高斯过程回归在航空瞬变电磁去噪中的应用

The Application of Gaussian Process Regression in Aviation Transient Electromagnetic Denoising

航空瞬变电磁法具有速度快、勘探面积大、成本相对较低、能克服复杂地形条件的限制等优点,已经在很多领域得到了广泛的应用。但是由于航空瞬变电磁二次场电磁响应属于宽频带信号,容易受到多种噪声的影响,从而降低数据的质量,影响后期反演解释的精度。因此,对航空瞬变电磁数据去噪方法的研究仍然是当前研究的热点之一。采用机器学习的思想,将高斯过程回归方法应用于航空瞬变电磁数据去噪。首先,利用模拟数据进行去噪研究,通过对模拟数据添加不同程度的高斯白噪声和天电噪声,其中高斯白噪声代表航空电磁数据中的随机噪声,而天电噪声是航空电磁中主要的噪声之一,经过模拟数据去噪效果发现,高斯过程回归对于其中的天电噪声、随机噪声都有很好的滤除效果。而后再对实测数据进行研究,无论是从单测点衰减曲线来看,还是从剖面数据上分析,都可以看出去噪后曲线相对更平滑,且幅值得到了相应地压制。

Aviation transient electromagnetic method has the advantages of high speed, large exploration area, relatively low cost, and ability to overcome the limitations of com- plex terrain conditions. It has been widely used in many fields. However, because the elec- tromagnetic transient response of the aviation transient electromagnetic field belongs to broadband signal, it is easy to be affected by various noises, thereby reducing the quality of the data and affecting the accuracy of late inversion interpretation. Therefore, the research on the denoising method of aviation transient electromagnetic data is still one of the hot topics in current research. This paper attempts to use the Gauss process regression method to denoise aeronautical transient electromagnetic data by using the theory of machine learning. Firstly, the simulation data are used for denoising research. Different degrees of Gaussian white noise and atmospheric noise are added to the simulation data, of which Gaussian white noise represents random noise in aviation electromagnetic data, and atmospheric noise is one of the main noises in aviation electromagnetic data. After denoising effect of analog data, Gaussian process regression has a good filtering effect on the atmospheric noise and random noise. Then, the measured data are studied. Whether it is from the single--point attenuation curve or the profile data, it can be seen that the curve after de--noising is relatively smooth and the amplitude is correspondingly suppressed.