A fast converging evolutionary neural network for the prediction of uplift capacity of suction caissons

Evolutionary neural networks (ENN) with genetic algorithm (GA) based weight determination have lead to better performance aspects such as fast convergence or reduced learning error of the network, or better solutions when compared to their specific counterparts. In this investigation, we propose a scheme of GA based weight determination employing a genetic inheritance operator termed short term reproduction expectancy (STRE) which when embedded in the scheme results in a better performing ENN. The ENN employing the STRE scheme of inheritance (NEU_GEN(STRE)) has been implemented and its performance compared with the same employing a generational GA for its weight determination (NEU_GEN). The performance analysis has been demonstrated on the problem of prediction of uplift capacity of suction caissons in the field of geotechnical engineering.