CN113791445A

CN113791445A - Frequency component wave field information extraction method based on FPGA full-waveform inversion

Info

Publication number: CN113791445A
Application number: CN202111127625.0A
Authority: CN
Inventors: 陈剑飞; 许康
Original assignee: Shanghai Xuehu Technology Co ltd
Current assignee: Shanghai Xuehu Technology Co ltd
Priority date: 2021-09-26
Filing date: 2021-09-26
Publication date: 2021-12-14

Abstract

The invention relates to the technical field of oil-gas exploration, in particular to a frequency component wave field information extraction method based on FPGA full waveform inversion.

Description

Frequency component wave field information extraction method based on FPGA full-waveform inversion

Technical Field

The invention relates to the technical field of oil and gas exploration, in particular to a frequency component wave field information extraction method based on FPGA full waveform inversion.

Background

The efficiency and the precision of an inversion algorithm are improved, the underground medium characteristics are acquired more quickly and accurately, and the method is a permanent target for oil and gas exploration and complex engineering geophysical exploration. Seismic exploration is to quantitatively estimate the related attributes of underground media by using seismic observation data and provide a basis for finding oil and gas resources, and inversion is one of the main means of oil and gas exploration.

Conventional inversion methods typically use only a portion of the information in the seismic response, which results in some useful information in the seismic response being ignored. With the deepening of the complexity of oil and gas exploration, a Full wave form Inversion (FWI) method is regarded as a method capable of objectively reflecting the seismic wave propagation law, and is more and more emphasized by people, which is suitable for any geological model and provides technical support for the research of the earth deep structure, the middle and deep mineral exploration and oil exploration, the investigation of the shallow surface geological environment, the engineering construction and other works. The full waveform inversion is a new seismic imaging technology developed based on the reverse time migration technology, and is different from the situation that the reverse potential migration technology reversely transmits seismic records, primary imaging is obtained in a model space, and the model is not needed to be updated; the full waveform inversion technology reversely transmits residual errors of seismic records, the model is continuously updated until the matching of field observation records and simulation records meets the technical requirements, and the final inversion model is the best model which can be obtained by people but needs to provide an initial model meeting certain precision. The development of reverse time migration techniques with similar theoretical frameworks has greatly facilitated the theoretical perfection and practical application of full waveform inversion techniques.

The full waveform inversion includes two processes, a forward process and an inversion process. The forward modeling is an important component of the inversion, and the accuracy and efficiency of the inversion are determined by the selection of the forward modeling method, so that the feasibility of the full waveform inversion is determined by the correct selection of the forward modeling method, especially in the case of 3D multi-shot exploration. In the three-dimensional situation, the forward inversion of the full waveform is carried out only by using a time domain algorithm, and the inversion process needs huge storage space and consumes longer CPU time; the forward process of the direct solution has extremely high requirement on memory storage, the memory required by LU decomposition processing of the shot set usually reaches the order of magnitude of hundreds of GB or even TB, and the parallel operation is difficult to realize, and the time consumption is long. In order to avoid the problem, a time domain forward method is adopted for full waveform inversion under the 3D processing condition, a specific frequency component extraction mode is flexibly applied after time domain seismic records are obtained through forward modeling, time domain wave fields are converted into frequency domains to achieve full waveform inversion of the frequency domains, therefore, wave field values of the frequency domains are obtained, required preprocessing can be carried out on data in the time domains, wave field values of multiple required frequencies can be obtained through one forward modeling, and inversion can be achieved in the time domains or the frequency domains.

The requirement of full waveform inversion on computing resources is one of the bottlenecks that restrict the development of the full waveform inversion. The large amount of computing and storage is a primary problem that needs to be solved at present. Seismic exploration has been a cross field for many years and involves multiple disciplines such as mathematics, physics and geophysics. In recent years, with the development of computer technology, the performance of computers is continuously improved, the data processing capability is also improved year by year, and the application of 3D full waveform inversion to actual data is started to appear in succession. At present, the difficulty of full waveform inversion mainly comes from huge calculation amount, huge storage space requirement and calculation instability caused by high unsuitability of problems, the demand of the problems on calculation power exceeds the development speed of the moore's law of general-purpose CPUs, the performance increase speed of heterogeneous calculation can meet emerging directions and trends, the heterogeneous calculation can well meet the calculation requirements of calculation intensive fields such as artificial intelligence, high-performance data analysis and financial analysis, and the technology can gradually replace the part which is not good at in the original general calculation.

Heterogeneous computing will occupy more and more shares in future high-new-energy computing; at present, the heterogeneous computing field is that a GPU processor occupies a mainstream position, and with the establishment and improvement of an ecological environment of an FPGA (field Programmable Gate array), i.e., a field Programmable Gate array, in the future, in addition, the FPGA has advantages of higher computing performance, higher hardware acceleration performance, lower device interconnection delay and the like compared with the GPU, the FPGA will play an irreplaceable role in high-performance computing.

Disclosure of Invention

In view of the technical problems, the invention provides a frequency component wave field information extraction method based on FPGA full-waveform inversion; the invention utilizes the advantages of parallel computation and hardware acceleration performance of the FPGA and combines the characteristics of full waveform inversion, flexibly applies specific frequency component factors to extract the wave field value of specific frequency components after forward calculation to obtain time domain seismic records, and converts a time domain wave field into a frequency domain to provide a required frequency domain wave field value for realizing full waveform inversion of the frequency domain.

