Depositional facies recognition method based on deep belief network

Abstract

The application discloses a sedimentary microfacies recognition method based on a deep belief network, and relates to the technical field of intelligent information processing, and comprises the following steps: firstly, combining and optimizing logging curves; then, using original data of the selected logging curves to construct binary images reflecting curve shape features, taking the binary images as sample feature data, and using a deep belief network based on a restricted Boltzmann machine to capture the mapping relationship between the sample features and the categories thereof. The trained network can be used for single-well profile sedimentary microfacies recognition. With the powerful learning ability and generalization ability of the deep belief network, the recognition precision of the sedimentary microfacies can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent information processing, and particularly relates to a sedimentary microfacies recognition method based on a deep belief network. BACKGROUND

[0002] Reservoir description is a qualitative or quantitative description of a reservoir, thereby comprehensively evaluating the reservoir, and determining the type and spatial distribution of sedimentary facies and sedimentary microfacies is one of the important research contents. The traditional sedimentary facies research mainly identifies the sedimentary environment by observing the rock composition, structure and sedimentary rhythm of the core. The prerequisite for carrying out the sedimentary facies research is to obtain sufficient coring data in the research well section, and the high cost of coring work cannot guarantee the provision of sufficient coring data. The increasingly developed logging technology gives each well rich logging data, and these logging data contain rich sedimentary environment information, so the combination of coring data and logging data for sedimentary facies research has become the current mainstream.

[0003] The curve shape features in well logging data have been used to identify sedimentary environment by many geologists in manual processing, but the results are subjective, time-consuming and vary from person to person. In order to make full use of well logging data, many scholars have introduced various advanced technologies into the identification of sedimentary microfacies, among which the most popular is neural network technology. Xu Shaohua et al. proposed a sedimentary microfacies automatic identification method based on the combination of neural network and image processing technology, which effectively improved the learning efficiency and generalization ability of neural network. They also applied regular fuzzy neural network to the identification of reservoir sedimentary microfacies, which has good adaptability and practicality in solving the problem of sedimentary microfacies identification. Luo Li, Lu Song, Wu Can-can, Cheng Liang-bing, Pang Guoyin, Ma Kui, Liu Lingyun, etc. used BP (BackPropagation) neural network, genetic neural network, probabilistic neural network and other technologies to automatically identify sedimentary microfacies from actual well logging data, with remarkable results. Zhang Fuming et al. extracted feature parameters from well logging curves, optimized the curve combination that was sensitive to sedimentary facies based on core observation and regional experience, and used neural network pattern recognition technology to automatically interpret single well section sedimentary microfacies. Li Wenbo et al. applied self-organizing neural network method to sedimentary microfacies identification, which extracted geometric and image feature parameters of well logging curves, used these feature parameters to establish the relationship between sedimentary microfacies types and well logging curve shapes, developed a microfacies pattern recognition system, and achieved good recognition results. Xia Zhen et al. studied an active layer-based sedimentary microfacies automatic identification algorithm, which obtained the variation characteristics of well logging curves from a set of well logging response data, divided the stratigraphic section into several electric facies using these characteristics, established a correlation model, and automatically identified sedimentary microfacies. Duan Dongpeng et al. used an improved BP neural network method to realize the automatic identification of sedimentary microfacies in high-density well spacing areas and a large number of well areas, which shortened the working time and improved the work efficiency. Mei Junwei et al. constructed a sedimentary microfacies identification system, which used principal component analysis and back propagation neural network based on genetic algorithm, reduced the amount of calculation and avoided falling into local optimal solution, and effectively improved the identification efficiency and maintained high recognition accuracy. Li Guorong et al. applied neural network fuzzy clustering analysis method to the multi-factor fuzzy comprehensive identification of sedimentary microfacies, combining the self-adaptability and fault tolerance of neural network with the fuzzy comprehensive judgment characteristics of fuzzy logic, in view of the multiple solutions and fuzziness of the corresponding relationship between sedimentary environment and sedimentary characteristics. Liu Hui et al. generated a deep process neural network by stacking a deep belief network and a BP classifier, and realized the automatic identification of sedimentary microfacies based on it.Luo Renze et al. proposed an intelligent sedimentary microfacies identification method based on characteristic structures (DMC) and bidirectional long short-term memory (BiLSTM) for well logging sedimentary microfacies, making the algorithm more robust. He Xu et al. established a well logging facies identification model by training a convolutional neural network. They then used wavelet basis functions of different scales and extreme value segmentation to process and segment the well logging data, ultimately achieving the classification of sedimentary units of different scales. Valentin et al. established a deep residual network to automatically identify reservoir lithofacies using single-point ultrasonic and microresistivity borehole imaging log data as input. Xiao He et al. used a hierarchical clustering method and Fisher discriminant to divide strata in uncored wells, thereby achieving automatic sedimentary microfacies classification. Camila Martins Saporetti et al. optimized an artificial neural network using an adaptive differential evolution (DE) algorithm. The improved ANN network architecture demonstrated promising results in solving the lithologic identification problem in the South Provence Basin, providing improved classification accuracy. Hong et al. designed a target detection strategy that uses integrated classifiers to detect the type and location of sedimentary microfacies. Their application to clastic strata in the Ordos Basin demonstrated the effectiveness and consistency of this method for automatic classification of sedimentary microfacies. Luo Xin et al. deepened their understanding of the internal structure of sandstone-conglomerate sedimentary microfacies through FMI image extraction and learning using a CNN network, providing evidence for the detailed characterization and effective prediction of sedimentary microfacies. While these techniques have achieved some success in sedimentary microfacies identification, they require the pre-process extraction of feature parameters from well logging curves, which results in a certain amount of information loss. Furthermore, the network learning and testing process uses relatively few sample features and input nodes, which directly affects the network's generalization ability and leads to low recognition accuracy.

