Seismic integration time migration method based on Spark

Abstract

The present invention provides a Spark-based seismic integration method time migration method, which includes: step 1, setting parameters for pre-stack time migration using the integration method; step 2, dividing input data and output imaging range according to the parameters to form an input RDD; step 3, performing integration method time migration processing on the input RDD; and step 4, merging temporary imaging result RDDs to obtain final imaging results. The Spark-based seismic integration method time migration method overcomes the shortcomings of the prior art in terms of temporary file processing, network communication, etc., further improves parallel efficiency, and can more stably support the processing of massive data.

Description

Technical Field

[0001] The invention relates to the technical field of oilfield development, and in particular to a Spark-based seismic integration method time migration method. Background Art

[0002] Seismic data processing is an important application field of big data high-performance computing technology, and the seismic data integration method time migration method is a frequently used and computationally intensive processing step. Especially with the development of seismic exploration technology, the volume of seismic data is increasing, the amount of computation in seismic migration processing is also increasing, and the calculation cycle is getting longer and longer. In order to meet the needs of actual production, many parallel methods have been applied to seismic migration algorithms, including CPU multi-process based on MPI, CPU multi-threading based on OpenMP, GPU multi-threading based on CUDA, and the combination of multiple methods.

[0003] In recent years, big data technology has developed rapidly and has also begun to be applied in the field of geophysical exploration. Hadoop is an emerging distributed infrastructure. Hadoop is mainly composed of the distributed file system HDFS and the distributed parallel computing framework MapReduce. HDFS supports massive data storage, and MapReduce provides parallel processing of massive data.

[0004] In the field of geophysical exploration, Hadoop was first applied to seismic data migration processing. In the Chinese patent application with application number: 201310519635.8, a method and system for processing seismic data using pre-stack time migration by integration method is involved. This method pre-analyzes the calculation method and data access characteristics of pre-stack time migration by integration method, adopts MapReduce framework in data management and access, and adopts CUDA programming to realize CPU / GPU heterogeneous cluster collaborative acceleration in specific migration calculation. The specific steps include: seismic data preparation, input data division, migration task operation, starting Reduce operation to merge temporary imaging results in the distributed file system of each node, and outputting the final imaging result. This method invents a method for processing seismic data using pre-stack time migration by integration method based on MapReduce programming model, supports the processing of large-scale data, and improves the computational efficiency of migration processing.

[0005] In the Chinese patent application with application number: CN202010502755.7, a variable offset VSP ray tracing integrated pre-stack time migration method and device are involved, which include the following steps: S1. Input variable offset VSP common detection point data, 2D velocity geological model and migration parameters; S2. According to the input data, establish the coordinates of the shot point and the detection point in the 2D geological model; S3. Ray tracing calculates the travel time from the shot point to the model grid node, and saves the travel time table of each shot point; S4. Ray tracing calculates the travel time from the detection point to the model grid node, and saves the travel time table of each detection point; S5. Calculate the time-depth table of the 2D geological model, and calculate the common scattering travel time based on the travel time table obtained in step S3 and step S4, so as to realize variable offset VSP common detection point integrated pre-stack time migration imaging; S6. Rearrange the integrated pre-stack time migration of step S5 into a common imaging gather, and obtain the variable offset VSP integrated pre-stack time migration stacking imaging by stacking the common imaging gather.

[0006] In the literature [Kang Yonggan, Zhao Gaijian, Wei Jia, et al. 2015], the parallel algorithm of prestack time migration is also discussed in detail based on the MapReduce parallel framework. These works have improved the parallel efficiency of migration processing, but are also limited by the shortcomings of the MapReduce framework itself, such as the simple abstraction of the framework, which is not friendly to complex logical algorithms such as migration processing; for example, the distribution and synchronization of data heavily rely on HDFS, which makes the reading and writing of temporary files very time-consuming; these shortcomings seriously affect the processing efficiency.

