A Prediction Model for Loess Collapse Coefficient Based on Genetic Algorithm Back Propagation Neural Network

As an important property of loess, the collapsibility coefficient is commonly used in engineering to evaluate the collapsibility of loess. The main methods for predicting the collapsibility coefficient of loess currently include fuzzy algorithm, principal component analysis, data mining, etc. Due to the numerous indicators that affect the collapsibility of loess and their mutual influence, current research has problems such as incomplete consideration of indicators or insufficient data used to fit prediction models, resulting in insufficient prediction accuracy. Using factor analysis method, factor analysis is conducted on the selected indicator data based on the theory of collapsible deformation structure. Water content, porosity, and plasticity indicators are selected as independent main factors, and a Back Propagation (BP) neural network prediction model is constructed to predict the collapsible coefficient. Genetic algorithm is used to optimize the initial network parameters of the model. Compared to the BP neural network prediction model, the model optimized by genetic algorithm has varying degrees of improvement in accuracy, high prediction accuracy, and better practicality in engineering.