CN112733440A

CN112733440A - Intelligent fault diagnosis method, system, storage medium and equipment for offshore oil-gas-water well

Info

Publication number: CN112733440A
Application number: CN202011608819.8A
Authority: CN
Inventors: 李伟; 郭凯伟; 李升�; 夏如君; 张海荣; 张风波; 杨佳丽; 金云智; 王继鹏
Original assignee: Beijing Zhihui Dacheng Technology Co ltd; Guangdong Ocean University; CNOOC China Ltd Zhanjiang Branch; Southern Marine Science and Engineering Guangdong Laboratory Guangzhou
Current assignee: Beijing Zhihui Dacheng Technology Co ltd; Guangdong Ocean University; CNOOC China Ltd Zhanjiang Branch; Southern Marine Science and Engineering Guangdong Laboratory Guangzhou
Priority date: 2020-12-30
Filing date: 2020-12-30
Publication date: 2021-04-30

Abstract

The invention belongs to the technical field of fault diagnosis of oil and gas wells, and discloses an intelligent fault diagnosis method, system, storage medium and equipment for offshore oil and gas wells, including: cleaning oil well data; data analysis; model establishment; model evaluation. The invention takes offshore oil and gas wells as research objects, and proposes an intelligent fault diagnosis method for offshore oil and gas wells, which performs data cleaning, feature analysis, fault diagnosis, and fault prediction on development well data, which is beneficial to improve collaboration efficiency and reduce communication costs. Through the research of the present invention, a development well history data integration and intelligent early warning system is designed, which mainly uses statistical analysis technology to monitor the production process of oil and gas wells in real time, and provides early warning for abnormal trends in the production process; meanwhile, it can identify the production practice of different oil and gas fields Possible faults and risks, so that production managers can take timely countermeasures, so as to achieve the purpose of improving production time and reducing maintenance costs.

Description

Intelligent fault diagnosis method, system, storage medium and equipment for offshore oil-gas-water well

Technical Field

The invention belongs to the technical field of oil-gas-water well fault diagnosis, and particularly relates to an intelligent fault diagnosis method, system, storage medium and equipment for an offshore oil-gas-water well.

Background

At present, with the production of new oil and gas fields and adjustment wells, the number of development wells of branch companies is gradually increased, and reaches 452 by 11 months in 2018, and meanwhile, with the aging of well shafts of the development wells and ground facilities, the failure probability is gradually increased. The method has the advantages that the production abnormity is captured quickly, the fault processing time is shortened, and the method plays a very key role in improving the production efficiency and guaranteeing the basic yield. The existing method for tracking production and dynamically capturing production abnormity of a development well through production management personnel has large workload and poor timeliness, and a set of intelligent system is urgently needed to assist in realizing quick and efficient early warning.

Through the above analysis, the problems and defects of the prior art are as follows: the existing method for tracking production and dynamically capturing production abnormity of a development well through production management personnel has large workload and poor timeliness.

The difficulty in solving the above problems and defects is: the collected oil well data are too messy, and the oil wells with the same type of faults (such as pipe column leakage) have different characteristic attributes, so that a large amount of time is spent on data cleaning, characteristic extraction and data set establishment.

The significance of solving the problems and the defects is as follows: the method is characterized in that fault diagnosis of the offshore oil-gas well is taken as a research object, and an offshore oil-gas well fault diagnosis model is established and an offshore oil-gas well fault intelligent early warning system is set up mainly based on a deep learning thought method. The method is beneficial to improving the management level of the offshore oil-gas well in China, improving the fault diagnosis efficiency of the offshore oil-gas well in China and reducing the maintenance cost of the offshore oil-gas well in China.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides an intelligent fault diagnosis method, system, storage medium and equipment for an offshore oil-gas-water well.

The invention is realized in this way, an offshore oil-gas-water well intelligent fault diagnosis method, which comprises the following steps:

firstly, acquiring offshore oil well data in a database, and providing data support for data cleaning;

step two, data cleaning: deleting abnormal and missing data, performing linear, quadratic spline, cubic spline or nearest neighbor interpolation on characteristic attributes such as daily liquid production amount of calculation and the like, and finally extracting data 3 months before and 15 months after each type of fault point to provide clean data for characteristic extraction;

step three, feature extraction: the offshore oil well data has more attributes, and some characteristics are not significant, so that the influence on the early warning research of the oil well is caused, the important attributes of each oil well are extracted by using the maximum information coefficient, the robust method and the random forest weight, and the technical support is provided for the establishment of a data set;

fourthly, performing data analysis on the important characteristics to establish a data set; then data transformation (normalization and standardization) is carried out, and data support is provided for model training;

step five, establishing a model: training the data set by using an LSTM model and a DBN model, and adjusting parameters;

evaluating the model by utilizing the evaluation indexes of the loss function, the RMSE, the MSE, the MAE, the MAPE and the SMAPE;

step seven, constructing an offshore oil-gas-water well early warning system by using MyEclipse + js + css + html5+ SQLserver, and providing technical support for the step eight;

and step eight, displaying the constructed early warning system on the Hadoop.

