Rock burst chaotic prediction on multivariate time series and LSSVR

State variables reconstructed by multivariate time series were used as LSSVR model inputs to predict the future value of rock burst monitor variables. First, the chaotic prediction principle on multivariate reconstruction and LSSVR was given. Then given the effects of reconstruction parameters on reconstruction results and LSSVR parameters on prediction error, genetic algorithm was adopted to determine reconstruction and LSSVR parameters simultaneously to ensure chaotic prediction accuracy. Finally, in Matlab2009b environment, based on the effectiveness verify of our method by Lorenz chaos system, Microseism time series were simulated to predict rock burst. The results show that the rock burst prediction method on multivariate time series reconstruction and LSSVR can accurately predict monitoring variables in advance to forecast rock burst even in the case of relatively short history data.