Research On Data-driven Supply Chain Performance Optimization

In recent years,with the development of globalization of the supply chain,more and more companies have begun to purchase raw materials and sell products all over the world.The number of members in the supply chain has increased a lot of partners across continents and time zones.These changes have made the structure and management of the supply chain extremely complicated.In such a global market full of uncertainty,competitive pressure and turbulence,current supply chain is vulnerable to prevent the risk of supply disruption and demand fluctuations.It has become the important issues among enterprises that how to optimize supply chain performance so that it can effectively reduce supply disruption and adapt to demand fluctuations.At present,enterprises have been inseparable from the support of big data technology when making supply chain management strategies.Big data has revolutionary changed the supply chain management.It has transformed traditional supply chain competition into a competition based on "data driven".Learning algorithms,as a kind of data-driven technology,such as k-nearest neighbor(k-NN),logistic regression(LR),artificial neural network(ANN)and some integrated algorithms,have become more and more popular.They can help us obtain effective information from a large amount of data and assist managers in making decisions.Therefore,exploring a kind of data-driven models to optimize supply chain performance has theoretical and practical significance.Based on the relevant literature review,we summarize the researches on the optimization of supply chain performance and propose a data-driven model framework.Subsequently,we choose resilient supplier selection and demand forecasting as the two core points of optimizing supply chain performance.At the same time,we combine with the data-driven model framework proposed in this article to solve the corresponding issues:(i)Combining simulation technology and machine learning algorithms,we develop a hybrid technique and examine its applications to data-driven decision-making support in resilient supplier selection.We consider on-time delivery as an indicator for supplier reliability.The results show that if the combination of machine learning algorithms and simulation is used properly,it can improve the reliability of supplier and reduce the risk of supply disruption.The premise of this model is the data-oriented and short-term cooperation relationship between customers and suppliers.(ii)Combining real data and deep learning algorithms,we develop a combined model based on LSTM and Light GBM and examine its applications to data-driven decision-making support in demand forecasting.Through the visual analysis of sales data of a large foreign chain store,we summarize methods of judging the effective features of data,data cleaning and constructing feature engineering;then,we used the combined model based on LSTM and Light GBM to predict representative product sales.Experiments show that the combined model can forecast the sales of commodities accurately and effectively,and the predicted results are more interpretable,which is of great significance to improve the production and operation mode,daily management,price management and precise marketing of enterprise supply chain.