Further, the model establishing method in the third step is as follows: and constructing the model through an LSTM long-time memory network and a DBN deep confidence network.

Further, in the long and short term memory network, the hidden layer of the original RNN has only one state, i.e., h, which is very sensitive to short term input. A state, c, is added to save a long-term state called cell state.

At time t, there are three inputs to the LSTM: an input value x _ t of a network at the current moment, an output value h _ t-1 of an LSTM at the last moment and a unit state c _ t-1 at the last moment;

the output of the LSTM is two: the LSTM at the current time outputs a value h _ t and a cell state c _ t at the current time.

Further, in the forward calculation of the LSTM, three control switches are used to control the long-term state:

the first switch is responsible for controlling the long-term state c to be continuously stored;

the second switch is responsible for controlling the input of the instant state into the long-term state c;

and the third switch is responsible for controlling whether the long-term state c is taken as the output of the current LSTM.

These three switches are implemented in the algorithm with gates:

defining: the gate is actually a fully connected layer, the input is a vector, the output is a real number vector between 0 and 1, and the formula is as follows:

g(x)＝δ(Wx+b)。

further, the gate has three types:

(1) forget the door: it determines how much of the cell state c _ t-1 at the previous time remains to the current time c _ t. In the calculation process, W _ f is a weight matrix of a forgetting gate, [ h _ t-1, x _ t ] represents that two vectors are connected into a longer vector, b _ f is a bias term of the forgetting gate, and sigma is a sigmoid function.

(2) An input gate: it determines how much of the network input x _ t is saved to the cell state c _ t input gate at the current time.

Calculation of the currently input cell state c _ t: the last unit state c _ t-1 is multiplied by the forgetting gate f _ t according to elements, then the currently input unit state c _ t is multiplied by the input gate i _ t according to elements, and then the two products are summed, so that the current memory c _ t and the long-term memory c _ t-1 can be combined together to form a new unit state c _ t.

(3) An output gate: how much the control unit state c _ t is output to the current output value h _ t of the LSTM.

Further, the DBN deep belief network is a generative model, and training data is generated by the whole neural network according to the maximum probability by training weights among neurons of the model.

Further, the intelligent fault diagnosis method for the offshore oil-gas-water well further comprises the following steps:

(1) underground oil reservoir simulation: analyzing the establishment of a reservoir seepage parameter dynamic change rule and the internal reason, and establishing an oil reservoir numerical simulator considering the reservoir seepage parameter change;

(2) and (3) injection and production intelligent optimization: establishing an oil reservoir production real-time optimization and automatic history fitting method system, and realizing intelligent optimization and injection-production dynamic adjustment of an oil-water well allocation scheme;

(3) single well intelligent production: the application of the artificial intelligence technology in oil and gas production is researched, and the requirements of real-time perception, cognitive reasoning and intelligent diagnosis are met.

(1) constructing a knowledge graph of the oil well subsystem: connecting different kinds of information to obtain a relationship network; and (3) cleaning concepts, attributes and relationships such as objects, activities and knowledge in the oil well subsystem by applying a domain ontology construction method, and establishing a knowledge driving framework.

(2) An oil well intelligent production decision mechanism and a method system are constructed: in a decision mechanism of the oil and gas exploitation cognitive computing system, a platform and land combined optimization decision mechanism is designed; in an oil well intelligent production decision adaptation method system, a human cognitive mode is simulated, a corresponding algorithm is researched, and a mixed decision combining knowledge and data is formed.

(3) The oil well Internet of things hybrid enhanced computing architecture design comprises the following steps: and designing a computing architecture combining the edge and the cloud end of a hybrid decision method system by using the architectures of edge computing and cloud computing.

(4) And (3) establishing an oil well cognitive computing system: and developing an oil well cognitive computing system, and realizing the whole-process self-learning intelligent closed-loop management of oil well detection, diagnosis, prediction, optimization and the like.

Further, the method for constructing the intelligent oil well production decision mechanism and method system further comprises the following steps: a dynamic and static optimization decision mechanism for offshore oil and gas production is provided, an edge and cloud cooperative intelligent decision model triggered by a service scene is established, and a knowledge and data bidirectional driving edge and cloud cooperative computing method is established.

It is a further object of the invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of:

collecting data;

data cleaning: interpolating partial oil well data by using linear, quadratic spline and cubic spline interpolation methods;

feature extraction: extracting important features of oil well data by methods such as a maximum information coefficient, a robust method, random forest weight and the like;

carrying out data analysis and establishing a data set;

establishing a model: training the data set by using an LSTM model and a DBN model, and adjusting parameters;

evaluating the model;

building an early warning system;

and displaying by the Hadoop cloud server.

It is another object of the present invention to provide a computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of:

collecting data;

carrying out data analysis and establishing a data set;

evaluating the model;

building an early warning system;

and displaying by the Hadoop cloud server.

The invention also aims to provide an information data processing terminal which is used for realizing the intelligent fault diagnosis method for the offshore oil-gas-water well.