[0004] Deep belief networks (DBNs) are a relatively mature model in the field of deep learning. Based on restricted Boltzmann machines (RBMs), they employ a layered training approach to effectively overcome the vanishing gradient problem inherent in deep networks. They have been successfully applied in fields such as classification, detection, monitoring, and identification. However, the application of DNBs in sedimentary microfacies identification has yet to be reported. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a sedimentary microfacies identification method based on deep belief network. The constructed samples can make full use of the morphological characteristics of small-layer microfacies to improve the identification accuracy of sedimentary microfacies. By utilizing the multi-layer mapping mechanism of deep learning, the identification accuracy and generalization ability of sedimentary microfacies can be improved.

[0006] To achieve the above objectives, the present invention provides a sedimentary microfacies identification method based on a deep belief network, comprising:

[0007] Convert the original data of the well logging curve combination of the target well into a binary image, which reflects the shape feature of the well logging curve;

[0008] Train the binary image by using a deep belief network, the deep belief network is composed of multiple layers of restricted Boltzmann machines, and the mapping relationship between the well logging curve shape and the sedimentary microfacies category is learned through unsupervised pre-training and supervised fine-tuning;

[0009] The well logging curve data of the well to be identified is classified by using the trained deep belief network.

[0010] The technical effect of the present application is that the sedimentary microfacies recognition method based on the deep belief network is disclosed, the original curve shape feature in the logging data is used as the input sample, the deep belief network is combined for sedimentary microfacies recognition, the geologist can be relieved from manual processing, the logging data processing and sedimentary microfacies recognition work can be completed automatically by the computer, the artificial workload can be reduced, the information integrity of the sampled logging data can be ensured, and the objectivity, efficiency and accuracy of the recognition result can be ensured. The results show that compared with the core analysis results, the sedimentary microfacies recognition rate of the method can reach more than 87%, so the method is suitable for single-well sedimentary microfacies recognition in non-coring areas. Compared with other artificial intelligence methods, the deep belief network method adopted by the present application has good adaptability, good network robustness, strong generalization ability and high recognition efficiency, and has good applicability to the processing of sedimentary microfacies recognition problems. BRIEF DESCRIPTION OF DRAWINGS

[0011] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the present application, and the illustrative embodiments of the present application and their description serve to explain the present application, and do not constitute improper limitations on the present application. In the drawings:

[0012] Figure 1 The restricted Boltzmann machine model is an embodiment of the present application;

[0013] Figure 2 The structure of the deep belief network is an embodiment of the present application;

[0014] Figure 3 The basic scheme diagram for identifying the sedimentary microfacies of the single-well profile by using the DBN is an embodiment of the present application;

[0015] Figure 4 The small layer data point number diagram is an embodiment of the present application;

[0016] Figure 5 The small layer logging curve data diagram is an embodiment of the present application, wherein (a) is GR, (b) is HAC, (c) is RLLD, and (d) is RLLS.

