The key of forecasting transmutation of wall rock correctly is to construct the reasonable mathematics model of time-distance curve from measuring data when distorting, which is hard to describe accurately with traditional method of recursive analysis. According to the characteristics of feed forward neural network of radial basis function to construct the forecast model of deformation of wall rock in multi-arch tunnel and cllso uses Matlab tool to solve the optimal problem. The engineering case at the end of this paper validates the method. For its fast solving the problem,more optimal results,and better forecasting effects,this method shows its advantages and feasibility.