Seismic attribute fusion method and seismic data interpretation method based on isolated forest
By integrating seismic attributes based on isolated forests, the multi-solution problem caused by redundancy of seismic attribute data is solved, and more accurate reservoir prediction is achieved.
Patent Information
- Application Number
- CN202311429726.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-31
- Publication Date
- 2025-05-02
AI Technical Summary
The redundant seismic attribute data information leads to multi-solution problems in seismic interpretation, making it difficult to accurately predict reservoir characteristics in oil and gas exploration and development.
The seismic attribute fusion method based on isolated forests is adopted to obtain more representative fusion attributes by constructing isolated forests and extracting anomaly data of isolated trees.
It effectively reduces the redundant information of seismic attribute data, optimizes the fusion process of seismic attributes, and improves the accuracy and consistency of seismic reservoir prediction.
Smart Images
Figure CN119916467A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of seismic data processing, and in particular to a seismic attribute fusion method and a seismic data interpretation method based on isolation forests. Background Art
[0002] As an important characteristic quantity describing geological information such as stratigraphic structure, lithology, and oil and gas content, seismic attributes are indispensable parameters in the interpretation of seismic data. Nowadays, with the maturity and widespread application of seismic data acquisition, processing, and interpretation workstation technology, more refined and complex requirements are put forward for research objectives in oil and gas exploration and development. Seismic attribute technology has developed rapidly and is widely used in reservoir prediction. It is very beneficial for describing the dynamic and static characteristics of oil and gas in underground production and detecting the deposition and migration of target reservoirs or geological bodies. However, with the increasing number of seismic attributes, due to the different sensitivities of different seismic attributes to the target research objects in different work areas and different reservoirs, and the seismic attributes represent a large amount of geological information, the comprehensive study of multiple attributes will cause the problem of multiple solutions of seismic interpretation due to the information redundancy of attribute data. Summary of the invention
[0003] The purpose of the present invention is to solve the problem of multiple solutions of seismic interpretation caused by redundancy of seismic attribute data information, and to propose a seismic attribute fusion method based on isolation forest to provide a research basis for seismic reservoir prediction.
[0004] The present invention is implemented by the following technical scheme: a seismic attribute fusion method based on isolation forest, comprising the following steps:
[0005] The sample set D is constructed using multiple seismic attributes;
[0006] Using the isolation forest algorithm to construct an isolation forest for the sample set D;
[0007] extracting abnormal data of isolated trees in the isolated forest as dominant features of earthquake attributes;
[0008] The advantageous features of different seismic attributes are integrated to obtain the fused attributes.
[0009] Furthermore, the method of equal-proportional superposition is adopted to integrate the advantageous features of different seismic attributes.
[0010] Furthermore, the isolation forest algorithm comprises the following steps:
[0011] Construct an isolated binary tree;
[0012] An isolation forest is constructed by a set number of isolated binary trees;
[0013] Instantiate Isolation Forest.
[0014] Furthermore, the steps of constructing an isolated binary tree include:
[0015] S1: Select the experimental data training sample set D, and arbitrarily select n sample data from it as the training sample set A that constitutes this isolated binary tree.
[0016] S2: Select an identification feature f and a segmentation value p from the training sample set A at random; if the maximum value of all sampled data contained in node N under the identification feature f is f max , the minimum value is f min , then p∈[f min , f max ];
[0017] S3: When the recognition feature is smaller than the segmentation value (ie, f<p), the sampling data of the point is divided to the left side of the node N; otherwise, it is divided to the right side.
[0018] S4: Repeat steps S2 and S3 to split the left and right branches of node N respectively to obtain an isolated binary tree.
[0019] Furthermore, when there is only one sample data in a branch node, or there are several identical sample data, or the isolated binary tree reaches the top of the height, the growth of the isolated binary tree is terminated.
[0020] Furthermore, the outlier value at point d is calculated according to the path length h(d) corresponding to the sample point d of each isolated binary tree, so as to grasp its abnormal situation. The calculation formula is as follows:
[0021]
[0022] In the formula, m represents the total number of sample points in the sample set A, E(h(d)) represents the average value of all path lengths E(h(d)), and the height of the isolated binary tree is normalized as Where H(k)=lnk+ξ, and the Euler constant is ξ=0.5772156649.