Another object of the present invention is to provide an offshore oil-gas-water well intelligent fault diagnosis system for implementing the offshore oil-gas-water well intelligent fault diagnosis method, the offshore oil-gas-water well intelligent fault diagnosis system comprising:

the data acquisition module is used for realizing data acquisition;

the data cleaning module is used for interpolating partial oil well data by using linear, quadratic spline and cubic spline interpolation methods;

the characteristic extraction module is used for extracting important characteristics of the oil well data by utilizing methods such as a maximum information coefficient, a robust method, a random forest weight and the like;

the data analysis module is used for establishing a data set;

the model establishing module is used for training the data set by using the LSTM and DBN models and adjusting parameters;

the model evaluation module is used for realizing the evaluation of the model;

the system building module is used for building the early warning system;

and the display module is used for realizing display of the Hadoop cloud server.

By combining all the technical schemes, the invention has the advantages and positive effects that: the invention provides an intelligent fault diagnosis method for an offshore oil-gas well by taking the offshore oil-gas well as a research object. Data cleaning, feature analysis, fault diagnosis and fault prediction are performed on the data of the development well. Through the research of the invention, a developed well history data integration and intelligent early warning system is designed, which mainly applies a statistical analysis technology to monitor the production process of the oil-gas well in real time and give an early warning to the abnormal trend of the production process; meanwhile, faults and risks which may exist are identified for production practices of different oil and gas fields, so that production management personnel can take corresponding measures in time, and the aims of improving production efficiency and reducing maintenance cost are fulfilled.

The present invention can use the DBN to not only identify features, classify data, but also generate data. On the basis of improving the original early warning accuracy and coverage degree, the invention can also realize the real-time diagnosis and optimization of the pump working condition, the evaluation and optimization of the oil well energy efficiency, and the pump potential prediction and the oil reservoir cooperative optimization.

The method takes methods of reliability maintenance analysis (RCMA), failure mode and influence analysis (FMEA) and the like as guidance to obtain knowledge such as mechanism models, expert experiences and the like of all levels of service fields such as equipment, service units, subsystems and the like; and obtaining a field data rule by adopting a deep learning technical means. And a domain ontology technical route is adopted, domain mechanism knowledge and data rule are fused based on subject knowledge, a continuous updating mechanism of the knowledge map is established, and a driving framework model is established for future knowledge and data dual-drive cognitive calculation.

The invention provides an edge cloud (end) collaborative computing framework meeting a dynamic and static optimization decision mechanism for offshore oil and gas production, fully considers the stability, the cooperativity and the expansibility of the system, and ensures the efficient collaborative optimization and the sustainable development of the whole system; and the cloud and mist mixed computing frame is adopted to be beneficial to improving the cooperative efficiency and reducing the communication cost on the basis of ensuring good sea and land intelligent interaction.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow chart of an intelligent fault diagnosis method for an offshore oil-gas-water well, which is provided by the embodiment of the invention.

FIG. 2(a) is a schematic diagram of an LSTM long and short term memory network according to an embodiment of the present invention.

Fig. 2(b) is a schematic diagram of an LSTM long and short term memory network according to the time dimension.

Fig. 3 is a schematic control process diagram of a long-term state c according to an embodiment of the present invention.

Fig. 4 is a schematic diagram of a calculation process for implementing three control switches in an algorithm according to an embodiment of the present invention.

Fig. 5 is a schematic diagram of a calculation process of a forgetting gate in a gate according to an embodiment of the present invention.

Fig. 6 is a schematic diagram of a calculation process of an input gate in a gate according to an embodiment of the present invention.

Fig. 7 is a schematic diagram illustrating a calculation process of an output gate in a gate according to an embodiment of the present invention.

Fig. 8 is a schematic structural diagram of a deep belief network DBN according to an embodiment of the present invention.

Fig. 9 is a schematic diagram of a general technical route provided by an embodiment of the present invention.

Fig. 10 is a schematic diagram of a general technical route provided by an embodiment of the present invention.

FIG. 11 is a schematic diagram of the construction of a knowledge-graph of an oil well subsystem provided by an embodiment of the present invention.

Fig. 12 is a schematic diagram of a decision mechanism for automatically optimizing the operating condition of the electric pump according to an embodiment of the present invention.

Fig. 13 is a schematic diagram of a decision method based on artificial intelligence according to an embodiment of the present invention.

Fig. 14 is a schematic diagram of a design of an oil well internet of things hybrid enhanced computing architecture provided by an embodiment of the invention.

FIG. 15 is a schematic diagram of a construction of a well cognitive computing system according to an embodiment of the present invention.

Fig. 16 is a schematic diagram of a marine oil and gas internet of things hybrid enhanced computing architecture provided by an embodiment of the invention.

FIG. 17 is a schematic diagram of a hybrid-driven intelligent decision making technique combining knowledge and data provided by an embodiment of the present invention.

Fig. 18 is a schematic diagram of linear interpolation, cubic spline, quadratic spline, and nearest neighbor interpolation provided by the embodiment of the present invention.

