Three-dimensional transient electromagnetic inversion method based on adaptive compression of vector field

Abstract

The application discloses a transient electromagnetic three-dimensional inversion method based on vector field adaptive compression, comprising: data acquisition: constructing a three-dimensional inversion objective function; designing a forward equation according to three-dimensional forward theory; information recoverable vector field compression, compressing electromagnetic field vector data, and extracting key point data information on electric field and magnetic field attenuation curves; information recoverable vector field compression storage; comprising: storage preparation and intermediate data storage; hierarchical storage of key points; storage connection of compressed data; mixed format storage and conversion; final storage of global format. The application adopts hierarchical storage of initial key points and screened target key points, avoids excessive compression waste storage, uses the least square method for secondary compression in a smooth section, accurately retains core data, effectively reduces data redundancy and storage demand, and significantly improves the calculation ability and processing efficiency of large-scale complex model three-dimensional inversion.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration, and in particular to a method for adaptive compression and high-precision recovery of vector fields, applicable to the field of transient electromagnetic three-dimensional inversion, aiming to improve the three-dimensional inversion calculation capability of large-scale complex models. Background Technology

[0002] Transient electromagnetic methods (TEM) are a key technology in geophysical exploration, offering significant advantages in subsurface resource detection, geological structure investigation, and geological hazard assessment. Their three-dimensional inversion is crucial for achieving detailed imaging of subsurface conductive media. Time-varying electromagnetic field vector data forms the basis for inverting spatial distribution, and its accuracy directly impacts the accuracy and reliability of subsurface structure interpretation. However, large-scale three-dimensional inversion faces severe storage and computational bottlenecks. The sensitivity matrix requires full-time-series forward modeling field values. Traditional storage schemes, centered on "full data retention and static adaptation," fail to consider data geometric characteristics and redundancy, leading to exponential increases in storage costs with model size and time step. Especially with limited GPU memory, overflow errors are easily triggered, hindering the feasibility of large-scale inversion. The requirements for miniaturized and low-power equipment in field exploration, coupled with bandwidth limitations in distributed computing, exacerbate the adaptation challenges.

[0003] Existing data compression methods have limitations: they neglect the characteristics of local abrupt changes in the electromagnetic field, leading to the loss of key information; and they struggle to balance compression efficiency and reconstruction accuracy, failing to meet the requirements for field value recovery. Therefore, it is necessary to develop a storage optimization technique that balances high compression efficiency, preservation of core features, and high-precision recovery.

[0004] The applicant's earlier application, CN120067514A, discloses a transient electromagnetic three-dimensional inversion method based on vector field geometric compression. This method adaptively compresses the forward field by controlling a threshold based on the geometric characteristics of the electromagnetic field data at all times in the three-dimensional forward modeling results, retaining only key feature points reflecting the shape of the magnetic field change rate curve. During the accompanying forward modeling process, the forward field values ​​at the required time are recovered through compression rules. This information-recoverable vector field timescale geometric compression method can significantly reduce data storage requirements, thus breaking the limitation of "small model, large storage." It enables large-scale three-dimensional inversion calculations with limited GPU memory resources. However, this method lacks adaptive compression capabilities during the compression process, employing a uniform compression and storage strategy for all data, which cannot be adaptively adjusted according to actual storage needs, resulting in low storage resource utilization. Furthermore, in sections where the curve approaches linearity, the forced selection of key points at a fixed sampling interval leads to data compression redundancy. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a transient electromagnetic three-dimensional inversion method based on vector field adaptive compression. Based on the geometric characteristics and physical response laws of the electromagnetic field, the method uses threshold-controlled adaptive compression to retain key feature points and accurately recovers the field values ​​through interpolation during the accompanying forward modeling. This aims to reduce storage requirements and GPU memory usage, and support the efficient implementation of large-scale three-dimensional inversion under limited hardware conditions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A transient electromagnetic three-dimensional inversion method based on vector field adaptive compression includes: 1. Data Acquisition: Time-domain response data of electromagnetic fields are acquired using transient electromagnetic method equipment to construct an electromagnetic field dataset containing multiple sampling times; 2. Construct a three-dimensional inversion objective function; 3. Design the forward modeling equations based on the three-dimensional forward modeling theory; 4. Information recovery vector field compression: compresses electromagnetic field vector data and extracts key point data information from the electric and magnetic field attenuation curves; 5. Information can be recovered after compression and storage of vector fields: 5.1. Storage Preparation and Intermediate Data Storage: ① Initialize key parameters, including time step, compression algorithm configuration, and storage statistics parameters; at the same time, set the number of key points per spatial point and its upper limit.

