Multivariate Time Series Correlation Extract and Prediction Based on Cluster

Based on general cluster method, this paper introduces a improved multivariate reduce dimension method. This method improves on the aspect of similar extend of multivariate variables, classifies variables more reasonable. Based on result of this classification, and uses it as the input of neural network, implement prediction for multivariate time series. Applying the method in this paper predict runoff yield of Yellow river, simulation results shows that compared to the prediction based on general cluster method, the method introduced in this paper behaves more effectively.