A method for identifying buried hill geological patterns while drilling based on multi-source information fusion

Abstract

The application discloses a buried hill geology pattern identification method while drilling based on multi-source information fusion and belongs to the technical field of oil exploration and development, and comprises the following steps: S1, data collection and processing; S2, stepwise regression analysis method is used to optimize logging element parameters and drilling engineering parameters; S3, a multi-parameter fusion technology based on a LightGBM algorithm is used to build a buried hill geology pattern identification model, namely a LightGBM algorithm model; in the manner, the method combines logging element data and drilling parameters, adopts the stepwise regression analysis method to optimize sensitive parameters, and establishes a geology model identification model through the LightGBM algorithm, so that the identification speed is faster and the accuracy is higher compared with a traditional method; the method can provide a strong basis for efficient on-site exploration and development decision and is favorable to promoting the oil and gas reservoir exploration and development process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil exploration and development, and particularly relates to a drilling-while-drilling buried hill geology pattern identification method based on multi-source information fusion. BACKGROUND

[0002] The research on buried hill geology pattern is very important for increasing reserves and production of oil and gas fields. However, in the research on buried hill geology pattern, the pre-drilling mainly uses seismic data for prediction, which often has a large error. The post-drilling mainly uses seismic, logging, core, field outcrop and production test data for identification, and lacks quick and intelligent evaluation techniques in the drilling process, which brings great challenges to the identification of buried hill geology pattern.

[0003] In order to provide a strong basis for efficient field exploration and development decision-making, and in view of the problem that logging data is less used in pre-drilling prediction and post-drilling evaluation, the present application designs a drilling-while-drilling buried hill geology pattern identification method based on multi-source information fusion to solve the above problems. SUMMARY

[0004] In view of the above shortcomings of the prior art, the present application provides a drilling-while-drilling buried hill geology pattern identification method based on multi-source information fusion.

[0005] To achieve the above purpose, the present application is realized by the following technical scheme:

[0006] A drilling-while-drilling buried hill geology pattern identification method based on multi-source information fusion comprises the following steps:

[0007] S1, data collection and processing;

[0008] S2, using stepwise regression analysis method to optimize logging element parameters and drilling engineering parameters;

[0009] S3, constructing a multi-parameter fusion technology based on LightGBM algorithm in the drilling-while-drilling buried hill geology pattern identification model, i.e. LightGBM algorithm model;

[0010] S4, verifying the LightGBM algorithm model by F1-score index.

[0011] Further, step S1 specifically comprises the following steps:

[0012] (1) using element logging technology to continuously analyze the cuttings to obtain the changes of constant elements and trace elements in the cuttings;

[0013] (2) pre-processing the samples to eliminate some trace elements with very low content;

[0014] (3) The drilling parameters are fused with the element parameters to form data samples after being denoised by quartile method and processed by five-point cubic smoothing, and finally used as the input of the model.

[0015] Further, the five-point cubic smoothing processing formula is as follows:

[0016]

[0017] wherein y represents the data after smoothing processing, x represents the sample data without smoothing processing, the value range is [1, n], n represents the sample quantity, and i represents the sample serial number.

[0018] Further, the step S2 specifically comprises the following steps:

[0019] 1) Given a significance level a, the critical value of the rejection region F a (m, n-m-1) is determined according to the degrees of freedom m (the number of characteristics) and n-m-1, for each characteristic parameter x i (1≤i≤m), m one-dimensional linear regression models are fitted, the statistic Q is regarded as an empty set, i.e. SSE(Q)=SST, then SST=(x i |Q)=SSR(x i ), MSE(Q, x i )=MSE(x i ), and the calculation is performed for each i:

[0020]

[0021] F i1 =max 1≤i≤m F i Formula 2

[0022] SSE-the sum of error squares of the corresponding points of the fitted data and the original data; SSR-the sum of squares of the difference between the predicted data and the mean of the original data; SST-the sum of squares of the difference between the original data and the mean; MSE-the mean square error;

[0023] If F im >F a, the regression model of the characteristic parameter x i is selected as the current model, otherwise the model does not introduce the independent variable;

[0024] 2) The remaining m-1 characteristic parameters are added to the current model respectively, m-1 binary regression models are obtained, and then:

[0025]

[0026] F i2 =max i≠i1 F i Formula 4

[0027] If F i2 If F > Fα, the selection ends, and the selected model parameters are optimal; if F i2 If <Fα, then the characteristic parameter x i2 Select the model, and calculate x according to the above formula. i2 Does it still have a significant impact on the type of geological model?

[0028]

[0029] If F i2 If ≤Fα, then x needs to be removed. i1 Contains x i2 The model is the optimal model;

[0030] 3) Based on the model selected in step 2, add the remaining m-2 feature parameters to the current model, fit the model and calculate its F value to determine whether to introduce the parameter.