[0007] ② Initiate the forward iterative calculation process to dynamically solve the real electric field and auxiliary electric field components in the x, y, and z directions, and record the current iteration step in real time.

[0008] ③ Perform iteration step verification: Determine whether the current iteration step has reached the preset iteration step interval threshold.

[0009] ④ If the threshold condition is met, the complete electromagnetic field data (including the values ​​of the six electric field components), the iteration step number, and the loop current parameters corresponding to the current iteration step are written into the temporary buffer.

[0010] ⑤ If the threshold condition is not met, skip the data storage operation and continue to execute the forward iteration until the termination time step is reached or the amount of data in the buffer triggers the sampling window condition.

[0011] 5.2. Hierarchical storage of key points: ① Counting Threshold Verification: Read the number of keypoints stored in the current single spatial point and determine whether it has reached the preset keypoint count threshold. If the counting threshold is met, update the keypoint count, store the keypoint corresponding to the current count, then adjust the counting rules, reset the keypoint count to 1, and restart the next round of counting accumulation to avoid excessive concentration of keypoints, and simultaneously update the maximum number of keypoints in a single spatial point; if the counting threshold is not met, directly store the initial keypoint (i.e., the first field value point corresponding to the starting time step) as the curve baseline feature point, and the keypoint count continues to iterate, entering the subsequent data verification stage. ② Least Squares Secondary Optimization: For the currently accumulated field value data set in the buffer, linear interpolation is used to calculate the interpolated field value corresponding to the intermediate time step. To avoid redundant compression caused by forcibly selecting key points at fixed intervals when the curve tends to a straight line, the time series field value data in the current buffer is subjected to straight line segment detection and least squares optimization compression: the linear correlation coefficient is used to determine whether the curve segment is approximately a straight line. If the straight line determination condition is met and the relative error of each point is small, the least squares method is used to perform global linear fitting on the entire data segment, retaining only the starting point of the segment and the fitting endpoint of the segment as key feature points, directly eliminating intermediate redundant data points, and no longer entering point-by-point error verification; if the straight line determination condition is not met, the conventional error threshold verification is continued.

[0012] ③ Error Threshold Verification: Calculate the relative error between the interpolated field value and the original field value according to a preset formula, and compare the relative error with the compression error threshold. If the relative error is greater than the compression error threshold, the original field value is identified as a target key point, marked, and stored in a temporary cache, and the key points are updated synchronously. If the relative error is less than or equal to the compression error threshold, the original field value is identified as a non-key point and removed, retaining only the current starting point to prepare for the next round of data verification.

[0013] 5.3 Storage connection of compressed data: ① Compressed data volume verification: Determine whether the total data volume after compression reaches the preset data volume threshold.

[0014] ② Handling of non-compliance: Update the iteration step and determine whether the iteration step has reached the end time step; return to the forward iteration link of step (6), continue to calculate new field value data and store intermediate results; repeat the hierarchical screening logic of step (7) for the newly generated intermediate data, supplement and store new key points until the total amount of compressed data is ≥ the preset data amount threshold or the iteration step is equal to the end time step; ③ Compliance processing: If the total amount of compressed data is greater than or equal to the preset data amount threshold, stop the current iteration accumulation and proceed to the next step of mixed format storage and conversion; if the iteration step is equal to the end time step, directly proceed to the format conversion step to avoid data omission.

