Self-supervised inversion method based on transient electromagnetic for deep learning

A self-supervised inversion method based on transient electromagnetic for deep learning, comprising:

• inputting electromagnetic response data to be inverted into a pre-trained inversion network for inversion to obtain inverted resistivity values;
• wherein an inversion network comprises an encoder part, a decoder part, and a smooth constraint part;
• and the pre-trained inversion network comprises:
• generating resistivity values and corresponding electromagnetic response data through a numerical simulation method, and constructing a dataset based on the resistivity values and the corresponding electromagnetic response data;
• wherein generating the resistivity values and the corresponding electromagnetic response data through the numerical simulation method comprises:
• setting basic parameters of underground stratum, comprising thickness of a top layer and boundary depth of a last layer;
• dividing a stratum into a plurality of layers of a same thickness for each layer;
• setting a plurality of control points within a preset stratum depth range, wherein one control point is set at a top and a bottom of a representative stratum;
• randomly generating a resistivity value for each of the control points and performing interpolation using a B-spline interpolation method;
• generating resistivity-thickness sample data;
• and
• using a central loop device, and substituting generated resistivity-thickness sample data into a one-dimensional (1D) numerical simulation method for transient electromagnetics to calculate corresponding electromagnetic response data;
• and
• wherein inputting the electromagnetic response data to be inverted into the pre-trained inversion network for the inversion comprises:
• inputting the electromagnetic response data to be inverted into the encoder part, and extracting features of the input electromagnetic response through three convolutional layers, wherein each convolutional layer comprises a 1D convolutional layer+a pooling layer+a hyperbolic tangent function (Tanh) activation function;
• in the decoder part, restoring and reprocessing features obtained by the encoder part through three feature upsampling layers, wherein the feature upsampling layer comprises one 1D upsampling layer and two 1D convolutional layers;
• and
• in the smooth constraint part, performing a further smoothing constraint to the decoder part by using a convolutional layer with a 1×1 convolution kernel, and finally outputting the inverted resistivity values;
• inputting the inverted resistivity values into a pre-trained forward network for forward calculation to obtain forward electromagnetic response data;
• updating parameters of the inversion network by minimizing a loss between output forward electromagnetic response data and input electromagnetic response data to be inverted;
• and
• using an inversion network with updated parameters to re-invert the electromagnetic response data to be inverted until a preset number of iterations is reached, and finally outputting inverted resistivity values.