Reservoir parameters are important for reservoir evaluation. Aiming at the difficulties of the traditional reservoir parameters prediction method to get rid of the constraint of a linear equation and the low prediction accuracy，a model combined with convolutional neural network（CNN）and gated recurrent unit（GRU）is proposed. The model not only has the local perception characteristics of CNN but also has the long-term memory function of GRU，thus having the ability to express the spatio-temporal features of data. The CNN-GRU porosity prediction model is established based on the well logging data of Well A to predict the porosity of unknown depth segment in this well area，and to further propose a variable learning rate training method. Compared with CNN or GRU models，experimental results show that CNN-GRU model can extract data features more effectively and can improve the reservoir parameters prediction accuracy，which provides a new idea to predict reservoir parameters.