[0015] 5.4. Mixed Format Storage and Conversion: ① Local format conversion: Identify local data blocks stored in CSR format in the temporary cache; since this type of data is relatively small and the advantages of column dimension become apparent over time, convert it to CSC format (and store it to adapt to the data organization requirements of column dimension and reduce the storage overhead of single data block).

[0016] ② Global format integration: The adaptive storage judgment logic is activated, and multiple local CSC format data blocks are merged based on the current storage fragmentation rate, data block size distribution, and remaining capacity of the total system storage capacity. All merged data is converted into a unified global CSC format, realizing centralized management of full compressed data, reducing the frequency of format switching, and improving the efficiency of subsequent batch processing.

[0017] ③ Storage status monitoring: During the conversion process, the system storage usage is monitored in real time to ensure that it does not exceed the total system storage capacity. If it is close to the capacity limit, small data blocks are merged first to release fragmented space and avoid memory overflow.

[0018] 5.5. Final storage of global format: ① Final iteration step verification: Confirm that the number of iteration steps has reached the preset termination time step, and ensure that the key points of all spatial points have been filtered and formatted to ensure that no data is missing.

[0019] ② Final format conversion: Given that subsequent adjoint field calculations and inversion analysis need to be based on row-first data organization, the full compressed data in Global Compressed Sparse Columns (CSC) format will be uniformly converted to Global Compressed Sparse Rows (CSR) format to ensure that the data storage structure is consistent with the calculation logic.

[0020] ③ Optimize storage and archiving: Use single-precision floating-point format to store the converted CSR data, reducing memory usage by 50% while ensuring calculation accuracy; write the final data to the specified storage medium and record storage logs, including key parameters such as compression threshold, total data volume after compression, maximum number of key points per space point, and storage capacity usage, to provide data support for subsequent parameter optimization; ④ Process Closure: After storage is complete, a "Data compression and storage successful" status indicator is output. At this point, the data can be directly accessed in subsequent calculation stages, marking the completion of the entire compression and storage process.

[0021] The beneficial effects of this invention are as follows: The core of this invention is the innovative application of information-recoverable vector field compression technology in 3D inversion, constructing a collaborative "compression-storage-recovery" technical system. Key feature points of the vector curve are extracted using a specific algorithm. Using a dual criterion of "key point count threshold + compression error threshold," initial key points and selected target key points are stored hierarchically. To avoid excessive compression and wasted storage, a second compression is performed in the smooth segment using the least squares method, accurately preserving core data. A compression curve is constructed based on the feature points to achieve geometric compression. During the accompanying forward modeling, interpolation is used to recover the full-time field values ​​and control errors, ensuring inversion accuracy and data interpretability. Simultaneously, an adaptive storage strategy combining CSR and CSC is adopted. Through dynamic format conversion of "small CSR to CSC, multiple small CSCs merged into a large CSC, and large CSC to CSR," storage resources are allocated on demand. Finally, storage and compression algorithms are deeply bound together, with storage triggering and data selection responding synchronously to the compression process results, forming a collaborative filtering mechanism. This effectively reduces data redundancy and storage requirements, significantly improving the computational power and processing efficiency of 3D inversion of large-scale complex models. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a curve feature point extraction method for recoverable information; Figure 2 Measurements before and after compression recovery under different compression error thresholds. Ex The inversion results for the points are as follows: (a) has a compression error threshold of 0.03%; (b) has a compression error threshold of 0.1%; (c) has a compression error threshold of 1%; and (d) has a compression error threshold of 3%. Figure 3 This is a graph showing the usage of video memory under different compression error thresholds. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] The structures, proportions, and sizes illustrated in the accompanying drawings are merely for illustrative purposes and to aid those skilled in the art in understanding and reading the invention. They are not intended to limit the scope of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, provided they do not affect the effectiveness or purpose of the invention, should still fall within the scope of the technical content disclosed herein. Furthermore, the terms "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.

