CN-121997731-A - Air gun array configuration optimization method and system based on SHAP explanatory guidance and particle swarm optimization depth fusion

CN121997731ACN 121997731 ACN121997731 ACN 121997731ACN-121997731-A

Abstract

The invention belongs to the technical field of intelligent optimization of marine geophysical exploration equipment, and discloses an air gun array configuration optimization method and an air gun array configuration optimization system based on SHAP explanatory guide and particle swarm optimization depth fusion. Experiments show that compared with the original configuration, the initial foam is improved by 73.3 percent in 20 iterations, compared with PSO, the initial foam is improved by 15.4 percent, the variation coefficient is as low as 0.10 (optimal robustness), the performance breakthrough is realized synchronously, an explanatory chart is output, and a decision basis of 'why optimal' and 'how to fine tune' is provided for engineers. The method is suitable for off-line high-precision array design, and provides a high-efficiency and reliable technical scheme for air gun array optimization. The method is successfully applied to the exploration project of a certain oil field in Bohai sea, effectively improves the quality of seismic data, and has remarkable industrial application value.

Inventors

YANG BO
JIN XIANGFENG
CHEN BIN
LIU LUSI
HUANG XUEQIN

Assignees

海南经贸职业技术学院

Dates

Publication Date: 20260508
Application Date: 20260120

Claims (10)

1. An array parameter configuration optimization method based on explanatory contribution guidance is characterized by comprising the following control mechanisms: When the array configuration parameters are subjected to group search, marginal contribution of each parameter to the performance index is used as an independent control quantity to be introduced into a search process, so that the search direction is simultaneously constrained by a historical optimal solution, a group optimal solution and the marginal contribution change trend, and the search process is iteratively converted into an interpretable guide search process constrained by a parameter contribution mechanism from a black box which is purely dependent on an objective function.
2. The method of claim 1, wherein the marginal contribution is calculated from a neighbor fitting model constructed based on current population history samples, the model characterizing the direction and magnitude of the impact of a single parameter change on the overall performance index change.
3. The method of claim 1, wherein the performance metrics include at least an array far-field response amplitude distribution characteristic and a target frequency band energy distribution characteristic.
4. The group search control method based on the parameter interaction strength self-adaptive grouping is characterized by comprising the following mechanisms: And constructing a parameter interaction network by utilizing the correlation among contribution sequences formed in the group searching process of each parameter, automatically dividing the parameters into a plurality of interaction subgroups according to the community structure of the network, and restricting a searching individual to perform collaborative searching only in the corresponding parameter subgroup, thereby avoiding the decline of searching efficiency caused by incorrect splitting of the strong coupling parameters.
5. The method of claim 4, wherein nodes in the parameter interaction network represent parameters and side weights represent covariance intensities between parameter contribution sequences.
6. The method of claim 4, wherein partitioning of the sub-groups of parameters is accomplished by maximizing community modularity such that intra-sub-group parameter association strengths are greater than inter-sub-group parameter association strengths.
7. An optimization method based on the direct participation of an explanatory gradient in the population dynamics update is characterized in that in the population searching process, an explanatory gradient term formed by the parameter contribution change trend is directly introduced into a state update equation of a searching individual, so that the searching individual moves to a historical optimal position and performs self-adaptive propulsion along the parameter direction with the fastest performance improvement, and the active searching based on the mechanism direction is realized.
8. The method of claim 7, wherein the interpreted gradient is obtained by linear trend fitting a relationship between parameter changes and performance changes in a neighborhood sample.
9. The method of claim 7, wherein the update rate of searching for individuals is comprised of inertial terms, individual optimal guides, population optimal guides, and interpreted gradient guides.
10. An optimization system for implementing the method of any one of claims 1 to 9, comprising: The parameter simulation module is used for generating an array response corresponding to the given parameter configuration; The performance evaluation module is used for calculating and configuring corresponding comprehensive performance indexes; the explanatory analysis module is used for calculating marginal contribution of each parameter to performance and variation trend thereof; The interactive modeling module is used for constructing a parameter interactive network and completing parameter grouping; The search control module is used for executing explanatory guidance, grouping collaboration and state update; And the result output module is used for outputting the optimal parameter configuration result.

