Prediction of Rock Strain Using Hybrid Approach of Ann and Optimization Algorithms

Prediction of strain is one of the important factors for assessment of characteristics of rock material. Rock strata are mostly more brittle in nature. Most of the natural ground vibrations are caused due to release of the strain energy stored in the rock strata. The strain governs the development of crack pattern and failure of rock strata. This sudden failure of rock strata may lead to catastrophic disaster. This paper investigates to develop a new prediction approach of strain in rock sample using Artificial Neural network (ANN) with optimization techniques, namely Equilibrium Optimizer (EO), Differential Evolution (DE), Grasshopper Optimization Algorithm (GOA) and Whale Optimization Algorithm (WOA). For this purpose, a total of 3000 data with input and output data are collected from experimental investigations carried out in the laboratory. In this study, the strains are predicted in lateral and longitudinal dimension of the rock sample. The accuracy of the proposed prediction models, ANN-EO, ANN-DE, ANN-GOA and ANN-WOA, is studied by using different statistical indices. Furthermore, the proposed models are assessed for selecting the robust model for accurate prediction of rock sample by using boxplot, accuracy matrix, Taylor diagram, rank analysis, Objective Function (OBJ) Criterion, Akaike Information Criterion (AIC), Developed Discrepancy Ratio (DDR) criterion, external validation, and uncertainty analysis. The results show that the ANN-EO model outperforms other proposed models in the training and testing stages. The OBJ of ANN-EO and ANN-DE for lateral strain modelling are 0.065 and 0.064 respectively, while, for longitudinal strain modelling, the ANN-EO and ANN-WOA work best with OBJ of 0.037 and 0.075 respectively. Hence, ANN-EO can be proposed as a robust model when compared to other model for prediction of strain in rock sample.