基于CT-GRNN模型的采场顶板位移预测

将混沌学理论与广义神经网络相结合构建了基于CT-GRNN的采场顶板位移预测模型。首先应用Matlab混沌工具箱, 对顶板位移数据进行混沌判别, 得出顶板位移数据混沌时间序列的特点, 进而对顶板位移数据进行相空间重构, 最后采用广义回归神经网络对采场顶板位移进行预测。以新桥矿E15、E27采场顶板位移预测为例, CT-GRNN模型的预测误差分别为2.1%和2.6%, 相比传统BP神经网络预测（预测误差分别为5.7%和4.8%）, 精度得到大幅度提高, 可作为采场顶板位移预测的一种新手段。

Prediction of Stope Roof Displacement Based on CT-GRNN

The chaos theory combined with GRNN was adopted to establish a model to predict the stope roof displacement. The Matlab toolbox for chaotic system was firstly used for chaos identification of roof displacement data, resulting in the chaotic time order. Then the phase-space reconstruction was used to analyze the data and predict the displacement of stope roof along with GRNN. By applying this new model to predict the roof displacement of E15 and E27 stopes in Xinqiao Mine, it is found that CT-GRNN model has great improvement in prediction accuracy compared with BP Neural Network, with prediction error reduced from 5.7% and 4.8% to 2.1% and 2.6%, respectively. Thus it can be regarded as a new approach for stope roof displacement prediction.