The Research And Application Of Improved B-P Neural Network In Urban Land Evaluation Information System

With China's accession to WTO and the further economic reforms, the importance of the urban land classification and evaluation would be manifested. The urban land classification and land price evaluation work with the of characteristics large volume of information, timeliness, quality, and wider use etc. Researching and developing the Land Classification and Evaluation Information System is an important means of realizing the scientific and automatic in land classification and evaluation. And, the model of intelligent land evaluation could further improve the scientific level of land evaluation information.The current problems of the Urban Land Classification and Evaluation Information System (ULCEIS) are specificity functional, large subjective factors, lower intelligent degree in China. In practical work, how to improve the system to make the land classification and evaluation more objective, more scientific, and more intelligent is urgent to solve. Artificial Neural Network (ANN) is an appropriate way to resolve large subjective factors and lower intelligent degree. ANN is applicable to solve the problems which need to consider many factors and conditions, imprecise and fuzzy information particularly, so applying it in land classification and evaluation could make use of its advantage sufficiently, reduce the subjective influence on land classification and land evaluation, and make the result more objective At the same time, it is a new solution and idea for land classification and land evaluation.In this paper, the B-P Neural Network (B-PNN) was used to establish the intelligent model of Land Evaluation. In the ArcGIS development platform, developed the model of ANN Land Evaluation based on the improved B-P algorithm. And the data used to validate the model was the productions of the land classification and evaluation of Sanmenxia city, China in 2003a.Firstly, the two elected improved algorithm is on the basis of the research of improved B-P algorithm in the world. And their values of parameter were adjusted to make them more suitable for the land evaluation.Secondly, this paper analyzed the principal theory and application of land classification, land evaluation and elaborates the ANN. It also explored the workflow, the characteristics, and the scope of ANN in land classification and land evaluation[0]. The object-oriented model of spatial data which is called GeoDatabase was used to store the spatial data and attribute data in the same database. In this paper, the two improved B-PNNs, was programmed in the ArcGIS development platform. And then the tradition B-P algorithm, the variable step-size B-P algorithm, and the mutation of weights B-P algorithm were compared.Finally, by using of the production of Sanmenxia city land classification and land evaluation, 2003a, the model of land evaluation based on the B-PNN was established, trained and tested. And then concluded that the appropriate B-P algorithm for Sanmenxia city was the mutation of weights B-P algorithm. At then, prized the sample land, and contrasted the result with the market comparison. Comparative results showed that the mean relative deviation of the model of improved B-PNN land evaluation was less than the market comparison.The research concludes that the application of improved B-PNN is feasible, and the adjustment of the parameter of the variable step-size B-P algorithm and the mutation of weights B-P algorithm is availability in land evaluation. This paper solved the issue which is more infection of subjective factors in land evaluation with ANN. And it improved the intelligent degree of the ULCEIS in a certain extent.