Fig. 19 is a schematic diagram of important features extracted by the random forest weight according to the embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Aiming at the problems in the prior art, the invention provides an intelligent fault diagnosis method, a system, a storage medium and equipment for an offshore oil-gas-water well, and the invention is described in detail below with reference to the attached drawings.

As shown in fig. 1, the intelligent fault diagnosis method for the offshore oil-gas-water well provided by the embodiment of the invention comprises the following steps:

and S101, collecting oil well data in a database.

S102, cleaning the oil well data, deleting abnormal and missing data, and interpolating partial oil well data by using linear interpolation, cubic spline and quadratic spline

S103, extracting characteristics, namely extracting important characteristics such as production time, the diameter of a nozzle tip, pump frequency, wellhead temperature and the like by using a maximum information coefficient, a robust method and a random forest weight method;

s104, analyzing data and establishing a data set;

s105, establishing a model: training the data set by using an LSTM model and a DBN model, and adjusting parameters;

s106, evaluating the model;

s107, constructing a set of offshore oil-gas-water well early warning system by using the technologies of MyEclipse + js + css + html5 and the like;

and S108, displaying the constructed system on the Hadoop cloud server.

The technical solution of the present invention will be further described with reference to the following explanation of terms.

Intelligent early warning system: an intelligent computer program that utilizes knowledge and reasoning processes to predict problems in the production process of a development well.

Statistical Process Control (SPC): a process control tool by means of mathematical statistical methods (GB/T4901-. The method analyzes and evaluates a production process, timely discovers the sign of the systematic factors according to the feedback information, and takes measures to eliminate the influence, so that the production process is maintained in a controlled state only influenced by the random factors, and the purpose of controlling the quality is achieved.

The technical solution of the present invention is further described with reference to the following examples.

Example 1: project background

1. Summary of the invention

The well history data integration and intelligent early warning system (called as early warning system) mainly applies statistical analysis technology to monitor the production process of the oil-gas well in real time and give early warning to the abnormal trend of the production process; meanwhile, faults and risks which may exist are identified based on production practices of different oil and gas fields, so that production management personnel can take corresponding measures in time, and the aims of improving production efficiency and reducing maintenance cost are fulfilled.

The early warning system is based on the marine oil and gas field development well history compilation specification (Q/HS 2010-2002) of the marine oil enterprise standard, and aims at the problems of data integrity, timeliness and consistency existing in the production data and data management of the development well, a data integration mode throughout the whole life cycle of the development well is provided and realized, data integration and integration of a cross-heterogeneous database (including a structured database and a non-structured database) are realized, and business personnel are helped to quickly master all the production conditions of the development well.

On the basis of data provided by a well history integration platform, aiming at the limitation of a conventional early warning method of an offshore oil and gas well, an early warning system applies and optimizes a Statistical Process Control (SPC) technology to monitor the production Process of the oil and gas well in real time, and as a mature mathematical Statistical analysis technology, the early warning system can distinguish the random fluctuation and the abnormal fluctuation of production parameters in the production Process, thereby mastering the production dynamics of each well in real time and providing early warning for the abnormal trend of the production Process.

After the SPC control chart is analyzed to obtain the parameter abnormal state, the method is closely combined with the production practices of different oil and gas fields, firstly, a decision tree is proposed and applied to generate classification rules so as to intelligently identify possible faults and assist production management personnel to take corresponding measures in time.

The invention fully utilizes the support of the Internet of things technology, completes the butt joint of the early warning system, the production real-time management support system and the remote monitoring system of the operation data of the mechanical production well electric submersible pump, and provides a technical basis for realizing the production early warning function of the intelligent oil field in the future.

Finally, the invention accumulates cases and data in the system operation process by establishing the intelligent analysis process of closed-loop control, continuously corrects the model and finally improves the prediction effect of the model.

The system is a software product with complete intellectual property rights for the Zhonghai oil, and obtains one item of software copyright (with the authorization number of 2013SR 070179).

The system was brought on-line in 2014 and operated on-line in all fields in the west of south sea. Through effect analysis of online operation of the system, in order to timely develop the next operation and maintenance of the system and optimization work, the early warning method is improved based on the existing system operation data, the intellectualization and the accuracy of the early warning means are improved, the system is promoted to be smoothly popularized and exert application value in Zhanjiang branches, and an experienced and professional-background facilitator is required to intervene to provide technical consultation, technical guidance and technical support services.

2. Scope of counseling service

2.1 SPC control chart based production parameter Pre-Warning

Statistical Process Control (SPC) is a Process Control tool that relies on mathematical Statistical methods. In a well history integration and early warning expert system, the invention creatively introduces the mature mathematical statistical method and improves the mathematical statistical method so as to better identify the abnormal state of the production parameters.