The specific technical scheme of the invention is as follows:

a frequency component wave field information extraction method based on FPGA full waveform inversion is characterized by comprising the following steps:

step S1: the method comprises the steps that a CPU loads a database file, wherein the database file comprises speed field data, observation system data and harmonic wave seismic source data of a plurality of frequency components, and the data are transmitted into an FPGA after the data are loaded;

step S2: performing wave field continuation calculation in the FPGA by adopting a finite difference method according to the wave field continuation process;

step S3: performing data interaction on the wave field value calculated by the FPGA and a CPU (central processing unit) through a PCIE (peripheral component interface express) interface, and recording the wave field value at the moment in the CPU according to the position relation of the data of the observation system;

step S4: extracting a wave field value of the frequency component at the moment according to a preset frequency factor;

step S5: the iteration time is updated and the time is deduced backwards, and the step S2 is jumped to until the wave field value of the whole frequency component is completed.

The method for extracting the wave field information of the frequency components based on the FPGA full-waveform inversion is characterized in that in step S4, the wave field value of the frequency components at the moment is extracted according to a preset frequency factor, the wave field value is interacted with a CPU through a PCIE interface, and data is stored on a CPU platform.

The technical scheme has the following advantages or beneficial effects:

according to the frequency component wave field information extraction method based on FPGA full waveform inversion, required preprocessing can be flexibly carried out on data in a time domain on a CPU platform by utilizing a heterogeneous architecture mode, a plurality of wave field values of required frequency components can be obtained through forward modeling, and convenience is brought to the realization of subsequent inversion in the time domain or the frequency domain. The invention adopts a pipeline data processing mode, completely submerges the data preparation and transmission time in the calculation, and greatly shortens the calculation time of the algorithm.

Drawings

The invention and its features, aspects and advantages will become more apparent from reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings. Like reference symbols in the various drawings indicate like elements. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention.

FIG. 1 is a flow chart of a frequency component wave field information extraction method based on FPGA full waveform inversion.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention utilizes a frequency domain inversion combined time domain forward hybrid algorithm, and takes the advantages of the frequency domain inversion and the time domain forward hybrid algorithm to realize the wave field information of the frequency required by the 3D full waveform inversion by using an FPGA and a CPU heterogeneous architecture platform. Through parameter setting of a CPU side, FPGA is used for relevant wave field value calculation, after the wave field value is calculated, a wave field value of specific frequency is extracted by using specific frequency factors at the moment, data are interacted to a CPU platform end through a wave field of the corresponding frequency and stored in a disk, and wave field calculation of the next moment is carried out. FIG. 1 is a flow chart of a frequency component wave field information extraction method based on FPGA full waveform inversion, and the specific method comprises the following steps:

step S1: the CPU loads a database file, the database file comprises speed field data, observation system data and harmonic wave seismic source data of a plurality of frequency components, and the data are transmitted into the FPGA after the loading is finished;

step S2: performing wave field continuation calculation in FPGA by finite difference method according to wave field continuation process (i.e. wave field extrapolation calculation wave field value step in FIG. 1);

In the whole calculation process, the FPGA is packaged into a calculation operator library for wave field continuation and specific frequency wave field information extraction. The CPU is used as a scheduling tool and a data preparation tool of the whole calculation to complete the preparation and storage of the whole flow data.

In the embodiment of the present invention, in step S4, the wave field value of the frequency component at that time is extracted according to the preset frequency factor, the wave field value is interacted with the CPU through the PCIE interface, and the data is stored on the CPU platform.

According to the invention, by utilizing the parallel computing advantage of the FPGA, a time domain data preprocessing module is reserved on the CPU side, the advantage of the algorithm is realized by utilizing an FPGA hardware logic circuit, a flow data processing mode is adopted, the data preparation and transmission time is completely submerged in the computation, the computation time of the algorithm is greatly shortened, the wave field value information calculated in the time domain can be utilized in the frequency inversion method of the large-scale three-dimensional full waveform inversion, the operator library packaged by the FPGA aiming at the earthquake forward modeling is greatly utilized, and the utilization rate of bits developed by the FPGA is increased. The computational power required by full waveform inversion can be satisfied by utilizing an FPGA heterogeneous mode, so that the problem of insufficient computational power is greatly reduced. Meanwhile, the data can be flexibly preprocessed in a time domain on a CPU platform by utilizing a heterogeneous architecture mode, and a plurality of wave field values of required frequency components can be obtained through one forward modeling. And the method provides convenience for realizing the subsequent inversion in a time domain or a frequency domain. In addition, the scheme provided by the invention is tested, and the frequency factor is transmitted into the FPGA in a parameter form, so that the flexibility and the programmability of the whole system are greatly enhanced.

Those skilled in the art will appreciate that those skilled in the art can implement the modifications in combination with the prior art and the above embodiments, and the details are not described herein. Such variations do not affect the essence of the present invention and are not described herein.

The above description is of the preferred embodiment of the invention. It is to be understood that the invention is not limited to the particular embodiments described above, in that devices and structures not described in detail are understood to be implemented in a manner common in the art; those skilled in the art can make many possible variations and modifications to the disclosed embodiments, or modify equivalent embodiments, without affecting the spirit of the invention, using the methods and techniques disclosed above, without departing from the scope of the invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention are still within the scope of the protection of the technical solution of the present invention, unless the contents of the technical solution of the present invention are departed.

Claims

1. A frequency component wave field information extraction method based on FPGA full waveform inversion is characterized by comprising the following steps:

2. The method for extracting the wave field information of the frequency components based on the FPGA full-waveform inversion of claim 1, wherein in step S4, the wave field value of the frequency components at that time is extracted according to a preset frequency factor, the wave field value is interacted with a CPU through a PCIE interface, and data is stored on a CPU platform.