Study on the spatial distribution and advanced prediction of dolomite sandification grades using multiple detection methods

The problem of dolomite sandification has gradually become one of the key factors affecting tunnel construction safety. In practical engineering, accurately predicting the dolomite sandification grades and their spatial distribution, especially during the advanced forecasting stage of tunnel construction, has become a crucial technical issue for improving construction safety, optimizing design parameters, and enhancing support efficiency. This study first establishes a systematic multi-parameter index system for dolomite sandification grade classification. By integrating preliminary excavation indicators, structural characteristics of the rock mass, and quantitative physical–mechanical property indices, a comprehensive method for sandification grade classification from qualitative to quantitative is proposed. The fuzzy analytic hierarchy process and membership degree calculation methods are employed to eliminate the interference of subjective human factors, thereby improving the scientific and objective nature of sandification grade classification. Next, this study designs a multi-source data fusion framework for predicting sandification grades, incorporating multiple detection methods, such as seismic waves, transient electromagnetic methods, geological radar methods, and advanced drilling methods. By combining deep learning techniques, a high-precision spatial prediction model for sandification grades is constructed. The prediction results of different detection methods are fused using weighted averaging and ensemble learning techniques, which enhance the model's stability and accuracy. This is particularly effective in complex dolomite strata, where the spatial distribution patterns of sandification zones can be effectively identified. Finally, to ensure the scientific and practical nature of the predictions, this paper proposes and implements a “prediction–validation–feedback” closed-loop system. In the prediction phase, the multi-source data fusion framework is used to predict the spatial distribution of sandification grades. In the validation phase, the accuracy of the prediction is verified by comparing the lithological exposure after excavation in actual engineering projects. In the feedback phase, based on actual excavation results, model parameters are optimized, and the weights of different detection methods are adjusted, thus enabling dynamic improvements in prediction accuracy and dynamic optimization of support design. Through the application of this closed-loop system, dynamic optimization of sandification grade prediction was successfully achieved in the Yuxi section of the Yunnan Central Water Diversion Project, significantly improving the prediction accuracy.