[0031] 4) Repeat the above steps until all feature parameters are selected or eliminated. The remaining parameters are the optimal parameters.

[0032] Furthermore, a stepwise regression analysis method was adopted to optimize drilling engineering parameters. With geological model type as the dependent variable, drilling time, drilling pressure, drilling speed, torque, Dc index, and fracture pressure gradient were analyzed stepwise in sequence. The parameters finally selected were fracture pressure gradient, drilling time, drilling speed, torque, and drilling pressure.

[0033] Furthermore, in step S3, the selected logging elements and drilling engineering parameters are used as inputs to the model, and the geological pattern type after label encoding is used as the output of the model, thus establishing a geological pattern identification model based on the LightGBM algorithm.

[0034] Furthermore, in step S4, the F1-score is used for comprehensive evaluation. The formula for calculating the F1-score is as follows:

[0035] precision = TP / (TP + FP) (Formula 6)

[0036] recall = TP / (TP + FN) (Formula 7)

[0037] F1 score ==2×precision×recall / (precision+recall) Formula 8

[0038] Precision – Detection rate; Recall – Detection rate; TP – True Positive Instances; FP – False Positive Instances; FN – False Negative Instances.

[0039] Advantages

[0040] The method of the present application utilizes the combination of logging element data and drilling parameters, adopts stepwise regression analysis method to optimize sensitive parameters, and establishes a geological model identification model through LightGBM algorithm, which is faster and more accurate than traditional methods; the method can provide strong basis for efficient field exploration and development decision, and is conducive to promoting the process of oil and gas reservoir exploration and development. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0042] Figure 1 is a regression standardized residual plot;

[0043] Figure 2 is a measured cumulative probability plot;

[0044] Figure 3 is a comparison chart of prediction accuracy of three models;

[0045] Figure 4 is a geological model prediction chart of well N1 in Bozhong A gas field. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0047] The present application will be further described in combination with the embodiments.

[0048] Embodiment 1

[0049] The present embodiment provides a drilling buried hill geological model identification method based on multi-source information fusion, comprising the following steps:

[0050] S1, data collection and processing;

[0051] Step S1 specifically comprises the following steps:

[0052] (1) Using elemental logging technology to analyze the cuttings continuously, and obtaining the changes of constant elements and trace elements in the cuttings;

[0053] The reservoir space is developed in the weathering zone, and the formation is easy to drill. The drilling engineering parameters have good correlation with the development of effective reservoirs.

[0054] In this embodiment, 17 elements measured by elemental logging can be selected: Na, Mg, Al, Si, P, S, Cl, K, Ca, Ba, Ti, Mn, Fe, V, Ni, Sr and Zr, containing main elements such as Na, Mg, Al, Ca and Si, and some trace elements such as P, Cl, Sr and V. Combined with basic drilling parameters such as torque, fracture pressure gradient, drilling pressure, drilling speed and drilling time, the drilling potential buried hill geological model can be identified;

[0055] (2) Preprocessing the sample, and removing some trace elements with very low content;

[0056] In this embodiment, since the content of some trace elements measured by elemental logging is very low, the elements V, Ni, Sr and Zr are removed, and the main elements and secondary elements Na, Mg, Al, Si, P, S, Cl, K, Ca, Ba, Ti, Mn and Fe are analyzed and optimized to establish the evaluation model. In addition, since the sampling interval of elemental logging is 5m per data point, and the corresponding drilling parameter sampling interval is 1m, in order to strictly ensure the consistency of the data in the depth scale, the average value of the drilling engineering parameters at the depth of 5m corresponding to the elements is taken as the drilling parameter value at the depth;

[0057] (3) The drilling parameters are denoised by quartile method and five-point cubic smoothing processing, and then fused with the element parameters to form data samples, which are finally used as the input of the model;

[0058] The five-point cubic smoothing processing formula is as follows:

[0059]

[0060] Wherein, y represents the data after smoothing processing, x represents the sample data without smoothing processing, the value range is [1, n], n represents the sample quantity, and i represents the sample serial number;

[0061] S2, using stepwise regression analysis method to optimize logging element parameters and drilling engineering parameters;

[0062] Through stepwise regression analysis method, the parameters that have lost importance after F check when introducing new variables are removed, so that important variables are always maintained in the model, the dimension of the model input is reduced, and the risk of overfitting is prevented;

[0063] Step S2 specifically comprises the following steps:

[0064] 1) Given a significance level a, determine the critical value of the rejection region F (m, n-m-1) according to the degrees of freedom m (number of characteristics) and n-m-1, for each characteristic parameter xi (1≤i≤m), fit m one-dimensional linear regression models, take the statistic Q as an empty set, i.e. SSE(Q)=SST, then SST=(x i |Q)=SSR(x i ), MSE(Q, x i )=MSE(x i ), calculate for each i:

