CN114139458A - Drilling parameter optimization method based on machine learning - Google Patents

Drilling parameter optimization method based on machine learning Download PDF

Info

Publication number
CN114139458A
CN114139458A CN202111482929.9A CN202111482929A CN114139458A CN 114139458 A CN114139458 A CN 114139458A CN 202111482929 A CN202111482929 A CN 202111482929A CN 114139458 A CN114139458 A CN 114139458A
Authority
CN
China
Prior art keywords
representing
output
value
stratum
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111482929.9A
Other languages
Chinese (zh)
Inventor
付建红
陈一凡
彭炽
白璟
刘伟
张超越
董广建
李兆丰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southwest Petroleum University
Original Assignee
Southwest Petroleum University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southwest Petroleum University filed Critical Southwest Petroleum University
Priority to CN202111482929.9A priority Critical patent/CN114139458A/en
Publication of CN114139458A publication Critical patent/CN114139458A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a drilling parameter optimization method based on machine learning, which comprises the following steps: collecting stratum characteristic parameters of an area where a well is drilled, and preprocessing the stratum characteristic parameters; carrying out stratum characteristic clustering on the pretreated stratum characteristic parameters; and combining the clustered stratum characteristics and the drilling parameters, predicting the mechanical drilling speed by using a BP (back propagation) neural network or a circulating neural network, and constructing and obtaining a drilling parameter optimization model. Through the scheme, the method has the advantages of simple logic, accuracy, reliability and the like, and has high practical value and popularization value in the technical field of drilling.

