基于主成分分析的BP神经网络在稀土价格预测的应用:以氧化镝为例

The research on pricing forecast of rare earth products with multi-factor PCA-BP combination model:take dysprosium oxide price for example

稀土是我国重要的战略资源,对其价格趋势的准确预测意义重大。本文从稀土资源价格影响因素出发,设计并采用基于主成分分析的BP神经网络(PCA-BP)组合模型对稀土产品价格进行预测。鉴于影响稀土产品价格波动的因素众多,利用主成分分析(PCA)消除稀土价格预测影响因素之间存在的冗余信息,降低BP神经网络输入数据的维数,提高预测精度。本文以氧化镝价格为预测对象,选取2010年1月~2018年2月的月度数据,构建多因素PCA-BP组合模型。预测结果表明多因素PCA-BP组合模型在仿真能力、误差水平、收敛精度等方面优于主流的神经网络模型,能更加准确地预测氧化镝价格走势。

Rare earth is an important strategic resource in China.Understanding the price fluctuation of rare earth is very important for efficient utilization of resources.The BP neural network model based on principal component analysis is adopted on the basis of the price factors of rare earth resources.The prices of rare earth products are predicted according to the combined model.There are many factors that affect the price of rare earth products.In this paper,principal component analysis is used to eliminate the redundant information among the influencing factors.The dimension of BP neural network input data is reduced to improve the prediction accuracy.This paper takes dysprosium oxide price as the forecasting object,chooses the monthly data from January 2010 to February 2018,and constructs a multi-factor PCA-BP combination model.The prediction results show that the combined model is superior to the traditional BP neural network model in simulation capability,error level and convergence speed.Combined models can predict dysprosium oxide prices more accurately.