[0007] Spark technology has developed the MapReduce framework, providing a high-speed and general parallel framework for various computing methods such as batch processing and interactive processing of massive data. It has many superior features such as high scalability, high storage efficiency, in-memory computing, etc. In addition, the interface operators provided by Spark are simple, diverse, flexible, and efficient, which greatly reduces the complexity of development.

[0008] Most existing parallel methods are developed based on the MPI interface and MapReduce framework, which have the following disadvantages:

[0009] (1) The logic of the seismic integration time migration method is complex, and the development based on the above parallel technology is difficult.

[0010] (2) MPI and MapReduce have their own shortcomings in temporary file processing and network communication. The seismic integration time migration method developed based on them inherits these shortcomings and is not efficient enough in processing massive data.

[0011] To this end, the present invention proposes a Spark-based seismic integration method time migration parallel method to solve the above technical problems. Summary of the invention

[0012] The purpose of the present invention is to provide a Spark-based seismic integration method time migration method that can support parallel processing of massive data by algorithms and further improve processing efficiency.

[0013] The object of the present invention can be achieved by the following technical measures: a Spark-based seismic integration method time migration method, the Spark-based seismic integration method time migration method comprising:

[0014] Step 1, setting parameters for prestack time migration using the integral method;

[0015] Step 2: According to the parameters, the input data and the output imaging range are divided to form an input RDD (RDD is the abbreviation of Resilient Distributed Dataset);

[0016] Step 3, perform time shift processing on the input RDD using the integral method;

[0017] Step 4: Merge the temporary imaging results RDD to obtain the final imaging result.

[0018] The purpose of the present invention can also be achieved by the following technical measures:

[0019] In step 1, the parameters for prestack time migration using the integral method are set, including the output imaging range and the migration aperture.

[0020] Step 2 includes:

[0021] (2.1) Divide and number the output imaging range;

[0022] (2.2) Divide and number the input data;

[0023] (2.3) Divide the input data into blocks I j and imaging range block O i Several data slices are obtained by combining them in pairs to form the input RDD.

[0024] In step 2.1, the amount of memory required for the offset processing of the unit imaging range is calculated according to the offset parameters; the amount of available memory of the computing node is obtained to calculate the size of the imaging range that can be processed by a single node; according to the size of this range, the output imaging range is divided into several blocks, which are respectively denoted as O i , i is the block number of the imaging range.

[0025] In step 2.1, if subsequent calculations need to be performed in parallel using multiple threads, the memory usage of multiple threads needs to be considered when dividing the output imaging range.

[0026] In step 2.2, the input data is divided into several blocks according to the amount of data in a single block, which are denoted as I j , j is the input data block number; the data volume of a single block is set through parameters or calculated based on the total parallel number available in the computing cluster.

[0027] In step 3, the flatMap operator is applied, and the following processing is performed in the flatMap operator:

[0028] (3.1) Apply the integral time migration algorithm to each data slice of RDD, and obtain an imaging result data slice for each slice;

[0029] (3.2) The obtained imaging result data slices are sorted, and the imaging result data on the same vertical line are sorted into one data slice and marked with a vertical line number.

[0030] In step 3.1, when the integral time migration algorithm is applied, the input data range required for the data channel in the imaging result block is calculated based on the migration aperture and other parameters. If the data of the input data block is not within the input data range, no migration calculation is performed.

[0031] In step 3.1, if CPU multi-threaded parallelism is adopted, the input data blocks in the input RDD data slice are further divided according to the number of threads available on the computing node, and multi-threading technology is applied to implement thread parallel computing of the integral method time shift algorithm.

[0032] In step 3.1, if GPU multi-threaded parallelism is used, the input data of the data slice is copied to the GPU device in blocks, and the thread parallel calculation of the integral method time migration algorithm is performed on the GPU; after the calculation is completed, the result data is copied back from the GPU device.

[0033] In step 3.2, the flatMap operator flattens the processing results, that is, expands the result data slices, and the obtained RDD is recorded as the temporary imaging result RDD.

