基于深度学习的大蒜价格预测研究

Research on garlic price prediction based on deep learning

为了精准预测大蒜价格,基于大蒜时间序列价格和存量数据,再结合其他随机因素,提出了一种融合注意力机制(Attention)的门控逻辑单元(Gated Recurrent Unit,GRU)神经网络模型.GRU网络模型对大蒜价格数据集进行训练,归纳大蒜价格波动规律.Attention对GRU网络输出的数据进行注意力处理,使其价格波动特征更为明显,输出优化预测结果.将提出的Attention-GRU模型与GRU、CNN-GRU、BP 3种常见神经网络模型进行预测对比实验.实验表明,以全部样本为例Attention-GRU模型在均方误差(Mean Square Error,MSE)方面,比其他3种模型分别减少了4%、15%和18%.在4种网络模型预测试验中Attention-GRU预测精度也最高.

In order to accurately predict garlic price,based on garlic time series price and stock data combined with other random factors,a Gated Recurrent Unit(GRU)neural network model incorporating Attention is proposed.The GRU network model trains the garlic price data set and summarizes the garlic price fluctuation rule.Attention is paid to the data output by GRU network to make its price fluctuation characteristics more obvious and output the optimized prediction results.The proposed Attention-GRU model was compared with common neural network models such as GRU,CNN-GRU and BP3.The experiment shows that the Mean Square Error(MSE)of the Attention-GRU model is reduced by 4%,15%and 18%compared with the other three models,respectively.Attention-GRU also has the highest prediction accuracy among the four kinds of network model prediction tests.