Synthesis Well Logs Generation in a Naturally Fractured Reservoir Using Multi-Layer Perceptron Networks

Well logs which are considered as robust tools for the reservoir description cost a lot in the petroleum industry. The challenges in this process result in missing or incomplete data in some cases. Generating synthesis logs have already been proposed to fix this problem. This study presents a methodology to develop the synthesis logs for a naturally fractured reservoir. In this approach, multi-layer perceptron neural networks are used with available conventional wireline logs data from a naturally fractured oil reservoir to develop the missing or incomplete logs. In this study, three different approaches were used to utilize the available data including depth, Gamma Ray, Resistivity, Density and Sonic logs of five wells for training, testing and verification stages to predict the missed logs. The results showed that the generated synthesis Sonic and Density logs have very good accuracy with 0.93 and 0.92 average R2 values, respectively. The precision of the generated Gamma Ray is satisfactory with 0.82 average R2 value. Furthermore, the average R2 value for the prediction of the Resistivity log is 0.76 and the designed neural network failed to predict the Resistivity log in certain circumstances well. Therefore, care must be taken in this regard.