Highway Construction Cost Forecasting Model Based On Main Characteristic Factors And BP-GEP Network

The main characteristics of project cost prediction is that there are many affecting factors but less information. Project cost related to its own characteristics, but it also affected by many uncertain factors, the system is a rather complex nonlinear system. Therefore, how to reflect the nonlinear relation between the project characteristics and its cost is the key for building engineering cost predicting model. Gene expression programming as a new mathematical modeling method of artificial intelligence, have very strong function ability and high search efficiency. At the same time, the computing power of neural network has two characteristics:(1) massive parallel distributed structure;(2) neural network learning ability and generalization ability from it(generalization refers to the neural network t can produce reasonable output to data is not in the training set). Both information processing ability make neural network can solve some complex large problems that current can’t deal with. According to the characteristics of the highway engineering cost prediction and the deficiency of existing prediction model,it solve highway engineering cost prediction model based on main characteristic factors and BP-GEP networkOn the basis of model building using BP neural network to select scientific12main characteristic factors of highway construction cost,18characteristic factors and12main characteristic factors separately as input variables combine with BP neural network and GEP network respectively, the application of information and data of the15groups of cost and the characteristic factors of highway construction to empirical research. Evaluation by R square, mean square error, the root mean square error, the average relative error and the maximal relative error five standard shows that, main characteristic factors as model input variables can improve the prediction precision greatly; the highway construction cost prediction model using12main characteristic factors as input variable and GEP network as modeling method is best, GEP strong function searching ability in nonlinear space and high search efficiency verified again; the highway engineering cost prediction model based on GEP make up for the deficiency of the previous model in quite extent. The selected main characteristic factors and the best model based on the combination of selection of main characteristic factors with prediction method overcome the subjectivity and lack of scientific characters of the selection of characteristic factors and the problem that the selection of characteristics factors influences the selection of prediction method in the existing study, it has better value in the highway construction cost forecasting.