A PSO-CNN-LSTM Model for Seismic Facies Analysis: Methodology and Applications

Seismic facies analysis, as a crucial step in the study of depositional facies, effectively delineates the distribution patterns of depositional facies between wells. To address the limitations of conventional manual interpretation methods, particularly their low efficiency and strong subjectivity, this study proposes a hybrid CNN-LSTM model integrated with Particle Swarm Optimization (PSO-CNN-LSTM). The model systematically extracts spatial features of seismic reflections through CNN architecture while capturing temporal waveform dependencies via LSTM networks, with PSO automatically optimizing critical parameters including initial learning rate and LSTM neuron count. Experimental results demonstrate that PSO-CNN-LSTM achieves a classification accuracy of 89.74%, surpassing CNN (81.48%), LSTM (81.63%), and basic CNN-LSTM (84.61%) models by 8.26%, 8.11%, and 5.13% respectively. The model exhibits superior performance on the SEG 2020 benchmark dataset, confirming that automated parameter optimization effectively reduces manual intervention while enhancing convergence stability. Practical applications reveal consistent interpretation outcomes between the model’s predictions (using limited training samples) and expert analyses, providing reliable evidence for identifying favorable zones in heterogeneous carbonate reservoirs. The established intelligent waveform classification workflow validates PSO-CNN-LSTM model’s robustness and offers an efficient solution for seismic facies analysis, particularly in complex geological settings.