SPC analyzes and evaluates the production process, finds the signs of systematic factors in time according to the feedback information, and takes measures to eliminate the influence, so that the process is maintained in a controlled state only influenced by random factors, thereby achieving the purpose of controlling quality. When the process is only affected by random factors, the process is in a statistical control state (called a controlled state for short); when the influence of system factors exists in the process, the process is in a statistical runaway state (simply called a runaway state). Because process fluctuations are statistically regular, when the process is controlled, the process characteristics generally follow a stable random distribution; and out of control, the process profile will change. SPC is the analytical control of a process using the statistical regularity of process fluctuations. The production parameters of the offshore oil and gas wells are analyzed by applying SPC, the out-of-control state of a single production parameter can be found in advance, and whether early warning is given or not is determined according to actual conditions.

The primary statistical tool for the implementation of SPC is the control chart. A control chart is a graphical method that gives information on the sequence of samples that characterize the current state of the process and compares this information to control limits that have been established taking into account the inherent variability of the process.

According to the service requirements and the current data acquisition situation, 14 oil and gas well production parameters such as oil pressure are selected for analysis, the actual working experience of offshore oil and gas field development well production is combined, in order to better judge parameter abnormality, the SPC theoretical criterion is refined and revised in the construction process of the early warning system, and finally a plurality of key abnormality judgment rules and algorithms are formulated.

By selecting appropriate model parameters and exception rules for a single well, and applying a data sniffing engine, various abnormal states of production parameters can be analyzed.

2.2 decision Tree based Fault identification

Decision Tree (Decision Tree) is a typical inductive classification method, which is a Tree structure similar to a flow chart, each branch representing a test output, and each leaf node representing a class or class distribution. The decision tree can extract the knowledge expressed by the decision tree through induction and pruning processing, and the decision tree is converted into a classification rule expressed in an IF-THEN form. According to the practical work experience of the service expert, when the state of a certain group of parameters is abnormal, a corresponding certain type of fault occurs, and the group of parameters can be called as sensitive parameters of the fault. Decision trees for determining various types of faults can be generated through the sensitive parameters. Because the states of the fault constitution parameters are determined, the method can directly extract the state of the sensitive parameter when each fault occurs without estimating the accuracy of each discrimination rule, and summarize to form the classification rule based on the combined parameter.

The state change of the single parameter recorded by the single parameter early warning model can be used for matching the combined parameter state for each fault, early warning is carried out when all parameter state matches accord with a decision tree, if the parameter state matches accord with the decision tree, a fault diagnosis result is recorded, and if the parameter state does not accord with the decision tree, the diagnosis is finished.

2.3 real-time data Access

The access of the real-time data of the early warning system is based on the implementation of a Zhanjiang branch company production real-time management support system and a mechanical production well electric submersible pump operation data remote monitoring system. Take a production real-time management support system as an example: on the basis of a GE company iFix + Historian real-time data acquisition scheme, the production real-time data is acquired into an MES real-time library, and then the real-time data is transferred to a data center relational library according to sampling intervals by using an autonomously developed data conversion interface. The real-time data access can be realized by establishing a data interface with the data center relational database. Through the access of real-time data, the timeliness and the accuracy of single-parameter early warning and fault occurrence judgment of the oil-gas well are improved, a foundation is laid for deep data mining and analysis in the later period, and the application value of the early warning system is further improved.

2.4 Intelligent analysis Process with closed-Loop control

The existing models of the intelligent early warning system, such as single-parameter SPC control charts, multi-parameter fault decision trees and the like, are established based on the existing production historical data. In the invention, the intelligent analysis process of closed-loop control is provided and realized, cases and data in the operation process of the system are accumulated, the model is continuously corrected, and the prediction effect of the model is finally improved. The closed-loop control is mainly realized by analyzing an actual fault case, and the threshold value and the judgment rule of the single-parameter SPC control chart can be adjusted and optimized through a parameter fluctuation mode before the fault occurs, so that more accurate parameter state abnormity capture is realized; the multi-parameter fault decision tree can be adjusted and optimized, so that more intelligent fault identification is realized; the trend model can be adjusted and optimized, so that more accurate abnormal analysis of the liquid production amount decline and water content rise trend can be realized. And the other, such as the disposal of the user for early warning, early warning false alarm and missing report, and the like, is also the feedback of the system output quantity to the model input quantity.

3. Pre-research of new early warning method

Research on data mining technology: the method comprises the steps of selecting indexes such as oil production of an oil field as a definite excavation target function, selecting influence factors such as well opening rate, water content rising speed and oil extraction speed as warning sign indexes, combining a current mainstream data excavation method, conducting specialized processing on massive data, searching implicit relations among parameters, intelligently finding potential problems, analyzing reasons, conducting prediction and early warning, and providing a new solution idea and scheme for fine management of the oil field.

Example 2: offshore oil-gas-water well

1. Project background

With the production of new oil and gas fields and adjustment wells, the number of development wells of branch companies is gradually increased, and reaches 452 ports by 11 months in 2018, and meanwhile, the failure probability is gradually increased with the aging of well shafts of the development wells and ground facilities. The method has the advantages that the production abnormity is captured quickly, the fault processing time is shortened, and the method plays a very key role in improving the production efficiency and guaranteeing the basic yield. The existing method for tracking production and dynamically capturing production abnormity of a development well through production management personnel has large workload and poor timeliness, and a set of intelligent system is urgently needed to assist in realizing quick and efficient early warning.

