S100: acquiring basic data of a target well location as well as basic data and a logging-while-drilling (LWD) resistivity (Rt) of an adjacent well, and dividing the basic data of the target well location and the basic data of the adjacent well into lithological group data, porosity group data, drilling group data, and logging group data, wherein the lithological group data comprises: neutron logging (NEU), photoelectric factor (PEF), and gamma ray (GR); the porosity group data comprises: acoustic (AC), compensated neutron log (CNL), and density (DEN); the drilling group data comprises drilling rate, rotary speed, and bit pressure; and the logging group data comprises content of methane (C1), content of ethane (C2), mud density in all (MDIA), and formation pore pressure graduation (FPPG);

S200: calculating a first correlation value between each type of the basic data of the adjacent well and an LWD Rt curve of the adjacent well; and retaining data with a maximum first correlation value in each of the lithological group data, the porosity group data, the drilling group data, and the logging group data, to acquire a lithological group significant parameter, a porosity group significant parameter, a drilling group significant parameter, and a logging group significant parameter;

S300: eliminating, through an isolated forest algorithm, outliers of the lithological group significant parameter, the porosity group significant parameter, the drilling group significant parameter, and the logging group significant parameter to acquire standardized significant parameters;

S400: selecting, based on the standardized significant parameters, the lithological group significant parameter for geological stratification to acquire a number of depth units between stratigraphic abrupt positions, and acquiring a two-dimensional input feature map based on the number of the depth units, the standardized significant parameters, and the first correlation value;

wherein, a method for acquiring the depth units comprises:

setting ten data points (x_g1, x_g2 . . . x_g10) as one depth unit based on the lithological group data of the basic data of the target well location;

calculating an average value x_lit of the lithological group data of each depth unit:

wherein, g denotes a group number;

calculating a difference d_u between average values x_lit of the lithological group data of adjacent depth units:

setting, when d_u>10, a lower depth unit of corresponding adjacent depth units as a lithological abrupt position, and calculating a number h_t of depth units between each two abrupt positions to represent a stratum thickness, wherein u denotes a serial number of the depth unit;

taking a product of the standardized significant parameters and the first correlation value as a weight parameter Z_cate=S_im1cateC_cate, wherein cate denotes a category; and

constructing the two-dimensional input feature map Z_cate×4×h_t by taking the number h_t of the depth units as a stride;

S500: acquiring, based on the two-dimensional input feature map, an LWD Rt prediction curve through a trained LWD Rt prediction model; and

wherein acquiring the LWD Rt prediction curve through the trained LWD Rt prediction model comprises:

forming the LWD Rt prediction model, comprising a t-channel image recognition network with 2t convolutional layers and 2t average pooling layers, wherein each channel comprises a first convolutional layer, a first average pooling layer, a second convolutional layer, and a second average pooling layer that are sequentially connected; each convolutional layer has a different size; an f-th channel comprises a (4×f−1)× (4×f−1) first convolutional layer, a (4×f+4)×(4×f+4) second convolutional layer, and 2×2 pooling layers; and all channels are connected together to one fully connected layer and one naive Bayesian decision maker;

denoting the first convolutional layer of a first channel, the first convolutional layer of a second channel, and the first convolutional layer of a third channel as a C1 layer, a C3 layer, and a C5 layer, respectively, wherein the C1 layer is configured to convolve an input image through 8 h_t×h_t convolution kernels; the C3 layer is configured to convolve the input image through 8 3h_t×3h_t convolution kernels; and the C5 layer is configured to convolve the input image through 8 5h_t×5h_t convolution kernels;

wherein, con_{p, q}^l denotes a convolution result at a position (p, q); l denotes a current layer number; CON denotes a matrix covered by convolution kernels; L and W denote a length and a width of the matrix covered by the convolution kernels, respectively; m1 and m2 respectively denote a serial number of a length and a serial number of a width of the convolution kernels, ranging from 1 to L; ker denotes a kernel function; and b denotes a corresponding bias term;

fitting the convolution result through a rectified linear unit (ReLU) function to acquire a fitted convolution result;

performing upsampling pooling on the fitted convolution result:

wherein, con_{m3, m4}^l denotes a two-dimensional matrix representation of a pooled convolution result; and m3, m4 denote a convolution result at a position (m3, m4);

converting the pooled convolution result into a tiled one-dimensional feature vector through a tiling layer;

integrating the tiled one-dimensional feature vector through the fully connected layer:

wherein, con_key^l denotes a one-dimensional matrix representation of a feature after the integration through the fully connected layer; I denotes a current layer number; key denotes an index value of a one-dimensional matrix; R denotes a length of a feature vector in an (l−1)-th layer; w denotes a weight matrix; b denotes the corresponding bias term; and r denotes a serial number of a data point of the feature vector;

introducing the naive Bayesian decision maker into the fully connected layer; and

taking the integrated tiled one-dimensional feature vector as the LWD Rt prediction curve; and

S600: constructing sliding units based on the depth units; traversing an initial stratigraphic framework model through the sliding units; calculating a hydrocarbon parameter in a window based on the LWD Rt prediction curve; and locating an area with a hydrocarbon potential based on the hydrocarbon parameter at each position, thereby acquiring a formation physical property evaluation result.