Euclidean distance and second derivative based widths optimization of radial basis function neural networks

摘要: The design of radial basis function widths Radial Basis Function Neural Network (RBFNN) is thoroughly studied in this paper. Firstly, the influence on performance RBFNN illustrated with three simple approximation experiments. Based conclusions drawn from experiments, we find that two key factors including spatial distribution training data set and nonlinearity should be considered width design. We propose to use Euclidean distances between center nodes second derivative measure these respectively. Secondly, a step method proposed based information about aforementioned obtained comprehensive analysis given set. In first features are analyzed according points, each node estimated finite difference method. an initial heuristic equation. optimization techniques used optimize which can effectively optimum good baseline. Thirdly, one mathematical example taken verify efficiency method, followed by conclusions.