基于Wolfe搜索机制的随机梯度地震反演方法

Seismic inversion method based on SGO-WS

地震反演技术能够最有效地从地震信号中挖掘地层参数和岩性信息,一直是储层预测研究的焦点.传统线性地震反演算法缺乏全局搜索能力,反演结果精度较低.本研究以全局寻优为出发点,将一种结构简单和寻优能力强的全局优化算法——梯度优化算法(Gradient-Based Optimizer,GBO),引入地震反演.相比于差分进化等其他全局优化算法,GBO算法通过梯度随机搜索机制和局部逃逸算子进行全局搜索,能有效降低地震反演的多解性.但是,GBO算法收敛速度慢和局部随机性强,难以满足大批量的地震反演计算需求.因此,本文在GBO算法迭代过程中引入Wolfe线性局部搜索机制,提出基于Wolfe搜索的随机梯度优化算法(Stochastic—Gradient Optimization Based on Wolfe's Search,SGO-WS).在全局搜索过程中,通过线性搜索算子,充分挖掘当前迭代解周围的局部最优,既保证了反演解精度,又大幅提高了原GBO算法的计算效率,同时还有效降低了反演解的局部随机性.Marmousi-2模型测试验证了SGO-WS算法的可行性和准确性,厄瓜多尔Tapir油田地震资料也验证了SGO-WS算法的实用性.

Seismic inversion technologies are pivotal for effectively extracting stratigraphic parameters and lithological information from seismic signals, consistently focusing on reservoir prediction research. Conventional linear seismic inversion algorithms, which lack comprehensive global search capabilities, tend to produce suboptimal results. To mitigate this, we introduce the Gradient-Based Optimizer (GBO), a sophisticated global optimization algorithm, into the domain of seismic inversion. Unlike other global optimizers such as Differential Evolution (DE), the GBO utilizes a hybrid mechanism combining gradient-based random searches with a local escape strategy, markedly enhancing the uniqueness of inversion outcomes. However, the GBO's slow convergence rates and inherent randomness limit its efficiency in large-scale applications. To address these challenges, we integrate a Wolfe line search mechanism during the GBO's iterative process, developing the Stochastic Gradient Optimization based on Wolfe's Search (SGO-WS). This strategy employs a linear search operator to effectively harness local optima, significantly boosting GBO's computational speed while ensuring the accuracy and reliability of the results. The performance and precision of the SGO-WS algorithm were confirmed through validation tests using the Marmousi-2 model, and its applicability was further verified with seismic data from Ecuador's Tapir oil field.