基于神经网络的选址区位评价模型分析应用

针对设施选址区位评价问题,分析了分级评分、回归分析等传统方法的特点和不足,并由此提出基于人工神经网络的选址区位评价模型;系统地分析了利用神经网络模型进行选址区位评价的思路和特点,提出基于遗传算法的网络结构和网络学习优化算法;通过某大型银行网点的选址实例分析,验证了该评价模型的可行性和实用性,同时也总结了其在实际应用中的若干规律。

Research on the Evaluating Models in SDSS for Location Selection

Location selection is a very complex decision process which relates to many influencial factors. The evaluating models are used to express the relationship between the location and the influencial factors,and are the foci in the spatial decision support system for location selection. The traditional models include the Analytical Hierarchy Process (AHP),the Delphi Method and the regressive analysis model,etc. So,firstly,the author introduces these traditional models in brief: the principles,the processes and their characters. Some limitations of these modes are concluded,and then the author analyzes that the neural network (NN) model is useful and may cope with the troubles the traditional methods are faced,hence the NN model will be a perfect method to evaluate the location.Based on the characters of location selection,some difficulties in building the evaluating models are identified: (1) it’s very difficult to express the relationship among the complex factors; (2) according to the location selection problem for most establishments,there are limited criterions to describe the concerns and the rules affecting the factors; (3) it’s difficult to confirm the accurate formulae to express the real relationship be-tween the factors and the evaluating results; (4) some of the factors are very difficult to be expressed by accurate number; (5) some uncertainties hide in the analysis swatches. Then,the famous Error Back—Propagation network is introduced，and there are two very important points of it:optimizing the structure of the network and optimizing the learning algorithm of it.Genetic algorithm is a kind of stochastic search methods that mimic the metaphor of natural biological evolution.The author discusses the method to integrate the genetic algorithm (GA) and the neural network，and discusses how to use the genetic algorithm to optimize the structure and connected power of the network.Finally,integrating with a project of breach planning of a certain bank，this paper researches the process of evaluating the location by the neural network model，including the steps of implementing the method，and some concrete important outlines. It’s turned out that the evaluating model can play a steady role in evaluating the location. Based on analyzing the results in detail，some useful conclusions are also put forward in the paper.