| 摘要: |
| 【目的】农产品价格变动关乎国计民生，由于农产品的价格受到多方面因素的共同影响，其价格预测也一直是研究中的难点。只有充分分析农产品价格的变化趋势才能提高价格预测精度，更好地指引农产品产业健康发展。【方法】文章以菠菜、大白菜、番茄、辣椒和马铃薯5 种蔬菜为研究对象，基于2013 年1 月至2018 年12 月共72 组月度价格数据，研究农产品价格变动趋势，并基于小波变换和BP 神经网络构建农产品价格组合预测模型。首先利用小波变换对价格进行db5 的3 尺度分解，其次采用BP 神经网络模型对分解出的趋势部分和细节部分分别进行预测，最后对各分量的预测结果进行组合重构。【结果】采用预测精度指标对5 种蔬菜的价格预测结果进行评价分析，其平均绝对误差最小值为0.083 元/kg，平均百分比误差最小为3.95%，均方根误差最小值为0.102。【结论】将小波变换和BP 神经网络结合起来的组合预测模型具有较好的农产品价格预测性能，该组合方法能适应多种蔬菜的价格预测，具有普适性。但农产品价格波动幅度和强度会对该模型的预测精度产生影响。 |
| 关键词: 小波变换 BP 神经网络 农产品 价格预测 |
| DOI：10.12105/j.issn.1672-0423.20190609 |
| 分类号: |
| 基金项目:河南省重大科技专项（171100110600-01）；河南省现代农业产业技术体系（S2010-01-G04） |
|
| Agricultural product price forecast based on wavelet transform and BP neural network |
|
Liu Hebing, Han Jingjing, Xi Lei※
|
|
School of Information and Management Sciences，Henan Agricultural University，Zhengzhou 450002，China
|
| Abstract: |
| ［Purpose］The price change of agricultural products is related to the national economy and people’s livelihood. As the price of agricultural products is affected by many factors， its price is difficult to predict. Fully analyzing the changes in the price of agricultural products can improve the accuracy of price prediction and promote the healthy development of the agricultural products industry. ［Method］This article focuses on the prices of spinach，Chinese cabbage， tomato，pepper and potato as the research objects. Based on a total of 72 sets of monthly price data from January 2013 to December 2018，this article studies the trend of agricultural product price changes by the combining methods of wavelet transform and BP neural network. A combined forecast model for agricultural product price is constructed. Firstly，the price is decomposed by db5 to a 3-scale using wavelet transform. Secondly，the BP neural network model is used to predict the decomposed trend and details separately. Finally，the prediction results of each component are combined and reconstructed.［Result］The prediction accuracy indexes were used to evaluate the price prediction results of five vegetables. The minimum absolute error was 0.083 yuan/kg，the minimum percentage error was 3.95%，and the minimum root mean square error was 0.102.［Conclusion］The combined forecasting model based on wavelet transform and BP neural network has a better performance for agricultural product price forecasting. The combined method can be adapted to the price forecasting of various kinds of vegetables and is universal. However，the amplitude and intensity of agricultural product price fluctuations will affect the prediction accuracy of the model. |
| Key words: wavelet transform BP neural network agricultural products price forecast |
