Research On Cost Prediction Of Subway Civil Engineering Based On Gene Expression Programming

Cost prediction is one of the important contents of subway civil engineering cost control and management.Its accuracy directly affects the scientificity of project investment decision,construction scale determination and engineering design scheme formulation.Therefore,its research is of great significance.There are many factors affecting the cost of subway civil engineering,and there is great uncertainty,and the cost has a highly nonlinear relationship with its own characteristics.It can be seen that the key problems in the research on the cost prediction of subway civil engineering are mainly the selection of influencing factors,the selection of prediction methods and the optimization thereof.In this paper,the variable selection method based on the contribution analysis of neural network and the GEP algorithm with strong ability to deal with highly nonlinear systems are selected to construct the cost prediction model of subway civil engineering.Since subway civil engineering is mainly composed of subway station and interval tunnel,this paper takes subway station and interval tunnel as independent prediction units to construct models respectively.Firstly,through literature review,expert interviews and expert scoring,the characteristic factors that affect the cost of subway station and interval tunnel civil engineering are preliminarily selected,and then the main characteristic factors of subway station and interval tunnel cost are further selected by using the variable selection method of neural network contribution analysis.And all factors and main characteristics of the selected factors combined BP neural network and GEP respectively two kinds of forecast method to establish cost prediction model,the history of the application of cost and characteristics of the two factors of the example of engineering data,through R2,MSE,RMSE,MaxRE four indicators to evaluate the model,the results show that,with the main characteristics of the factors for the model input variables can significantly improve the prediction accuracy of model;The cost prediction model established by the combination of main feature factors and GEP prediction method is the best.GEP’s powerful function mining ability and extremely high global search efficiency are verified again,which overcomes the problem of insufficient generalization ability of BP neural network(high prediction accuracy in training period,low prediction accuracy in verification period).By studying the main characteristic factors selected and combining the selection of the main characteristic factors with the selection of the prediction method,an optimal model for solution was constructed,which well solved the problem of the lack of subjectivity and scientificity of the selection of the characteristic factors in the existing researches and the influence of the selection of the characteristic factors on the selection of the prediction method.