Forecast for vehicle components demand and order planning

Challenge

• Development and operation of a system to support the planning process of vehicle properties (equipment characteristics)
• Weekly provision of a model to forecast the demand quantities on component level with a 12-month time horizon
• The model is to consistently integrate the following existing data:
  • Rules defined by sales and logistics
  • History of vehicles built in the past
  • Specifications based on capacities, orders, forecasts etc.

Solution

Markov networks integrate structural and probability information: 

• The structure of the Markov network is determined based on logic and sales rules
• The initial distribution of the network is “learned” based on the history of built vehicles (dependencies, relevance)
• Planning specifications are consistently incorporated through changes to the probability distribution

Benefit

• Forecasting demand quantities on component level is possible with high quality
• Component requirement bottlenecks are detected at an early stage
• Inconsistencies within the rules or between rules and specifications are detected
• Inconsistencies in the planer specifications are automatically compensated for and can be analyzed
• Order planning can be based on a consistent model