Rockburst prediction based on multivariate time series reconstruction and GRNN

Given chaotic characteristics of rockburst data, the state variables reconstructed by multivariate time series were taken as prediction model input to predict the variables of monitoring rockburst, where generalized regression neural network (GRNN) was adopted as prediction model. According to reconstruction parameters computed through mutual information method and false nearest neighbor method, phase space is reconstructed by multivariate time series to overcome noise´s influence on the data of monitoring rockburst. In view of the limited sample, chaotic prediction using GRNN model that the smoothing parameter is selected by holdout method. Finally, two examples, electromagnetic radiation and microseismic time series, were simulated in MATLAB2010a environments. The results show that our prediction method can fast and accurately predict monitoring variables.