[0065]

[0066] F i1 =max 1≤i≤m F i Equation 2

[0067] SSE - sum of squared errors of fitted data and corresponding points of original data; SSR - sum of squared differences between predicted data and mean of original data; SST - sum of squared differences between original data and mean; MSE - mean squared error;

[0068] If F im >Fα, select the regression model of characteristic parameter x i as the current model, otherwise the model does not introduce independent variables;

[0069] 2) Add the remaining m-1 characteristic parameters to the current model respectively, obtain m-1 binary regression models, then:

[0070]

[0071] F i2 =max i≠i1 F i Equation 4

[0072] If F i2 >Fα, end selection, and the selected model parameters are optimal; if F i2 <Fα, select the characteristic parameter x i2 into the model, calculate the introduction of x i2 according to the above equation, and determine whether the characteristic parameter x i2 still has a significant impact on the geological pattern type (dependent variable y):

[0073]

[0074] If F i2 ≤Fα, x i1 needs to be removed, and the model containing x i2 is the optimal model;

[0075] 3) On the basis of the model selected in step 2, the remaining m-2 characteristic parameters are added to the current model, fitting and calculating the F value to determine whether to introduce the parameter;

[0076] 4) The above steps are cycled until all characteristic parameters are selected or rejected, and the remaining parameters are the preferred parameters;

[0077] In this embodiment, the geological pattern type is taken as the dependent variable, and the screened elements are taken as the independent variables for sequential regression analysis. Stepwise regression analysis is to introduce one variable at a time according to the importance of the variable. The first introduced variable is Al. At this time, the sum of squares of the mean difference between the predicted type and the actual type SSR=23.669, the degree of freedom is 1, the sum of squares of the error between the predicted type and the actual type SSE=163.277, the degree of freedom is 756, and the mean square error can be obtained Average residual sum of squares Therefore, the result of F test Due to the introduction of Al element, the multiple determination coefficient R 2 of the model is 0.127, the significance P=0.000b<0.05, so the regression coefficient is significant, and according to the regression principle, Al should be introduced in the optimal variable. Similarly, one element is introduced at a time, and the above steps are cycled. The finally selected parameters are Al, S, Si, Mn, Na, Ca, K, Fe, and the remaining elements are rejected. Figure 1 is the regression standardized residual plot; Figure 2 is the measured cumulative probability plot; it can be seen that the final characteristic distribution generally obeys the normal distribution, and the curve is close to the diagonal line, indicating that the model fitting effect is good.

[0078] In this embodiment, the stepwise regression analysis method is also used to optimize the drilling engineering parameters. The geological pattern type is taken as the dependent variable, and the drilling time, drilling pressure, drilling speed, torque, Dc index, and fracture pressure gradient are taken as the independent variables for sequential analysis. The finally selected parameters are fracture pressure gradient, drilling time, drilling speed, torque, and drilling pressure.

[0079] S3, construct a multi-parameter fusion technology based on the LightGBM algorithm for the while-drilling buried hill geology pattern identification model, i.e. the LightGBM algorithm model;

[0080] The LightGBM algorithm model based on gradient boosting tree is constructed, not only supports multi-class features, but also saves time and space overheads in single-side gradient sampling compared with traditional machine learning algorithms in traversing all feature values, supports efficient parallel training, has faster training speed, lower memory consumption, better accuracy, supports distribution and can quickly process massive data and the like;

[0081] In the embodiment, 608 data combinations of 10 wells are used as training samples, the preferred logging elements (Al, S, Si, Mn, Na, Ca, K, Fe) and drilling engineering parameters (fracture pressure gradient, drilling time, drilling speed, torque, drilling pressure) are used as inputs of the model, and the labeled coded geological pattern type is used as the output of the model, and a geological pattern identification model based on the LightGBM algorithm is established.

[0082] S4, verify the LightGBM algorithm model through the F1-score index;

[0083] In order to better verify the accuracy of the model, a set of comprehensive evaluation indexes of the model need to be established; in general, the pattern recognition evaluation index in model verification is measured by prediction accuracy, but it is unreasonable to make prediction model evaluation according to the index only, especially for uneven sample quantity, the accuracy cannot truly reflect the prediction ability of the model, therefore, the F1-score index is proposed for comprehensive evaluation; the calculation formula of the F1-score is as follows:

[0084] precision=TP / (TP+FP) Formula 6

[0085] recall=TP / (TP+FN) Formula 7

[0086] F1 score =2×precision×recall / (precision+recall) Formula 8

[0087] precision-precision rate; recall-recall rate; the classification results of pattern recognition can generally be divided into four categories, TP-true example, FP-false positive example, FN-false negative example and TN-true negative example;