Therefore, the invention provides an intelligent fault diagnosis method for the offshore oil-gas well by taking the offshore oil-gas well as a research object. Data cleaning, feature analysis, fault diagnosis and fault prediction are performed on the data of the development well. Through the research of the invention, a developed well history data integration and intelligent early warning system is designed, which mainly applies a statistical analysis technology to monitor the production process of the oil-gas well in real time and give an early warning to the abnormal trend of the production process; meanwhile, faults and risks which may exist are identified for production practices of different oil and gas fields, so that production management personnel can take corresponding measures in time, and the aims of improving production efficiency and reducing maintenance cost are fulfilled.

According to the requirements made by the client, the following steps are carried out:

1. collecting data

2. Data cleaning;

3. feature extraction

4. Establishing a data set;

5. establishing a model;

6. evaluating the model;

7. building system

8. And (5) displaying the platform.

2. Model construction

2.1LSTM long and short memory network

The LSTM neural network is an improved recurrent neural network, can solve the problem that RNNs cannot handle long-distance dependence, and is popular at present.

2.1.1 memorizing the idea of the network in long and short time:

the hidden layer of the original RNN has only one state, h, which is very sensitive to short-term input. A state, c, is added to save a long-term state called cell state. As shown in fig. 2(a), fig. 2(a) is developed in the time dimension to obtain fig. 2 (b).

2.1.2LSTM Forward calculation

1. How to control the long-term state c using three control switches (as shown in fig. 3)

2. How to implement these three switches in the algorithm: the method comprises the following steps: use the door (gate)

Defining: the gate is actually a fully connected layer, the input is a vector, and the output is a real vector between 0 and 1. The formula is as follows:

g(x)＝δ(Wx+b)。

the calculations are shown in fig. 4.

gate gates have three types:

1. forget gate (forget gate): it determines how much of the cell state c _ t-1 at the previous time remains to the current time c _ t. The calculation is shown in fig. 5, where W _ f is the weight matrix of the forgetting gate, [ h _ t-1, x _ t ] denotes concatenating the two vectors into one longer vector, b _ f is the bias term of the forgetting gate, and σ is the sigmoid function.

2. Input gate (input gate): it determines how much of the network input x _ t is saved to the cell state c _ t input gate at the current time, the calculation is shown in fig. 6.

Calculation of the currently input cell state c _ t: multiplying the last unit state c _ t-1 by an element to a forgetting gate f _ t, then multiplying the currently input unit state c _ t by an element to an input gate i _ t, and then adding the two products:

thus, the current memory c _ t and the long-term memory c _ t-1 can be combined to form a new cell state c _ t.

It can keep information long before because of forgetting the control of the gate, and it can avoid the current irrelevant content to enter into the memory because of inputting the control of the gate.

How much the control unit state c _ t is output to the current output value h _ t of the LSTM. The output gate calculation is shown in fig. 7.

2.2DBN deep belief network

Deep Belief Networks (DBN) were proposed by Geoffrey Hinton in 2006. The method is a generative model, and training data can be generated by the whole neural network according to the maximum probability by training the weight among the neurons of the model. The present invention can use the DBN to not only identify features, classify data, but also generate data (as shown in fig. 8).

Example 3: intelligent diagnosis and production management optimization key technology research for offshore oil-gas-water well

1. Object of study

By applying the artificial intelligence technologies such as cognitive computation, machine learning and edge computation to oil and gas production management and constructing intelligent cognitive decision-making mechanisms such as real-time perception, cognitive inference and intelligent diagnosis of offshore oil and gas production, the intelligent diagnosis precision and the intelligent diagnosis range of oil and gas production are improved, and the production management mode conversion requirements of platform unmanned, less-humanized and land-based operation of intelligent oil fields in the future are met.

2. Content of research

The conventional development well early warning is low-frequency data based on days, the accuracy is difficult to improve after reaching about 80%, after real-time data is accessed in the future, the data volume is large, the frequency is high, and how to improve the accuracy of the early warning better is a prominent problem. The intelligent closed-loop management and self-learning continuous optimized production of the oil well is realized by taking intelligent production of an oil well subsystem as a research target, designing a knowledge graph of intelligent production of the oil well, researching an oil well production optimization method based on the knowledge graph and data mining and realizing intelligent closed-loop management and self-learning continuous optimized production of oil well production monitoring, diagnosis, analysis, optimization and the like. The method specifically comprises the following steps:

(1) single well subsystem knowledge graph construction

An application domain ontology construction method; combing the core service of the oil well subsystem; concepts such as objects, activities, knowledge, etc. in the system are clarified.

(2) Research on intelligent production decision mechanism and method system of oil well

Designing a platform (edge) and terrestrial (cloud) joint optimization decision mechanism; the intelligent production decision of the oil well is suitable for the research of a method system.

(3) Oil well Internet of things hybrid enhanced computing architecture design

By utilizing architectures such as edge computing and cloud computing, computing architectures which meet the requirements of a hybrid decision method system and are combined with a cloud end are researched.

