基于堆叠式LSTM与熵值法的苹果价格组合预测模型研究

为提高苹果价格的预测精度,提出一种将长短期记忆模型(LSTM)与熵值法相结合的新方法。发挥LSTM的自身学习特性,以及熵值法的客观拟合特征信息特点,研究构建堆叠式LSTM与熵值法结合的组合模型。对比分析多个组合模型的预测性能,试验结果显示,苹果价格存在较为明显的空间传导效应,空间传导效应对价格波动产生显著影响。加入价格空间传导效应的堆叠式多层LSTM与熵值法相结合的组合模型预测精度对比LSTM提高18.81%,在苹果价格预测方面的效果表现良好。最后,基于优化后的组合模型对苹果价格进行预测,验证组合模型的有效性。

Study on forecasting model of apple price combination based on stacked LSTM and entropy method

In order to improve the prediction accuracy of apple price, a new method combining long term and short term memory models (LSTM) with the entropy method was proposed. By exploiting the self learning characteristic of LSTM and the objective fitting feature information characteristic of the entropy method, a combination model combining stacked LSTM and entropy method was constructed in this paper. Through comparing and analyzing the prediction performance of several combination models, the experimental results showed that: apple price had obvious spatial conduction effect, and spatial conduction effect had a significant impact on price fluctuation. After considering the price spatial conduction effect, the prediction accuracy of the combined model of stacked multi layer LSTM and entropy method was 18.81% higher than LSTM, showing the great performance in price prediction. Finally, based on the optimized combination model, the apple price was predicted to verify the effectiveness of the combination model.