[0023] Further, the steps of instantiating the isolation forest include:
[0024] First, you need to import the libraries you want to use, including: numpy library, pandas library, seaborn library and matplotlib library; you also need to import the isolation forest algorithm used in sklearn.ensemble.
[0025] After the library import is complete, read the data from the csv format to generate a pandas data frame and check the first ten rows of the data;
[0026] Next, create a model volume and instantiate the Isolation Forest.
[0027] Furthermore, the following parameter values were input into the Isolation Forest algorithm run to create the model: number of evaluators, maximum number of samples, expected proportion of outliers in sample set D, maximum features.
[0028] Further, train the created model, call the predict() function on the trained model, find the value of the anomaly column, and add the values of the two columns to the data column. Negative values displayed in the score column and -1 in the anomaly column indicate anomaly data, and 1 in the anomaly column indicates normal data.
[0029] The present invention also provides a seismic data interpretation method, which performs interpretation based on the fused attributes obtained by the seismic attribute fusion method based on isolation forest described in the present invention.
[0030] Compared with the prior art, the beneficial effects of the present invention include:
[0031] The present invention performs feature screening on numerous seismic attributes, realizes seismic attribute fusion optimization, and verifies the applicability of seismic attribute fusion optimization in seismic reservoir prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is a flowchart of the seismic attribute fusion method based on isolation forest;
[0033] Figure 2 is the original attribute slice map;
[0034] Figure 3 It is a fused attribute slice map. DETAILED DESCRIPTION
[0035] As the types of seismic attributes increase, different seismic attributes have different sensitivities to the target research objects in different work areas and reservoirs, and seismic attributes represent a large amount of geological information. When conducting comprehensive research on multiple attributes, the information redundancy of attribute data will cause multiple solutions to seismic interpretation. Based on this, the isolation forest algorithm is introduced to conduct fusion research on seismic attributes to provide technical support for future seismic reservoir research.
[0036] A seismic attribute fusion method based on isolation forest includes the following steps:
[0037] The sample set D is constructed using multiple seismic attributes;
[0038] Using the isolation forest algorithm to construct an isolation forest for the sample set D;
[0039] extracting abnormal data of isolated trees in the isolated forest as dominant features of earthquake attributes;
[0040] The advantageous features of different seismic attributes are integrated to obtain the fused attributes.
[0041] Preferably, the advantageous features of different seismic attributes are integrated by using an equal-proportional superposition method.
[0042] Preferably, the isolation forest algorithm comprises the following steps:
[0043] Construct an isolated binary tree;
[0044] An isolation forest is constructed by a set number of isolated binary trees;
[0045] Instantiate Isolation Forest.
[0046] Preferably, the steps of constructing an isolated binary tree include:
[0047] S1: Select the experimental data training sample set D, and arbitrarily select n sample data from it as the training sample set A that constitutes this isolated binary tree.
[0048] S2: Select an identification feature f and a segmentation value p from the training sample set A at random; if the maximum value of all sampled data contained in node N under the identification feature f is f max , the minimum value is f min , then p∈[f min , f max ];
[0049] S3: When the recognition feature is smaller than the segmentation value (ie, f<p), the sampling data of the point is divided to the left side of the node N; otherwise, it is divided to the right side.
[0050] S4: Repeat steps S2 and S3 to split the left and right branches of node N respectively to obtain an isolated binary tree.
[0051] Preferably, when there is only one piece of sample data in a branch node, or there are several identical sample data, or the isolated binary tree reaches the top of the height, the growth of the isolated binary tree is terminated.
[0052] Preferably, the outlier value at d is calculated according to the path length h(d) corresponding to the sample point d of each isolated binary tree, so as to grasp its abnormal situation. The calculation formula is as follows:
[0053]
[0054] In the formula, m represents the total number of sample points in the sample set A, E(h(d)) represents the average value of all path lengths E(h(d)), and the height of the isolated binary tree is normalized as Where H(k)=lnk+ξ, and the Euler constant is ξ=0.5772156649.
[0055] Preferably, the step of instantiating the isolation forest includes:
[0056] First, you need to import the libraries you want to use, including: numpy library, pandas library, seaborn library and matplotlib library; you also need to import the isolation forest algorithm used in sklearn.ensemble.
[0057] After the library import is complete, read the data from the csv format to generate a pandas data frame and check the first ten rows of the data;
[0058] Next, create a model volume and instantiate the Isolation Forest.
