ARTIFICIAL INTELLIGENCE IDENTIFICATION METHOD FOR SHALE PLATINUM BOX BASED ON ELEMENT LOGGING

Description

TECHNICAL FIELD

The present disclosure relates to the technical field of geosteering of horizontal well drilling, in particular to an artificial intelligence identification method for shale platinum box based on element logging.

BACKGROUND

Shale oil and gas, as an important unconventional oil and gas resource, has a relatively thin high-quality sub-layer, so it is necessary to determine the relative positions of the current bottom hole and high-quality shale sub-layer in real time during horizontal well drilling for shale oil and gas, thus adjusting the well trajectory in time, and improving the drilling rate of the high-quality shale oil and gas reservoirs. In engineering, the high-quality shale sub-layer is defined as a platinum box, and whether the current bottom hole is inside the platinum box can be determined in real time during the horizontal well drilling, so the shale platinum box identification can be transformed into the identification of the shale sub-layer to which the bottom hole belongs. At present, the main mode to identify the shale sub-layer to which the bottom hole belongs is as follows: engineers at the site analyze the characteristics of shale sub-layers adjacent to a completely-drilled horizontal well and a vertical well on the element logging curve, and establish the element characteristic patterns of the shale sub-layers qualitatively or quantitatively, which can be used as the identification basis for the current horizontal well drilling.

Visibly, the current technology is mainly manual analysis. Apparently, the manual analysis is difficult to grasp the change characteristics of all element curves. In addition, due to measurement errors and irregular operation of workers at the site, the element content curve usually has some noise, while the manual analysis method is easily affected by the noise. On the other hand, the recognition rate of manual analysis is low, so it is difficult to separate layers with thin thickness from the element logging curve, and it is even more difficult to identify the shale sub-layer for each sample point.

In conclusion, the method for identifying the shale sub-layer by element logging is subjective, leading to low reliability of identification results and recognition rate, and poor degree of automation of identification process.

SUMMARY

An objective of the present disclosure is to provide an artificial intelligence identification method for a shale platinum box based on element logging. The artificial intelligence identification method for the shale platinum box based on element logging is designed by taking a deep neural network as the core for shale sub-layer identification and the grading and progressive calculation as an implementation mode for accurately locating shale sub-layer By means of the method, the automation of shale sub-layer identification can be achieved, the objectivity and reliability of an identification result are improved, and the recognition rate of the shale sub-layer identification is increased.

The technical solution of the present disclosure is as follows:

An artificial intelligence identification method for a shale platinum box based on element logging includes the following steps:

• • Step S1, extracting a logging-while-drilling element curve of a horizontal well;
  • Step S2, designing a structure of a shale platinum box quantitative identification method; and
  • Step S3, designing application process of the shale platinum box quantitative identification method.

Specifically, the Step S2 further includes the following sub-steps:

• • Step S21, designing the structure of the shale platinum box quantitative identification method based on a three-factor orthogonal method;
  • Step S22, a generalizability of a vertical well on the horizontal well in a shale platinum box quantitative identification method is evaluated; and
  • Step S23, the structure of the shale platinum box quantitative identification method is optimized.

Specifically, the structure in the Step S21 is a neural network structure which includes a framework, a number of modules, and an activation function.

The framework includes a convolutional neural network, a recurrent neural network, or a perceptron neural network.

The number of modules is 2, 3 or 4.

The activation function comprises Rectified Linear Unit (ReLU), Leaky Rectified Linear Unit (LeakyReLU), or Hyperbolic tangent (Tan h).

Specifically, the Step S22 includes screening a structure design scheme with excellent performance based on a sensitive element curve of a pilot well corresponding to an adjacent completely drilled horizontal well and a validation accuracy of a shale platinum box identification result to prediction data, that is, the generalizability of the method is taken as an evaluation index.

In some embodiments, the structure design scheme with excellent performance is a recurrent neural network with three recurrent layers, and the activation function is Tan h, or LeakyReLU.

Specifically, the optimization in the Step S23 includes introducing a Dropout method to achieve positive feedback.

Specifically, the Step S3 further includes the following sub-steps:

• • Step S31, designing a grading and progressive calculation route of the shale platinum box quantitative identification method; and
  • Step S32, designing a prediction route of a “pilot well-landing section-horizontal section” three-stage quantitative identification method prediction route.

Specifically, the Step S31 includes: further predicting a position of a next-level platinum box based on a prediction result of a previous-level subdivision, so as to forming a progressive method calculation route.

