Shulou(Shulou.com)06/03 Report--

In the era of artificial intelligence and big data, more and more cloud data and more and more intelligent models begin to assist people to make all kinds of optimal decisions, from operational efficiency, cost savings, optimal allocation and other aspects to achieve cost reduction and efficiency. Further improve business efficiency. JD.com, Meituan, Didi, Shunfeng and many other well-known manufacturers, through operational optimization platform, transform their supply chain, intelligent dispatch, ride matching, intelligent sorting and so on.

There are many links in the retail industry, from production to warehouse to offline stores in the supply chain, even if the demand for the final product is very stable, the bullwhip effect often occurs. The reason is that when each node in the supply chain makes production or supply decisions only according to its adjacent demand information, the unauthenticity of demand information will be magnified step by step along the countercurrent of the supply chain. More accurate demand forecasting is only a step of decision-making, as well as inventory order decision, price fluctuation decision, shortage game decision and other decision-making problems in the process of business changes such as sales volume and process management. Bullwhip effect shows that even if the prediction is more accurate, if the subsequent process decision-making process is not effectively managed, the benefits of accurate prediction will be offset by the loss caused by unreasonable safety inventory.

The decision-making process of many enterprises often rely too much on the personal experience of the corresponding positions, and on the one hand, the information obtained by employees is incomplete, on the other hand, there are a lot of repeated work of estimation and comparison in the decision-making process, which leads to the inefficiency and instability of the output of decision-making schemes. Employees' repeated work limits personal growth, and enterprises consume human resources and valuable decision-making time. In order to meet the needs of rapid replication, centralized and efficient decision-making, rapid information feedback and planning decision effect prediction of excellent planning decision-making methods in enterprises, singularity cloud introduces the application of decision engine on the basis of data center.

Singular point cloud decision engine

Perfect data collection and management, extracting information from data and understanding the laws of things can not release the great value of data. In order to produce practical value, data must really improve the quality of decision-making and realize the automation, flow and standardization of decision-making.

After completing the development of the data center for customers, the intelligent decision-making service based on data assets is provided. According to the different scenarios, the decision-making methods such as maximum income expectation decision, maximum and minimum income decision, minimum and maximum regret value decision, Markov game decision are selected, and the decision objectives are solved by combining operation optimization algorithm and reinforcement learning.

In real life, there are many problems can be described as optimization problems, and then use the knowledge of operational optimization to solve them.

The two core steps of comparison are modeling and solve. According to the mature software toolkits (cplex, gurobi, glpk,lpsolve, scip...), the singularity cloud gives the baseline solution of the classical operation optimization problem, which can be put into trial operation quickly. In the process of operation, according to the core index of result evaluation, combined with operation optimization algorithm and reinforcement learning, the algorithm and solving process are further optimized. so that the planning decision-making model, solving process, evaluation system can meet the planning and decision-making process needed by customer business development.

The singularity cloud prediction engine takes demand forecasting as the starting point, while the decision engine pays attention to the planning decision efficiency and decision quality in the implementation process. For the seasonal impact of commodities and the instability of market supply, we need to make reasonable follow-up decisions on replenishment; after the completion of delivery, the replenishment from warehouse to store and the transfer from store to store still require a lot of work from customer staff to generate replenishment and transfer plans for each phase. In order to complete the delivery, replenishment and transfer of goods at the same time to ensure the loose and tight balance of the market, we also need a reasonable plan.

The core of planning decision is the allocation of inventory, including warehouse inventory, in-transit inventory, store inventory and so on. Inventory management is to manage and control all kinds of goods, finished goods and other resources in the whole process of production and operation of manufacturing or service industry, so as to keep its reserves at an economic and reasonable level. Make use of historical data to realize real-time updated demand forecast and provide replenishment suggestions for enterprises. Reasonable design of storage shelf placement, commodity area division, high and low shelf placement, optimal path deployment in and out of storage, etc., can save huge costs and a lot of human labor costs for enterprises. It can reduce the occupation of funds, improve inventory turnover, improve automatic management, improve the utilization rate of personnel and equipment, and reduce the inventory burden.

Operational optimization to calculate the optimal dispatching strategy

Qidianyun, a major fashion customer, has thousands of offline stores, each with hundreds of sku. Using historical data to predict the future sales of each sku in each store, some stores are bound to have insufficient inventory, while some stores have the problem of inventory overstocking. Then the overall gross profit of the company will be improved by transferring the goods from overstocked stores to those with insufficient inventory. The logistics costs between stores are different, and the types of goods out of stock and backlog are also different. the optimal transfer strategy is calculated by the method of mixed integer programming in operation planning. The model of mixed integer programming can be abstractly modeled as follows:

Through the modeling and solution of the transfer and replenishment process, the corresponding repetitive workload of customer business personnel is reduced by 80%, and the planning and decision-making time is reduced by three days. Business people can see more data basis when making decisions, and the input and output of planning decisions are clear and efficient.

Conclusion

In the customer's business process, a large number of links will involve decision-making problems. How to efficiently use data to drive decision-making is the core of singularity cloud decision engine. In the last StartDT AI Lab column, we mentioned the importance of accurate demand forecasting, but in practice, forecasting is always biased and uncertain, and decisions need to be made in the case of multi-level uncertainty caused by different links. Combine demand forecasting and decision engine to make data decision more intelligent. In the future, we will continue to work in the field of demand forecasting and decision engine to help customers create greater value.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.