[0059] Preferably, the following parameter values are input into the isolation forest algorithm run to create the model: number of evaluators, maximum number of samples, expected proportion of outliers in the sample set D, maximum features.
[0060] Preferably, the created model is trained, and the predict() function is called on the trained model to find the value of the abnormal column. The values of the two columns are added to the data column respectively. Negative values displayed in the score column and -1 in the abnormal column are abnormal data, and 1 in the abnormal column indicates normal data.
[0061] The invention is further described in detail below with reference to the accompanying drawings and embodiments.
[0062] Example 1
[0063] Construct an isolation forest binary tree:
[0064] The isolation forest algorithm was proposed by Liu et al. in 2008. The algorithm obtains outliers by calculating the average path length of experimental data under the constructed isolation forest model. It is suitable for high-dimensional data. The isolation forest consists of several isolated binary trees. The algorithm focuses on constructing isolated binary trees. The construction process is as follows:
[0065] (1) Select the experimental data training sample set D and preprocess it. Randomly select n sample data as the training sample set A that constitutes this isolated binary tree.
[0066] (2) Select an identification feature f and a segmentation value p at random from the training sample set A. If the maximum value of all sample data contained in node N under the identification feature f is f max , the minimum value is f min , then p∈[f min , f max ].
[0067] (3) When the recognition feature is smaller than the segmentation value (i.e., f<p), the sampling data of the point is divided to the left side of the node N; otherwise, it is divided to the right side.
[0068] (4) Repeat steps (2) and (3) to split the left and right branches of node N respectively to obtain an isolated binary tree. When there is only one sample data in the branch node, or there are several identical sample data, or the isolated binary tree reaches the top of the height, the growth of the isolated binary tree is terminated.
[0069] (5) Finally, the operator sets the amount of data to construct an isolation forest using isolated binary trees. The outlier value at point d is calculated based on the path length h(d) corresponding to the sample point d of each isolated binary tree, and the abnormality is then understood. The calculation formula is as follows:
[0070]
[0071] Where m represents the sample set A s The total number of sample points, E(h(d)) represents the average value of all path lengths E(h(d)), and the height of the isolated binary tree is normalized as Where H(k)=lnk+ξ, and the Euler constant is ξ=0.5772156649.
[0072] Example 2
[0073] Implement the Isolation Forest Algorithm:
[0074] The purpose of using Python to analyze and explore experimental data is to understand the relevant information of the given experimental data. First, you need to import the libraries you want to use, including: numpy library, pandas library, seaborn library and matplotlib library. You also need to import the isolation forest algorithm used in sklearn.ensemble. After completing the import library operation, read the data from csv format to generate a pandas data frame and check the first ten rows of the data. Secondly, you need to create a model variable and instantiate IsolationForest. Enter the following parameter values into the isolation forest algorithm run:
[0075] (1) Number of estimators: n_estimators represents the number of sub-estimators in the ensemble (or the number of trees in the isolated forest). The number of estimators is an adjustable integer, and the default value is 50.
[0076] (2) Maximum number of samples: max_sample is the number of training samples for each base evaluator. When max_sample is larger than the number of samples, all samples will be used when training the tree. The default value is [auto]. When defined as [auto], max_sample = min(256, n_sample);
[0077] (3) Data contamination problem: The algorithm is very sensitive to the Contamination parameter, which represents the expected proportion of outliers in the experimental data set and is often used to define the critical value based on the sample score fit. The default value is [auto]. When [auto] is selected, the critical value is set based on the original isolation forest algorithm;
[0078] (4) Maximum feature: Not all sub-evaluators are trained with all the feature factors of the data set. The maximum feature is often selected from all the feature factors and used in the training process of the number of features of each sub-evaluator or tree in the isolated forest. Its default value is -1.
[0079] After defining the model, you can use the given experimental data to train the model, mainly through fit(). This method requires the input parameter - the usage data. After the model is trained correctly, the isolated forest instance is successfully output. After the model is defined and fitted, find the score and its anomaly columns. Call the decision_function function based on the trained model, pass the usage data into it, and find the value of the score column. Analogously to the above operation, call the predict() function on the trained model, and input the experimental data as a parameter to find the value of the anomaly column. Add the values of the two columns to the data column respectively. Negative values displayed in the score column and -1 displayed in the anomaly column indicate an anomaly. 1 displayed in the anomaly column indicates normal data.