[0017] Figure 6 Fig. 20 is a schematic view of 20 randomly selected image samples for an embodiment of the present application;

[0018] Figure 7 Fig. 21 is a schematic view of a result comparison of microfacies identification based on the DBN method and geological data for a certain well of an embodiment of the present application. DETAILED DESCRIPTION

[0019] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0020] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0021] The sedimentary microfacies identification method based on the deep belief network provided in the embodiment includes:

[0022] The original data of the well logging curve combination of the target well is converted into a binary image, and the binary image reflects the morphological features of the well logging curve;

[0023] The deep belief network is trained on the binary image, and the deep belief network is composed of multiple layers of restricted Boltzmann machines. The mapping relationship between the well logging curve morphology and the sedimentary microfacies category is learned through unsupervised pre-training and supervised fine-tuning.

[0024] The trained deep belief network is used to classify the sedimentary microfacies of the well logging curve data of the well to be identified.

[0025] Further, the well logging curve combination includes natural gamma, acoustic time difference, deep lateral resistivity, and shallow lateral resistivity.

[0026] Further, the generation process of the binary image includes: unifying the data length of each small layer well logging curve, cutting off the excess part, and zero padding the insufficient part; according to the value range of the well logging curve, the data points are mapped to the corresponding pixel positions of the binary image, wherein 1 represents the data point and 0 represents the background.

[0027] Further, the hidden layer nodes of the deep belief network are configured as 50, 50, and 250.

[0028] Further, the unsupervised pre-training includes sequentially training each layer of the restricted Boltzmann machine, wherein the first layer of RBM is iterated 10 times, the second layer of RBM is iterated 50 times, and the top layer of RBM is iterated 100 times.

[0029] Further, the supervised fine-tuning uses the back propagation algorithm to optimize the network parameters.

[0030] Furthermore, the sedimentary microfacies categories include distributary channels, non-main channel, sheet sand, sheet sand margin and distributary inter-channel mud.

[0031] Furthermore, the logging curve data also includes small layer thickness, which is mapped to a binary image after being magnified.

[0032] The Restricted Boltzmann Machine (RBM) is a probabilistic graphical model with only two layers. Strictly speaking, it is not a true deep learning model. It has attracted attention because it can be used to construct deep learning models such as autoencoders and deep belief networks. From the perspective of probabilistic graphical models, this article introduces the standard model and learning algorithm of RBM. Figure 1 As shown in the figure Is the visual layer node, is the visual layer bias; is a hidden layer node, is the hidden layer bias.

[0033] First, the RBM energy function is defined as follows:

[0034] (1);

[0035] Based on this function, the joint probability of the RMB visual layer and hidden layer can be further defined:

[0036] (2);

[0037] According to formula (2), we can get the following two edge laws:

[0038] (3);

[0039] (4);

[0040] make , , the training algorithm of RBM is given below.

[0041]

[0042] Mathematically, the Deep Belief Net (DBN) is a hybrid graph model consisting of two parts: undirected and directed. Figure 2 As shown in Figure 2. The top two layers are undirected graphs, forming an associative memory network, and the remaining layers are directed graphs. These layers are essentially restricted Boltzmann machines. Figure 2 middle, is the visible layer vector, is the visible layer vector, . is the weight between the visible layer and the first hidden layer, is the weight between the first hidden layer and the second hidden layer, is the weight between the second hidden layer and the label layer, is the weight between the label layer and the first hidden layer, is the bias of the visible layer, is the bias of the first hidden layer, is the bias of the label layer.

[0043] Let , , the joint probability distribution of the deep belief network is:

[0044] (5) ;

[0045] where , and can be calculated by the restricted Boltzmann machine with as the visible vector and as the hidden vector. By eliminating the label vector through the probability summation operation, formula (5) can be simplified as follows:

[0046] (6) ;

[0047] In the label layer , only one component is 1 and the rest are 0. The subscript of the component with a value of 1 in indicates the class of the sample. In summary, the input-output relationship of the deep belief network is as follows.

[0048] (7) ;

[0049] where is the vector with the th component being 1 and the rest being 0, . .

[0050] The training of the deep belief network adopts a combination of unsupervised and supervised methods. First, the RBM algorithm is used to perform unsupervised pre-training on the restricted Boltzmann machines of each layer, and then the backpropagation algorithm is used for supervised optimization.

