Abstract: Porosity is one of the key indicators to characterize the por structure of shale reservoirs. Quantitative prediction research on porosity is an important step in reservoir evaluation. The accurate value of shale porosity must be obtained through core analysis. How to obtain an accurate prediction of the porosity of the entire well based on a very small amount of coring well data is a significant problem. This paper proposes a new transfer deep neural network model, based on a small amount of core and logging data, to achieve accurate prediction of porosity. Firstly, according to Pearson correlation coefficient method, the logging parameters suitable for the source well deep neural network are selected as the input for the model. Secondly, a new method is proposed to calculate the similarity of the well logging data distribution between the source well and the target well, quantitatively measure the geological difference between two wells. Thridly, retrain the source well prediction network with a small amount of target well logging data similar to the source well logging data distribution, and build a transfer deep neural network for predicting the porosity migration of the target well. The test results of A2 and B2 show that: 1) This method requires only 10% of the data volume, and reaches the performance of an absolute mean error of 0.032 9 and a coefficient of determination of 0.841 6; 2) The proposed method for calculating the similarity of two wells can effectively measure the difference between wells. The more similar the distribution of the source well logging data and the target well logging data, the higher the accuracy of porosity prediction of the transfer learning network. The proposed model can effectively reduce the dependence on logging and core data, and greatly reduce shale gas exploration and development costs.

Key words: shale gas, well logging, shale porosity prediction, difference between wells, deep transfer learning