[0080] The abnormal data in each isolated tree corresponds to a dominant feature, and a sample in the sample set D corresponds to a seismic attribute.
[0081] Example 3
[0082] Construct an isolation forest binary tree:
[0083] The isolation forest algorithm was proposed by Liu et al. in 2008. The algorithm obtains outliers by calculating the average path length of experimental data under the constructed isolation forest model. It is suitable for high-dimensional data. The isolation forest consists of several isolated binary trees. The algorithm focuses on constructing isolated binary trees. The construction process is as follows:
[0084] (6) Select the experimental data training sample set D and preprocess it. Randomly select n sample data as the training sample set A that constitutes this isolated binary tree.
[0085] (7) Select any identification feature f and segmentation value p from the training sample set A. If the maximum value of all sample data contained in node N under the identification feature f is f max , the minimum value is f min , then p∈[f min , f max ].
[0086] (8) When the recognition feature is smaller than the segmentation value (i.e., f<p), the sampling data of the point is divided to the left side of the node N; otherwise, it is divided to the right side.
[0087] (9) Repeat steps (2) and (3) to split the left and right branches of node N respectively to obtain an isolated binary tree. When there is only one sample data in the branch node, or there are several identical sample data, or the isolated binary tree reaches the top of the height, the growth of the isolated binary tree is terminated.
[0088] (10) Finally, the operator sets the amount of data to construct an isolation forest using isolated binary trees. The outlier value at point d is calculated based on the path length h(d) corresponding to the sample point d of each isolated binary tree, and the abnormality is then understood. The calculation formula is as follows:
[0089]
[0090] Where m represents the sample set A s The total number of sample points, E(h(d)) represents the average value of all path lengths E(h(d)), and the height of the isolated binary tree is normalized as Where H(k)=lnk+ζ, and the Euler constant is ξ=0.5772156649.
[0091] Implement the Isolation Forest Algorithm:
[0092] The purpose of using Python to analyze and explore experimental data is to understand the relevant information of the given experimental data. First, you need to import the libraries you want to use, including: numpy library, pandas library, seaborn library and matplotlib library. You also need to import the isolation forest algorithm used in sklearn.ensemble. After completing the import library operation, read the data from csv format to generate a pandas data frame and check the first ten rows of the data. Secondly, you need to create a model variable and instantiate IsolationForest. Enter the following parameter values into the isolation forest algorithm run:
[0093] (5) Number of estimators: n_estimators represents the number of sub-estimators in the ensemble (or the number of trees in the isolated forest). The number of estimators is an adjustable integer, and the default value is 50.
[0094] (6) Maximum number of samples: max_sample is the number of training samples for each base evaluator. When max_sample is larger than the number of samples, all samples will be used when training the tree. The default value is [auto]. When defined as [auto], max_sample = min(256, n_sample);
[0095] (7) Data contamination problem: The algorithm is very sensitive to the Contamination parameter, which represents the expected proportion of outliers in the experimental data set and is often used to define the critical value based on the sample score fit. The default value is [auto]. When [auto] is selected, the critical value is set based on the original isolation forest algorithm;
[0096] (8) Maximum feature: Not all sub-evaluators are trained with all the feature factors of the data set. The maximum feature is often selected from all the feature factors and used in the training process of the number of features of each sub-evaluator or tree in the isolated forest. Its default value is -1.
[0097] After defining the model, you can use the given experimental data to train the model, mainly through fit(). This method requires the input parameter - the usage data. After the model is trained correctly, the isolated forest instance is successfully output. After the model is defined and fitted, find the score and its anomaly columns. Call the decision_function function based on the trained model, pass the usage data into it, and find the value of the score column. Analogously to the above operation, call the predict() function on the trained model, and input the experimental data as a parameter to find the value of the anomaly column. Add the values of the two columns to the data column respectively. Negative values displayed in the score column and -1 displayed in the anomaly column indicate an anomaly. 1 displayed in the anomaly column indicates normal data.
[0098] The abnormal data in each isolated tree corresponds to a dominant feature, and a sample in the sample set D corresponds to a seismic attribute.
[0099] Seismic attribute extraction and seismic attribute fusion based on isolation forest algorithm
[0100] Seismic attributes are extracted for the study area, and the characteristic values of seismic attributes are isolated and fused through the operation of the isolation forest algorithm, emphasizing the dominant characteristic parts of seismic attributes. After extracting the dominant characteristics of multiple seismic attributes, the seismic attributes are superimposed and fused in equal proportions.
