The invention relates to a
lithology identification method for a
hydrate reservoir in a frozen soil region. The method comprises the following steps: collecting
hydrate drilling
well logging data in a frozen soil region, and establishing a sample set; performing windowing
processing, standardized preprocessing and
lithology label coding on the sample set to form a sequence sample suitable for
deep learning; constructing a
hybrid neural
network model combining a
convolutional neural network and a long-
short term memory network (CNN-LSTM), extracting local spatial features in the
logging data by using a convolutional layer, and extracting
time sequence features of lithologic evolution by combining an LSTM layer; key hyper-parameters of the model are globally optimized based on a
Bayesian optimization algorithm, the number of
convolution kernels, the number of LSTM units, the Dropout proportion and the learning rate are included, and the classification performance and generalization ability of the model are improved; and utilizing the trained optimal CNN-LSTM model to carry out
lithology automatic identification and classification on the frozen soil region
hydrate reservoir. According to the method, through combined extraction of
convolution and
time sequence features, the recognition precision and robustness of complex lithology of the hydrate reservoir in the
permafrost region are effectively enhanced, meanwhile, the
Bayesian optimization method is introduced to improve the
model parameter adjustment efficiency and stability, and reliable
technical support can be provided for fine description and resource evaluation of the hydrate reservoir.