2-211 予測・診断モデル構造決定のためのクラスタリング適用の検討

概要

論文の詳細を見る
Artificial neural network techniques have been applied for various forecasting and diagnosis problems because of high ability for nonlinear model construction. Generally, it is difficult to obtain the accurate forecasting results using only one neural network model. However, it is also difficult to decide the configuration of appropriate forecasting models. This paper proposes a forecasting model construction method using clustering methods. Initial clusters are obtained using a general hierarchical clustering method or k-means method. Input factors for clustering are selected using separation degree for clusters. A forecasting model is constructed for each cluster using a neural network separately. The clusters are adjusted by merging operation in order to improve correlations between input and output factor of forecasting models. Forecasting models are re-constructed for the adjusted clusters. Consequently, appropriate forecasting models with best forecasting accuracy for all cluster are decided. The simulation results for an electric load forecasting problem reveal the effectiveness of the proposed method.
一般社団法人日本機械学会の論文
2002-11-14