Evolutionary Engineering Optimization Using Recursive Regional Neural Network and Genetic Algorithm

This study presents a soft computing based evolutionary optimization for engineering applications with the constraint of sample size. Existing field data or experimental designs are often applied as training samples to establish a simulated network model for the engineering system following by an optimum search. However, possible biased distribution of field data and scarce samples from QA experiments might compromise modeling generality. The proposed methodology defines the Reliable Radius to confine the genetic algorithm search in the hyper-spheres surrounding the training samples for a reliable quasi-optimum. The verification of the optimum is added to the learning samples to retrain the regional network model that evolves intelligently according to the prediction accuracy using a fuzzy inference. Instead of a dense sample distribution to increase global accuracy, the design iteration will provide additional samples in the most probable regions of the optimum, and thus increase sampling efficiency.