Research Of Concrete Performance On Support Vector Machine

As the important bases concrete structure design and construction，concreteperformance study is becoming highly valued.According to the traditional methods,the compressive strength of concrete specimens are usually obtained after28daysstandard maintenance;the durability of concrete specimens are predicted under the thesimulation environmen.So, in order to solve the shortcomings,such as the long testtime and imperfect environmental simulation, machine learning method is becominga popular reserch direction.Basing on the firm mathematic theory foundation and the strict theory analysis，Support Vector Machine（SVM）is a kind of new machine learning method，which hasthe advantages of global optimization,strong adaptability and generalizationcapability. SVM uses the statistics theory of Structural Risk Minimization (SRM) andtakes account of the training error and generalization ability. Comparing to NeuralNetwork (NN) used empirical risk minimization rule, SVM is a great progress.Aiming at the "excessive learning" issue of neural network, this paper uses SVMmodel to forecast the expansion of sulfate concrete. Comparing with the results ofSVM and NN model, the results show that SVM can commendably forecast theinflation problem of sulfate concrete.In allusion to the long-term compressive strength of silicate concrete, this paperuses SVM model to forecast. The consequences show that the SVM model is moresuitable for forecasting the silicate concrete’s long-term compressive strength than theNN model and the fuzzy algorithm model.On the superiority of the feasibility of SVM from the two above examples, thispaper uses SVM and Statistic Analysis to forecast the compressive strength and theslumps of concrete. The outcomes show that SVM in the small sample premise canalso well forecast the compressive strength and the slumps of concrete. In the lastplace, combing the forecasting results and the sensitivity’s size of the concrete’sparameters of compressive strength and slumps, this paper has provided valuableopinions for the actual projects.The concrete28-day compressive strength has been predicted based on Genetic expressionprogramming and support vector machine. The predictions of the two methods are compared withthe the one of neural network model. The results show that the accuracy and error of SVM are better than the other two methods. It clearly demonstrates that the support vector machine issuitable for the prediction of compressive strength of concrete.