[0025] A transient electromagnetic three-dimensional inversion method based on vector field adaptive compression includes: 1. Data Acquisition: Time-domain response data of the electromagnetic field are acquired using a transient electromagnetic method device to construct an electromagnetic field dataset containing multiple sampling times. This device includes a transmitting coil powered by a transmitter to excite the primary field, and a receiving coil deployed on the ground or mounted on a UAV. The data acquisition process includes: powering the transmitting coil to excite the primary field with a 50% duty cycle; during the current-off phase, recording the rate of change of the secondary induced magnetic field generated by the underground eddy current through the receiving coil. This dataset characterizes the rate of change of the secondary magnetic field induced by the underground eddy current.

[0026] 2. Constructing the 3D Inversion Objective Function: Transient electromagnetic 3D inversion is an objective function optimization process. First, an initial model is constructed as a perturbation; then, the model parameters are updated through iterative calculations until the objective function meets the convergence condition. At this point, the forward simulation results and the actual observation data are considered to meet the fitting requirements. In 3D inversion, the objective function typically consists of two parts, namely… , (1) in, φ d and φ m Subscripts represent the data fitting term and the model constraint term, respectively. m The dimension of the data is represented by m, which represents the electrical parameter (conductivity in this invention), and λ is the regularization factor for balancing the data terms and model terms.

[0027] 3. According to the three-dimensional forward modeling theory, the forward modeling equations can be written in the following form: Kx=b, (2) In the formula, K represents the coefficient matrix, x represents the electric and magnetic fields in the governing equations of the three-dimensional forward modeling algorithm, and b represents the right-hand side term.

[0028] 4. The core term for calculating the gradient of the objective function is the product of the transpose of the Jacobian matrix and the data residual vector, J. T r can be represented as: , (3) Where r is the residual vector between the observed data and the forward modeling data, J is the Jacobian matrix, v is called the adjoint field, and L represents the temporal and spatial interpolation operators. , In the time-domain method, the order of K is related to the model size and the number of computation time points.

[0029] The residual vector r and the forward field x are obtained by solving the forward model in one step, and then J Tr is used as the source term. After one adjoint forward calculation, v is obtained. Finally, x and v are substituted into equation (3) to calculate J. T r.

[0030] 5. Information can be recovered via vector field compression and storage: First, the electromagnetic field vector data is compressed. Based on the field value variation curve, key feature points representing the curve shape are selected and extracted, while redundant data points are removed. During the calculation process, the required electromagnetic field values ​​are reconstructed segment by segment according to the key feature points and data compression rules. Compression algorithms can include the Douglas-Pock algorithm, the vertical distance limit method, or the angle limit method. Compression strategies can employ lossy or lossless compression modes.

[0031] The specific steps for information recovery vector field compression and storage include (1)-(10). (1) Set the compression error threshold of the vector curve. Combine the compression ratio and the scale accuracy to set the compression error threshold of the vector curve and control the compression error range.

[0032] (2) Extraction of feature points of vector curves: such as Figure 1 As shown, retaining the first point A0 of the curve as the feature point, and connecting the first point A0 and the third point A2 from the initial step to form a line segment. Using interpolation: To calculate the intermediate value and determine the relative error.

[0033] (3) Feature Point Judgment and Retention: Compare the relative error with the compression error threshold, and handle two cases: If the relative error ≤ the compression error threshold, retain the current point as the key feature point and set it as the new starting point, connect the next point to form a line segment, and calculate the error by interpolation for comparison; If the relative error > the compression error threshold, do not retain the current point, retain the original starting point, connect the initial point and the fourth point to form a line segment, calculate the error by interpolation, and if the error > the threshold, retain the previous point as the key feature point. Figure 1 For example, starting from the current starting point A0, the system first connects to the next sampling point A2 and calculates the relative error. If the relative error meets a preset threshold, A1 is determined to be a key feature point, and the starting point is updated to A1. If the relative error does not meet the threshold, the starting point remains unchanged at A0, and the system then connects to A3 and calculates the error again. If the result of this calculation still does not meet the requirements, the previous point A2 is determined to be a key feature point. The sequence A0-A1-A2-A3-A4-A5 shown in the figure intuitively illustrates this process of "trial and error—judgment—updating the starting point".

