面向农产品期货价格预测的改进LSTM方法

Research on the improved LSTM method for agricultural futures prices forecasting

期货市场在金融领域具有重要的地位,而期货价格走势的预测对投资者和决策都十分关键。目前,期货价格走势预测模型使用的方法较为单一,且预测模型的精度不够理想。文章以农产品期货价格预测为研究对象,围绕数据预处理、模型构建、集成决策等展开相关工作,针对农产品期货价格具有的典型时序性特征以及其非线性、非平稳等特点,提出一种基于LSTM模型的改进LSTM预测方法。该改进方法引入EEMD方法,先对原始期货价格序列进行分解,再对分解所得的每个子序列进行建模、预测、叠加子序列预测结果以得到最后的预测结果。实验结果表明,该改进方法与LSTM,SVR等传统的机器学习预测模型相比,精度明显提升。

Considering the significance of futures market in the financial field, the forecasts of futures price movementsis critical to investors and policy makers. At present, the method of futures price trend prediction model is relativelysingle, and the accuracy of the prediction model is not ideal. Taking the prediction of agricultural product futures priceas a research sample, this paper focuses on data processing, model construction, integration decision-making, etc. andproposes an improved LSTM based the price forecasting method considering the typical characteristics of time seriesfeatured by futures price series, such as nonlinearity and non-stationarity. In this improved method, EEMD method isintroduced. Firstly, the original futures price sequence is decomposed, and then each subsequence is modeled, predictedand superimposed to obtain the final prediction results. The results of experiments show that the accuracy of thisimproved method is significantly promoted when it’s compared with the traditional machine learning prediction modelssuch as LSTM and SVR.