Predicting Construction Cost Using Fuzzy Logic And Neural Networks

The cost estimation of construction project is a key data reference for the investment decision of the project in the early stage.The cost estimation also runs through the project investment estimation,design budget,construction drawing budget and other stages,and is essential for the feasibility study report and design specification.In recent years,the widely used engineering cost estimation methods include quota calculation method,multiple regression analysis method,gray prediction method,time series method,fuzzy mathematics method and BP neural network prediction method.At present,these methods have their own advantages and disadvantages.Rapid and accurate engineering cost estimation method can not only increase the speed of project economic comparison and improve the efficiency of investment decision,so as to do a better job in the control of engineering cost,but also guide and promote the establishment of engineering cost information database,so it is particularly important.This paper aims to improve the calculation accuracy of the project cost prediction model and advocates the fast estimation method.Firstly,the characteristics of construction cost are analyzed,the basic principles of fuzzy logic method and neural network method are introduced,and the rationality and applicability of model method selection are explained qualitatively in theory.Then,a two-step method based on fuzzy logic and neural network construction engineering cost estimate model,the first step in using fuzzy logic analysis method to calculate the influence degree coefficient of each construction project cost factors,select the input vector of the prediction model,the second step is to establish a neural network,the network topology,activation function,training function and error,call the MATLAB toolbox to network for training.Finally,the same sample data is used in fuzzy logic + neural network prediction model,multiple linear regression prediction model and traditional BP neural network prediction model.Conclusions: There are many factors affecting the construction cost,some of which are related and redundant.The same prediction model is predicted with different input vectors,and the prediction accuracy and generalization ability of the obtained model are different;According to the influence degree coefficient,the indicators that have a great impact on the construction cost are selected,which effectively simplifies the prediction model,accelerates the calculation speed and improves the prediction accuracy;The prediction accuracy of multiple linear regression model is the worst,Neural network model has poor generalization ability and unstable performance,PNN + neural network method has the highest prediction accuracy,and overcomes the disadvantages of low accuracy of multiple linear regression model and poor generalization ability of neural network model.