Abstract:We apply the gray relational algorithm to determine the major factors directly influencing surface subsidence. The Gaussian kernel function and the polynomial kernel function are constructed, we determine an optimization of model parameters by genetic algorithm, and a prediction model of ground surface subsidence is established. The experimental data shows that: the gray relational algorithm can quantitatively reflect the degree of correlation between many factors of the system and the change of surface subsidence prediction; furthermore, the information of the gray system can be processed effectively. The reasonable combination of the weighted kernel functions can be transformed into a linearly separable map of the high-dimensional feature space by the low-dimensional linearly separable map; the genetic algorithm is simple to calculate and shows adaptive iterative optimization; the relevance vector machine model can greatly reduce the computational burden of the kernel function and the probabilistic interpretation of the process and results. The model based on relevance vector machine has good predictive effects. In addition, the accuracy of this method is better than that the precision of BP neural network, GR-SVM method.