Specifically, the Step S32 includes:

• • for the pilot well, obtaining a detailed platinum box identification result through a well log after finishing drilling, and training level by level a plurality of identification method models at different subdivision levels by using an optimized identification method according to the grading and progressive calculation route and identification accuracy requirements; and
  • for the landing section, predicting a position of the platinum box by using the multiple identification method models which are trained by the pilot well, and using a prediction result and an identification result of the pilot well to jointly train the plurality of identification method models at different subdivision levels, thereby identifying a platinum box in the horizontal section.

The present disclosure has the beneficial effects that:

The artificial intelligence identification method for the shale platinum box based on element logging designed by the present disclosure can achieve automation of shale sub-layer identification, improve the objectivity and reliability of identification results, and increase the recognition rate of shale sub-layer identification.

BRIEF DESCRIPTION OF THE DRAWINGS

To describe the technical solutions of the present disclosure or in the prior art more clearly, the following briefly introduces the accompanying drawings required for describing the embodiments or the prior art. Apparently, the accompanying drawings in the following description show merely some embodiments of the present disclosure, and those of ordinary skill in the art may still derive other drawings from these accompanying drawings without creative efforts.

FIG. 1 is a diagram of technical route according to the present disclosure;

FIG. 2 is a schematic diagram of a sensitive element curve of a pilot well corresponding to a case horizontal well according to the present disclosure;

FIG. 3 is a diagram of a structure of a shale platinum box identification method for the case horizontal well according to the present disclosure;

FIG. 4 is a diagram of a grading and progressive calculation route according to the present disclosure;

FIG. 5 is a diagram of a three-stage quantitative identification method prediction route;

FIG. 6 is a diagram of a platinum box identification result for the case horizontal well according to the present disclosure.

DETAILED DESCRIPTION OF THE EMBODIMENTS

It should be understood that specific embodiments described here are only used to explain rather than limiting the present disclosure.

To understand the technical features, objectives and beneficial effects of the present disclosure more clearly, the technical solution of the present disclosure is described in detail below. Apparently, the described embodiments are merely a part rather than all of the embodiments of the present disclosure, and thus cannot be construed as the limitation of implementable range of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

As shown in FIG. 1, the present disclosure provides an artificial intelligence identification method for a shale platinum box based on element logging. According to this method, firstly, the sensitivity of all element logging curves to the change of a shale sub-layer is analyzed to extract a sensitive element curve. Secondly, a three-factor orthogonal method is used to design a deep neural network structure from three aspects: a layer type, the number of layers, and the number of layer nodes. After learning an element curve of a vertical well and the distribution characteristics of a platinum box through the deep neural network formed by different structural design schemes, the performance of different design schemes is evaluated based on the generalizability of a horizontal well, thus selecting an optimum structural design scheme and optimizing the network performance by combining different functional modules of the deep neural networks. Finally, a calculation route of the identification algorithm is designed in a grading and progressive way, and a prediction route of a “pilot well-horizontal well landing section-horizontal well horizontal section” three-stage shale platinum box quantitative identification algorithm is designed according to the actual horizontal well drilling engineering process. By means of the shale platinum box artificial intelligence identification method, the automation of shale sub-layer identification can be achieved, the objectivity and reliability of an identification result are improved, and the recognition rate of the shale sub-layer identification is increased.

1. Extraction of Logging-while-Drilling Element Curve of Horizontal Well

According to a shale platinum box identification result of an adjacent completely drilled horizontal well, the sensitivity of each element curve to the shale platinum box identification is analyzed to screen sensitive elements as the data basis for identification, and a pilot well corresponding to a horizontal well to be drilled is used to verify the effectiveness of the sensitive elements.

A case horizontal well is located in a shale gas block in Luzhou, southern Sichuan. Through the element curves of the adjacent completely drilled horizontal well and a platinum box identification result, nine sensitive element curves of aluminum, silicon, calcium, thorium, uranium, magnesium, nickel, zinc and manganese are determined. FIG. 2 shows change characteristics of these nine sensitive element curves on the pilot well corresponding to the horizontal well to be drilled. Apparently, these curves have significant change characteristics at an interface of each sub-layer, and there are also significant numerical differences within the sub-layer. Therefore, it is feasible to carry out shale platinum box identification during horizontal well drilling by using these nine sensitive element curves.

