential safety hazards

investigation and safety pre-control, accident

emergency rescue and safety accident handling, etc.

The knowledge demand of these tasks is the driving

factor of knowledge support. The key tasks such as

knowledge structure modelling, automatic

knowledge extraction and intelligent knowledge

selection should be based on knowledge demand.

(2) Based on the general and expandable

knowledge structure (Sawant 2019). Whether

knowledge can meet the needs of supporting security

management is closely related to the availability of

knowledge, which depends on the organization and

structure of knowledge. Therefore, the basic work of

constructing a domain knowledge base is knowledge

structure modelling. In order to integrate multi-source

heterogeneous knowledge to the greatest extent, it is

required to design a general and extensible

knowledge structure, which directly determines the

content and structure of the domain knowledge base.

(3) Taking multi-source heterogeneous data as

raw materials. At present, there is no general domain

knowledge base suitable for oilfield safety

supervision and management, and the structured

knowledge is relatively limited. Extracting

knowledge from a large number of multi-source

heterogeneous data has become a practical and

effective way to obtain knowledge. These data

include various management and technical materials

from within the organization. A large part of these

materials is stored in unstructured text, which not

only provides rich data sources for knowledge

extraction, but also increases the difficulty of

knowledge acquisition.

(4) Utilizing automatic knowledge extraction. It is

precise because the existing data has the

characteristics of being multi-source, heterogeneous

and large quantity, and the efficiency of using

artificial methods is too low. Therefore, it is

necessary to continuously extract valuable

knowledge from various data sources using automatic

knowledge extraction with the help of relevant

technologies and methods in the field of artificial

intelligence, so as to enrich and improve the

knowledge base in the field of safety management

and to realize the continuous accumulation of

knowledge.

(5) Based on the knowledge base in the field of

safety management. The best way to realize

knowledge integration and reuse is to structure the

knowledge. The knowledge extracted from different

sources is stored through a unified structure mode to

form a domain knowledge base. Through long-term

accumulation and improvement, the knowledge base

can provide rich knowledge sources for safety

managements and basic conditions for safety

management knowledge support.

(6) Taking knowledge intelligent selection as the

approach. Different safety management problems or

the same management problems have different

requirements for the scope and granularity of

knowledge in different situations. Knowledge

intelligent selection is to automatically select the

knowledge that users may be interested in from the

knowledge base in the field of security management,

or to provide accurate knowledge that can solve

problems to support security management decisions.

(7) The result is to obtain the potential interest

knowledge set or precise demand knowledge unit.

When the requirements are not clear or the

knowledge that meets the requirements is not unique,

the results of the intelligent selection will be provided

to the knowledge set which users are potentially

interested in. When the demand is clear and the

knowledge is unique, the result of the intelligent

selection will be provided to the user’s accurate

knowledge unit. Through the interaction with users,

we can further obtain accurate knowledge units from

the potential interest knowledge set.

(8) The ultimate goal is safety management

decision support. Assisting in safety management

decision-making is not only the starting point of

knowledge support, but also the ultimate goal of

knowledge support. Through knowledge support, we

can accurately and quickly obtain the knowledge

required by the security management business, so as

to improve the efficiency of security management

decision-making and improve the scientificity and

rationality of decision-making.

Analysis of Intelligent Question-answer Technology for Oilﬁeld Safety Supervision based on Knowledge Graph

2.2 Frame Structure

In the process of intelligent knowledge support, three

core problems need to be solved to achieve three

support objectives (Nguyen 2020, Agibetov 2020): 1)

knowledge structure modelling is used to solve the

problem of knowledge structure, and the expandable

knowledge structure mode is studied to support the

effective integration of multi-source heterogeneous

knowledge; 2) automatic knowledge extraction

solves the problem of knowledge transformation

efficiency, and constructs automatic knowledge

extraction methods and models to support the

continuous accumulation of domain knowledge; 3)

knowledge intelligent selection solves the problem of

efficiency and accuracy of knowledge acquisition,

and supports safety management decision-making

quickly and accurately by studying intelligent

knowledge application scenarios and methods.