[0101] Figure 2 The dominant features of each attribute are selected based on the isolation forest algorithm. It can be seen that the dominant features displayed by each attribute only account for Figure 3 The fusion of the two parts will lead to incomplete and inaccurate results in reservoir prediction. However, after fusion, the attribute advantages it contains are more comprehensive and can also improve the accuracy of reservoir prediction in the future.
[0102] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0103] In the description of the present invention, unless otherwise specified, the terms "upper", "lower", "left", "right", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limitations on the present invention.
[0104] The above technical solution is only one implementation mode of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the principles disclosed in the present invention, and it is not limited to the technical solution described in the above specific embodiments of the present invention. Therefore, the above description is only preferred and does not have a restrictive meaning.
Claims
1. A seismic attribute fusion method based on isolation forest, characterized in that: The following steps are involved: The sample set D is constructed using multiple seismic attributes; Using the isolation forest algorithm to construct an isolation forest for the sample set D; extracting abnormal data of isolated trees in the isolated forest as dominant features of earthquake attributes; The advantageous features of different seismic attributes are integrated to obtain the fused attributes.
2. The seismic attribute fusion method based on isolation forest according to claim 1, characterized in that: The advantageous features of different seismic attributes are integrated by using the method of equal-proportional superposition.
3. The seismic attribute fusion method based on isolation forest according to claim 1, characterized in that: The isolation forest algorithm comprises the following steps: Construct an isolated binary tree; An isolation forest is constructed by a set number of isolated binary trees; Instantiate Isolation Forest.
4. The seismic attribute fusion method based on isolation forest according to claim 1, characterized in that: The steps to construct an isolated binary tree include: S1: Select the experimental data training sample set D, and arbitrarily select n sample data from it as the training sample set A that constitutes this isolated binary tree. S2: Select an identification feature f and a segmentation value ρ from the training sample set A at random; if the maximum value of all sampled data contained in node N under the identification feature f is f max , the minimum value is f min , then p∈[f min , f max ]; S3: When the recognition feature is smaller than the segmentation value (ie, f<p), the sampling data of the point is divided to the left side of the node N; otherwise, it is divided to the right side. S4: Repeat steps S2 and S3 to split the left and right branches of node N respectively to obtain an isolated binary tree.
5. The seismic attribute fusion method based on isolation forest according to claim 4, characterized in that: When there is only one sample data in the branch node, or there are several identical sample data, or the isolated binary tree reaches the top of the height, the growth of the isolated binary tree is terminated.
6. The seismic attribute fusion method based on isolation forest according to claim 4, characterized in that: According to the path length h(d) corresponding to the sample point d of each isolated binary tree, the outlier value at d is calculated to grasp its abnormal situation. The calculation formula is as follows: In the formula, m represents the total number of sample points in the sample set A, E(h(d)) represents the average value of all path lengths E(h(d)), and the height of the isolated binary tree is normalized as Where H(k)=lnk+ξ, and the Euler constant is ξ=0.5772156649.
7. The seismic attribute fusion method based on isolation forest according to claim 3 is characterized in that: The steps to instantiate an Isolation Forest are: First, you need to import the libraries you want to use, including: numpy library, pandas library, seaborn library and matplotlib library; you also need to import the isolation forest algorithm used in sklearn.ensemble. After the library import is complete, read the data from the csv format to generate a pandas data frame and check the first ten rows of the data; Next, create a model volume and instantiate the Isolation Forest.
8. The seismic attribute fusion method based on isolation forest according to claim 7, characterized in that: The following parameter values were input into an Isolation Forest algorithm run to create the model: number of evaluators, maximum number of samples, expected proportion of outliers in the sample set D, maximum features.
9. The seismic attribute fusion method based on isolation forest according to claim 8, characterized in that: Train the created model, call the predict() function on the trained model, find the value of the anomaly column, and add the values of the two columns to the data column. Negative values displayed in the score column and -1 in the anomaly column indicate anomaly data, and 1 in the anomaly column indicates normal data.
10. A method for interpreting seismic data, characterized in that: The interpretation is performed based on the fused attributes obtained by the seismic attribute fusion method based on isolation forest as described in any one of claims 1 to 9.