Description

Drilling parameter optimization method based on machine learning
Technical Field
The invention relates to the technical field of drilling, in particular to a drilling parameter optimization method based on machine learning.
Background
The drilling process is a complex process influenced by multiple parameters of audiences, the drilling rate of the machine is influenced by a plurality of factors, and the drilling rate can be divided into two categories of controllable factors and uncontrollable factors according to the nature of the factors. The controllable factors are factors which can be influenced by artificial regulation, such as mechanical parameters (bit pressure, rotating speed and the like), hydraulic parameters (discharge capacity, riser pressure and the like) and drilling fluid parameters (drilling fluid density, viscosity and the like); uncontrollable factors are objective factors that are not affected by human regulation and control, and mainly include lithology characteristics of the drilled formation.
In the drilling process, the optimization of the drilling parameters is based on the matching relation between the drilling parameters and the formation lithology characteristics, the higher the matching degree is, the higher the mechanical drilling speed is in actual drilling, the less energy is consumed, meanwhile, the drilling period can be shortened, and the drilling cost can be saved. Therefore, establishing the matching relationship between the drilling parameters and the formation characteristics has important significance for optimizing the drilling parameters and improving the mechanical drilling speed.
In the actual engineering operation process, drilling parameter optimization is mainly carried out according to stratum system classification, but the thickness of the stratum system is often hundreds or even thousands of meters, and the stratum characteristics of the same stratum system are different. In general, the effects of drilling parameters on rate of penetration are similar for formation environments with similar characteristics.
At present, there are also studies on optimization of drilling parameters in the prior art, for example, chinese patent with patent application No. 201911075820.6 entitled "a drilling rate prediction method based on BP neural network and a drilling rate optimization method based on BP neural network and particle swarm algorithm" which collects torque, drilling pressure, pump pressure and displacement of drilling tool and predicts drilling rate based on BP neural network, specifically comprising the steps of 1: measuring parameters of torque Nm, drilling pressure Pm, pump pressure Pb and discharge capacity Qm of the drilling tool through a sensor according to a sampling period; step 2: sequentially normalizing the parameters to determine an input layer vector x of the three-layer BP neural network, wherein x is { x1, x2, x3 and x4 }; wherein x1 is the torque coefficient of the drilling tool, x2 is the weight-on-bit coefficient, x3 is the pump pressure coefficient, and x4 is the displacement coefficient; and step 3: the input layer vector is mapped to an intermediate layer; and 4, step 4: obtaining an output layer vector o ═ { o1 }; o1 is the drilling rate prediction coefficient; and 5: predicting the drilling speed of the drilling tool as follows: and outputting a drilling rate prediction coefficient of a layer vector for the ith sampling period, wherein the omegamlmax is the maximum drilling rate of the drilling tool, and the omegammam (i +1) is the predicted drilling rate of the drilling tool in the ith sampling period. The technology is optimized from the angle of the drilling machine, the condition of an actual place is not combined, the optimization is over theoretical, and the technology cannot be applied to complex and various scenes.
Therefore, a drilling parameter optimization method based on machine learning, which is simple in logic, accurate and reliable, is urgently needed to be provided.
Disclosure of Invention
Aiming at the problems, the invention aims to provide a drilling parameter optimization method based on machine learning, and the technical scheme adopted by the invention is as follows:
a method for machine learning based optimization of drilling parameters comprising the steps of:
collecting stratum characteristic parameters of an area where a well is drilled, and preprocessing the stratum characteristic parameters;
carrying out stratum characteristic clustering on the pretreated stratum characteristic parameters;
and combining the clustered stratum characteristics and the drilling parameters, predicting the mechanical drilling speed by using a BP (back propagation) neural network or a circulating neural network, and constructing and obtaining a drilling parameter optimization model.
Preferably, the formation characteristic parameters include formation lithology parameters, shale content, compressive strength, shear strength, internal friction angle, internal friction force, rock hardness, drillability extrema, and rock abrasiveness parameters.
Preferably, the formation characteristic parameters are preprocessed, including data cleaning, discretization, normalization, and data dimensionality reduction.
Further, stratum feature parameters are clustered by adopting one of a K-means algorithm, a MeanShif algorithm, a BDSCAN algorithm, fuzzy clustering and expectation maximization clustering based on a Gaussian mixture model.
And further, comparing the stratum rock mechanical standard deviation and the variation coefficient of the stratum characteristics before and after clustering to obtain data corresponding to the same type of stratum characteristics.
2. Preferably, the BP neural network is adopted for predicting the drilling rate of the machine, and the method comprises the following steps:
firstly, carrying out normalization processing on data, wherein the expression is as follows:
Figure BDA0003396101730000021
wherein, XkRepresenting an input sample;
Figure BDA0003396101730000031
a feature representing an input sample; ckRepresenting the normalized input sample;
Figure BDA0003396101730000032
a feature representing the normalized input sample; k represents the number of input samples;
weight ω of signal source to any input sampleijThreshold value μjtRandomly assigning values;
secondly, the input and the output of any neuron of a hidden layer and an output layer in the BP neural network are obtained, and the expression is as follows:
Figure BDA0003396101730000033
Figure BDA0003396101730000034
wherein the content of the first and second substances,
Figure BDA0003396101730000035
an output representing hidden layer neurons;
Figure BDA0003396101730000036
representing the output of the output layer neurons; m represents the total number of hidden layer neurons; n represents the total number of output layer neurons.
Thirdly, solving the error between the output layer and the hidden layer, wherein the expression is as follows:
Figure BDA0003396101730000037
Figure BDA0003396101730000038
wherein the content of the first and second substances,
Figure BDA0003396101730000039
representing the error of the output layer;
Figure BDA00033961017300000310
representing a hidden layer error;
the fourth step, updating the weights and thresholds, which are expressed as:
Figure BDA00033961017300000311