[0051] The sedimentary microfacies types determined by the core data expert analysis results and the optimized well logging curve combination original morphological data are used as learning samples, which are input into the deep belief network model for learning, so as to identify the sedimentary microfacies types of a single well. The basic scheme for identifying the sedimentary microfacies of a single well profile by using the deep belief network is as followsFigure 3 are shown.

[0052] As one of the important models of deep learning, DBN is essentially still a neural network, and it needs to be trained with standard small layers before being used for microfacies identification. To construct the training samples, it is necessary to first determine all the microfacies types contained in the target oil layer, that is, to construct a microfacies type set, and each element in the set will be the expected output of the training sample.

[0053] Taking the experimental oil layer in the experimental oilfield as an example, according to the expert analysis results of coring and electrical logging data, the study area belongs to a large river-controlled delta front subfacies, mainly including five kinds of sedimentary microfacies: distributary channel, non-main channel, sheet sand, sheet sand edge and interdistributary mud.

[0054] In actual facies identification work, each well has its own multiple logging curves, and the contribution of these logging curves to the identification of sedimentary microfacies is not the same. In addition, different oilfield areas have different sedimentary environments, so when studying sedimentary microfacies, it is necessary to select the logging curve combination according to the characteristics of the region. The significance of logging curve combination selection is to facilitate the construction of high-quality training samples, thereby improving the training efficiency and generalization ability of the network.

[0055] The basic method of logging curve combination selection is to evaluate the correspondence between the logging curve combination features and the known sedimentary microfacies types according to the existing coring data, and then select the logging curve combination with high sensitivity to sedimentary microfacies identification to construct the small layer feature sample. Taking the experimental oil layer in the experimental oilfield as an example, through analysis and comparison, the logging curve combination with a larger correlation with the sedimentary microfacies type is as follows: natural gamma, acoustic time difference, deep lateral resistivity and shallow lateral resistivity; in addition, the formation thickness is also an important factor affecting the microfacies type.

[0056] The key to identifying sedimentary microfacies using artificial intelligence methods is: how to let the computer automatically identify each logging curve, analyze the morphological difference of each curve, and then determine the sedimentary microfacies type corresponding to the logging curve combination according to the curve combination features. The advantage of deep belief network is that it can directly use the original features of the sample data for training and identification, therefore, after selecting the logging curve combination, only simple processing of the original curve data is needed.

[0057] Unlike the sample construction method of traditional neural networks, in the recognition method based on DBN proposed by the invention, all the data of the selected logging curves are directly used to construct the training samples, so that the feature data can be fully utilized to improve the recognition effect, which is also the outstanding advantage of this method. The specific method is to map the logging curves corresponding to each small layer into a binary image, and the image is the input sample of DBN, and the microfacies class of the small layer is the corresponding output sample of DBN.

[0058] First, count the number of data points of each small layer on the well logging curve for all wells participating in the training. According to the statistical results, adjust the data points of all small layers to the same length . If the length is more than the length, cut off the processing, if less than the length, zero processing.

[0059] Then count the maximum and minimum values of each well logging curve of all small layers. Suppose the preferred well logging curve determined by combination is , and the maximum and minimum values of each well logging curve are , , and the maximum and minimum values of all small layer thicknesses are .

[0060] For each small layer, first generate a binary empty image of , at this time all pixel gray values are 0. Then map the first well logging curve to the first region from left to right in the image. The specific method is that for the first data point on the first curve, first round to an integer , and then set the pixel in the binary image to 1. For the small layer thickness, the same method is used to set the corresponding pixel to 1, the only difference is that the data points of the small layer, the thickness is the same value. Thus the construction of the input sample is completed. For the expected output sample, it can be constructed as a dimensional vector , which only has the first dimension as 1, and the rest are 0, where is the number of microfacies categories.

[0061] First, train the DBN with the training sample to learn the highly complex nonlinear mapping relationship between the small layer features and the microfacies types. After the training is completed, it can be used for single well sedimentary microfacies type identification. The specific method is: according to the construction method of the training sample, construct the recognition sample in the form of binary image for the target small layer in the well to be identified, and then submit these recognition samples to the DBN. At the output end of the DBN, the sedimentary facies type of the target small layer in the well can be obtained.

