Industrial Network Topology Generation with Genetic Algorithms

Christoph Fischer; Maximilian Berndt; Dennis Krummacker; Janis Zemitis; Daniel Fraunholz; Hans Dieter Schotten

In: YEF ECE. IEEE Young Engineers Forum (ECE-2020), IEEE, 2020.


Networks of industrial plants are engineered manually as of today. This covers the configuration of a network as well as the cabling and the design of the network topology itself. Hereby, usually multiple subnetworks using different transmission technologies are involved. In our proposed solution the process of deriving a feasible network topology from communication demands for a given set of interacting distributed applications is automated. This describes a Constraint Satisfaction Problem (CSP) of which this paper approaches a solution with a Genetic Algorithm (GA). The strategy is generating an own search space for a solution, extending it iteratively based on knowledge ascertained in previous iterations and finally selecting the best solution. Presented is the description of an algorithm, an encoding format for network topologies as well as a methodology for evaluating solutions. This evaluation includes Key Performance Indicators and how to describe and measure them, e.g. to calculate a fitness score based on the number of hops. Additionally, an explicit feasibility check is introduced to prevent the selection of solutions that might have a high fitness score but are unable to serve the use case requirements, e.g. due to an incomplete interconnection graph. Finally, an evaluation of the developed algorithm in terms of quality of a found solution and performance of execution is shown.


Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence