CN-121502491-B - Drilling condition identification method and device, computer equipment and medium

CN121502491BCN 121502491 BCN121502491 BCN 121502491BCN-121502491-B

Abstract

The present disclosure relates to the field of oil and gas field exploration and development technologies, and in particular, to a method, an apparatus, a computer device, and a medium for identifying drilling conditions. The method comprises the steps of carrying out initial classification on key characteristic parameters in real-time drilling parameters according to preset classification standard parameters to obtain initial working condition categories, respectively inputting the key characteristic parameters corresponding to the initial working condition categories into working condition identification models corresponding to the initial working condition categories to obtain drilling working condition identification results, wherein the working condition identification models are constructed based on state space recursion updating, a gating selection mechanism and dynamic convolution kernels. The application obviously improves the identification accuracy and interpretation ability under long tail working condition, provides a feasible technical path for high-precision and real-time working condition identification in the drilling operation process, and has stronger engineering application value and popularization prospect.

Inventors

PANG ZHAOYU
WANG YIFAN
SONG XIANZHI
WANG HAIZHU
LI GENSHENG
HUANG ZHONGWEI
ZHU ZHAOPENG
ZHANG CHENGKAI

Assignees

中国石油大学(北京)

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (8)

1. A method of identifying drilling conditions, the method comprising: According to preset classification standard parameters, carrying out initial classification on key characteristic parameters in real-time drilling parameters, and determining initial working condition categories corresponding to the key characteristic parameters, wherein the classification standard parameters are determined by determining correlation coefficients of each key characteristic parameter and each drilling working condition to obtain a correlation coefficient calculation result; The method comprises the steps of respectively inputting key characteristic parameters corresponding to each initial working condition category into a working condition identification model corresponding to each initial working condition category, and outputting to obtain time sequence characteristics; normalizing all the working condition label sequences according to the integral scoring function to obtain conditional probability distribution, wherein the conditional probability distribution formula is as follows: ; Wherein, the Representing the overall scoring function as a whole, The method comprises the steps of (1) representing a distribution function, wherein x represents an observation sequence of drilling real-time parameters, y represents a working condition label sequence corresponding to the input observation sequence x, and y' represents all possible labels; establishing probability distribution of a modeling tag sequence according to the conditional random field; traversing all paths to iterate the distribution function until the value of the objective function is minimum; Determining a final transfer score according to a state transfer mask matrix and an initial transfer score, and constraining the time sequence feature based on the final transfer score, wherein the working condition identification model is obtained based on state space recursion update, a gating selection mechanism and dynamic convolution kernel construction, and the working condition identification model is obtained according to sample key feature parameter training.
2. The method of claim 1, wherein the structured sequence decoding module comprises a conditional random field and a state transition matrix.
3. The method of claim 1, wherein the condition recognition model is trained by: Acquiring key characteristic sample parameters in historical drilling sample parameters; Inputting the key characteristic sample parameters into a state space for recursion to obtain an initial hidden representation; Inputting the initial hidden representation to a gating selection mechanism to obtain a gating weighted representation; Inputting the gate control weighted representation into a dynamic convolution kernel to obtain emission characteristics; And determining a loss function value according to the emission characteristic and the working condition label, iteratively updating an initial model formed by the state space recursion, the gating selection mechanism and the dynamic convolution kernel according to the loss function value until the iteration stopping condition is met, and constructing to obtain the working condition identification model.
4. The method of claim 1, wherein the correlation coefficient of each key feature parameter with each working condition is determined by: performing rank conversion on the value of each key characteristic parameter to generate a rank sequence; calculating a correlation coefficient between the order sequences according to the label information of each working condition class; and establishing a characteristic working condition correlation matrix, wherein each element of the matrix represents a correlation coefficient calculation result of the characteristic and the working condition label.
5. The method of claim 1, wherein inputting key feature parameters corresponding to each initial operating condition category to the operating condition identification model corresponding to each initial operating condition category, respectively, obtaining a drilling operating condition identification result comprises: Performing state space recursion on the key characteristic parameters, and determining the relation between the key characteristic parameters and the hidden state; Nonlinear transformation is carried out on the key characteristic parameters and the historical state based on the gating unit, and gating weight is generated; generating a dynamic convolution kernel according to the hidden state parameters; And carrying out pooling operation on the hidden state sequence to obtain a context representation vector, and outputting the working condition prediction probability according to the context representation vector.
6. A drilling condition identification device, the device comprising: The system comprises an initial classification unit, a classification unit and a classification unit, wherein the initial classification unit is used for carrying out initial classification on key characteristic parameters in real-time drilling parameters according to preset classification standard parameters and determining initial working condition categories corresponding to the key characteristic parameters, and the classification standard parameters are determined by determining the correlation coefficient of each key characteristic parameter and each drilling working condition to obtain a correlation coefficient calculation result; The system comprises a working condition identification unit, a structural sequence decoding module, a time sequence analysis unit and a time sequence analysis unit, wherein the working condition identification unit is used for respectively inputting key characteristic parameters corresponding to each initial working condition category into a working condition identification model corresponding to each initial working condition category to obtain a drilling working condition identification result and outputting to obtain a time sequence characteristic; normalizing all the working condition label sequences according to the integral scoring function to obtain conditional probability distribution, wherein the conditional probability distribution formula is as follows: ; Wherein, the Representing the overall scoring function as a whole, The method comprises the steps of (1) representing a distribution function, wherein x represents an observation sequence of drilling real-time parameters, y represents a working condition label sequence corresponding to the input observation sequence x, and y' represents all possible labels; establishing probability distribution of a modeling tag sequence according to the conditional random field; traversing all paths to iterate the distribution function until the value of the objective function is minimum; And determining a final transfer score according to the state transfer mask matrix and the initial transfer score, and constraining the time sequence characteristic based on the final transfer score, wherein the working condition identification model comprises a state space recursion update, a gating selection mechanism and a dynamic convolution kernel, and is obtained by training according to sample key characteristic parameters.
7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any one of claims 1 to 5 when the computer program is executed.
8. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program which, when executed by a processor, implements the method of any one of claims 1 to 5.

