Seismic Checkpointing Strategy for Reverse Time Migration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Seismic data processing for reverse time migration requires accessing forward-in-time simulations in reverse order, which leads to impractical storage and access time issues due to large data volumes, especially in applications like pre-stack Reverse Time Depth Migration (RTM), where storing all time steps exceeds available memory and access times are problematic.
Innovation Solution
Implementing a checkpointing strategy that stores correlation checkpoints at carefully chosen time steps, allowing for recomputation of forward simulations in reverse time order, and optimizing storage and computation by distinguishing between full state and correlation checkpoints, with the latter requiring less storage space and computation than traditional methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If all time steps of forward simulation are stored for reverse time migration, then complete access to simulation data is achieved, but storage requirements become impractically large (3.36 petabytes for standard test project)
Solution Approach 1:
The patent segments the continuous simulation data into discrete checkpoints at specific time steps. Instead of storing all time steps continuously, only selected checkpoints are stored, reducing storage volume while maintaining the ability to access simulation data at any time through recomputation between checkpoints.
Solution Approach 2:
The patent performs preliminary computation and storage of checkpoints at strategically selected time steps before the actual reverse time migration. This preliminary action allows the system to later access simulation data without storing the entire continuous record, as the checkpoints serve as pre-computed reference points.
2Reliability
If full state checkpoints are stored at all time steps, then simulation can be restarted from any point, but storage and computational overhead become excessive
Solution Approach 1:
The patent applies different storage requirements to different parts of the simulation. Full state checkpoints are stored only at specific critical time steps where restart capability is most needed, while intermediate time steps use reduced correlation buffers. This local differentiation maintains reliability where needed while reducing overall storage requirements.
Solution Approach 2:
The patent uses partial action by storing only the necessary correlation data (not full state) at most time steps. Full state checkpoints are stored selectively rather than continuously, providing sufficient restart capability at key moments while avoiding excessive storage of complete simulation states at all times.
3Loss of information
If correlation checkpoints are stored at every time step, then reverse time access is enabled, but storage requirements increase significantly
Solution Approach 1:
The patent segments the time series into intervals between checkpoints. Instead of storing correlation data continuously at every time step, only at discrete checkpoint intervals is full correlation data stored. Between checkpoints, the system recomputes correlations using stored full states, reducing storage while maintaining reverse time access capability.
Solution Approach 2:
The patent introduces full state checkpoints as intermediary reference points that enable reconstruction of correlation data between checkpoint intervals. These intermediaries allow the system to access reverse time information without storing complete correlation records at all times, as the full states serve as mediators for recomputing intermediate correlations.
4Quantity of substance
If more computation is performed to minimize input/output operations, then storage requirements are reduced, but computational time increases
Solution Approach 1:
The patent changes the parameter of checkpoint frequency and type (full state vs. correlation only) to optimize the balance between storage and computation. By adjusting these parameters, the system can reduce storage requirements through selective checkpointing while managing computational overhead through strategic placement of full states for efficient recomputation.
Data Source
AI summary
Method and system for more efficient checkpointing strategy in cross correlating (316) a forward (328) and backward (308) propagated wave such as in migrating (326) or inverting seismic data. The checkpointing strategy includes storing in memory forward simulation data at a checkpointed time step, wherein the stored data are sufficient to do a cross correlation at that time step but not to restart the forward simulation. At other checkpoints, a greater amount of data sufficient to restart the simulation may be stored in memory (314). Methods are disclosed for finding an optimal combination, i.e. one that minimizes computation time (1132), of the two types of checkpoints for a given amount of computer memory (1004), and for locating a checkpoint at an optimal time step (306, 1214, 1310). The optimal checkpointing strategy (1002) also may optimize (1408) on use of fast (1402) vs. slow (1404) storage.