[0034] (4) Curve reconstruction: Repeat the above process to traverse all data points, connect the retained key feature points according to the time series, and complete the curve compression reconstruction.

[0035] (5) Information recovery verification: The compressed data is recovered and verified. The compression effect is evaluated by comparing the relative error between the recovered data and the original data. During the calculation of the accompanying field, the full field data is recovered based on feature point interpolation, and the interpolation accuracy is controlled by the compression error threshold.

[0036] (6) Storage preparation and intermediate data storage: ① Initialize key parameters, including time step, compression algorithm configuration, and storage statistics parameters; at the same time, set the number of key points per spatial point and its upper limit.

[0037] ② Initiate the forward iterative calculation process to dynamically solve the real electric field and auxiliary electric field components in the x, y, and z directions, and record the current iteration step in real time.

[0038] ③ Perform iteration step verification: Determine whether the current iteration step has reached the preset iteration step interval threshold.

[0039] ④ If the threshold condition is met, the complete electromagnetic field data (including the values ​​of the six electric field components), the iteration step number, and the loop current parameters corresponding to the current iteration step are written into the temporary buffer.

[0040] ⑤ If the threshold condition is not met, skip the data storage operation and continue to execute the forward iteration until the termination time step is reached or the amount of data in the buffer triggers the sampling window condition.

[0041] (7) Hierarchical storage of key points: ① Counting Threshold Verification: Read the number of keypoints stored in the current single spatial point and determine whether it has reached the preset keypoint count threshold. If the counting threshold is met, update the keypoint count, store the keypoint corresponding to the current count, then adjust the counting rules, reset the keypoint count to 1, and restart the next round of counting accumulation to avoid excessive concentration of keypoints, and simultaneously update the maximum number of keypoints in a single spatial point; if the counting threshold is not met, directly store the initial keypoint (i.e., the first field value point corresponding to the starting time step) as the curve baseline feature point, and the keypoint count continues to iterate, entering the subsequent data verification stage. ② Least Squares Secondary Optimization: For the currently accumulated field value data set in the buffer, linear interpolation is used to calculate the interpolated field value corresponding to the intermediate time step. To avoid redundant compression caused by forcibly selecting key points at fixed intervals when the curve tends to a straight line, the time series field value data in the current buffer is subjected to straight line segment detection and least squares optimization compression: the linear correlation coefficient is used to determine whether the curve segment is approximately a straight line. If the straight line determination condition is met and the relative error of each point is small, the least squares method is used to perform global linear fitting on the entire data segment, retaining only the starting point of the segment and the fitting endpoint of the segment as key feature points, directly eliminating intermediate redundant data points, and no longer entering point-by-point error verification; if the straight line determination condition is not met, the conventional error threshold verification is continued.

[0042] ③ Error Threshold Verification: Calculate the relative error between the interpolated field value and the original field value according to a preset formula, and compare the relative error with the compression error threshold. If the relative error is greater than the compression error threshold, the original field value is identified as a target key point, marked, and stored in a temporary cache, and the key points are updated synchronously. If the relative error is less than or equal to the compression error threshold, the original field value is identified as a non-key point and removed, retaining only the current starting point to prepare for the next round of data verification.

[0043] (8) Storage connection of compressed data: ① Compressed data volume verification: Determine whether the total data volume after compression reaches the preset data volume threshold.