[0088] The identification accuracy of the weathering zone and the inner zone by the LightGB algorithm model verifies the advancement and superiority of the LightGBM algorithm model;

[0089] In the embodiment, 150 data combinations of 2 wells are used as test samples to verify the effect of the LightGBM algorithm model; Figure 3The contrast chart of the prediction accuracy of the three models; from the F1-score value of the LightGBM algorithm model, it can be seen that the accuracy of the LightGBM algorithm model for identifying weathering zone is as high as 96.7%, and the accuracy of identifying the inner zone is 95.8%; through verification, it can be found that the LightGBM algorithm model performs well, which shows that the tree model based on gradient boosting fully embodies the advancement and superiority in small layer fine division;

[0090] The method combines logging element data and drilling parameters, uses stepwise regression analysis to optimize sensitive parameters, and establishes a geological model identification model through the LightGBM algorithm, which is faster and more accurate than traditional methods; the method can provide strong basis for efficient exploration and development decision-making, and is conducive to promoting the process of oil and gas reservoir exploration and development;

[0091] Taking Bohzhong A gas field as an example, a multi-parameter while-drilling buried hill geological pattern identification model is established by the method, first, the collected logging element data and basic drilling parameters are preprocessed; second, the stepwise regression analysis method is used to optimize the parameters; then the LightGBM machine learning model is established; finally, the LightGBM model is verified by F1-score index. Figure 4 The geological pattern prediction chart of N1 well in Bohzhong A gas field.

[0092] The above examples are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying buried hill geologic patterns while drilling based on multi-source information fusion, characterized in that, Comprise the following steps: S1, data collection and processing; S2, using stepwise regression analysis method to optimize logging element parameters and drilling engineering parameters; 1) Given a significance level , determine the critical value of the rejection region and based on the degrees of freedom , for each feature parameter ( ), fit one-dimensional linear regression models, treating the statistic Q as an empty set, i.e. then , calculate for each : Formula 1 ; Formula 2; — sum of squares of errors of corresponding points of fitted data and original data; — sum of squares of differences of predicted data from mean of original data; — sum of squares of differences of original data from mean; — mean square error; If , the regression model of the selected characteristic parameter is the current model, otherwise the model does not introduce independent variables; 2) the remaining characteristic parameters are added to the current model, obtaining binary regression models, then: Formula 3: Formula 4: If , the selection is finished and the selected model parameter is the optimum; if , the characteristic parameter is selected into the model and the introduced is calculated according to the above formula, and whether there is still a significant influence on the geological pattern type: Equation 5; If , then the pixel needs to be culled , the model containing is the optimal model; 3) On the basis of the model selected in step 2, the remaining characteristic parameters are added to the current model, fitting is performed and the F value is calculated, thereby determining whether the parameter is introduced or not. 3) On the basis of the model selected in step 2, the remaining characteristic parameters are added to the current model, fitting is performed and the F value is calculated, thereby determining whether the parameter is introduced or not. 4) cycle the above steps until all characteristic parameters are selected or rejected, the remaining parameters are the preferred parameters; Take stepwise regression analysis method to optimize drilling engineering parameters, take drilling time, drilling pressure, drilling speed, torque, Dc index, fracture pressure gradient as independent variables, and take geological pattern type as dependent variable, and analyze step by step, finally, the optimized parameters are fracture pressure gradient, drilling time, drilling speed, torque and drilling pressure; S3, constructing a multi-parameter fusion technology based on LightGBM algorithm in the geologic pattern identification model while drilling buried hill, namely LightGBM algorithm model; S4, verify the LightGBM algorithm model by F1-score index.

2. The method according to claim 1, wherein, Step S1 specifically comprises the following steps: (1) continuously analyze the rock debris by using element logging technology, and obtain the changes of constant elements and trace elements in the rock debris; (2) pretreat the sample, and remove some trace elements with extremely low content; (3) after the drilling parameters are denoised by quartile method and processed by five-point cubic smoothing, the element parameters are fused to form data samples, which are finally used as the input of the model.

3. The method according to claim 2, wherein, The formula of five-point cubic smoothing processing is as follows: ； Wherein, y represents the data after smoothing processing, x represents the sample data without smoothing processing, the value range is [1, n], n represents the sample quantity, and i represents the sample serial number.

4. The method according to claim 1, wherein, In step S3, the optimized logging element and drilling engineering parameter are used as the input of the model, the labeled coded geological pattern type is used as the output of the model, and the geological pattern identification model based on LightGBM algorithm is established.

5. The method according to claim 1, wherein, In step S4, F1-score index is used for comprehensive evaluation, and the calculation formula of F1-score is as follows: Formula 6: Formula 7: Formula 8: - precision; - recall; - true positives, - false positives, - false negatives.

Images