Application of the Improved RBFNN Based on DPC in Monthly Rainfall Forecasting

Abstract

Generally, it is still a challenge to determine the basis function center of the Radial Basis Function Neural Network (RBFNN). In order to address this problem, an improved RBFNN prediction method based on Density Peak Clustering (DPC) is proposed in this paper. In this approach, we first use cosine similarity to compute distances between different points. Then, by considering both of the density peak and distance factors, the errors neighbor method is introduced to automatically identify the data center value and the clustering number, which will serve as the initial parameters of RBFNN and the hidden layer nodes number of the RBFNN, respectively. Finally, we use gradient descent method to optimize the RBFNN’s structure and its various parameters to establish the monthly rainfall forecasting model. Compared with several other models, e.g., Back Propagation Neural Network (BPNN), the results show that the proposed model has gained higher prediction accuracy and stability.