[0044] ② Handling of non-compliance: Update the iteration step and determine whether the iteration step has reached the end time step; return to the forward iteration link of step (6), continue to calculate new field value data and store intermediate results; repeat the hierarchical screening logic of step (7) for the newly generated intermediate data, supplement and store new key points until the total amount of compressed data is ≥ the preset data amount threshold or the iteration step is equal to the end time step; ③ Compliance processing: If the total amount of compressed data is greater than or equal to the preset data amount threshold, stop the current iteration accumulation and proceed to the next step of mixed format storage and conversion; if the iteration step is equal to the end time step, directly proceed to the format conversion step to avoid data omission.

[0045] (9) Mixed format storage and conversion: ① Local format conversion: Identify local data blocks stored in CSR format in the temporary cache; since this type of data is relatively small and the advantages of column dimension become apparent over time, convert it to CSC format (and store it to adapt to the data organization requirements of column dimension and reduce the storage overhead of single data block).

[0046] ② Global format integration: The adaptive storage judgment logic is activated, and multiple local CSC format data blocks are merged based on the current storage fragmentation rate, data block size distribution, and remaining capacity of the total system storage capacity. All merged data is converted into a unified global CSC format, realizing centralized management of full compressed data, reducing the frequency of format switching, and improving the efficiency of subsequent batch processing.

[0047] ③ Storage status monitoring: During the conversion process, the system storage usage is monitored in real time to ensure that it does not exceed the total system storage capacity. If it is close to the capacity limit, small data blocks are merged first to release fragmented space and avoid memory overflow.

[0048] (10) Final storage of global format: ① Final iteration step verification: Confirm that the number of iteration steps has reached the preset termination time step, and ensure that the key points of all spatial points have been filtered and formatted to ensure that no data is missing.

[0049] ② Final format conversion: Given that subsequent adjoint field calculations and inversion analysis need to be based on row-first data organization, the full compressed data in Global Compressed Sparse Columns (CSC) format will be uniformly converted to Global Compressed Sparse Rows (CSR) format to ensure that the data storage structure is consistent with the calculation logic.

[0050] ③ Optimize storage and archiving: Use single-precision floating-point format to store the converted CSR data, reducing memory usage by 50% while ensuring calculation accuracy; write the final data to the specified storage medium and record storage logs, including key parameters such as compression threshold, total data volume after compression, maximum number of key points per space point, and storage capacity usage, to provide data support for subsequent parameter optimization; ④ Process Closure: After storage is complete, a "Data compression and storage successful" status indicator is output. At this point, the data can be directly accessed in subsequent calculation stages, marking the completion of the entire compression and storage process. Example

[0051] This embodiment takes the calculation of a three-dimensional inversion model as an example, and compares the results using the adaptive compressed vector field method proposed in this invention with those not using the compressed vector field method. The comparison data comes from the forward calculation and adjoint forward calculation results in the first inversion iteration process.

[0052] 1. Assessment of compressed data recovery: To assess the difference between the original data and the recovered data in the model, Figure 2 The calculated electric field at the center of the emission source is shown. The orange solid line in the figure represents the original... E x The curve, with the blue dashed line representing the curve recovered using vector compression, is shown. E x The curves, with the purple solid line representing the relative error, show that the recovered data curves highly overlap with the original data curves, fully reproducing the core characteristics of the electromagnetic field decaying over time and effectively preserving the key evolutionary trend of the vector field. The maximum values ​​of the relative errors do not exceed the corresponding compression thresholds. The errors are mainly concentrated in the earlier time stages, where the rate of change of the electromagnetic field is high and the curve shape is complex, especially in the 10⁻⁻⁶ time range. 4 After s, the curve tends to be smoother, and the relative error is significantly reduced.

[0053] The maximum error of the field value is generally within the range of 1μs-100μs. Beyond 100μs, the relative error is much smaller than the initial threshold. This is because the curve shape is relatively smooth and satisfies the non-critical point rule over a long period. Therefore, to avoid this situation, a maximum time step interval (Ns) for the "critical point" is introduced, which is set to 50 in this paper.

