Novel Project Evaluation Method Based On BP, RBF And GAAA-RBF Neural Network

Early in the construction project，it has a very important significance toestimate the cost of the projects qu ickly and accurately for the investment decision ofthe project.With the in-depth study of intelligent algorithm, it come out of many newprediction methods in the field of engineering valuation,such as multiple regression,neural networks, case-based reasoning, genetic algorithms, wavelet analysis. Thepaper presents an improved intelligent algorithm which make use of ant colony toassemble and optimize the structure parameters of RBF neural network.And thisalgorithm would fit for the Quick estimates cost of the construction project.The article first collected70successful engineering data, based on fullunderstanding and learning data preprocessing technology, combined with thespecific characteristics of the project cost data, proposed Data preprocessing methodsand processes including data cleansing, data conversion and data reduction.we finallyget55available and process data as a Projects library which is the basis for thesubsequent modeling.Secondly, the paper established a valuation model based on improved BPneural network engineering, the RBF neural network which use the NEWRB functionand RBF neural network based on K-means clustering, then compare and analyzethese three models.The analysis results show that: the BP neural network is relativelydifficult to build，we need to define the complex parameters such as learning rate,momentum factor size, and the number of hidden layer nodes, cause there is noRigorous theoretical guidance to design these parameters，we need to be estimatedthem。AS for RBF neural network，it is relatively simple whatever constructionmethods we use to build this model.When use NEWRB function to build RBF, itsperformance depends primarily on the distribution width，RBF neural network basedon K-means clustering depends primarily on the overlapping coefficients and hiddenlayers. Overall, RBF neural network based on K-means clustering has a fasterlearning speed, higher prediction accuracy, but the learning effect needs to beimproved. Finally, in order to further improve the learning effect and the predictionaccuracy of the RBF neural network based on K-means clustering, This paperintroduced a genetic-ant colony hybrid algorithm to assemble and optimize the mainstructure parameters such as Center vector, wide base vector network weights,thenbuild valuation models RBF neural network based on improved ant colony algorithmengineering.We combined successful engineering data in Xiamen to carry out thelearning simulation, The simulation results show that when use the parameter of TheRBF neural network through optimizing to forecast the engineering cost， theprediction error is less than5%，and The generalization ability is superior, auxiliaryestimates for the actual project cost，it can be used for the auxiliary of the actualproject cost estimate.