Research On Data-driven Lean Delivery Demand Prediction

With the rapid growth of the demand for delivery,people have increasingly higher requirements for the level of delivery services,and the speed of delivery has become the most concerned factor.For companies,establishing a rapid response system has become an inevitable requirement.However,faster means higher costs.From the perspective of delivery,how to maintain the balance between reducing costs and responding quickly to customer needs has become an issue that enterprises need to solve urgently.Lean delivery is an important entry point to solve this problem.Lean thoughts originate from lean production,introduces it into the logistics process,and identifies the product delivery requirements of each distribution site based on the customer needs at the end of logistics.By predict future demand,distribution site can prepare goods in advance to improve the delivery efficiency,and reducing the cost of each operation of logistics.The key to implementing lean delivery is to identify distribution needs in advance.Distribution demand identification can be refined into two aspects:demand object identification and demand volume prediction.Demand object identification refers to prospecting potential customers in advance to achieve precise marketing and insight into future needs.Demand prediction refers to forecasting site sales in advance based on historical data to optimize warehouse management and reduce distribution costs.At present,the demand management of distribution sites generally uses traditional methodssuch as empirical prediction and replenishment on schedule.There are problems such as large uncertainty and strong dependence on experts.Mining valuable information from massive data and using data-driven lean delivery demand identification can solve these problems scientifically and reasonably.The core of data-driven refinement distribution demand identification is to use the corresponding algorithm model to do potential customer mining and sales predict of data.However,the data systems currently used by enterprises are still traditional facilities based on IOE.In the face of using machine learning algorithms to process massive amounts of data,the server's single-node database cannot carry data mining and computing tasks.In response to this problem,this article introduces big data technology in a distributed environment,builds a data system independent of the business system through corresponding components,and builds two models of lean delivery demand identification based on this.The identification of delivery demand objects solves the problem of whom to send to,and mines potential customers based on external data obtained by web spider.The machine learning methods used in the mining process include constructing one-hot encoding and classification regression tree(CART).The distribution demand is identified to solve the problem of when and how much to send,and the sales predict is based on the historical sales data of the distribution site.The specific method is to build Ridge Regression and eXtreme Gradient based on feature engineering.Finally,according to the lean delivery needs of a Guangzhou LPG company,a demand object identification model and a demand prediction model were actually constructed in the data system,and the evaluation set was evaluated to select the model with the best operating effect.The model code is mainly written by python.The libraries used in the process include native XGBoost and sklearn.Through the distribution demand prediction model,the warehousing and order management of each distribution site is optimized,so that the site is always in a state of balanced supply and demand,which can meet the real-time needs of customers,and improve the efficiency of distribution and customer satisfaction.