Today, the value network has become the dominant value creation structure in the area of production. For manufacturing companies in such networks the order management is a central task which faces an environment of growing challenges. On the one hand, shorter product life cycles and increasing individualisation are putting companies under pressure. In this context, multi-variant production has gained in importance. On the other hand, the probability and impact of disruptions are increasing, especially in networks, which challenges the capabilities of production systems. A promising approach in this context is the use of dynamic pricing via a continuous price-lead-time function. By using a dynamic pricing it would be possible to bring the demands in the network in line with the capabilities of the production system. The idea is to shift demand peaks and to supply customers according to their individual preferences. This way, the overall profit can be increased although the capacities remain constant.
To achieve and examine the described effects, the aim of this thesis is to develop a
methodology for dynamic pricing for the make to order and assemble to order produc-
tion. In the first step, key performance indicators and an objective function are estab-
lished. Second, an actor critic reinforcement learning method is motivated, since rein-
forcement learning has already proven its potential to handle complex optimisation
tasks efficiently in comparable domains. For the training and the later testing of the
resulting learning agent, a simulation model is developed as an environment and a procedure for the execution and evaluation of the experiments is worked out.
The approach has been discussed and applied within the BMBF research projects
ReKoNeT and BaSys4SupplyQ. The results from two use cases show that the approach
is able to achieve significantly higher profits with the same capacity and to smooth demand peaks to a certain extent. At the same time, the results indicate that saving on capacities does not seem to be a dominant strategy. Instead, it is reasonable to maximise service for selected customers in order to justify higher margins. This reflects the application-specific consideration of various targets, such as service, capacity costs and margin. Thus, the results also reveal the central importance of aligning the parameterisation of the method with the strategy and goals of the applying company