基于深度神经网络的卷烟智能投放模型构建方法

Modeling for intelligent cigarette release based on deep neural network

为解决卷烟产品精准投放等问题,提出了一种基于LSTM(Long Short-Term Memory)和BP(Back Propagation)神经网络的卷烟智能投放模型,包括产品销量预测和投放策略生成两大步骤。首先,通过LSTM提取卷烟销量时序特征,结合专家提取特征进行产品销量模型预测;其次,根据销量预测值和人工选择的投放方式,由模型自动推算出卷烟产品投放策略。以山东青岛烟草有限公司189种卷烟规格的历史销售数据和25种主销卷烟为对象对投放模型进行验证,结果表明:189种卷烟销量预测准确率为95.67%,25种卷烟投放准确率为92.40%。该技术可为实现卷烟产品投放策略的智能化生成提供支持。

In order to put cigarette products into markets precisely,an intelligent cigarette release model based on LSTM(Long Short-Term Memory)and BP(Back Propagation)neural network was proposed,which included two procedures:product sales volume prediction and product release strategy generation.First,extract the time sequencing characteristics of cigarette sales volume by LSTM,and predict the sales volume by the model combining with the characteristics extracted by experts.Second,calculate the release strategy of cigarette products automatically by the model on the basis of the predicted sales volume and the artificially selected ways of product release.The model was validated with the historical sale data of the cigarettes of 189 different versions and 25 major cigarette brands manufactured by Shandong Qingdao Tobacco Company Limited.The results showed that the prediction accuracy for the sales volume of the 189 cigarette versions was 95.67%and that for the release of the 25 major cigarette brands was 92.40%.This technology provides a support for the intelligent generation of cigarette product release strategy.