[0062] Using Matlab to program a deep belief network, we simulated sedimentary microfacies identification for 10,400 substrata in the experimental oil formations of 800 wells in an experimental oil field. As mentioned previously, these substrata have five types of sedimentary microfacies. After optimizing well log combinations, we ultimately selected gamma ray (GR), acoustic transit time (HAC), deep lateral resistivity (RLLD), shallow lateral resistivity (RLLS), and substratum thickness as substratum characteristic data.

[0063] According to the statistical results, the number of data points contained in all 10400 small layers is as follows Figure 4 As shown, the maximum value is 240 and the minimum value is 12. The four logging curve data are as follows Figure 5 As shown, (a) is GR, (b) is HAC, (c) is RLLD, and (d) is RLLS. The maximum values ​​of the four curves are 1075, 471, 361, and 306 after rounding, and the minimum values ​​are 0, 0, 0, and -70 after rounding. To simplify the calculation, according to Figure 4 , the present invention uniformly sets the number of points of all small layers to ;according to Figure 5 The four curves are restricted to the ranges of GR: [0, 300], HAC: [150, 400], RLLD: [0, 100], and RLLS: [0, 100]. Data outside this range are truncated. All sub-layer thicknesses are between 0 and 8 cm. To improve discrimination, all sub-layer thicknesses are multiplied by 10 and rounded to the nearest integer. The sub-layer thickness range is now [0, 80].

[0064] Based on the sample construction method, each small layer feature data can be converted into a magnitude 20 images randomly selected from 10400 image samples are shown in the figure below. Figure 6 These binary images can be directly submitted to the deep belief network for network training.

[0065] In order to fully verify the excellent performance of the deep belief network, the following experimental scheme is adopted: the small layer data of all 800 wells are divided into training samples and test samples according to different proportions. and Represent the number of training samples and the number of test samples respectively. This scheme will The specific settings are: 7:1, 6:2, 5:3, 4:4, 3:5, 2:6, and 1:7. The DBN parameters are set as follows: 150300 visible layer nodes, 200 first hidden layer nodes, 200 second hidden layer nodes, 1000 top layer nodes, and 6 label layer nodes. The number of iterations for each layer is 10 for the first RBM, 50 for the second RBM, and 100 for the top RBM.

[0066] To ensure the objectivity of the experimental results, the DBN network was trained 10 times for each ratio, and the recognition effect after each training was statistically analyzed. The specific results are shown in Table 1.

[0067] Table 1

[0068]

[0069] According to the experimental results, when The experimental results are best when , with both recognition rates being the highest, as shown in bold in Table 1. This demonstrates that DBN can achieve high recognition results for sedimentary microfacies types even with a small number of training samples. Furthermore, the DBN's recognition performance varies little across various ratios, indicating that DBN is minimally affected by sample ratio, demonstrating the robustness of the DBN recognition method.

[0070] In order to comprehensively examine the recognition performance of the DBN method for sedimentary microfacies types, based on the above experimental results, the recognition effect of DBN under different hidden layer node configurations is further examined. The specific scheme is: The hidden layer nodes were set to the following seven configurations: (10, 10, 50), (30, 30, 150), (50, 50, 250), (100, 100, 500), (150, 150, 750), (200, 200, 1000), and (250, 250, 1250). The number of iterations for each RBM was set the same as before. For each node configuration, the DBN was trained independently 10 times, and the recognition performance after each training session was analyzed. The specific results are shown in Table 2.

[0071] Table 2

[0072]

[0073] The experimental results show that the best performance is achieved when the hidden layer node configuration is (50, 50, 250), achieving the highest recognition rates for both methods. This is shown in bold in Table 2. This demonstrates that a relatively simple node configuration can achieve ideal results in identifying sedimentary microfacies using DBNs. Furthermore, the DBN's recognition performance is similar across various hidden layer node combinations, indicating that DBNs are minimally affected by the node configuration, further demonstrating the robustness of the DBN recognition method.