Description

Drilling condition identification method and device, computer equipment and medium Technical Field The specification relates to the field of oil and gas field exploration and development, in particular to a drilling working condition identification method, a device, computer equipment and a medium. Background Hydrocarbon drilling is one of the most critical links in hydrocarbon exploration and development, which provides a path for hydrocarbon flow to the surface by creating a wellbore between the surface and the reservoir. To ensure wellbore quality, the drilling process requires the cooperation of a variety of equipment and tools, and is accomplished in concert by a variety of professionals. The construction process is complex, endless and generally continuous. According to statistics, the drilling cost occupies more than 60% of the total oil and gas exploration and development cost, and in order to increase the oil and gas exploration and development force and ensure the energy safety, the cost reduction and synergy are needed in the drilling process. Therefore, it is important to analyze performance indexes in drilling construction in time, provide construction suggestions, and perform risk early warning, and the basis of these works is to accurately identify current drilling conditions and to determine construction operation states. The time duty ratio of each working condition, the working condition conversion process and the tool state are accurately obtained, so that reliable data support is provided for efficient analysis of KPIs, basis is provided for next key construction decision, and drilling risks and accidents are reduced. Currently, most of working condition identification methods are logical judgment models based on expert experience. The models can identify most of the conventional working conditions by setting specific thresholds and rules and analyzing drilling parameters. The method has the advantages of being relatively simple to realize and having a good recognition effect on most standard working conditions. However, for some working procedures, such as rotary lifting, pump starting and drill tripping (tripping), and the like, which are complicated, involve various parameter change interleaving and are difficult to describe by simple logic rules, the traditional logic judgment model is often difficult to accurately identify. In recent years, with the development of artificial intelligence technology, researchers begin to introduce neural network models to realize automatic identification of drilling conditions. Deep learning models such as Convolutional Neural Network (CNN), cyclic neural network (RNN/LSTM) and the like achieve a certain result on the problem of working condition classification. However, in the existing research and industrial practice, the models still have a plurality of key problems that (1) the models generally lack expert knowledge fusion capability, auxiliary judgment cannot be carried out by utilizing experience rules accumulated by field engineers for many years, prediction results which are contrary to engineering logics easily appear, (2) the traditional model based on logic threshold values or deep learning is difficult to capture local features and global features at the same time, the model accuracy is low, and (3) the deep model has a good classifying effect on main flow working conditions, but has low recall rate and serious long tail problems when few samples such as ' pump start and drill (drill start and drill off) ' rotation and lifting up ' are identified, so that the integrity and practicability of an identification system are limited. (4) The traditional model is sensitive to fluctuation and is easy to cause error recognition, and (5) strict operation sequence constraint exists between drilling working conditions, and the traditional model lacks modeling of working condition transfer logic and is easy to generate unreasonable jump. Disclosure of Invention The method comprises the steps of initially classifying key characteristic parameters in real-time drilling parameters according to preset classification standard parameters to obtain initial working condition types, wherein the classification standard parameters are determined by determining correlation coefficients of each key characteristic parameter and each drilling working condition to obtain a correlation coefficient calculation result, determining classification standard parameters according to the correlation coefficient calculation result, respectively inputting real-time drilling parameters corresponding to each initial working condition type into a working condition recognition model corresponding to the initial working condition to obtain a drilling working condition recognition result, and the working condition recognition model is obtained based on state space recursion updating, a gate control selection mechanism and dynamic convolution kernel construction. According to one aspe