2. Design of a Structure of a Shale Platinum Box Quantitative Identification Algorithm

A three-factor orthogonal method is used to design an algorithm structure from three aspects: a framework, the number of modules and an activation function, thus determining a verification route of the structural scheme. Using the generalizability of the vertical well on the shale platinum box identification in the horizontal well as a criterion for scheme evaluation, the performance of the structural design schemes is compared to select the structural design scheme with the best generalizability; and algorithm performance is synthesized to optimize the function module, thus forming a deep-learning shale platinum box identification algorithm.

{circle around (1)} Design a Structure of Shale Platinum Box Quantitative Identification Algorithm Based on three-Factor Orthogonal Method

The identification algorithm structure is designed from three levels of three factors: the framework used in the algorithm, the number of modules corresponding to the framework, and the activation function. According to the influence degree of each factor on the algorithm performance, a verification route of the design scheme is determined.

The factors affecting the performance of the neural network include: a framework, the number of modules, the number of module nodes, an activation function, an optimizer, training parameters, data preprocessing modes, etc. Three relatively important factors which are the framework, the number of modules and the activation function are selected from the factors as the basic elements of structural design. Aiming at the framework of the neural network algorithm, convolutional neural networks, recurrent neural networks and perceptron neural networks are mainly considered. For the number of modules, three levels are mainly considered: 2, 3 and 4. For the activation function, Rectified Linear Unit (ReLU), Leaky Rectified Linear Unit (LeakyReLU) and Hyperbolic tangent (Tan h) are mainly considered. Direct comparison of twenty-seven structural design schemes with the three factors and three levels significantly increases the time cost of the whole set of method, so an orthogonal design method is used to formulate the structural design scheme. According to the importance of the influence on the performance of the neural network, the performance of three basic frameworks is compared first, and the influences of three module numbers on the performance of the neural network are further compared on the basis of the determined framework. Finally, the activation function type is determined on the basis that the framework and the number of modules are clear.

{circle around (2)} Evaluation of Generalizability of Vertical Well on Horizontal Well by Shale Platinum Box quantitative Identification Algorithm

The sensitive element curve of the pilot well corresponding to the adjacent completely drilled horizontal well and the shale platinum box identification result are used as training data, the sensitive element curves and the identification result of the completely drilled horizontal well are used as prediction data, and after training the identification algorithm according to the training data, a validation accuracy of the prediction data, i.e., generalizability of the algorithm, is used as an evaluation index to screen the structural design scheme with excellent performance.

As shown in Table 1, in order to compare the performance of the neural network under each scheme, the element data of the pilot well corresponding to an adjacent completely drilled well and the platinum box identification result are used as training data to investigate the generalizability of different schemes on the horizontal well, which is used as a screening indicator. According to the structural scheme designed in {circle around (1)}, the validation results shown in Table 1 are obtained, from which it can be seen that the recurrent neural network is used as the basic framework of the network and three recurrent layers are constructed. When the activation function of each recurrent layer is Tan h or LeakyReLU, the performance of the identification algorithm reaches the maximum. Although the performance of the neural network using LeakyReLU is slightly lower than that using Tan h, the difference therebetween is small, and if the performance is the average of the maximum values for many times, the performance of the neural network using LeakyReLU is equivalent to that of the neural network using Tan h.

{circle around (3)} Optimization of the Structure of Shale Platinum Box Quantitative Identification Algorithm

Compared with the improvement degree of the algorithm by the commonly used functional module with optimized generalizability, an appropriate functional module is embedded in the algorithm to form a final shale platinum box identification algorithm.

On the basis of {circle around (2)}, the performance changes of the algorithm are compared when adjustment modules such as Dropout, residual connection and layer normalization are introduced into the neural network, thus selecting an adjustment module with a positive feedback to the algorithm performance to form the final shale platinum box identification algorithm. Comparison results show that the Dropout module can slightly change the performance of the identification algorithm, while the other modules have no significant impact on the performance of the algorithm. The finally formed identification algorithm structure is as shown in FIG. 3.

3. Design of Application Process of Shale Platinum Box Quantitative Identification Algorithm

The platinum box is subdivided level by level according to specified subdivision levels. For each subdivision level, an identification algorithm is trained separately for prediction, thus forming a grading and progressive identification algorithm calculation route. According to the field horizontal well drilling process, the grading and progressive identification algorithm calculation is carried out for a pilot well, a landing section of the horizontal well and a horizontal section of the horizontal well, thus forming a three-stage shale platinum box prediction.

