Since their introduction, Physics-guided Neural Networks (PGNN), a novel class of hybrid models, have already been successfully implemented in several domains of application. As a result, both synergetic effects as well as physically sound models were obtained. Within the context of this thesis, for the first time, the potential of PGNNs for the identification of dynamic systems is investigated from a control engineering point of view. Various approaches are identified and first investigations are performed. By combining a recurrent neural network with a physical dynamics model, a Physics-guided Recurrent Neural Network (PGRNN) is constructed. It is demonstrated that the PGRNN generally outperforms a purely data-driven approach by a substantial margin. Further investigations indicate that the quality of the introduced dynamics model is merely of minor importance to the resulting performance benefits. Consecutively, the PGRNN is augmented by a physics-based constraint, inciting energy conserving solutions as well as a function library of nonlinear terms. The latter resulted in the full compensation of prediction error deficiencies due to inaccurate dynamic models.

Full Text (in German)

Bayesian Optimization is a powerful tool for providing solutions to complex optimization problems, e.g. as part of setup processes. It enables the automated setting of basic tuning parameters, especially when the collection of data is very time- and cost-intensive. The optimization of both processes and devices usually takes place without any understanding of the regarding systems in the form of a black box optimization. In the context of technical systems, however, prior knowledge is typically available in the form of a physical dynamics model, which could potentially result in greater efficiency of the optimization process. In order of investigating this connection in more detail, the effect of a priori knowledge in the context of the optimization process is examined on the basis of an exemplary application. It can be shown that the utilization of sufficiently accurate prior knowledge results in a significant increase in the efficiency of the Bayesian optimizer. In addition to the conditioning of the optimization problem, the dimension of the problem also plays a particularly important role. In general, however, the use of a priori knowledge should not be neglected, since the number of required iterations can already be reduced by specifying rough baselines and performance losses only occur in special cases due to the introduction of misleading information.

Full Text (in German)