Figure BDA00033961017300000312
Figure BDA00033961017300000313
Figure BDA00033961017300000314
wherein r represents the number of error corrections; η represents the learning rate; α represents a momentum coefficient;
the fifth step, setting k as k +1,
Figure BDA00033961017300000315
Figure BDA00033961017300000316
repeating the second step to the fifth step until all k values are trained, and then executing the sixth step;
a sixth step of adding an error function value, which is expressed as:
Figure BDA00033961017300000317
wherein Loss represents the error function value, EkRepresenting a single error value;
and seventhly, repeating the second step to the sixth step until the error function value is smaller than a preset error value.
Further, the mechanical drilling speed prediction is carried out by adopting a recurrent neural network, and the method comprises the following steps:
step one, sequentially inputting data corresponding to the clustered stratum features to an update gate of a recurrent neural network, wherein the expression is as follows:
zt＝σ(Wz·[ht-1,xt]+bf) (3-1)
wherein z istThe presentation finger updates the door; h ist-1An output signal representing a neuron on the same layer; h istAn output signal representing the neuron at that time; x is the number oftRepresenting the input of the current neuron; wzRepresents the weight of the update gate; sigma represents a sigmoid function; []Representing a table join vector;
and secondly, inputting the data passing through the updating gate into a resetting gate, wherein the expression is as follows:
rt＝σ(Wr·[ht-1,xt]) (3-2)
wherein r istRepresents a reset gate; wrRepresenting the weight of the reset gate;
thirdly, obtaining an undetermined output value, wherein the expression of the undetermined output value is as follows:
Figure BDA0003396101730000041
wherein the content of the first and second substances,
Figure BDA0003396101730000042
represents a pending output value;
Figure BDA0003396101730000043
representing the weight occupied by the undetermined output value;
Figure BDA0003396101730000044
a compensation value representing the output value to be determined;
the fourth step, find the final output value and the output h of the signaltThe expression is as follows:
Figure BDA0003396101730000045
wherein the content of the first and second substances,
Figure BDA0003396101730000046
representing the value of the output to be determined.
Compared with the prior art, the invention has the following beneficial effects:
(1) the invention can extract, screen and classify the characteristics of complex and various label-free data by applying unsupervised learning. The invention obtains the clustering information of the drilled stratum by clustering the similar stratum characteristics. And judging the cluster class of the target layer according to the cluster class, and analyzing the drilling parameters used by the cluster class to provide basis for the prediction of the next mechanical drilling speed and the optimization of the drilling parameters.
(2) The invention adopts unsupervised learning-based stratum feature clustering, aims to provide preparation work for predicting the drilling rate of the machine based on supervised learning in the later period, and the essence of clustering lies in data mining. Because stratum features are more and category labels cannot be added artificially, the mutual connection among the data features is found through clustering, and then the samples are divided into different clusters by utilizing Euclidean distance or Manhattan distance, so that the data divided into the same cluster have higher similarity on the data structure, namely, the similar data are added with the labels.
(3) The invention clusters disordered stratum characteristics through unsupervised learning, and sequentially classifies points with similar stratum characteristics, and aims to make a data set for predicting the drilling rate of the machine more accurate. Because the mechanical drilling speed is mainly influenced by uncontrollable stratum characteristic factors and controllable drilling mechanical parameters, the clustering of the stratum characteristics eliminates the influence of the uncontrollable stratum characteristics on data analysis, so that the stratum characteristics in the data sets of each cluster obtained by clustering are similar and the mechanical parameters are different, and the influence of stratum characteristic noise is eliminated for the prediction of the mechanical drilling speed. And then, selecting mechanical parameters which mainly affect the drilling rate through researching the mechanical parameters of the clusters, and predicting the drilling rate by using a neural network method.
(4) The method skillfully adopts the BP neural network or the cyclic neural network to predict the mechanical drilling speed, wherein the BP neural network has strong nonlinear mapping capability, and can approximate any nonlinear continuous function with any precision aiming at the complete mapping process from an input layer to an output layer; in addition, the signal output each time by the recurrent neural network should be influenced by the previous output signal, and the signal transmission between the neurons in the same layer endows the neural network with a memory function; therefore, the stratum characteristic clustering model can be ensured to acquire data similar to the target stratum characteristic, and a data set is provided for the prediction of the mechanical drilling speed in the next step.
In conclusion, the invention has the advantages of simple logic, accuracy, reliability and the like, and has very high practical value and popularization value in the technical field of drilling.
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 will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention, and therefore should not be considered as limiting the scope of protection, and it is obvious for those skilled in the art that other related drawings can be obtained according to these drawings without inventive efforts.
FIG. 1 is a graph of the HT1 well compressive strength as a function of well depth in the present invention.
FIG. 2 is a graph of HT1 in-well friction angle as a function of well depth in accordance with the present invention.
FIG. 3 is a prediction convergence diagram of neural network learning in the present invention.
FIG. 4 is a comparison graph of the direct prediction result and the actual value in the present invention.
FIG. 5 is a comparison graph of the predicted values and actual values of the test sets of each cluster in the present invention.
FIG. 6 is a flow chart of a drilling parameter optimization model of the present invention.
FIG. 7 is a data space distribution diagram after dimension reduction of the data set in the present invention.
FIG. 8 is a graph of the elbow method for determining K value according to the present invention.
FIG. 9 is a K-value chart of the contour coefficient method of the present invention.
FIG. 10 is a clustered distribution diagram according to the present invention.
FIG. 11 is a distribution diagram of clusters to which the Lianmu Qin group belongs.
FIG. 12 is a diagram of the test results of the cluster 3 optimal neural network model test set in the present invention.
