Trained BPNN Method and Application in Tight Gas Sandstone Formation Lithology classification
Zheng Wang, Yang Gao, Jing Zhang, R. He, Haoan Dong · 2021
Summary Tight gas sandstone formation has been becoming a critical reservoir. Precisely classifying the rock facies boundaries from borehole data is significant step in the reservoir characterization. In this work, BPNN based networks are applied to identify rock lithology from borehole data. The trained BPNN networks show that optimizer selection and parameter determination are two critical factors influencing the effectiveness of BPNN. By qualitatively comparing the predicted results from BPNN with different optimizers and different parameter setup, a well-trained BPNN is proposed to predict the rock facies from borehole logs. In this paper, the five log curves of acoustic (AC), caliper (CAL), density (DEN), gamma ray (GR) and spontaneous potential (SP) were selected as the input of the model. The results indicate that trained BPNN with proper parameters can precisely identify the rock facies.