农产品价格变动关乎国计民生，农产品市场价格波动频率快，波动幅度大，受多方面因素共同影响，并呈现出非平稳、非线性等不规律波动特征，这给农产品市场带来极大的挑战。只有充分地分析农产品价格的变化趋势，提高价格预测精度，才能更好地指引农产品产业健康发展。

文章以马铃薯为研究对象，基于2014年1月至2019年12月共72组月度价格数据，提出一种经验模态分解（EMD）和极限学习机（ELM）的农产品价格组合预测模型。该组合预测模型充分利用了经验模态分解的自主分解能力和极限学习机设置较少的参数的优势。首先利用经验模态分解将原始价格序列分解为若干个本征模态函数（IMF）和余项，然后将这些分量分别用极限学习机进行预测，最后把各个分量的预测结果进行组合得到原始序列的预测值。

将该方法实际应用于马铃薯价格进行预测并对其预测结果进行评价分析，结果表明其平均绝对误差为0.093元/kg，平均百分比误差为4.265%，均方根误差为0.148，并与单独的极限学习机、BP神经网络和ARIMA方法进行比较，结果表明EMD-ELM组合预测模型具有较好的农产品价格预测性能，能够为预测农产品价格波动提供一种思路，为行业和政府主管部门保障农产品供应决策提供参考依据。

The price fluctuation of agricultural products is related to both the national economy and personal livelihoods. The price fluctuation frequency for the agricultural products market is fast, and the fluctuation range is large and is affected by various factors. The fluctuation range presents non-stationary, nonlinear, and other irregular characteristics, which brings great challenges to the agricultural products market. Only by fully analyzing the change trend of agricultural product prices and improving the accuracy of price prediction can we better guide the healthy development of the agricultural product industry.

Taking potatoes as the research object and using 72 sets of monthly price data from January 2014 to December 2019, this paper proposed an Empirical Mode Decomposition and Extreme Learning Machine agricultural product price combination prediction model. The combinatorial prediction model used the autonomous decomposition ability of Empirical Mode Decomposition and the advantage of the Extreme Learning Machine’s fewer parameters. Firstly, the original price series is decomposed into several Intrinsic Mode Functions and remainder terms by Empirical Mode Decomposition, then these components are separately predicted by Extreme Learning Machines, and finally the prediction results of each component are combined to obtain the original sequence Predicted Value.

This method is applied to the prediction of potato prices and the evaluation and analysis of their prediction results. The results showed that the average absolute error was 0.093 yuan/kg, the average percentage error is 4.265%, and the root mean square error was 0.148. We compared our results with the results obtained by using the single Extreme Learning Machine, the BP Neural Network, and the ARIMA method. The results of this comparison showed that the prediction model combined with Empirical Mode Decomposition and Extreme Learning Machine performed better at agricultural product price prediction than the other methods. The results provide a method of predicting the price fluctuation of agricultural products, and they provide a reference for the industry and government department in charge to use in guaranteeing the decision of agricultural product supply.