FIG. 13 is a graph of sensitivity of cluster 3 mechanical parameters to rate of penetration analysis in accordance with the present invention.
FIG. 14 is a diagram of the test results of the cluster 4 optimal neural network model test set of the present invention.
FIG. 15 is a diagram of the test results of the cluster 9 optimal neural network model test set in the present invention.
FIG. 16 is a distribution diagram of clusters belonging to the clear water river group in the invention.
FIG. 17 is a test result chart of the test set of the optimal neural network model in the present invention (one).
FIG. 18 is a second test effect diagram of the test set of the optimal neural network model in the present invention.
Detailed Description
To further clarify the objects, technical solutions and advantages of the present application, the present invention will be further described with reference to the accompanying drawings and examples, and embodiments of the present invention include, but are not limited to, the following examples. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Examples
As shown in fig. 1 to 16, the present embodiment provides a drilling parameter optimization method based on machine learning, which includes the following steps:
the method comprises the steps of firstly, collecting stratum characteristic parameters of an area where a well is drilled, and preprocessing the stratum characteristic parameters. In this embodiment, the formation characteristic parameters include formation lithology parameters, shale content, compressive strength, shear strength, internal friction angle, internal friction, rock hardness, drillability extremes, and rock abrasiveness parameters. In this embodiment, the preprocessing of the formation characteristic parameters includes data cleaning, discrete processing, normalization and data dimension reduction. Among them, the most common method for data dimension reduction is Principal Component Analysis (PCA), which aims to map high-dimensional data into a low-dimensional space for representation by some linear projection, and expects the variance of the data to be the largest in the projected dimension, thereby using less data dimensions while preserving the characteristics of more raw data points. Meanwhile, the principal component analysis algorithm is a linear dimensionality reduction method which loses the original data information at least and does not try to explore the internal structure of the data. With the reduction of dimensionality, the relation between data points can be observed more intuitively, meanwhile, the relation between the data points is simplified, and the data fitting of an algorithm is facilitated.
And secondly, carrying out stratum characteristic clustering on the pretreated stratum characteristic parameters.
In the embodiment, stratum feature parameters are clustered by adopting one of a K-means algorithm, a MeanShif algorithm, a BDSCAN algorithm, fuzzy clustering and expectation maximization clustering based on a Gaussian mixture model.
In drilling engineering, the discussion of geologic features is often categorized according to traditional stratigraphic series, i.e., stratifying in geologic stratification according to the formation time of the stratigraphic layer. The purpose of the clustering of the stratum features is to divide the strata with similar rock features into the same group from the perspective of the similarity of the stratum features.
In the embodiment, taking 1 well for respiratory exploration in south area of Xinjiang as an example, the five-open vertical well body structure is adopted, the total well depth is 7593m, and four quartos are drilled in the same place to meet 5 strata (wood-connected Qin group K)1l, Shengjin Kou group K1s, Hu map wall river group K1h, clear water river group K1q, Kkara-zao group J3k) The depth and thickness of each stratum are shown in Table 1.
TABLE 1-Sinkiang respiratory 1-well five-well stratigraphic and lithology description
Formation of earth Depth of bottom boundary Layer thickness Description of lithology
Lianmu Qin group K1l 5968 630 Dark brown, grey brown sandy mudstone, mudstone
Shengjin Kou group K1s 6110 142 Silty sandstone, silty mudstone, mudstone
Hu map wall river group K1h 6760 650 Silty mudstone, mudstone
Clear water river group K1q 7510 750 Silty mudstone, argillaceous siltstone, conglomerate
Karez-binding group J3k 7730 220 Fine sandstone of conglomerate, argillaceous fine sandstone and siltstone
As can be seen from Table 1, the layer thickness difference of 5 groups of strata encountered by five drilling cuttings is large, the smallest layer is a winning metal group, the layer thickness is only 142m, the largest layer is a clear water river group, and the layer thickness is 750 m. The difference of lithology among the stratums and in the stratums is large, and mudstone, sandstone and conglomerate are staggered.
The parameters of the partial stratum of the stratum encountered by the well five, namely compressive strength and internal friction angle, are shown in the figures 1 to 2 along with the depth. From the figure, it can be found that the average compressive strength of the five-open-hole section is 139.14MPa, the variation range of the compressive strength relative to the average compressive strength is larger and larger as the well depth is increased, the maximum compressive strength is 199.61MPa at 7542m, the minimum compressive strength is 84.01MPa at 6090m, the amplitude reaches 137.6%, the standard deviation is 18.33, and the coefficient of variation is 0.134 (the coefficient of variation is used for representing the dispersion degree of the sample, and the larger the coefficient of variation is, the larger the dispersion degree is, the more irregular the sample is). The average internal friction angle of the five-well-opening section is 39.30 degrees, the change range of the internal friction angle relative to the average internal friction angle is larger and larger along with the increasing of the well depth, the maximum internal friction angle is 44.99 degrees at 74672m, the minimum internal friction angle is 32.89MPa at 6037m, the amplitude reaches 36.8 percent, the standard deviation is 1.83, and the variation coefficient is 0.047. Therefore, the lithology difference of the well sections of 5700m to 7593m is large.
In this embodiment, a K-means algorithm is used for clustering, and the calculation results of the before-and-after-clustering variation coefficients are as follows:
TABLE 2 calculation of coefficient of variation before and after clustering
Figure BDA0003396101730000081
Figure BDA0003396101730000091
Through comparison of statistical data such as stratum rock mechanical standard deviation, coefficient of variation and the like before and after clustering, data of the same cluster after clustering are more concentrated, and the fact that rock characteristics of the same cluster after clustering are more similar is shown, and the same cluster belongs to the same category from the aspect of rock characteristics. Under the condition, the drill bit is in the stratum of the same cluster, the influence of stratum rock characteristics on the mechanical drilling rate is basically the same, and the factors influencing the mechanical drilling rate mainly comprise controllable parameters such as mechanical parameters and the like, so that the precision of the mechanical drilling rate prediction in the later period can be greatly improved.
