Optimization Potential for Adapting Symbolic Regression Models Applied to Energy Flow Control

In recent publications, symbolic regression models developed using genetic programming have been introduced for the optimization of the energy flows in households with photovoltaic (PV) systems that include a battery. These models pursue the objective of minimizing the energy costs of a household. The globally optimal solution for energy flow control is determined in this paper using dynamic programming, in order to provide a performance assessment of the resulting energy costs when the symbolic regression model is applied. In the performance assessment, the resulting energy costs after simulation using energy tariffs and power measurements for two different households are compared. The results show that the symbolic regression models achieve cost values close to the globally optimal solution for the data they have been trained with. For different time spans, the remaining optimization potential takes values of up to approximately ten percent. Moreover, the symbolic regression models have been applied to energy flow control of a new system, for which the original model had not been trained. The optimization potential on this new system makes up more than thirty percent.