Time consuming numerical model calibration using Genetic Algorithm (GA), 1-Nearest Neighbor (1NN) classifier and Principal Component Analysis (PCA)

Single Objective Genetic Algorithm (SGA) optimization process usually needs a large number of objective function evaluations before converging towards global optimum or a near-optimum. The SGA is used as automatic calibration method for a wide range of numerical models. However, the evaluation of the quality of solutions is very time-consuming in many real-world numerical model calibration problems. The algorithm SGA-1NN-PCA, an effective and efficient dynamic approximation model to reduce the number of actual fitness evaluations, is presented in this paper. Training data of INN classifier are produced from early generations. 1-Nearest Neighbor (INN) classifier is used to predict objective function values for evaluations. Principal Component Analysis (PCA) linearly transforms high-dimensional optimization parameters into low-dimensional optimization parameters to save test time for INN. The test results show that the proposed method only requires about 25 percent of actual fitness evaluations of the SGA.