And thirdly, combining the clustered stratum characteristics and the drilling parameters, predicting the mechanical drilling speed by using a BP (back propagation) neural network or a circulating neural network, and constructing and obtaining a drilling parameter optimization model. The drilling parameters comprise standard well depth (sounding, inclined depth), vertical depth, well inclination angle, azimuth angle, well mouth coordinates, hook load, drilling time, drilling pressure, torque, rotary table rotating speed, riser pressure, inlet (outlet) port flow, inlet (outlet) port temperature, inlet (outlet) port density, inlet (outlet) port conductivity, inlet (outlet) port density, drilling fluid viscosity, dynamic shear force, well bore size, mechanical drilling speed and the like.
In this embodiment, a BP neural network or a recurrent neural network is used to predict the penetration rate of the machine, wherein the core idea of the BP neural network is as follows: forward propagating signals, backward propagating errors. The algorithm implementation process is divided into two stages: firstly, a signal passes through a hidden layer from an input layer, and analyzed data finally reaches an output layer; the second stage is the back propagation of error, from the output layer to the hidden layer and finally to the input layer, and the weights and offsets from the hidden layer to the output layer and from the input layer to the hidden layer are adjusted in turn.
Specifically, the basic steps of the algorithm based on the three-layer BP neural network are as follows:
(1) carrying out normalization processing on the data, wherein the expression is as follows:
Figure BDA0003396101730000092
wherein, XkRepresenting an input sample;
Figure BDA0003396101730000093
a feature representing an input sample; ckRepresenting the normalized input sample;
Figure BDA0003396101730000094
a feature representing the normalized input sample; k represents the number of input samples;
weight ω of signal source to any input sampleijThreshold value μjtRandomly assigning values;
(2) obtaining the input and the output of any neuron of a hidden layer and an output layer in the BP neural network, wherein the expression is as follows:
Figure BDA0003396101730000095
Figure BDA0003396101730000101
wherein the content of the first and second substances,
Figure BDA0003396101730000102
an output representing hidden layer neurons;
Figure BDA0003396101730000103
representing the output of the output layer neurons; m represents the total number of hidden layer neurons; n represents the total number of output layer neurons.
(3) And obtaining the error of the output layer and the hidden layer, wherein the expression is as follows:
Figure BDA0003396101730000104
Figure BDA0003396101730000105
wherein the content of the first and second substances,
Figure BDA0003396101730000106
representing the error of the output layer;
Figure BDA0003396101730000107
representing a hidden layer error;
(4) update weights and thresholds, which are expressed as:
Figure BDA0003396101730000108
Figure BDA0003396101730000109
Figure BDA00033961017300001010
Figure BDA00033961017300001011
wherein r represents the number of error corrections; η represents the learning rate; α represents a momentum coefficient;
(5) let k be k +1,
Figure BDA00033961017300001012
Figure BDA00033961017300001013
repeating the steps (2) to (5) until all k values are trained, and then executing the step (6);
(6) adding an error function value, which is expressed as:
Figure BDA00033961017300001014
wherein Loss represents the error function value, EkRepresenting a single error value;
(7) and (6) repeating the steps (2) to (6) until the error function value is smaller than the preset error value.
In addition, the present embodiment may also adopt a recurrent neural network for data processing, which includes the following steps:
(1) sequentially inputting the data corresponding to the clustered stratum features to an updating gate of a recurrent neural network, wherein the expression is as follows:
zt＝σ(Wz·[ht-1,xt]+bf) (3-1)
wherein z istThe presentation finger updates the door; h ist-1An output signal representing a neuron on the same layer; h istAn output signal representing the neuron at that time; x is the number oftShow bookInput of a sub-neuron; wzRepresents the weight of the update gate; sigma represents a sigmoid function; []Representing a table join vector;
(2) inputting the data passing through the updating gate into a resetting gate, wherein the expression is as follows:
rt＝σ(Wr·[ht-1,xt]) (3-2)
wherein r istRepresents a reset gate; wrRepresenting the weight of the reset gate;
(3) obtaining an output value to be determined, wherein the expression is as follows:
Figure BDA0003396101730000111
wherein the content of the first and second substances,
Figure BDA0003396101730000112
represents a pending output value;
Figure BDA0003396101730000113
representing the weight occupied by the undetermined output value;
Figure BDA0003396101730000114
a compensation value representing the output value to be determined;
(4) determining the final output value and the output h of the signaltThe expression is as follows:
Figure BDA0003396101730000115
wherein the content of the first and second substances,
Figure BDA0003396101730000116
representing a value of an output to be determined
In this embodiment, for a constructed neural network, the data set is divided into two parts: training set and test set. The data of the training set is used to train the model and find correlations in the prediction data, the test set data follows the same probability distribution as the training set data, and the test set is the data used to independently evaluate the prediction training set. The training set and the testing set are randomly divided according to a preset proportion, the training set is generally 80 percent, the testing set is 20 percent, the testing set does not participate in model training, the training set does not participate in model evaluation, and if the data suitable for the training set is also suitable for the testing set, minimum overfitting occurs, so that the testing set is in the best state of the model.
In the embodiment, the clustering algorithm is firstly used for clustering and distinguishing data points of different stratum characteristics of stratum rock characteristics, then the neural network algorithm is used for training mechanical parameters corresponding to the stratum rock characteristics of the same cluster, and then the mechanical drilling speed is predicted according to the training model. The neural network models of the two approaches are different: the input parameters of the former are complex and comprise parameters with various properties; the latter has fewer input parameters and more single parameter property.
Comparing the two methods, wherein the two methods adopt the same data set which contains 14000 groups of data, and the two methods adopt the same BP neural network structure which is a hidden layer 1 layer, and the optimal neural network model is obtained by changing the number of neurons in the hidden layer. Fig. 3 is a diagram of the direct prediction convergence process, where the input parameters include formation characteristic parameters such as compressive strength, shale content, etc., and mechanical parameters such as weight on bit, rotational speed, etc., and it can be seen from the diagram that the data set finally converges to 5979 learning times, and the loss value is 0.272. FIG. 3 is a comparison of predicted results and actual values, in which case the rate of penetration accuracy for the randomly partitioned test set is 89.7% based on the calculated results.
