Optimisation study of carbon dioxide geological storage sites based on GIS and machine learning algorithms

Abstract Comparison is a crucial stage of site-level selection process. This study integrates the geographic information system (GIS) techniques and analyses the stability of predictions based on five machine learning models to identify key indices for site selection. The study results reveal that: (1) the relevant site selection index system was improved. The precision of predictions using the five machine learning models all reached 95%, with the deep neural networks (DNN) model achieving the highest precision at 96.4%, indicating its broader applicability for site selection. (2) A machine learning index optimisation process is proposed. Based on the results of index importance, indices are categorised as important, less important, and general. Using only the important indices yields satisfactory evaluation results. (3) A rapid assessment model was developed. In the DNN model, the results could be predicted more accurately by using approximately 25% of the data and 50% of the indices. This provides a reference for subsequent site selection for difficult-to-obtain data. This study aims to accumulate extensive data via future research to establish a model database. The database will help refine geological models for different types and stages of engineering projects and incorporate more site-specific models. The ultimate goal is to provide more convenient theoretical guidance and recommendations for subsequent site selection processes.