Intelligent identification for subgrade disease based on multi-source data

Abstract

Subgrade disease reduces the load-bearing capacity and service life of the road, increasing safety risks and maintenance costs. To address this issue, a novel method for identifying subgrade diseases using surface dynamic load recognition has been proposed. This method employs the iTransformer-LSTM algorithm, which incorporates an attention mechanism to regress soil pressure within the road structure based on surface dynamic load data. Subsequently, the classified soil pressure data obtained through an elastic neural network enables the identification of specific types of subgrade diseases, thereby achieving the objective of recognizingsubgrade disease types directly from surface dynamic load data. This study enhances the regression and classification capabilities for subgrade diseases by integrating the iTransformer-LSTM regression module based on attention mechanisms and an elastic neural network classification algorithm. The results indicate that the model effectively regresses the distribution and developmental trends of soil pressure data based on surface load and displacement data, with a regression accuracy exceeding 95%. Furthermore, the predicted soil pressure data is classified with an accuracy of over 96%, enabling the identification of specific subgrade disease types. Simultaneously, this algorithm allows for the rapid and accurate recognition of subgrade disease types. This research presents a new approach for the prevention and treatment of subgrade diseases, which has significant implications for the sustainable development of road infrastructure.