As shown in fig. 4 to 5, the stratigraphic features in the data set are clustered by a clustering algorithm, and finally the data set is divided into 6 cluster classes. For each cluster, 20% of test sets are divided for training and test set prediction, the convergence process graphs of 6 clusters are different, the learning frequency of each cluster is different, the learning frequency of cluster 1 is the minimum and is 194, the corresponding data is the minimum, the learning frequency of cluster 5 is the maximum and is 926, and the corresponding data is the maximum. And comparing the prediction results of the 6 cluster test sets with actual values. The predicted value and the actual value of each cluster are relatively close, and the accuracy rate of the prediction results of all the test sets is 96.7%.
As shown in fig. 8, which is a graph of the result of determining the optimal k value by using the elbow method, as the clustering number k increases, the Sum of Squared Errors (SSE) decreases, and when the decrease of the sum of squared errors suddenly slows down, the corresponding k is the optimal clustering number k value. As can be seen from the figure, the decrease of the k value in the range of 12-14 is suddenly slowed down, and when the k value is larger than 14, the k value is still in a decreasing trend, but the decrease is not obvious and is relatively gentle. Therefore, the condition is satisfied by taking any value of k from 12 to 14 by the elbow method. Meanwhile, the contour coefficient method is used as an auxiliary method to verify the result of the simultaneous supplementation of the elbow method. Fig. 9 is a result graph of determining the optimal k value by using the contour coefficient method, where the larger the contour coefficient is, the clearer the contour among clusters is, and the better the clustering effect is. It can be seen from the figure that the contour coefficients of the k values at 5, 14 and 15 have small differences and are all at maximum values, but when k is 5, the square sum of errors still drops significantly, and when k is 15, the square sum of errors already has a significantly slow down trend, so that the optimal cluster number can be determined to be 14.
As shown in fig. 10, the present embodiment clusters the stratigraphic rock features into 14 cluster classes by clustering, and this figure shows a distribution diagram of the dimensionality reduction of each cluster class in a 3-dimensional space after clustering. Fig. 10 is a comparison graph of the number of data points of each cluster after clustering, and it can be seen from the graph that the number of data points of each cluster is greatly different, the cluster 12 has the largest number of data points, and 1275 data are provided, and the cluster 5 has the smallest number of data points, and only 232 data are provided.
Case 1
In this embodiment, for example, the wood connection Qinqing group (5700-. The stratum data points are known to be related to more clusters through clustering, but the number of clusters 3, 4 and 9 is the largest, and the total percentage accounts for 91%, so that the continuous wood Qin group mainly uses the cluster 3, 4 and 9 as main bodies to carry out drilling parameter optimization. Wherein, the cluster 3 is mainly distributed on the upper part of the stratum 5700m-5755m, and the data points account for 17%; cluster 4 is mainly distributed in the upper part 5760m-5882m of the stratum, and the data points account for 32%; clusters 9 were distributed primarily in the lower portion of the formation 5890m-6032m with 42% data points. Other cluster types are distributed more discretely, the data point proportion is smaller, and the parameter optimization analysis is not participated.
And (3) performing mechanical drilling speed prediction on each data point of the wood-connected Qin group belonging to the cluster class 3 by using a corresponding optimal neural network model under each parameter combination, normalizing the prediction results of each data point, and then averaging the results, wherein the results are shown in FIG. 12. And (3) performing mechanical drilling speed prediction on each data point of the wood-connected Qin group belonging to the cluster 4 by using a corresponding optimal neural network model under each parameter combination, normalizing the prediction results of each data point, and then averaging the results, wherein the results are shown in FIG. 14. And (3) performing mechanical drilling speed prediction on each data point of the wood-connected Qin group belonging to the cluster 9 under each parameter combination by using a corresponding optimal neural network model, normalizing the prediction results of each data point, and then averaging the results, wherein the results are shown in fig. 15.
Case 2
In this case, the depth of the calling wall group is 6062-7203m and 1141 data points are taken as an example, the stratum data points relate to more clusters, but the cluster 1, the cluster 6 and the cluster 8 are the most, and the total accounts for 95%, so that the calling wall group performs drilling parameter optimization by taking the cluster 1, the cluster 6 and the cluster 8 as main bodies. Wherein the cluster 8 is mainly distributed on the upper part 6062-6422m of the stratum, and the data points account for 32 percent; cluster 6 is mainly distributed in the upper part 6430 and 7074m of the stratum, and the data points account for 52 percent; cluster 8 is distributed primarily in the lower middle portion 7074 and 7203m of the formation, with 11% data points. Other cluster types are distributed more discretely, the data point proportion is smaller, and the parameter optimization analysis is not participated.
And (3) predicting the drilling rate of the machine under each parameter combination by using each data point of the call wall group belonging to the cluster 8 through a corresponding optimal neural network model, normalizing the prediction result of each data point, and then averaging the result, wherein the result is shown in fig. 16. And (3) predicting the drilling rate of the machine under each parameter combination by using each data point of the calling wall group belonging to the cluster class 6 through a corresponding optimal neural network model, normalizing the prediction result of each data point, and then averaging the result, wherein the result is shown in fig. 17.
Therefore, the method can accurately and reliably predict and is highly consistent with the actual value; compared with the prior art, the method has outstanding substantive features and remarkable progress.
The above-mentioned embodiments are only preferred embodiments of the present invention, and do not limit the scope of the present invention, but all the modifications made by the principles of the present invention and the non-inventive efforts based on the above-mentioned embodiments shall fall within the scope of the present invention.

