Study On The Residential Project Cost Estimation Model Based On The Improved Back Propagation Neural Network

With the steady development of the construction market, the investment in fixed assets of the whole society increase gradually, and the uncontrolled investment getting more severe, thus the reasonable control of the project investment is particularly important. In the whole life cycle of construction projects, the influence degree of the early stage of investment control is higher than the construction phase, and construction enterprise do not pay enough attention to the early stage of investment decision, therefore the focus of investment control should shift from construction stage to early investment decision-making stage. But in the investment decision-making stage, using the traditional project cost estimation can’t adapt to the trend of the development of market economy well at present, accuracy of investment estimation become the key factor affecting investment control of construction project. So, estimating the project cost quick and accurate is of great significance for the investment control of construction project.First of all, sorting and extracting the engineering characteristic factors influencing residential construction cost from the main separated parts of a project based on the primary and secondary factors analysis method, and setting up a hierarchical structure model to calculate the weights of each characteristic factors’ impact degree of project cost, screening the main characteristic factors according to the primary and secondary factors analysis method, meanwhile analyzed the engineering characteristic factors influence mechanism of the project cost, and use it as input vectors of estimation model.Secondly, according to the learning process of standard BP neural network, set up the initial connected weights between hidden layer and input layer with the method of interval grey number based on the relevant knowledge of grey number in grey theory, in the process of training by constantly adjusting the taking value interval of interval gray numbers to approximate the optimal value of weights, improved the learning process of standard BP neural network.Thirdly, using the MATLAB software, and using the BP neural network toolbox functions to achieve the establishment and initialization of the model. Train the improved BP neural network according to the collected and collated 115 groups of the training sample data, building the initial estimation model which error is within 10%, and optimized the initial weights and whitening targeting coefficient of the model with the genetic algorithm, improved the accuracy of the estimation model.Finally, taking the 2# floor of a residential district in Xuzhou as the calculating object, complete the training of the improved BP neural network estimation model through the sample data on the basis of the collection and collation of the similar project cost data, and calculate the data of 2# floor such as the cost of per square meter and consumption of manpower. By comparing the prediction of estimation model and calculated results on the basis of the investment estimation index, verify the applicability and reliability of the optimized and improved BP neural network estimation model.