[0054] 2. Memory usage under different compression error thresholds: This section records the peak memory usage during the forward and accompanying forward calculations in the inversion process, and provides a statistical analysis of video memory consumption, as shown in Table 1 and... Figure 3 As shown, when the compression threshold is set to 0%, the memory usage using double-precision storage reaches 23.6 GB, while single-precision storage uses 11.8 GB. It should be noted that forward modeling still uses double-precision storage. The results show that memory usage dynamically changes under different compression thresholds, gradually decreasing as the compression threshold increases. This is because a larger compression threshold discards more non-critical data points, thus reducing memory consumption. Furthermore, the storage scheme used in this study allocates memory in real time, causing dynamic changes in memory usage.

[0055] Table 1. Memory usage under different compression error thresholds

[0056] It is worth noting that the video memory usage of compressed field value data is related to the number of keypoints (Ns) and the total number of storage points (Na). When both Ns and Na are large, the allocated video memory may even exceed the usage when the compression threshold is set to 0%.

[0057] While using a compression threshold can effectively reduce model memory usage, it can also introduce data errors, thus affecting the accuracy of the calculation results. Therefore, the choice of compression threshold requires a trade-off between memory usage and computational accuracy, rather than simply pursuing a higher threshold.

[0058] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A transient electromagnetic three-dimensional inversion method based on vector field adaptive compression, comprising: 1) Data Acquisition: 2) Construct the three-dimensional inversion objective function; 3) Design the forward modeling equations based on the three-dimensional forward modeling theory; 4) Information recovery vector field compression: compresses electromagnetic field vector data and extracts key point data information from the electric and magnetic field attenuation curves; Its characteristics include: It also includes: 5) The information can be recovered after compression and storage; specifically: 5.1) Storage preparation and intermediate data storage; 5.2) Hierarchical storage of key points; 5.3) Storage connection of compressed data; 5.4) Mixed format storage and conversion; 5.5) Final storage of global format.

2. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 1, characterized in that, Step 5.1) Storage preparation and intermediate data storage specifically includes: ① Initialize key parameters, including time step, compression algorithm configuration, and storage statistics parameters; at the same time, set the number of key points per spatial point and its upper limit; ② Initiate the forward iterative calculation process to dynamically solve the actual electric field and auxiliary electric field components in the x, y, and z directions, and record the current iteration step in real time; ③ Perform iteration step verification: Determine whether the current iteration step has reached the preset iteration step interval threshold; ④ If the threshold condition is met, write the complete electromagnetic field data, iteration step number and loop current parameters corresponding to the current iteration step into the temporary buffer area. ⑤ If the threshold condition is not met, skip the data storage operation and continue to execute the forward iteration until the termination time step is reached or the amount of data in the buffer triggers the sampling window condition.

3. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 1, characterized in that, The hierarchical storage of the key points in step 5.2) is specifically as follows: ① Counting threshold verification: Read the number of key points stored in the current single spatial point and determine whether the preset key point number threshold has been reached; ② Least squares quadratic optimization: For the current accumulated field value data set in the buffer, the linear interpolation method is used to calculate the interpolated field value corresponding to the intermediate time step; ③ Error threshold verification: Calculate the relative error between the interpolated field value and the original field value according to the preset formula, and compare the relative error with the compression error threshold.

4. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 3, characterized in that, Step ①, the counting threshold verification, is as follows: if the counting threshold is met, the number of key points is updated, the key points corresponding to the current count are stored, the counting rules are then adjusted, the number of key points is reset to 1, and the next round of counting accumulation is restarted to avoid excessive concentration of key points, and the maximum number of key points in a single spatial point is updated synchronously; if the counting threshold is not met, the initial key points are directly stored as the curve reference feature points, the number of key points continues to iterate, and the subsequent data verification stage begins.

5. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 3, characterized in that, Step ②, least squares secondary optimization, specifically involves: to avoid redundant compression caused by forcibly selecting key points at fixed intervals when the curve approaches a straight line, performing straight line segment detection and least squares optimization compression on the time series field value data in the current buffer: judging whether the curve segment is approximately a straight line by the linear correlation coefficient; if the straight line judgment condition is met and the relative error of each point is small, then the least squares method is used to perform global linear fitting on the entire data segment, retaining only the starting point of the segment and the fitting endpoint of the segment as key feature points, directly eliminating redundant data points in the middle, and no longer entering point-by-point error verification; if the straight line judgment condition is not met, then the conventional error threshold verification continues.

6. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 3, characterized in that, Step ③, error threshold verification, specifically involves: if the relative error is greater than the compression error threshold, the original field value is determined as the target key point, marked and stored in a temporary cache area, and the key point is updated synchronously; if the relative error is less than or equal to the compression error threshold, the original field value is determined as a non-key point and removed, and only the current starting point is retained to prepare for the next round of data verification.

7. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 1, characterized in that, The storage connection of the compressed data in step 5.3) is specifically as follows: ① Compressed data volume verification: Determine whether the total data volume after compression reaches the preset data volume threshold; ② Handling of non-compliance: Update the iteration step and determine whether the iteration step has reached the end time step; Returning to the forward iteration in step 5.2), continue to calculate new field value data and store intermediate results; repeat the hierarchical filtering logic of step 5.3) on the newly generated intermediate data, supplement and store new key points until the total amount of compressed data is greater than or equal to the preset data amount threshold or the iteration step is equal to the end time step; ③ Compliance processing: If the total amount of compressed data is greater than or equal to the preset data amount threshold, stop the current iteration accumulation and proceed to the next step of mixed format storage and conversion; if the iteration step is equal to the end time step, directly proceed to the format conversion step to avoid data omission.

8. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 1, characterized in that, The mixed format storage and conversion in step 5.4) specifically includes: ① Local format conversion: Identify local data blocks stored in the temporary cache in a compressed sparse row format; since this type of data is relatively small and the advantages of column dimension become apparent over time, convert it to a compressed sparse column format and store it to adapt to the needs of column dimension data organization and reduce the storage overhead of a single data block. ② Global format integration: The adaptive storage judgment logic is activated, and multiple locally compressed sparse column format data blocks are merged based on the current storage fragmentation rate, data block size distribution, and remaining capacity of the total system storage capacity; all merged data are converted into a unified global compressed sparse column format, realizing centralized management of full compressed data, reducing the frequency of format switching, and improving the efficiency of subsequent batch processing. ③ Storage status monitoring: During the conversion process, the system storage usage is monitored in real time to ensure that it does not exceed the total system storage capacity. If it is close to the capacity limit, small data blocks are merged first to release fragmented space and avoid memory overflow.

9. The transient electromagnetic three-dimensional inversion method based on vector field adaptive compression as described in claim 1, characterized in that, The final storage of the global format in step 5.5) is specifically as follows: ① Final iteration step verification: Confirm that the number of iteration steps has reached the preset termination time step, and ensure that the key points of all spatial points have been filtered and formatted to ensure that no data is missing; ② Final format conversion: Given that subsequent adjoint field calculations and inversion analysis need to be based on row-major data organization, the full compressed data in the globally compressed sparse column format will be uniformly converted into the globally compressed sparse row format to ensure that the data storage structure is consistent with the calculation logic. ③ Optimize storage and archiving: Use single-precision floating-point format to store the converted compressed sparse row data, reducing memory usage by 50% while maintaining calculation accuracy; The final data is written to the specified storage medium and a storage log is recorded, including the compression threshold, the total amount of data after compression, the maximum number of key points in a single space, and the key parameters of storage capacity usage, to provide data support for subsequent parameter optimization. ④ Process closed loop: After storage is completed, output the status "Data compression and storage successful"; at this point, the data can be directly called to the subsequent calculation stage, marking the completion of the entire compression and storage process.

Images