基于多变量混沌时间序列的冲击地压预测

考虑到冲击地压的混沌特征及其监测数据含噪且长度有限，基于多变量时间序列重构和GRNN模型来预测冲击地压监测变量。给出了多变量时间序列相空间重构理论和GRNN混沌预测原理，并提出采用遗传算法同时确定最佳重构参数和GRNN的光滑因子以保证预测精度。在Matlab2010a仿真环境下，将本文方法用于Lorenz系统以验证对含噪且长度有限的混沌序列的适用性，最后对微震能量和电磁辐射两类数据进行预测研究。结果表明：即使历史数据有限，多变量混沌序列预测方法也能提前预测出多个监测变量，从而实现冲击地压预报。

Rock burst prediction on multivariate chaotic time series

Given rock burst chaotic characteristics and its limited length monitor data containing noise，the multiple rock burst monitor variants were predicted on multivariate time series reconstruction and generalized regression neural network (GRNN).The theories of multivariate phase space reconstruction and GRNN prediction were introduced，and the method was proposed that adopting genetic algorithm to simultaneously determine reconstruction parameters and GRNN smoothing parameter，to ensure prediction precision.In Matlab2010a environments，the method was simulated on Lorenz system to verify its effectiveness for limited length multivariate series containing noise.Finally the method was used to microseism energy and electromagnetic radiation signal monitor data，and the results show that the prediction method on multivariate chaotic series can predict multiple monitor variants and therefore forecast rock burst even in the case of relatively limited history data.