Study of an SSA-BP Neural Network-Based Strength Prediction Model for Slag–Cement-Stabilized Soil

• Published in: Materials Vol. 18; no. 15; p. 3520
• Main Authors: Zhang, Bei, Tao, Xingyu, Zhang, Han, Yu, Jun
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 27.07.2025; MDPI
• Subjects: Algorithms; Back propagation; Back propagation networks; Cement hydration; Composite materials; Compressive strength; Design engineering; Expected values; Industrial wastes; Mechanical properties; Moisture content; Neural networks; Neurons; Optimization; Prediction models; Sandy soils; Search algorithms; Slag; Soil properties; Soil strength; sparrow search algorithm; prediction model; prediction of compressive strength; cement-stabilized soil; SSA-BP neural network
• ISSN: 1996-1944; 1996-1944
• DOI: 10.3390/ma18153520

As an industrial waste, slag powder can be processed and incorporated into cement-based materials as an additive, significantly improving the engineering properties of cement–soil. The strength of slag–cement-stabilized soil is subject to nonlinear interactions among multiple factors, including cement content, slag powder dosage, curing age, and moisture content, forming a complex influence mechanism. To achieve accurate strength prediction and mix proportion optimization for slag–cement-stabilized soil, this study prepared cement-stabilized soil specimens with different slag powder contents using typical sandy soil and clay from the Nantong region, and obtained sample data through unconfined compressive strength tests. A Back Propagation (BP) neural network prediction model was also established. Addressing the limitations of traditional BP neural networks in prediction accuracy caused by random initial weight thresholds and susceptibility to local optima, the sparrow search algorithm (SSA) was introduced to optimize initial network parameters, constructing an SSA-BP model that effectively enhances convergence speed and generalization capability. Research results demonstrated that the SSA-BP model reduced prediction error by 53.4% compared with the traditional BP model, showing superior prediction accuracy and effective characterization of multifactor nonlinear relationships. This study provides theoretical support and an efficient prediction tool for industrial waste recycling and environmentally friendly solidified soil engineering design.