[0034] In step 4, the reduceByKey operator is applied to accumulate and merge the partitions with the same tag in the temporary imaging result RDD.

[0035] Step 4 also includes sorting the merged result data.

[0036] Compared with the prior art, the Spark-based seismic integration method time migration method in the present invention overcomes the shortcomings of the prior art in temporary file processing, network communication, etc., further improves the parallel efficiency, and can more stably support the processing of massive data. Under the trend of seismic exploration technology with increasing data volume, it has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The flowchart is a specific embodiment of the Spark-based seismic integration method time migration method of the present invention. DETAILED DESCRIPTION

[0038] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0039] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0040] In the specific embodiment 1 of the present invention, the Spark-based seismic integration method time migration method of the present invention includes the following steps:

[0041] (1) Set the parameters for integral prestack time migration, including output imaging range, migration aperture, etc.

[0042] (2) According to the parameters, the input data and output imaging range are divided to form the input RDD. The specific implementation steps include:

[0043] (2.1) Divide the output imaging range and number them. According to the offset parameter, calculate the amount of memory required for offset processing of the unit imaging range. Obtain the available memory of the computing node and calculate the size of the imaging range that can be processed by a single node. According to the size of this range, divide the output imaging range into several blocks, which are respectively denoted as i , i is the block number of the imaging range.

[0044] Furthermore, if subsequent calculations need to be performed in parallel using multiple threads, the memory usage of multiple threads needs to be considered when dividing the output imaging range.

[0045] (2.2) Divide the input data and number them. According to the amount of data in a single block, divide the input data into several blocks, denoted as I j , j is the input data block number. The amount of data in a single block can be set through parameters or calculated based on the total parallel number available in the computing cluster.

[0046] (2.3) Divide the input data into blocks Ij and imaging range block O i Several data slices are obtained by combining them in pairs to form the input RDD.

[0047] (3) Perform time shift processing using the integral method on the input RDD. The specific implementation steps are: Apply the flatMap operator and perform the following processing in the flatMap operator:

[0048] (3.1) The integral time migration algorithm is applied to each data slice of RDD, and each slice obtains an imaging result data slice.

[0049] When the integral time migration algorithm is applied, the input data range required for the data channel in the imaging result block needs to be calculated according to parameters such as the migration aperture. If the data of the input data block is not within the input data range, no migration calculation is performed.

[0050] If CPU multi-threaded parallelism is adopted, the input data blocks in the input RDD data slice can be further divided according to the number of threads available on the computing node, and multi-threading technology can be applied to realize thread parallel computing of the integral method time shift algorithm.

[0051] If GPU multi-thread parallelism is used, the input data of the data slice can be copied to the GPU device in blocks, and the thread parallel calculation of the integral method time migration algorithm can be performed on the GPU. After the calculation is completed, the result data is copied back from the GPU device.

[0052] (3.2) The obtained imaging result data slices are sorted, and the imaging result data on the same vertical line are sorted into one data slice and marked with a vertical line number.

[0053] The flatMap operator flattens the processing results, that is, expands the result data slices, and the obtained RDD is recorded as the temporary imaging result RDD.

[0054] (4) Merge the temporary imaging result RDD to obtain the final imaging result. Apply the reduceByKey operator to accumulate and merge the partitions with the same tag in the temporary imaging result RDD.

[0055] Furthermore, the merged result data may be sorted.

[0056] In a specific embodiment 2 of the present invention, Figure 1 As shown, Figure 1 The flowchart of the Spark-based seismic integration method time migration method of the present invention is shown in FIG. The Spark-based seismic integration method time migration method includes the following steps:

[0057] (1) Set the parameters for prestack time migration using the integral method.

[0058] (2) According to the parameters and cluster resources, the data and imaging range are divided respectively, and the slices divided in the two ways are combined to form the input RDD.

[0059] (3) Apply the flatMap operator to the input RDD. In the operator, apply the integral time migration algorithm to each partition data of the input RDD, and obtain an imaging result data slice for each slice.

