Weights and biases of ANN trained by hybrid of EHO and LM-BP

Published: 11-03-2019| Version 2 | DOI: 10.17632/dn3v95wcdr.2
Prashanth Janardhan,
Harish Narayana,
Sukomal Mandal


Even though ANN trained by Levenberg-Marquardt Back-Propagation (LM-BP) provided good results, it was observed that the performance would fluctuate on each trail and also gives poor performance at times, due to getting stuck in local optima. Elastic Modulus being a very sensitive parameter of structural concrete, it would be risky to adopt such a model. Here, ANN is trained by swarm-based optimization technique namely, Elephant Herding Optimization (EHO). Further, the weights and biases are exported and fed as pre-defined weights and biases for ANN trained by LM-BP. This takes advantage of complimentary features of either technique and improvises results. It was also observed that the performance was consistent over various trails and provides a good alternative for predicting elastic modulus of recycled aggregate concrete. The various weights and biases of ANN trained by hybrid of EHO and LM-BP as they pass through the network with two Hidden layers are given in the dataset.