The invention discloses a well-to-seismic joint initial lithologic model construction method based on deep learning, is applied to the field of three-dimensional geological modeling, and aims to solvethe problems that in the prior art, the frequency of well logging data is too low due to filtering, a lot of high-frequency effective information is lost, and seismic data cannot be effectively controlled in the interpolation process. According to the invention, the convolutional neural network is used to extract characteristics of long and short periods, namely high and low frequencies, contained in data; different features are classified and learned by adopting a long-term and short-term memory network, so that the relationship between seismic data and logging data is learned accurately, accurate prediction of rock attributes is achieved, a lithology initial model is constructed, a basis is provided for inversion of lithology parameters, and exploration and development of oil and gas and reservoir description of oil and gas reservoirs are guided.