Claims (7)

1. The drilling parameter optimization method based on machine learning is characterized by comprising the following steps of:
collecting stratum characteristic parameters of an area where a well is drilled, and preprocessing the stratum characteristic parameters;
carrying out stratum characteristic clustering on the pretreated stratum characteristic parameters;
and combining the clustered stratum characteristics and the drilling parameters, predicting the mechanical drilling speed by using a BP (back propagation) neural network or a circulating neural network, and constructing and obtaining a drilling parameter optimization model.
2. The machine-learning based drilling parameter optimization method of claim 1, wherein the formation characteristic parameters comprise formation lithology parameters, shale content, compressive strength, shear strength, internal friction angle, internal friction force, rock hardness, drillability extreme values, and rock abrasiveness parameters.
3. The machine-learning based drilling parameter optimization method of claim 1, wherein formation characteristic parameters are preprocessed, including data cleaning, discretization, normalization, and data dimensionality reduction.
4. The machine learning-based drilling parameter optimization method of claim 1, wherein the formation feature parameters are clustered using one of a K-means algorithm, a MeanShif algorithm, a BDSCAN algorithm, fuzzy clustering, gaussian mixture model-based expectation maximization clustering.
5. The machine learning-based drilling parameter optimization method according to claim 4, further comprising comparing the standard deviation of formation rock mechanics and the coefficient of variation of the formation features before and after clustering to obtain data corresponding to the same category of formation features.
6. The machine learning based drilling parameter optimization method of claim 1 or 2 or 3 or 4, wherein a BP neural network is used for rate of penetration prediction, comprising the steps of:
firstly, carrying out normalization processing on data, wherein the expression is as follows:
Figure FDA0003396101720000011
wherein, XkRepresenting an input sample;
Figure FDA0003396101720000012
a feature representing an input sample; ckRepresenting the normalized input sample; c. Cm kA feature representing the normalized input sample; k represents the number of input samples;
weight ω of signal source to any input sampleijThreshold value μjtRandomly assigning values;
secondly, the input and the output of any neuron of a hidden layer and an output layer in the BP neural network are obtained, and the expression is as follows:
Figure FDA0003396101720000013
Figure FDA0003396101720000021
wherein the content of the first and second substances,
Figure FDA0003396101720000022
representing outputs of hidden layer neurons；
Figure FDA0003396101720000023
Representing the output of the output layer neurons; m represents the total number of hidden layer neurons; n represents the total number of output layer neurons;
thirdly, solving the error between the output layer and the hidden layer, wherein the expression is as follows:
Figure FDA0003396101720000024
Figure FDA0003396101720000025
wherein the content of the first and second substances,
Figure FDA0003396101720000026
representing the error of the output layer;
Figure FDA0003396101720000027
representing a hidden layer error;
the fourth step, updating the weights and thresholds, which are expressed as:
Figure FDA0003396101720000028
Figure FDA0003396101720000029
Figure FDA00033961017200000210
Figure FDA00033961017200000211
wherein r represents the number of error corrections; η represents the learning rate; α represents a momentum coefficient;
the fifth step, setting k as k +1,
Figure FDA00033961017200000212
Figure FDA00033961017200000213
repeating the second step to the fifth step until all k values are trained, and then executing the sixth step;
a sixth step of adding an error function value, which is expressed as:
Figure FDA00033961017200000214
wherein Loss represents the error function value, EkRepresenting a single error value;
and seventhly, repeating the second step to the sixth step until the error function value is smaller than a preset error value.
7. The machine learning based drilling parameter optimization method of claim 1 or 2 or 3 or 4, wherein a recurrent neural network is used for rate of penetration prediction, comprising the steps of:
step one, sequentially inputting data corresponding to the clustered stratum features to an update gate of a recurrent neural network, wherein the expression is as follows:
zt＝σ(Wz·[ht-1,xt]+bf) (3-1)
wherein z istThe presentation finger updates the door; h ist-1An output signal representing a neuron on the same layer; h istAn output signal representing the neuron at that time; x is the number oftInput representing the neuron at this time；WzRepresents the weight of the update gate; sigma represents a sigmoid function; []Representing a table join vector;
and secondly, inputting the data passing through the updating gate into a resetting gate, wherein the expression is as follows:
rt＝σ(Wr·[ht-1,xt]) (3-2)
wherein r istRepresents a reset gate; wrRepresenting the weight of the reset gate;
thirdly, obtaining an undetermined output value, wherein the expression of the undetermined output value is as follows:
Figure FDA0003396101720000031
wherein the content of the first and second substances,
Figure FDA0003396101720000032
represents a pending output value;
Figure FDA0003396101720000033
representing the weight occupied by the undetermined output value;
Figure FDA0003396101720000034
a compensation value representing the output value to be determined;
the fourth step, find the final output value and the output h of the signaltThe expression is as follows:
Figure FDA0003396101720000035
wherein the content of the first and second substances,
Figure FDA0003396101720000036
representing the value of the output to be determined.
CN202111482929.9A 2021-12-07 2021-12-07 Drilling parameter optimization method based on machine learning Pending CN114139458A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111482929.9A CN114139458A (en) 2021-12-07 2021-12-07 Drilling parameter optimization method based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111482929.9A CN114139458A (en) 2021-12-07 2021-12-07 Drilling parameter optimization method based on machine learning

