【作者】 张宁；

【导师】 张涛；

【作者基本信息】 北京工业大学 ， 软件工程（专业学位）， 2019， 硕士

【摘要】 在新的零售业蓬勃发展下,连锁便利店具有营业时间长,空间小,资产模式轻的优势,在消费者中越来越受欢迎。在消费升级和新零售的两轮驱动下,中国的便利店行业正在全面展开,各大零售巨头纷纷投入大量资金参与其中,苏宁店和天猫店等新的便利小店如雨后春笋般涌现并迅速扩大,如何在激烈的市场竞争中脱颖而出是一个值得思考的问题。商品销量准确预测可以有效指导便利店的后端运营,进行合理的库存管理,及时调整商品定价策略,满足周边居民日常购物需求,有效地提高品牌知名度和市场竞争力。由此可见,销量预测技术是便利行业市场竞争中的关键。连锁便利店在地区上分布广泛、商品种类多样、季节影响复杂、市场需求难以预测,这些因素给商品销量预测增加了困难。通过分析参考文献的相关研究,发现传统的回归预测算法,对于受多因素影响的零售商品销量的预测效果并不理想,本文为了提高便利店商品销量预测的准确性,基于机器学习和深度学习的相关理论,将深度信念网络(DBN)、神经网络(NN)和支持向量回归(SVR)方法相结合,建立一种新的回归预测组合模型。实验中以便利店历史销量数据集为研究对象,对数据进行预处理和特征提取等操作,使用组合模型与传统的神经网络(NN)、支持向量回归(SVR)和深度信念网络(DBN)做对比实验,实验结果表明,组合模型具有更好的预测效果。最后,本文从应用的角度出发,分析潜在的便利店用户需求,设计了一种便利店销量预测系统,包括系统的架构设计、功能设计和数据库设计,通过相关技术将组合模型整合到系统中,并使用相关开发技术对系统进行了实现。更多还原

【Abstract】 With the vigorous development of new retail industry,chain convenience stores have the advantages of long business hours,small space and light asset model,which are becoming more and more popular among consumers.Driven by two rounds of consumer upgrading and new retailing,China’s convenience store industry is being fully launched.Major retail giants have invested a lot of money to participate in it.New convenience stores such as Suning Store and Tianmao Store have sprung up and expanded rapidly.How to stand out in the fierce market competition is a question worthy of consideration.The location of each store in chain convenience stores is different,the surrounding environment is different,and the residents’ demand for different goods will also vary.Therefore,the store scientifically and reasonably allocates the stock of store goods,prevents the overstock of goods,more in line with the residents’ shopping needs,and improves brand awareness and market competitiveness.The accurate sales forecast of convenience stores can guide the back-end operation and make reasonable resource matching and Optimization in advance.Accurate forecasting of commodity sales can effectively guide the back-end operation of convenience stores,carry out reasonable inventory management,timely adjust commodity pricing strategies,meet the daily shopping needs of neighboring residents,and effectively improve brand awareness and market competitiveness.Thus,sales forecasting technology is the key to facilitate market competition in the industry.Chain convenience stores are widely distributed in the region,the variety of commodities,the complex seasonal impact,and the market demand is difficult to predict.These factors increase the difficulty of commodity sales forecasting.Through the analysis of the relevant research of references,it is found that the traditional regression prediction algorithm is not ideal for the prediction of retail sales affected by multiple factors.In order to improve the accuracy of the prediction of convenience store sales,based on the theory of machine learning and deep learning,the deep belief network(DBN),neural network(NN)and support vector regression(SVR)are applied.A new combination model of regression and prediction was established by combining the methods.In order to facilitate the historical sales data set of stores as the research object,pretreatment and feature extraction are carried out on the data.The combined model is compared with traditional neural network(NN),support vector regression(SVR)and deep belief network(DBN).The experimental results show that the combined model has better prediction effect.Finally,from the application point of view,this paper analyses the potential needs of convenience store users,and designs a convenience store sales forecast system,including system architecture design,function design and database design.Through related technologies,the combination model is integrated into the system,and the system is implemented by using related development technologies.更多还原