Predictive Analysis of the Mechanical Properties of Biopolymer-Fiber-Reinforced Composite-Stabilized Soil Based on Genetic Algorithm-Optimized Back Propagation Neural Networks.

Abstract

The limitations imposed by the inherent complexity of multi-component composition ratios in biological polymer-stabilized soils have hindered rapid and accurate performance prediction. To enhance the predictive accuracy for biopolymer-fiber-stabilized soils, an optimized GA-driven backpropagation (BP) neural network was developed. Three key factors influencing mechanical strength (guar gum (GG), xanthan gum (XG), and polybutylene succinate (PBS)) were identified. The global optimization capability of GA was utilized to construct an integrated GA-BP model, with these factors serving as inputs and 7d compressive strength as the output. Support vector machine (SVM) was also incorporated to provide a benchmark comparison of predictive performance. Validation was performed using 80% of the dataset, with the remaining 20% used for testing. The optimal biopolymer dosage was found to be within the range of 0.5% to 1.0%, and the maximum 7d compressive strength achieved was 466.67 kPa at the 0.5% XG-0.5% GG combination, representing a 273% increase over untreated soil. The GA-BP model demonstrated superior performance in terms of prediction accuracy and stability, as indicated by an R² of 0.887-significantly higher than those of the BP (0.714) and SVM (0.554) models. The mean squared error was substantially reduced to 1413, compared to 2130 and 3113 for BP and SVM, respectively. Although MAPE approached those of the GA-BP, the overall predictive efficacy of SVM was found to be inferior. A reliable and robust methodology for forecasting the mechanical behavior of stabilized soils has thus been provided by this model, supporting advanced applications within geotechnical engineering.