Publications (1)

Publication Number Publication Date
CN114139458A true CN114139458A (en) 2022-03-04

Family

ID=80384284

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111482929.9A Pending CN114139458A (en) 2021-12-07 2021-12-07 Drilling parameter optimization method based on machine learning

Country Status (1)

Country Link
CN (1) CN114139458A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117211969A (en) * 2023-10-17 2023-12-12 江苏省无锡探矿机械总厂有限公司 Energy-saving control method and system for hydraulic drilling machine
CN117211758A (en) * 2023-11-07 2023-12-12 克拉玛依市远山石油科技有限公司 Intelligent drilling control system and method for shallow hole coring
CN117763466A (en) * 2024-02-22 2024-03-26 中石化经纬有限公司 stratum drillability evaluation method and system based on clustering algorithm
CN117910538A (en) * 2024-03-19 2024-04-19 成都三一能源环保技术有限公司 Downhole drilling instrument running state monitoring system based on machine learning

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110807557A (en) * 2019-11-06 2020-02-18 中法渤海地质服务有限公司 Drilling rate prediction method based on BP neural network and drilling rate optimization method based on BP neural network and particle swarm optimization
CN112162316A (en) * 2020-09-28 2021-01-01 北京中恒利华石油技术研究所 High-resolution well-seismic fusion prestack inversion method driven by AVO waveform data
US20210047910A1 (en) * 2018-05-09 2021-02-18 Landmark Graphics Corporation Learning based bayesian optimization for optimizing controllable drilling parameters
CN112796746A (en) * 2021-02-26 2021-05-14 西安石油大学 Drilling method for petroleum geological exploration
CN112901137A (en) * 2021-03-08 2021-06-04 西南石油大学 Deep well drilling mechanical drilling speed prediction method based on deep neural network Sequential model

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210047910A1 (en) * 2018-05-09 2021-02-18 Landmark Graphics Corporation Learning based bayesian optimization for optimizing controllable drilling parameters
CN110807557A (en) * 2019-11-06 2020-02-18 中法渤海地质服务有限公司 Drilling rate prediction method based on BP neural network and drilling rate optimization method based on BP neural network and particle swarm optimization
CN112162316A (en) * 2020-09-28 2021-01-01 北京中恒利华石油技术研究所 High-resolution well-seismic fusion prestack inversion method driven by AVO waveform data
CN112796746A (en) * 2021-02-26 2021-05-14 西安石油大学 Drilling method for petroleum geological exploration
CN112901137A (en) * 2021-03-08 2021-06-04 西南石油大学 Deep well drilling mechanical drilling speed prediction method based on deep neural network Sequential model

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117211969A (en) * 2023-10-17 2023-12-12 江苏省无锡探矿机械总厂有限公司 Energy-saving control method and system for hydraulic drilling machine
CN117211969B (en) * 2023-10-17 2024-03-29 江苏省无锡探矿机械总厂有限公司 Energy-saving control method and system for hydraulic drilling machine
CN117211758A (en) * 2023-11-07 2023-12-12 克拉玛依市远山石油科技有限公司 Intelligent drilling control system and method for shallow hole coring
CN117211758B (en) * 2023-11-07 2024-04-02 克拉玛依市远山石油科技有限公司 Intelligent drilling control system and method for shallow hole coring
CN117763466A (en) * 2024-02-22 2024-03-26 中石化经纬有限公司 stratum drillability evaluation method and system based on clustering algorithm
CN117910538A (en) * 2024-03-19 2024-04-19 成都三一能源环保技术有限公司 Downhole drilling instrument running state monitoring system based on machine learning

Similar Documents

Publication Publication Date Title
CN114139458A (en) Drilling parameter optimization method based on machine learning
Otchere et al. Application of supervised machine learning paradigms in the prediction of petroleum reservoir properties: Comparative analysis of ANN and SVM models
CN111783825A (en) Well logging lithology identification method based on convolutional neural network learning
US7433851B2 (en) System and method for inferring geological classes
CN112083498B (en) Multi-wave earthquake oil and gas reservoir prediction method based on deep neural network
CN109815223A (en) A kind of complementing method and complementing device for industry monitoring shortage of data
CN113610945B (en) Ground stress curve prediction method based on hybrid neural network
CN109407173A (en) Lithology fining and automatic identification method based on Logging Curves
Yan et al. A real-time intelligent lithology identification method based on a dynamic felling strategy weighted random forest algorithm
CN114723095A (en) Missing well logging curve prediction method and device
CN113378998B (en) Stratum lithology while-drilling identification method based on machine learning
Abdel Azim Application of artificial neural network in optimizing the drilling rate of penetration of western desert Egyptian wells
CN116384554A (en) Method and device for predicting mechanical drilling speed, electronic equipment and computer storage medium
Kumar et al. Electrofacies classification–a critical approach
Verma et al. Quantification of sand fraction from seismic attributes using Neuro-Fuzzy approach
CN113033637A (en) Integrated learning lithofacies prediction method based on rock physical sample enhancement
CN110552693A (en) layer interface identification method of induction logging curve based on deep neural network
Kakouei et al. Lithological facies identification in Iranian largest gas field: A comparative study of neural network methods
CN112147679B (en) Lithology prediction method and device based on elastic parameters under fuzzy logic framework
CN110674868A (en) Stratum lithology identification system and method based on high-dimensional drilling parameter information
Wang et al. Auto recognition of carbonate microfacies based on an improved back propagation neural network
CN112034515A (en) Volcanic channel identification method based on unsupervised neural network
Fan et al. Quantitative Prediction of Low-Permeability Sandstone Grain Size Based on Conventional Logging Data by Deep Neural Network-Based BP Algorithm
Wu et al. Classification of reservoir fracture development level by convolution neural network algorithm
de Oliveira et al. Ensemble of heterogeneous classifiers applied to lithofacies classification using logs from different wells

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination