Prediction of high-performance concrete compressive strength through novel structured neural network

The difficulties in determining the compressive strength of concrete are inherited due to the various nonlinearities rooted in the mix designs. These difficulties raise dramatically considering the modern mix designs of high-performance concrete. Presents study tries to define a simple approach to link the input ingredients of concrete with the resulted compressive with a high accuracy rate and overcome the existing nonlinearity. For this purpose, the radial base function is defined to carry out the modeling process. The optimal results were obtained by determining the optimal structure of radial base function neural networks. This task was handled well with two precise optimization algorithms, namely Henry's gas solubility algorithm and particle swarm optimization algorithm. The results defined both models' best performance earned in the training section. Considering the root mean square error values, the best value stood at 2.5629 for the radial base neural network optimized by Henry's gas solubility algorithm, whereas the same value for the the radial base neural network optimized by particle swarm optimization was 2.6583 although both hybrid models provided acceptable output results, the radial base neural network optimized by Henry's gas solubility algorithm showed higher accuracy in predicting high performance concrete compressive strength.