Compressive Strength Prediction and Mix Ratio Optimization Design of Oil Palm Shell Concrete Based on Hybrid Intelligent Algorithm

Abstract This study innovatively proposes a hybrid intelligent model based on the seahorse optimizer (SHO), eXtreme gradient boosting (XGBoost), and gated recurrent unit (GRU) for the high-precision prediction of the compressive strength of oil palm shell (OPS) concrete. Unlike traditional methods, this study employs XGBoost to construct new features and combines it with GRU to achieve deep learning of nonlinear characteristics. Additionally, SHO is applied for the first time to optimize model parameters, significantly enhancing performance, with the model achieving R 2 = 0.9839 , root mean square error ( RMSE ) = 1.3948 MPa , and mean absolute error ( MAE ) = 1.1956 MPa on the test set. To further improve the practical value of OPS concrete, this study adopts the NSGA-II algorithm to perform mix ratio optimization under dual objectives of compressive strength and cost, generating a Pareto front that balances performance and economy. The optimized mix ratio reduced costs by up to 10.9% and improved compressive strength by up to 2.7% compared to the initial scheme. This study is the first to integrate the SHO-XGBoost-GRU model with NSGA-II for the prediction and optimization design of OPS concrete, providing an innovative method and theoretical support for the performance prediction and mix ratio optimization of green, low-cost concrete.