Predicting the thermal conductivity of soils using integrated approach of ANN and PSO with adaptive and time-varying acceleration coefficients

This study aims to propose hybrid adaptive neuro swarm intelligence (HANSI) techniques for predicting the thermal conductivity of unsaturated soils. The novel contribution is made by integrating artificial neural networks (ANNs) and particle swarm optimisation (PSO) with adaptive and time-varying acceleration coefficients. Two HANSI techniques, namely ANN-IPSO (ANN optimised with improved PSO) and ANN-APSO (ANN optimised with adaptive PSO) were constructed. To train and validate the proposed models, a dataset of 257 measurements of the thermal conductivity of soils featuring 6 influencing factors was used. The proposed ANN-APSO model has the best prediction performance in both the training and testing stages, according to the results. Also, during the validation phase, the suggested ANN-APSO model attained the most accurate prediction with Adj. R2 = 0.9265, R2 = 0.9442, and RMSE = 0.0515. These results are significantly better than those obtained from other hybrid ANN models constructed with standard PSO, IPSO, new self-organizing hierarchical PSO, modified new self-organizing hierarchical PSO, genetic algorithm, biogeography-based optimisation, and firefly algorithm. Based on the experimental results, the newly constructed HANSI models, especially ANN-APSO can be considered as an alternate solution for predicting real-time engineering problems, including the thermal conductivity of unsaturated soils.