An Uncertainty Estimation Method of Support Vector Machine Surrogate Model Assisting for Expensive Optimization

In data-driven evolutionary optimization, surrogate model is usually used to replace expensive physical experiments or simulations to reduce the evaluation cost. When selecting promising individuals for exact evaluation, those with a large degree of uncertainty in approximation also need to be taken into account, which can strengthen the exploration ability of algorithms and improve the accuracy of surrogates. Support vector machine (SVM) is suitable for expensive small sample training set. However, it can only predict the mean value without the uncertainty. In order to solve this problem, the zero mean Gaussian random process is used to measure the uncertainty of support vector machine. The method’s efficiency is demonstrated on a set of benchmark problems. The results show that the uncertainty measurement method used in this paper improves the exploration ability of the evolutionary algorithm and obtains a better solution.