(4) Oil well cognitive computing system set-up

And developing an oil well cognitive computing system, and realizing the whole-process self-learning intelligent closed-loop management of oil well detection, diagnosis, prediction, optimization and the like.

3. Research scheme and key technology

3.1 general technical route (as shown in FIG. 9)

(1) Underground oil reservoir simulation: analyzing the establishment of a reservoir seepage parameter dynamic change rule and the internal reason; establishing reservoir numerical simulator considering reservoir seepage parameter change

(2) And (3) injection and production intelligent optimization: and establishing an oil reservoir production real-time optimization and automatic history fitting method system, and realizing intelligent optimization and injection-production dynamic adjustment of an oil-water well allocation scheme.

(3) Single well intelligent production: the application of the artificial intelligence technology in oil and gas production is researched, the requirements of real-time perception, cognitive reasoning and intelligent diagnosis are met, and the production intelligent diagnosis precision and the production intelligent diagnosis range are improved.

3.2 study protocol (as shown in FIG. 10)

1. Construction of oil well sub-system knowledge maps

1) Connecting different kinds of information to obtain a relationship network;

2) and (3) cleaning concepts, attributes and relationships such as objects, activities and knowledge in the oil well subsystem by applying a domain ontology construction method, and establishing a knowledge driving framework.

As shown in fig. 11, using methods such as reliability maintenance analysis (RCMA) and failure mode and impact analysis (FMEA) as guidance, knowledge such as mechanism models and expert experiences in various levels of service fields such as devices, service units and subsystems is obtained; and obtaining a field data rule by adopting a deep learning technical means. And a domain ontology technical route is adopted, domain mechanism knowledge and data rule are fused based on subject knowledge, a continuous updating mechanism of the knowledge map is established, and a driving framework model is established for future knowledge and data dual-drive cognitive calculation.

2. Research on intelligent production decision mechanism and method system of oil well

1) Researching a decision mechanism of an oil and gas exploitation cognitive computing system, and designing a platform (edge) and onshore (cloud) joint optimization decision mechanism;

2) the intelligent production decision adaptation method system research of the oil well: and (3) simulating a human cognitive mode, researching a corresponding algorithm, and forming a mixed decision combining knowledge and data.

And (3) providing a dynamic and static optimization decision mechanism for offshore oil and gas production, and establishing an edge and cloud cooperative intelligent decision model triggered by a service scene. And (3) constructing a knowledge and data bidirectional driven edge (edge) cloud (end) cooperative computing method, and realizing the autonomous iteration and evolution capability from data to knowledge and from knowledge to decision.

The decision mechanism for automatic optimization of the pump operation is shown in fig. 12.

The artificial intelligence based decision method is shown in fig. 13.

At the heart of perception, understanding, interpretation, learning and interaction of cognitive computing systems is to establish an iterative, evolving methodology from data to knowledge and from knowledge to decision (intelligence).

3. Oil well internet of things hybrid enhanced computing architecture design (as shown in figure 14)

An edge cloud (end) collaborative computing framework meeting a dynamic and static optimization decision-making mechanism for offshore oil and gas production is provided, the stability, the cooperativity and the expansibility of the system are fully considered, and the efficient collaborative optimization and the sustainable development of the whole system are ensured.

4. Construction of oil well cognitive computing system (as shown in figure 15)

Based on the research results, an oil well cognitive computing system is built, under the oil well Internet of things big data environment, the knowledge graph based on oil well production is used as a main line, and intelligent closed-loop management of accurate sensing, intelligent control and continuous optimization in the whole process of monitoring, diagnosis, prediction, optimization and the like is realized through knowledge and data dual-drive and edge (edge) cloud (terminal) cooperative intelligence.

On the basis of improving original early warning accuracy and coverage degree, key hoisting points are as follows:

firstly, diagnosing and optimizing the working condition of the pump in real time;

evaluating and optimizing the energy efficiency of the oil well;

and thirdly, pump potential prediction and reservoir collaborative optimization.

3.2 Key technology

(1) Offshore oil and gas internet of things hybrid enhanced computing architecture (as shown in figure 16)

And the cloud and mist mixed computing frame is adopted to be beneficial to improving the cooperative efficiency and reducing the communication cost on the basis of ensuring good sea and land intelligent interaction.

(2) Knowledge and data combined hybrid driving intelligent decision technology (as shown in figure 17)

1) The basic method for expressing, analyzing and reasoning things from different granularities and different levels by researching and simulating human beings, namely the basic idea of particle calculation is to realize problem solving on different granularity levels;

2) multi-level knowledge evolution and reasoning are adopted to form more accurate decision;

3) and establishing an iterative and evolutionary method system from data to knowledge and from knowledge to decision (intelligence).

The above description is only for the purpose of illustrating the present invention and the appended claims are not to be construed as limiting the scope of the invention, which is intended to cover all modifications, equivalents and improvements that are within the spirit and scope of the invention as defined by the appended claims.

Claims

1. The intelligent fault diagnosis method for the offshore oil-gas-water well is characterized by comprising the following steps of:

collecting data;

carrying out data analysis and establishing a data set;

evaluating the model;

building an early warning system;

and displaying by the Hadoop cloud server.

