回帰とRBFを利用した近似最適化

概要

論文の詳細を見る
In our previous studies, we proposed approximate optimization system using RBF network as approximation tool and Genetic Range Genetic Algorithms as optimizer, we showed that it provides necessary data efficiently regardless of whether constrained or unconstrained and we can get high-precision approximation in small number of function call. Although, it is often pointed out that RBF network is effective in approximating complicated response, but inadequate to approximate simple response, especially one has linearity. We will propose approximation that combine RBF network and linear regression or polynomial regression, and use it in our optimization system. In this method, we make the regression model to get rough tendency of the response, and RBF network learns the difference between the response and the regression model. When the response has strong linearity, the regression model fits the response and RBF network value become near to zero, thereby we will get good approximation of the response. In addition, we can predict the response that is in no RBF bases region and it will work well to find next searching points. We will use proposed method to benchmark problems and show the effectiveness of proposed method
社団法人日本機械学会の論文
2004-12-08