A novel neuro genetic programming framework for modelling compressive strength of recycled aggregate concrete

Abstract

Recycled Aggregate Concrete (RAC) can represent as a sustainable alternative to conventional concrete; however compressive strength prediction of RAC is difficult due to the complex and nonlinear interactions between the materials in the mix. Researchers have been utilizing soft computing tools like: Genetic Programming (GP), Artificial Neural Network (ANN) etc. for predicting compressive strength, however the performance of these models is satisfactory. To leverage the characteristic of evolutionary optimization in Multi Gene Genetic Programming (MGGP): variant of GP and the effective nonlinear input and output mapping through ANN, the present study introduces a Novel Neuro Genetic Programming framework (NMGGP), an advanced hybrid approach that combines the strengths of MGGP and ANN. The NMGGP framework is a framework devised in three phases: (i) MGGP constructs a multigene model, where each gene represents a weighted mathematical expression derived from input parameters; (ii) High-weighted genes, are selected, and transformed into transformed inputs to enhance the learning process. (iii) These transformed inputs, along with the original output, are fed into an ANN enabling advanced nonlinear mapping and precision in prediction. The proposed model was tested for predicting the 28-day compressive strength of RAC which considered 8 inputs as the kg/m³ quantity of concrete mix. Further three highly weighted genes were transformed into transformed inputs and predictions were done in phase III. The results displayed an excellent model with r=0.967, RMSE as 4.744 and MAE as 3.944 as compared to the stand-alone models of MGGP and ANN with r=0.930, RMSE more than 5.0 and MAE mode than 5.0. The results also can be said to be robust through a balanced scatter plot and graphical comparison between predicted and observed 28-day strength of RAC.