Construction Land Demand Forecast in Nantong City Based off BP Neural Network
GUO Jie,OU Minghao,LIU Qiong,OU Weixin
DOI: https://doi.org/10.3321/j.issn:1007-7588.2009.08.014
2009-01-01
Abstract:Being the frontier of nonlinearity and artificial intelligence, Artificial Neural Networks (ANN) are widely used in many fields, such as trend analysis, pattern recognition, ecosystem simulation, ecological data processing, remote sensing-based extraction of ecological parameters, oil exploration, pest and disease control, and geological hydrology. However, ANN models are seldom used in land use change simulation. There are non-linear relationships between construction land scale and its driving factors such as total population, GDP, investment in fixed assets and others. Based on the data of socio-economic development and construction land from 1988 to 2006 in Nantong city of Jiangsu Province, the driving factors related to construction land expansion are selected by using binary variables correlation analysis method. Then, multiple regression analysis and BP neural network are employed to build the forecast model for construction land demand, and the optimal model is used to predict the demand of construction land. Finally, the gray system model is combined with trend evaluation to verify the credibility of the predictive results. The results show that the model using ENTER method is less credible, while the model using STEPWISE method has less indices, because most of the driving factors are deleted from the model as multi-collinearity eliminates. Nevertheless, the model based on BP neural network integrates various driving factors, and the variation coefficient is only 1.78%. The comparative analysis of the three methods proves that the credibility of the results from BP-neural- network-based model is higher, because: 1) by comparing the results of grey system and BP neural network, it can be seen that the differences between both predictive values of construction land are insignificant (relative error are less than 10%), but the simulation value in BP neural network in 2006 is much closer to the actual area(1.71×104hm2); 2) During the Ninth Five-Year-Plan and the Tenth Five-Year-Plan periods, the annual average net increase of construction land scale in Nantong were 5 654.28hm2 and 3 660.32hm2 respectively, and according to the gray system and neural network methods, the annual average net increase of construction land are 5 668hm2 and 1 907hm2 respectively. The predicting results of neural network reflect that the intensive degree of land use will increase gradually. The research proves that the use of BP neural network model can effectively improve the forecast accuracy, and the results can be used as the reference in social and economic planning.