Abstract: The key deployment of agricultural market work shows the determination of the country to adhere to the stable development of agricultural and rural markets, and also reflects the necessity of studying agricultural product price prediction. The fluctuation range of the price series of smallholder farmers’ products is large, and the phenomenon of price sudden increase and decrease occurs frequently, which produces shocking news in society, which is not conducive to the stable development of the agricultural product market. Because the price series of small-scale agricultural products have obvious nonlinear and non-stationary characteristics, the prediction effect of a single model is not good. To this end, this paper proposes a combined prediction model based on “decomposition and integration”.
Firstly, the Whale Optimization Algorithm (WOA) is applied to the Variational Modal Decomposition (VMD) algorithm, and the Sample Entropy is used to solve the problem. SampEn is used as the fitness function to screen out the optimal parameters. Then, the optimized variational mode decomposition method is used to realize the multi-mode decomposition of the agricultural product price series, solve the modal confusion problem in the complex small-scale agricultural product series, and obtain the modal components that can reflect the different characteristics of the original series. Secondly, the decomposed components and residual sequences are integrated into the Long Short-Term Memory (LSTM) neural network, which is trained as the feature quantities of the original agricultural product price series, so as to enhance the learning ability of the LSTM neural network and improve the prediction accuracy of the combined model. In this study, the daily average price data of potato, lotus root, white radish, Chinese cabbage, broccoli and cabbage in Henan Province from January 1, 2016 to December 31, 2021 are selected as the research object, and the combined model method is used to predict the price series of six small-scale agricultural products.
The root mean square error (RMSE) and coefficient of determination (R2)are used as the evaluation indicators of the prediction effect of the model. The experimental results show that the RMSE of the WOA-VMD-LSTM combination model is 0.292, 0.381, 0.129, 0.125, 0.782, 0.142, respectively. The coefficient of determination is 0.755, 0.971, 0.947, 0.907, 0.911, and 0.973, respectively. The EMD-LSTM combination model and ARIMA model are used to predict six kinds of price series, and the prediction results of the three models are compared comprehensively. The RMSE values obtained by WOA-VMD-LSTM combination model for the price series of lotus root, white radish, Chinese cabbage, broccoli and cabbage are lower than those in the EMD-LSTM model and ARIMA model, and the determination coefficient values are higher than those of the other two prediction models. Although the coefficient of determination value obtained by the WOA-VMD-LSTM combined prediction model for potato sequences is not better than that in the ARIMA model, the RMSE value is48.1% lower than that in the EMD-LSTM model and 47.2% lower than that in the ARIMA model. In summary, it can be concluded that the method of using whale optimization algorithm to optimize the variational mode decomposition model for sequence decomposition and using neural network to complete the sequence prediction of agricultural product price can effectively improve the price prediction accuracy.
This study tries to explore the influence of meteorological temperature, economic policy, crop yield, planting area and other factors on the daily average price data of six agricultural products, but does not achieve good results. Therefore, this paper uses the method of sequence decomposition to achieve the purpose of extracting sequence features. In the subsequent research, the sequence components can be divided according to different frequencies, and the characteristics of high-frequency components can be deeply analyzed to achieve deep noise reduction. The combined prediction model proposed in this study can effectively improve the accuracy of small-scale agricultural product price prediction, which not only stabilizes the supply and demand relationship of agricultural product market, but also protects the interests of agricultural product suppliers and consumers, and has the value of popularization and application.

Key words: VMD, SampEn, WOA, price prediction

摘要： 针对小宗农产品价格序列波动特征中呈现出的非平稳、非线性等问题,提出了一种基于“分解与集成”的WOA-VMD-LSTM组合预测模型。首先利用样本熵作为鲸鱼优化算法(WOA)的适应度函数,对变分模态分解方法(VMD)的两个自由参数进行全局寻优;再使用优化后的变分模态分解方法对价格序列进行分解;最后将得到的多模态分量及残差作为输入特征集成到长短期记忆网络(LSTM)中,构建组合模型。将该方法应用于马铃薯、莲藕、白萝卜、大白菜、西兰花、卷心菜的日均价格数据进行预测,实验结果表明,WOA-VMD-LSTM组合模型的均方根误差分别为0.292,0.381,0.129,0.125,0.782和0.142,且与EMD-LSTM组合模型以及ARIMA模型进行对比,WOA-VMD-LSTM组合模型在多种农产品价格的预测上具有更明显的优势。本研究提出的组合预测模型有助于相关产业对市场进行合理配置。

关键词: 变分模态分解, 样本熵, 鲸鱼优化算法, 价格预测