**3. Reduction of the Solution Effort**

In order to realize a stable convergence of the optimization environment with a solution space with sufficiently many degrees of freedom for the variation of the machine geometry, many iterations and large populations are required due to the used stochastic optimization methods. This yields a long computation time, particularly for machine models with a high number of degrees of freedom. Depending on the required level of detail of the optimization problem, which influences the model resulting from the model selection methodology, the problem may thus not be solvable within a few days. A reduction of the solution effort is desirable. The solution effort can be reduced by a preselection of the individuals to be simulated. In this work, the preselection is realized by estimating the fitness values of the individuals using an ANN . This estimates for each individual those decision parameters of the optimization environment that require a time-intensive operating map simulation, such as the mean losses over a given drive cycle. Geometry-dependent decision parameters, such as the volume or mass of the machine, on the other hand, are computed in a regular manner, so that a fast estimate of an individual's fitness can be obtained from the estimate of the ANN as well as from the individual geometry parameters. Based on these estimates, solutions are discarded and only those individuals promising better fitness than the current minimum are simulated. In the following, the ANN and its application will be discussed.
