Self-Compacting Concrete (SCC) Strength Prediction via Optimized Duple-Deep-Learning Model and Distance Ranked Fire-Hawk Optimizer (DRFO)

Abstract

Despite the years of widespread usage of self-compacting concrete in construction, there is still no reliable quantitative approach that can accurately forecast its strength. This restriction results from the very nonlinear relationship between the “compressive strength (CS) of the SCC (Self-Compacting Concrete)” and the mixed materials. In this research work, a novel SCC prediction model is developed using the new Optimised Duple--Deep Learning Model. Data gathering, pre-processing, feature extraction, feature fusion, and strength prediction are the five primary steps of the suggested model. The min-max normalization method is initially used to normalize the obtained data. Higher-order statistical features (Variance & skewness), Statistical features (Min-Max, Mean, Median, and Standard Deviation), and Pearson's correlation coefficient-based features are then derived from the normalized data. The features that were extracted are concatenated. A new duple-deep-learning model is developed using the fused features. Bidirectional long short-term memory (Bi-LSTM) and an improved Convolutional Neural Network (CNN) will both be included in the dual deep learning model. The novel Distance Ranked Fire-Hawk Optimizer (DRFO) is employed to fine-tune the activation function of the dual-deep learning model to increase prediction accuracy. This DRFO model conceptually enhances the baseline Fire Hawk Optimizer (FHO). The dual-deep learning model provides the ultimate decision on the CS of SCC. The suggested model outperforms existing models based on the MATLAB results in terms of performance analysis % values such as Mean Square Error (MSE-0.282), Root Mean Square Error (RMSE-1.325), Mean Absolute Error (MAE-0.334), Mean Absolute Percentage Error (MAPE-0.112), Normalized Mean Square Error (NMSE-0.00), and correlation coefficient (0.999).