Prediction Research Of The Surrounding Rock Deformation In Engineering By The Support Vector Machine

Geotechnical engineering is a very complex system engineering. Due to the complexity of geotechnical medium, problems in geotechnical engineering are highly non-linear, uncertainty, randomness and fuzziness, the establishment of a precise mathematical model to simulate the stability of surrounding rock and the establishment of accurate prediction models to predict changes in rock over time the rules are very difficult. The appearance and development of Intelligent Rock Mechanics provide a new way to its prediction. Genetic algorithms are combined with support vector machine, to solve the difficult problem of the mathematical model described have excellent adaptation, and has a wide application prospect. In this paper, the basis of previous studies, genetic algorithms are combined with support vector machine, constituting the evolutionary support vector machine, to predict rock deformation in the project. The main work is as follows in this paper:(1) This paper discusses the theory and application of supported vector machine arithmetic method, and make further research for the problem of choosing parameter and the influenced laws of the coefficient of kernel function, penalty factor C and no-sensitivity coefficientεto support vector machine based on wall rock deformation predictive analytics to certain project cases, and the a better performance supported vector machine develops when each parameter attains top-notch combination.(2) This paper propose a sort of modified supported vector machine method and it improves the capability of regression precision and prediction, in order to make coefficient of kernel function, penalty factor C and no-sensitivity coefficientεof supported vector machine and avoid blindness of man-made selected parameter.(3) Search parameters of supported vector machine using the genetic algorithm, this paper combines genetic algorithm and supported vector machine, propose evolutional supported vector machine method, re-predict and analyze wall rock deformation of certain project cases, make further research for sensibility of supported vector machine parameter based on genetic evolution develops evolutional supported vector machine forecasting system using VC++, and get the conclusion:the highest sensitivity coefficient is not no-sensitivity coefficientε, but penalty factor C, and the lowest sensitivity coefficient is the coefficient of kernel function.(4) The creep test data in Gou Pi Tan Hydropower Project and the rock deformation data in Long Tan project are analysed deeply. Mining their law, we constitute the learning samples of evolutionary support vector machine according to the deformation data, using evolutionary support vector machine prediction system to optimize the parameters, establishing the prediction model of evolutionary support vector machine and predicting, analyzing the results and coming to the conclusion, verifying the feasibility of this method.