This thesis aims to evaluate the use of Nonlinear~Model~Predictive~Control (NMPC) as a control concept for production planning and balance control, for a fictitious combined power and district heating production portfolio, in a component-based modeling context. A comparison to Linear~Model~Predictive~Control (LMPC) is used as basis for this evaluation. The desire to use NMPC is due to the desire of eliminating the need for linearization; potentially loosing valuable information and minimizing the need for manual preconditioning labor.

An optimization-friendly first-principle nonlinear model of the production portfolio and consumers is constructed in Modelica, heavily relying on the component-based capabilities of the language. This model is linearized; thus both a nonlinear model usable in NMPC and a linear model usable in LMPC is obtained. Through simulations, the linear model was found comparable to the nonlinear model -- but with deviations when using the accumulator included in the production portfolio.

The MPC control scheme is designed around an economical cost function, derived through basic economical considerations of the use of production units and a simplified power market model. The resulting optimal control problem is shared between both NMPC and LMPC; the only difference being the model employed in the constraints enforcing system dynamics. To provide full state information an Extended Kalman Filter (EKF) is designed, under the assumption that consumer states are not measurable.

The optimal control problems are solved using JModelica.org, a framework that allows optimization directly on Modelica models. Thus, a simulation framework is designed and implemented on top of JModelica.org, allowing for simulations with both NMPC and LMPC.

Simulation studies show, that NMPC uses the accumulator more actively. The extensive use of the accumulator by NMPC, is performance-wise better, considering long simulations with historical power prices and ambient temperatures. The method of using first-principle nonlinear models directly in MPC is thus, at least on a conceptual level, highly encouraged.