The above intelligent knowledge support process

realizes the refining process from data to knowledge

(Figure 2): obtain multi-source heterogeneous data

sources (DS) - extract knowledge and store it in the

security management domain knowledge base (KB) -

identify the knowledge set of potential interest (IKS)

- accurately obtain the required knowledge unit

(RKU) according to the actual problems. Among

them, DS comes from the data related to internal

security management of the organization and the

security-related data resources obtained from the

Internet, which presents text, pictures, tables and

other forms. It has the characteristics of being multi-

source, heterogeneous and large quantity. Its value

density is low and needs further screening and

improvement to play its role. KB is the knowledge

expressed and stored in a specific structure, which

can be extracted from DS or edited manually, and can

be reused. IKs is a collection of knowledge

potentially interested by users selected from KB

according to user security management business

requirements. RKU is usually obtained by further

analysis and processing based on the knowledge units

in IKS and is directly used for security management

decision support. This process generally requires

human-computer interaction or automatic loading

and execution by the business system. In addition, in

the process of intelligent knowledge support, it needs

the support of relevant technologies and tools in the

field of artificial intelligence.

Figure 2: The framework of intelligent Q&A system for oilfield environmental protection and safety supervision based on

knowledge graph.

2.3 Fragmentation of Knowledge

Points

Intelligent knowledge support (Figure 3) aims to

obtain the required knowledge comprehensively,

quickly and accurately, to automatically analyse the

knowledge demands by perceiving user

characteristics and business situation characteristics,

and to realize the intelligent acquisition of domain

knowledge based on a knowledge base, rule base and

model library, thus supporting security management

decisions.

Access to knowledge demands mainly includes

the following aspects:

(1) Based on the interactive system interface,

actively input knowledge demand information. The

search conditions entered by users in various

knowledge search engines, the search questions put

forward by users in the question-answer system, and

the selection of knowledge modules in knowledge

CAIH 2021 - Conference on Artiﬁcial Intelligence and Healthcare

navigation all reflect the knowledge needs and

knowledge acquisition willingness actively expressed

by users.

(2) Obtain user context information based on

intelligent perception engines and automatically

analyse knowledge needs. Use the interfaces

provided by various system software to capture user

information, operation behaviours, and objects of

concern, and comprehensively analyse the user

situation to obtain potential knowledge needs.

(3) Automatically analyse the knowledge

requirements based on the data call requests of

various management systems. Various management

systems usually use the interface provided by the

database to make data call requests. The data call

requests of each business processing module in the

system directly reflect the knowledge requirements of

business processing.

Figure 3: Overall implementation framework of intelligent knowledge support.

2.4 Constructing Knowledge Graph

(1) Knowledge base

The knowledge base provides knowledge content

for intelligent knowledge support. The intelligent Q

& A section of oilfield safety supervision based on

knowledge graphs organizes and manages the

knowledge related to safety management in the form

of network structure, which is the core part of the

domain knowledge base, while other non-network

structured knowledge can be organized and managed

through traditional relational databases and document

management systems. For example, the user’s basic

information and system logs are still stored in the

traditional databases, and the documents, pictures,

videos, and other materials are still organized and

managed based on the traditional file management

modes.

(2) Model library

The model library provides intelligent knowledge

support with models required for knowledge

extraction and intelligent application. According to

different functions, it can be divided into knowledge

extraction model, analysis and statistics model,

prediction and early warning model, etc. According

to different implementation methods, it can be

divided into sequence annotation model,

classification prediction model, etc. According to

different algorithms, it can be divided into a rule-

based model, statistics-based model, deep learning

model, and so on. In the process of knowledge

extraction and recognition, multiple model libraries

will be established.

