基于扩张因果卷积模型的冷库商品销售量预测

Forecast of Cold Storage Commodity Sales Based on Extended Causal Convolution Model

疫情环境下,供应链受到不良影响,库存及市场投入量关乎着社会以及民生的稳定。但是供给与需求无法达到完全一致的现象普遍存在,这使存储管理上面临两方面难题:要么库存过剩增加成本,要么库存不足造成供给短缺。在这种情况下,对商品销售量预测进行深入的研究是一件非常重要的事情。传统的一维卷积神经网络(CNN)在销售量预测上存在信息泄露的问题,且其结构难以获取较长的记忆。文中提出扩张因果卷积神经网络(Dilated Causal Convolution)来优化模型解决问题,其中扩张卷积可以增加卷积模型的感受野大小,获取序列的长时记忆;同时引入因果卷积来解决信息泄露问题。实验结果表明文中提出的扩张因果卷积在销售量预测方面有着较好的预测效果。

Under the epidemic situation,the supply chain is adversely affected.Inventory and market input are related to the stability of society and people's livelihood.However,the phenomenon that supply and demand cannot be completely consistent is widespread,which makes storage management face two problems:Either excess inventory increases costs,or insufficient inventory causes supply shortage.In this case,it is very important to make an in-depth study on the forecast of commodity sales.The traditional one-dimensional convolutional neural network(CNN)has the problem of information leakage in sales forecasting,and its structure is difficult to obtain a long memory.In this paper,a modified causal convolution neural network is proposed to optimize the model to solve the problem.The expanded convolution can increase the receptive field size of the convolution model and obtain the long-term memory of the sequence;at the same time,causal convolution is introduced to solve the problem of information leakage.The experimental results show that the extended causal convolution proposed in this paper has a good prediction effect in sales volume prediction.