Evolutionary Neural Network-based Method for Constructing Surrogate Model with Small Scattered Dataset and Monotonicity Experience

Engineering design can be regarded as an iterative optimization process. This process is difficult because computer aided engineering (CAE) is time-consuming. In the research community, a surrogate model is proposed to deal with this problem. This work is an initial attempt to develop a method for building a surrogate model with only a small dataset and design experience, which is very common in practical scenarios. The basic idea is to integrate the small dataset with expert experience by an evolutionary neural network. Following this idea, the method simply compiles a neural network as a vector and takes it as individual. Expert experience is taken as fitness function of the evolutionary algorithm. Three groups of experiences are conducted to validate the proposed methods. The experimental results imply expert experience can be fused into the surrogate model. Besides, the incorporation of expert experience has potential to decrease the generalization error of the surrogate model, and the model capacity is an important meta-parameter that should be carefully decided.