(3) Method library

The method library provides various concrete

implementation methods for intelligent knowledge

support. These methods are usually embedded in

modular programs, select appropriate models

according to different needs, and process data or

knowledge.

2.5 Collecting Supervision Problems

Knowledge of different granularity can be acquired

as needed. According to the degree of user

interaction, standardized knowledge acquisition

approaches can be divided into three categories:

knowledge navigation, intelligent search and

knowledge recommendation. Knowledge navigation

displays the knowledge structure and scope through

the knowledge navigation menu or knowledge

graphs, and gradually guides users to select the

required knowledge content through user interaction.

Knowledge search, by analysing the search

conditions provided by users, accurately determines

the actual knowledge needs of users, then matches the

Analysis of Intelligent Question-answer Technology for Oilﬁeld Safety Supervision based on Knowledge Graph

knowledge in the knowledge graphs, and finally

provides users with a relevant knowledge list or

accurate knowledge content. Knowledge

recommendation infers the list of knowledge that the

user may be interested in according to the user’s

professional characteristics and historical search

behaviours, and the user can further select the

required knowledge content. Users still focus on

learning normative knowledge, and improve their

safety awareness and ability by mastering normative

knowledge related to safety management.

3 CONCLUSIONS AGIBETOV A.

& SAMWALD M.

Based on the knowledge modelling in the field of

oilfield safety supervision, the intelligent knowledge

support of oilfield safety supervision realizes the

automatic extraction and transformation of

knowledge elements into structured knowledge by

using the relevant technologies and methods of

artificial intelligence, achieves the intellectualization

of knowledge acquisition by accurately

understanding the knowledge requirements, and

provides decision-making support for scientific and

efficient safety management. Knowledge modelling

is the design process of knowledge structure mode in

the field of oilfield safety supervision and

management. A reasonable knowledge structure

mode is not only conducive to the integration and

expansion of knowledge, but also supports the

automatic reasoning and completion of knowledge,

which is the basis for realizing intelligent knowledge

support. Its design process should follow the

principle of “starting from demands and ending with

an application”. Knowledge element extraction is a

process of extracting key knowledge elements from

various data sources and organizing and representing

knowledge according to a specific structure. In the

face of a large number of multi-source heterogeneous

security data, it is particularly important to realize the

automatic extraction of knowledge elements with the

help of intelligent means. It is a process of testing the

knowledge structure model. The dynamic and

continuous accumulation of domain knowledge base

provides a rich source of knowledge for intelligent

knowledge selection. Knowledge intelligent selection

and decision-making support should first obtain and

understand the actual needs of users for knowledge.

On this basis, appropriate tools and methods are used

to realize the intelligent selection and processing of

knowledge, and finally to give feedback to users as

accurate knowledge as possible to assist in supporting

safety management decision-making. Oilfield safety

supervision is driven by the demands for safety

management knowledge, based on the structured

knowledge base in the field of safety management,

through automatic knowledge extraction and

intelligent selection of supply, and with the ultimate

goal of safety management decision-making support.

REFERENCES

Agibetov A. & Samwald M. (2020). Benchmarking neural

embeddings for link prediction in knowledge graphs

under semantic and structural changes. J. Web Semant.

64, 100590.

Kumar A. J., Schmidt C. & Koehler J. (2017). A knowledge

graph based speech interface for question answering

systems. Speech Commun. 92, 1–12.

Nguyen H. L., Dang T. V. & Jung J. J. (2020). Knowledge

graph fusion for smart systems: A Survey. Inform.

Fusion 61, 56–70.

Paulheim, H. & Cimiano P. (2017). Knowledge graph

refinement: A survey of approaches and evaluation

methods. Semant. Web 8(3), 489–508.

Sawant U., Garg S. & Chakrabarti S. (2019). Neural

architecture for question answering using a knowledge

graph and web corpus. Inform. Retrieval 22, 324–349.

CAIH 2021 - Conference on Artiﬁcial Intelligence and Healthcare