{circle around (1)} Design of Grading and Progressive Calculation Route of Shale Platinum Box Quantitative identification Algorithm

According to the actual engineering requirements, the subdivision levels of the shale platinum box are determined, and an identification algorithm is trained separately for each subdivision level. During prediction, the corresponding identification algorithm is used to calculate the position of the platinum box at the bottom hole at the current subdivision level according to the subdivision level from coarse to fine, thus obtaining an accurate platinum box position at the bottom hole.

In the actual horizontal well drilling engineering, the determination of the current bottom hole position usually needs to be accurate to upper, middle and lower positions in a specific layer, and the accuracy requirement is higher for a complex well section. Therefore, a grading and progressive prediction process with level-by-level subdivision is designed. One identification algorithm is used to predict the sub-layer to which the current bottom hole belongs, and the sub-layer can be regarded as a primary subdivision of the platinum box. On the basis of determining the sub-layer, another identification algorithm is used to further predict the upper, middle and lower positions inside the sub-layer (a secondary subdivision of the platinum box), and so on. Multiple identification algorithm models are used to subdivide the platinum box level by level, and the position of the next-level platinum box is further predicted based on the prediction result of the previous subdivision, thus forming a progressive algorithm calculation route, as shown in FIG. 4.

{circle around (2)} Design of the Prediction Route of the “Pilot Well-Landing Section-Horizontal Section” Three-Stage Quantitative Identification Algorithm

For the pilot well, a training on the grading and progressive identification algorithm is carried out. The trained algorithm is used to predict the landing section of the horizontal well, and a prediction result is added to a training set. The identification algorithm is further trained in grading, and the platinum box in the horizontal section is predicted using the trained identification algorithm.

Before horizontal well drilling, a pilot well is usually drilled as the basis for evaluating the scope of platinum box, the distribution characteristics of high-quality reservoirs and the storage and permeability performance of sub-layers of the horizontal well. During horizontal well drilling, a well trajectory usually goes through a deflecting section, the landing section and the horizontal section, in which the deflecting section is away from a platinum box covering horizon, so the platinum box prediction is usually carried out only for the landing section and the horizontal section.

For the pilot well, a detailed platinum box identification result can be obtained through well logs after finishing drilling, and multiple identification algorithm models at different subdivision levels are trained level by level by using an optimized identification algorithm according to the flow shown in FIG. 4 and identification accuracy requirements. As shown in FIG. 5, for the landing section of a horizontal well, the multiple identification algorithm models trained by the pilot well are used to predict a position of the platinum box, and a prediction result and an identification result of the pilot well are used to jointly train multiple identification algorithm models at different subdivision levels, thereby identifying the platinum box in the horizontal section.

After the above three technical links, the platinum box identification result of the horizontal well can be obtained. The final platinum box identification result of the case horizontal well is shown in FIG. 6. The platinum box is subjected to secondary subdivision by the well, in which 1 sub-layer and 2 sub-layer covered by the platinum box are both divided into two parts: A and B.

The basic principle, main features and advantages of the present disclosure have been shown and described above. It should be understood by those skilled in the art that the present disclosure is not limited by the above embodiments, and the description in the above embodiments and the specification are only the illustration of the principle of the present disclosure. There will be various changes and improvements without departing from the spirit and scope of the present disclosure, all of which shall fall within the scope of protection of the present disclosure. The scope of the present disclosure is defined by the appended claim and their equivalents.

It should be noted that for the sake of simple description, all the above method embodiments are expressed as a series of action combinations, but those skilled in the art should know that the present disclosure is not limited by the described action sequence, because some steps can be performed in other sequences or at the same time according to the present disclosure. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and units involved are not necessarily required for the present disclosure.

In the above embodiments, the description of each embodiment has its own emphasis. The parts that are not described in detail in an embodiment can refer to the relevant descriptions of other embodiments.

Those skilled in the art can understand that all or part of the processes in the method for implementing the above embodiment methods can be completed by instructing related hardware through a computer program, which can be stored in a nonvolatile computer-readable storage medium, and when executed, the program can include the processes of the above-mentioned embodiments. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random-access memory (RAM) and the like.

All the above are only the preferred embodiments of the present disclosure, which certainly should not be used to limit the scope of the present disclosure, so the equivalent changes made according to the claims of the present disclosure still fall within the scope of the present disclosure.