Wavelet transform-based support vector machine model for the prediction of residual settlement in old goaf

Multiresolution analyses based on wavelets and support vector machine were combined to establish a wavelet transform-based support vector machine (WT-SVM) model for the prediction of residual settlement in an old goaf. The stochastic volatility of the residual settlement in an old goaf is considered, and the test data of 3#monitoring point in an old goaf in Yanzhou are used. The results are compared with those obtained by the support vector machine (SVM) and back-propagation neural networks (BP-NN) models. According to the results, WT-SVM has many advantages in the aspects of prediction accuracy, step length, and stability over the other models. The WT-SVM model is feasible and effective in predicting residual settlement. WT-SVM model can effectively overcome the adverse effects of stochastic factors and fully reflect the temporal and spatial evolutions and their complicated non-linear relationship with the influencing factors. Thus, the existing problems in the SVM model, such as overdependence on parameter selection, and those that exist in the BP-NN model, such as low training rate and vulnerability to local minimization, are avoided. WT-SVM provides a new method to predict residual settlement in an old goaf and has a high practical value in the stability evaluation of the building foundation over an old goaf. © 2015 ejge.