2. The intelligent offshore oil-gas-water well fault diagnosis method of claim 1, wherein the cleaning of the well data comprises: interpolating the data with linear, cubic and quadratic splines; the method for extracting the important features comprises the following steps: carrying out oil well data by using a maximum information coefficient, a robust method and random forest weight; the establishment method of the model comprises the following steps: and constructing the model through an LSTM long-time memory network and a DBN deep confidence network.

3. The intelligent offshore oil-gas-water well fault diagnosis method as claimed in claim 2, wherein in the linear interpolation, the characteristics of oil well data such as the converted daily liquid production volume and the like are data recorded once in several days, and the linear interpolation is needed to interpolate the attribute;

in the cubic spline interpolation, the characteristics of the oil well data such as the reduced daily liquid production amount and the like are data recorded once in a plurality of days, and the attribute needs to be interpolated by using the cubic spline interpolation;

in the secondary spline interpolation, the characteristics of the oil well data such as the converted daily liquid production amount are data recorded once in several days, and the attribute needs to be interpolated by using the secondary spline interpolation;

in the long-term memory network, the hidden layer of the original RNN has only one state, namely h, which is very sensitive to short-term input; adding a state, namely c, and enabling the state to be stored for a long time, namely a unit state;

4. The intelligent offshore oil-gas-water well fault diagnosis method as claimed in claim 2, wherein in the forward calculation of the LSTM, three control switches are used for controlling the long-term state:

the third switch is responsible for controlling whether the long-term state c is taken as the output of the current LSTM;

these three switches are implemented in the algorithm with gates:

g(x)＝δ(Wx+b)；

the gate has three types:

(1) forget the door: it determines how much the cell state c _ t-1 at the previous time is reserved to the current time c _ t; in the calculation process, W _ f is a weight matrix of a forgetting gate, [ h _ t-1, x _ t ] represents that two vectors are connected into a longer vector, b _ f is a bias term of the forgetting gate, and sigma is a sigmoid function;

(2) an input gate: it determines how much of the network input x _ t is saved to the calculation of the cell state c _ t input gate at the present moment;

calculation of the currently input cell state c _ t: multiplying the last unit state c _ t-1 by the forgetting gate f _ t according to elements, multiplying the currently input unit state c _ t by the input gate i _ t according to elements, and adding the two products, so that the current memory c _ t and the long-term memory c _ t-1 can be combined together to form a new unit state c _ t;

5. The intelligent offshore oil-gas-water well fault diagnosis method of claim 1, further comprising:

6. The intelligent offshore oil-gas-water well fault diagnosis method of claim 1, further comprising:

(1) constructing a knowledge graph of the oil well subsystem: connecting different kinds of information to obtain a relationship network; cleaning concepts, attributes and relationships such as objects, activities and knowledge in the oil well subsystem by applying a domain ontology construction method, and establishing a knowledge driving framework;

(2) an oil well intelligent production decision mechanism and a method system are constructed: in a decision mechanism of the oil and gas exploitation cognitive computing system, a platform and land combined optimization decision mechanism is designed; in an oil well intelligent production decision adaptation method system, a human cognitive mode is simulated, a corresponding algorithm is researched, and a mixed decision combining knowledge and data is formed;

(3) the oil well Internet of things hybrid enhanced computing architecture design comprises the following steps: designing a computing architecture which meets the combination of the edge and the cloud of a mixed decision method system by using the architectures of edge computing and cloud computing;

(4) and (3) establishing an oil well cognitive computing system: developing an oil well cognitive computing system to realize the whole process of self-learning intelligent closed-loop management of oil well detection, diagnosis, prediction and optimization;

the construction method of the intelligent oil well production decision mechanism and method system further comprises the following steps: a dynamic and static optimization decision mechanism for offshore oil and gas production is provided, an edge and cloud cooperative intelligent decision model triggered by a service scene is established, and a knowledge and data bidirectional driving edge and cloud cooperative computing method is established.

7. A computer device, characterized in that the computer device comprises a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to carry out the steps of:

collecting data;

carrying out data analysis and establishing a data set;

evaluating the model;

building an early warning system;

and displaying by the Hadoop cloud server.

8. A computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of:

collecting data;

carrying out data analysis and establishing a data set;

evaluating the model;

building an early warning system;

and displaying by the Hadoop cloud server.

9. An information data processing terminal, characterized in that the information data processing terminal is used for realizing the intelligent fault diagnosis method for offshore oil-gas-water wells as claimed in any one of claims 1 to 6.

10. An offshore oil-gas-water well intelligent fault diagnosis system for implementing the offshore oil-gas-water well intelligent fault diagnosis method as claimed in any one of claims 1 to 6, wherein the offshore oil-gas-water well intelligent fault diagnosis system comprises:

the data acquisition module is used for realizing data acquisition;

the data analysis module is used for establishing a data set;

the model evaluation module is used for realizing the evaluation of the model;

the system building module is used for building the early warning system;