Application of GRNN in Time Series Prediction for Deformation of Surrounding Rocks in Soft Rock Roadway

During soft rock roadway construction, the deformation of surrounding rocks is a significant factor in stability evaluation. However, the deformation still has long duration, obvious nonlinear effect after the soft rock roadway construction accomplishment. Certainly, this potential stability change will make the maintenance cost increased in the future. We propose a Time Series Prediction model based on Generalized Regression Nerual Network(GRNN)to predict long-term potential deformation trend of surrounding rocks in soft rock roadway. To implement, first training samples which constructed scientifically based on observed local deformation at an interval of 15 days, while the model is trained circularly using MATLAB neural network toolbox, then using the well trained model to forecast long-term potential both roof-to-floor and side-to-side displacements of the surrounding rocks. The implementation results from this method shows that both the forecasting accuracy and efficiency are at satisfactory levels. The model has a great application value in both supporting design and maintenance of the soft rock roadway.