[0074] Based on the comparison between the microfacies results identified by the DBN method and the geological data, the microfacies results identified by the DBN method are basically consistent with the microfacies divided by the geological data, and only the sheet sand microfacies of PI11 small layer is identified as a non-main channel, and the reason is that the curve shape has the characteristics of the scouring surface of the river channel at the bottom, and the thickness of the sandstone is between the sheet sand and the non-main channel microfacies, which leads to the identification error. Finally, the sedimentary microfacies of 600 test wells are identified and compared with the sedimentary microfacies identification results of these logging experts as shown in the figure, and the results show that the sedimentary microfacies identification method based on the DBN method is feasible and effective. Figure 7

[0075] The network structure of the application is simple and has strong generalization ability. This is because the DBN is different from the ordinary neural network, adopts a new learning mechanism and a multi-layer structure, overcomes the gradient disappearance phenomenon caused by the BP algorithm when the number of layers is large, and thus effectively improves the identification ability of the DBN. On the other hand, the original sample data is converted into a binary image, the morphological characteristics of the logging curve are fully utilized for identification, which is also an important factor for improving the identification performance of the DBN, and is also a feature of the application.

[0076] The original curve morphological characteristics in the logging data are used as input samples, the deep belief network is combined for sedimentary microfacies identification, the geologists can be relieved from manual processing, the logging data processing and sedimentary microfacies identification work can be completed automatically by the computer, the artificial workload is reduced, the information integrity of the sampled logging data is guaranteed, the objectivity, efficiency and accuracy of the identification results are guaranteed. The results show that the sedimentary microfacies identification rate of this method can reach more than 87% compared with the core analysis results, so this method is suitable for single-well sedimentary microfacies identification in non-coring areas. Compared with other artificial intelligence methods, the deep belief network method used in the application has good adaptability, good network robustness, strong generalization ability and high identification efficiency, and has good applicability to the processing of sedimentary microfacies identification problems.

[0077] The above is only the preferred specific embodiment of the application, but the protection scope of the application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the application, which should be covered in the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.​

Claims

1. A sedimentary microfacies identification method based on deep belief network, characterized by: include: Converting raw data of the target well's logging curve combination into a binary image, wherein the binary image reflects the morphological characteristics of the logging curve; The well logging curve combination includes natural gamma, sonic transit time, deep lateral resistivity and shallow lateral resistivity; The binary image is trained using a deep belief network, where the deep belief network is composed of a multi-layer restricted Boltzmann machine and learns the mapping relationship between the well logging curve morphology and the sedimentary microfacies category through unsupervised pre-training and supervised fine-tuning; The trained deep belief network is used to classify sedimentary microfacies from the well logging data of the identified wells. The binary image generation process includes: unifying the data length of each small layer's logging curve, truncating the excess part, and filling the insufficient part with zero; mapping the data points to the corresponding pixel positions of the binary image according to the value range of the logging curve, where 1 represents the data point and 0 represents the background; Supervised fine-tuning uses the back-propagation algorithm to optimize network parameters; Sedimentary microfacies categories include distributary channels, non-mainstream channels, sheet sands, sheet sand margins, and distributary inter-channel muds; The well logging curve data also includes the thickness of the small layer, which is mapped to a binary image after being magnified; For all wells participating in the training, the number of data points in each small layer on the logging curve is counted; based on the statistical results, all small layer data points are adjusted to the same length L; if they exceed this length, they are truncated; if they are less than this length, they are padded with zeros; Count the maximum and minimum values ​​of each logging curve of all small layers; suppose that there are n logging curves determined by the combination optimization, and the maximum and minimum values ​​of each are Max i ,Minx i ,i=1,2,...,n;the maximum and minimum values ​​of all small layer thicknesses are Max n+1 ,Minx n+1 ; For each small layer, generate a A binary empty image, in which all pixel grayscale values ​​are 0; the i-th logging curve is mapped to the i-th region from left to right of the image; the specific method is as follows: for the j-th data point d on the i-th curve ij , d ij Rounded to [d ij ], the pixels in the binary image Set to 1; for the thickness of a small layer, set the corresponding pixel to 1, and the L data points of the small layer have the same thickness, completing the construction of the input sample; for the expected output sample, it is constructed as an m-dimensional vector e i , only the i-th dimension of this vector is 1, and the rest are all 0, where m is the number of microfacies categories.

2. The sedimentary microfacies identification method based on deep belief network according to claim 1, characterized in that: The hidden layer node configuration of the deep belief network is 50, 50, 250.

3. The sedimentary microfacies identification method based on deep belief network according to claim 1, characterized in that: The unsupervised pre-training involves training each layer of restricted Boltzmann machines in sequence, where the first layer RBM is iterated 10 times, the second layer RBM is iterated 50 times, and the top layer RBM is iterated 100 times.

Images