【摘要】 为帮助烟草企业更好地控制生产成本，科学制定采购价格，利用小波去噪和长短期记忆网络（LSTM）模型预测烟用薄膜上游主要原材料聚丙烯的价格。首先，借助小波分析对聚丙烯期货数据实施去噪；然后，构建LSTM模型，并与ARIMA、MLP以及RNN模型展开对比；最后，选取多组特征组合，利用预测精度最高的LSTM模型开展预测。结果表明，小波分析去噪法处理金融数据噪音的可靠性强，基于小波分析去噪后的收盘价、最大值及最小值特征组合的LSTM模型预测效果最优。更多还原

【Abstract】 In order to help tobacco enterprises control production cost better and make purchase price scientifically, this paper uses wavelet denoising and long short term memory network(LSTM)model to predict the price of polypropylene, the main raw material upstream of tobacco film. Firstly, using wavelet analysis to denoise polypropylene futures data. Then, an LSTM model was constructed and compared with ARIMA, MLP, and RNN models. Finally, using LSTM model and selecting multiple sets of feature combinations for price prediction. The results show that the effect of LSTM model prediction based on the combination of closing price,maximum value and minimum value after wavelet analysis is optimal, and wavelet analysis denoising method in dealing with financial data noise is verified to be valid.更多还原