TheGaussianprogressclassification(GPC)hasbeenintroducedtosurroundingrock classificationtotakeadvantageofits meritssuchasgoodgeneralizationandpredictionwith probability interpretation.However, due to the surrounding rock classification influenced by many factors, the key step in classification problem is to design or choose an appropriate kernel functionsoastoimprovethepredictionaccuracyofclassification.In this paper, a number of combined kernel functions are obtained by combining different standard kernel function based on its closure properties in mathematics. Furthermore, in order to get the optimal parameters of the GPCmodel,thegeneticalgorithm(GA)associatewithittoformtheGA-GPCmodelthat commendablyestablishestheoptimal mappingbetweentheinfluencingfactorsofsurrounding rockanditsclassificationindex.Consequently the GA-GPC model with combined kernel function isappliedintosurroundingrockclassification of Erlang Mountaintunnel on Sichuan- Tibet road compared with ART and BP respectively. The results show that the GA-GPC model with the combined kernels outperformsthetraditional GPC with standard kernelsin application of complexclassification problemand booststhe predictionaccuracyand generalization ofthe net, so it has certain superiority in application for future similar engineering.
During the construction process of expressway tunnel, the engineerneedto makereasonablesurroundingrock classification inordertoevaluatethestabilityof surrounding rockbyfullyconsideringtheengineering property ofsurroundingrock.Furthermore, thetunnel construction anddesignplancanbeconducted scientifically.How to select the classification index for boosting theprecisionofclassificationbecomesthehot topic ofundergroundengineering.At present, the development ofsurroundingclassificationmethodhas experienced fromsinglequalitativeindextothe comprehensive indexconsistingofmultivariate,multi index descriptionincombinationwithquantitative evaluation suchas,thecurrent<> (JTG D70—2004) and <> (SL279— 2002)in china, which are all combinational index standard. But the judgment results for classification using the specificationmentionedabovearemoredifferent between eachother,especiallywhendescribingthe qualitative index.Therefore, under the background of intelligence rockmechanics,manynewmethodsare come forthsuchasartificialneuralnetwork(ANN), fuzzy systemevaluationandsupportvectormachine (SVM)etc.The improved BP algorithm, fuzzy cluster analysis andSVMarerespectivelyusedtoclassthe tunnelsurroundingrock(Zhouetal., 2005; Yangetal., 2006; Zhu, et al.