Annual Runoff Prediction Based on Least Square Support Vector Machines-Markov Chain Combined Model

Abstract

A new annual runoff prediction model was proposed by combining the least squares support vector machines and markov chains named LSSVM-MC. Firstly, adopt the least squares support vector machines to predict annual runoff as the first step prediction. Then regarding the series of prediction errors as a process of Markov chain, adopt Markov chain method to predict the potential error as the second step prediction. At last, subtract the two-step prediction results to get the final prediction value. The case study shows that the combined model effectively improves the annual runoff prediction accuracy.