Dynamic optimal power flow control with simulation-based evolutionary policy-function approximation

In nowadays operations research, dynamic optimisation problems are a central and challenging research topic. Especially in complex real-world systems such as electric power grids, dynamic problems occur where robust solutions need to be found that enable (near-) optimal control over time in volatile as well as uncertain power grid operation. The authors of this work identified the application of policy-function approximation for suchlike problems. Here, an analytic function is aimed to be found, that takes a state of the dynamic system as input and directly derives control actions that lead to approximate optimal operation at runtime, without the need of doing imbedded optimisation. Applying this approach to two popular and scientifically challenging problem classes in power grids research, this work aims at providing a general view on this optimisation concept. Therefore, a dynamic generation unit control task will be experimentally treated on the one hand, while dynamic load control under uncertainty with electric vehicles represents the second use case. Both applications are related to dynamic stochastic optimal power flow problems. Hence, this paper shows the successful application of policy-function approximation to this problem domain.