[0060] The obtained imaging result data slices are sorted, and the imaging result data on the same vertical line are sorted into one data slice and marked with a vertical line number.

[0061] The flatMap operator flattens the processing results, that is, expands the result data slices, and the obtained RDD is recorded as the temporary imaging result RDD.

[0062] (4) Apply the reduceByKey operator to accumulate and merge the partitions with the same label in the temporary imaging result RDD to obtain the final imaging result.

[0063] The large rectangle in the figure represents the data RDD, and each small rectangle in the large rectangle represents a partition of the RDD. The figure only uses three partitions as an example. The middle of the two large rectangles represents the conversion of the RDD. The English words below indicate the operators used in the conversion, and the Chinese words above indicate the corresponding algorithm process. The arrows indicate the flow of data during the conversion process.

[0064] In the specific embodiment 3 of the present invention, the post-stack integration method time migration processing was performed on 2.6GB post-stack seismic data in a certain exploration area in the east using the patent technology. The parameter settings include line number range 1855-2274, point number range 1267-2203, migration aperture 5000m, etc. The running time is 3 hours and 38 minutes. When multi-threading technology is used in the method, the running time is 50 minutes.

[0065] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0066] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. The Spark-based seismic integration time migration method is characterized by: The Spark-based seismic integration time migration method includes: Step 1, setting parameters for prestack time migration using the integral method; Step 2: According to the parameters, the input data and output imaging range are divided to form the input RDD; Step 3, perform time shift processing on the input RDD using the integral method; Step 4, merging the temporary imaging results RDD to obtain the final imaging result; In step 1, the parameters for prestack time migration using the integral method are set, including the output imaging range and the migration aperture; Step 2 includes: 2.1 Divide the output imaging range and number it; 2.2 Divide and number the input data; 2.3 Divide the input data into blocks I j and imaging range block O i Combine two by two to get several data slices to form the input RDD; In step 2.1, the amount of memory required for the offset processing of the unit imaging range is calculated according to the offset parameters; the amount of available memory of the computing node is obtained to calculate the size of the imaging range that can be processed by a single node; according to the size of this range, the output imaging range is divided into several blocks, which are respectively denoted as O i , i is the block number of the imaging range; In step 2.2, the input data is divided into several blocks according to the amount of data in a single block, which are denoted as I j , j is the input data block number; the data volume of a single block is set by parameters or calculated based on the total parallel number available in the computing cluster; In step 3, the flatMap operator is applied, and the following processing is performed in the flatMap operator: 3.1 Apply the integral time migration algorithm to each data slice of RDD, and obtain an imaging result data slice for each slice; 3.2 Arrange the obtained imaging result data slices, and arrange the imaging result data on the same vertical line into one data slice and mark it with the vertical line number; In step 3.1, when the integral method time migration algorithm is applied, the input data range required for the data channel in the imaging result block is calculated according to the migration aperture and other parameters. If the data of the input data block is not within the input data range, the migration calculation is not performed; In step 3.2, the flatMap operator flattens the processing results, that is, expands the result data slices, and the obtained RDD is recorded as the temporary imaging result RDD; In step 4, the reduceByKey operator is applied to accumulate and merge the partitions with the same tag in the temporary imaging result RDD.

2. The Spark-based seismic integration method time migration method according to claim 1, characterized in that: In step 3.1, CPU multi-threaded parallelism is used to further divide the input data blocks in the input RDD data slice according to the number of threads available on the computing node, and multi-threading technology is applied to implement thread parallel computing of the integral method time shift algorithm.

3. The Spark-based seismic integration method time migration method according to claim 1, characterized in that: In step 3.1, GPU multi-threaded parallelism is used to copy the input data of the data slice to the GPU device in blocks, and the thread parallel calculation of the integral method time migration algorithm is performed on the GPU; after the calculation is completed, the result data is copied back from the GPU device.

4. The Spark-based seismic integration method time migration method according to claim 1, characterized in that: Step 4 also includes sorting the merged result data.

Images