A Development of Price Forecasting System by Neural Network

本研究的目的是将神经网络信息处理理论应用于果蔬需求结构分析，并检验神经网络在非线性分析中的能力及其内部表征。神经网络具有通过适当的学习过程再现一般难以理解的输入输出关系的能力。在学习之后，神经网络因为对样本数据有了足够的了解，所以被用于仿真。但如果神经网络不能准确地对样本数据进行内部表征，就没有进行仿真的依据。我们使用的神经网络是一种改进的前馈神经网络，其作用类似于多元回归分析。在分析中，我们输入鸟取县等6个产区的转发数量和转发时间作为输入数据，鸟取蔬菜生产的实际价格作为教学数据，以便了解这些数据之间的关系。学习结束后，我们模拟了转发数量与价格的关系。所得结果反映了需求结构的非线性。神经网络的内部表征准确地反映了水果和蔬菜的需求结构。从而为需求结构的非线性提供了逻辑依据。结果表明，蔬菜价格与数量之间不存在线性关系。

The aim of this research is to apply the neural network information processing theory to an analysis of demand structure of fruits and vegetables and examine the ability of neural network in non-linear analysis and its internal representations. Neural network has the ability to reproduce the input-output relationship, which generally hard to be understood, by appropriate learning process. After learning, the neural network used to be applied to simulation because it has obtained enough knowledge about sample data. But if the neural network can not make the internal representations of sample data precisely, there will be no grounds for simulation.We used MR neural network which is a kind of modified feed-forward neural network and acts like Multiple Regression analysis. For the analysis, we input the quantity forwarded from six producing areas including Tottori prefecture and forwarding time as input data, and the actual prices of the vegetables produced in Tottori as teaching data, so that it could learn the relationship of these data. After the learning ended, we simulated the relation between quantity of forwarded and price.The obtained results represented non-linearity of the demand structure. the internal representations of the neural network showed the demand structure of fruits and vegatables precisely. Thus, the non-linearity of the demand structure was given logical grounds.The results shows that the price of vegetables has no linear relationship with quantity.