Description

Air gun array configuration optimization method and system based on SHAP explanatory guidance and particle swarm optimization depth fusion Technical Field The invention belongs to the technical field of intelligent optimization of marine geophysical exploration equipment, and particularly relates to an air gun array configuration optimization method and system based on SHAP explanatory guide and particle swarm optimization depth fusion. Background In marine seismic exploration, an air gun array is a key device for exciting seismic waves, and the performance of the air gun array directly influences the quality of acquired data. The depth configuration of the array is a core parameter that affects far-field wavelet characteristics (e.g., bubble ratio, spectral characteristics). The traditional air gun array optimization method is mostly dependent on a trial-and-error method, and has the problems of low efficiency and difficulty in obtaining a globally optimal solution. Although Particle Swarm Optimization (PSO) algorithms are introduced to improve the problem, there are obvious limitations that the optimization results are easy to fall into local optimum due to lack of understanding of parameter-performance influence mechanism only by relying on random search and swarm iteration, and the converged optimization results are lack of physical interpretability-engineers cannot clearly determine the specific effect of depth adjustment of a certain air gun, so that popularization of the optimization results in actual production is limited. SHAP (SHAPLEY ADDITIVE exPlanations) is used as an interpretable AI tool for performing Post-hoc Interpretation on a trained model, namely, is used as an independent analysis tool, is usually combined with an optimization process only in a simple series (such as analysis after optimization), fails to realize deep interaction and cooperation with an optimization algorithm, does not guide the optimization process in real time, cannot exert the core value of revealing parameter-target mapping relation, and causes the optimization and interpretation to present a two-skin state. Therefore, how to deeply embed an interpretable artificial intelligence (XAI) technology into an optimization algorithm and realize an air gun array optimization technology with interpretable optimization process, multi-index collaborative optimization and performance-efficiency balance becomes a technical problem to be solved. Through the above analysis, the problems and defects existing in the prior art are as follows: SHAP is used as an independent analysis tool, and is combined with an optimization process only in a simple series (such as analysis after optimization), so that deep interaction and cooperation with an optimization algorithm cannot be realized, the optimization process is not guided in real time, the core value of the parameter-target mapping relation cannot be exerted, and the optimization and interpretation are in a two-skin state. Disclosure of Invention Aiming at the problems existing in the prior art, the invention provides an air gun array configuration optimization method and an air gun array configuration optimization system based on SHAP explanatory guiding and particle swarm optimization depth fusion. The invention is realized in such a way that an air gun array configuration optimization method based on SHAP explanatory guidance and particle swarm optimization depth fusion comprises the following steps: initializing and simulating, namely initializing a particle swarm, wherein each particle represents an air gun depth configuration scheme; simulating far-field pressure wavelets of the air gun array corresponding to each particle based on a physical mechanism model; Step 2, multi-objective fitness evaluation; step 3, SHAP value online calculation; Step 4, constructing and dynamically grouping parameter interaction matrixes, namely calculating interaction matrixes among parameters based on SHAP values of all particles in each dimension, wherein the matrixes reveal coupling relations among different air gun depth parameters; Step 5, SHAP gradient guided particle updating, which is to introduce SHAP gradient items into a standard PSO speed updating formula, explore the particles not only to individual history optimal and global history optimal learning, but also along the direction of the fastest performance improvement indicated by SHAP values (namely gradient direction); And 6, iterating and terminating, namely repeating the steps 2-5 until a termination condition is met (such as the maximum iteration number is reached), and outputting the global optimal depth configuration. Further, the calculation formula of the far-field pressure wavelet of the air gun array is as follows: Wherein, the Representing the synthetic pressure wavelet of the air gun array at a far-field receiving point at the moment t; summing sign from 1 st air gun To the Kth air gun) Accumulating, wherein K represents