基于分数阶最优控制网络的复杂勘探随机噪声消减方法

在实际勘探记录处理过程中，复杂随机噪声的出现严重影响了有效反射信息的提取，并对资料后续处理带来了不利影响。随着非常规油气资源开发，对勘探记录质量提出了更高的要求，常规方法在处理能力方面需要持续提升。为了解决复杂噪声消减问题，笔者将最优控制网络引入随机噪声消减领域。与传统的单一尺度消噪网络不同，FOC-NET具有分层结构，能够利用不同尺度信息并结合信息融合处理实现地震勘探数据潜在特征的高精度提取，克服了传统去噪网络单一尺度信息提取造成的有效特征损失问题。同时，在面对低信噪比勘探记录和弱反射同相轴时，多尺度特征交互方式同样可以有效提高噪声压制和信号恢复能力。合成记录和实际数据处理结果均表明，即使在低信噪比条件下，FOC-NET仍能有效地抑制随机噪声并准确重构出有效反射信息，极大提升勘探资料的质量。 

Random noise reduction method of complex exploration based on a fractional optimal control network

In field seismic prospecting, the existence of complex random noise strongly influences the detection of effective reflected signals and imposes adverse effects on the subsequent process of the data. With the exploration of unconventional petroleum resources, high-quality seismic records are needed, and the performance of conventional methods in processing needs to be further improved. To solve this problem, a fractional optimal control network (FOC-NET) is utilized to cope with complex noise reduction. Unlike conventional DnCNN, FOC-NET has a hierarchical architecture. It can cope with the feature loss caused by a single-scale feature extraction strategy. Then, the potential feature of the analyzed data can be accurately captured in combination with the different-scale information. Notably, the multi-scale fusion capability can also improve the recovery of weak reflection events and complex noise attenuation. Both synthetic and field experimental results demonstrate that FOC-NET can effectively suppress random noise and restore the desired signals, even under low-SNR conditions.