Division Supply Chain Services at Fraunhofer IIS
Division Supply Chain Services at Fraunhofer IIS

AI-based steel price forecast for efficient purchasing strategies

Price fluctuations as a challenge in steel trading

Raw material prices are sometimes subject to major fluctuations. There are various types of steel that are relevant for different sectors such as the construction and mechanical engineering industries. As the steels are manufactured using different production methods, different fluctuations can occur. These can be caused by price fluctuations in the raw materials, but also by price fluctuations in energy sources, market speculation and the like. It is difficult for planners to combine the complex interrelationships for the different types of steel and to correctly assess price trends - especially over a longer forecasting horizon. This is particularly important for wholesalers, as a reduction in purchase prices in the percentage range can mean major savings.

Support for planners through price forecasts

 

Artificial intelligence methods, and neural networks in particular, are particularly suitable for modelling complex relationships between several influencing variables. For this reason, a Historical Consistent Neural Network is used in the project to forecast the prices of various relevant steel grades. The aim is to support buyers with predicted prices that are as close as possible to the actual price. The forecast can provide further helpful insights. The model provides our customers' purchasing departments with an estimate of when prices will be at their lowest in the forecast period and whether the model is forecasting rising or falling prices. If the purchasing experts agree with the model forecast, the model confirms their opinion and thus increases certainty. If the model delivers different results, the analysis provides a deeper understanding of the company's steel market. This allows the timing of raw material procurement to be improved, which harbours potential for savings. The diagram shows an example of the forecasting process and the subsequent action aid for the purchasing departments. Various influencing variables are passed to the model, which then creates a forecast. For example, falling prices are forecast. Later procurement reduces the price and savings can be realised in purchasing.

This might also interest you

Data Science

Read more about methods, processes, algorithms and systems for extracting insights and patterns from data to generate information and derive recommendations for action here.

Group Data Science