Testing a model predictive control algorithm for a PV-CHP hybrid system on a laboratory test-bench.

• A test-bench for PV-CHP hybrid system has been build. • Input profiles (PV and load) are emulated with accuracies between 0.1% and 14%. • The CHP and gas burner should have modulated power supply for higher efficiency. • Hardware and metering generates additional deviations from forecast schedules. • A battery needs a faster control added to the MPC for higher accuracy. In order to reduce the global warming to less than two degrees, a large increase in renewable energy resources like photovoltaic systems is necessary. A business case based on feed-in tariffs does not exist in every countries and self-consumption is limited due to the unsteady nature of solar radiation on earth. A combination with other systems such as batteries, heat-pumps or even combined heat and power plants can enhance the use of generated power by photovoltaic systems, particularly in private households and small businesses. Rule-based controllers and optimization algorithms (model predictive control) can both realise the efficient operation of a photovoltaic system in combination with storage systems and a combined heat and power plant. However, different controllers and energy management systems have hitherto only been compared theoretically. A comparison of such controllers in a real, controllable hardware environment has not yet been carried out. In this study, a test-bench is introduced to test different control algorithms for photovoltaic systems in combination with storage systems and a micro- combined heat and power plant. The operation has been tested for a one day period. Key performance parameters have been derived and compared for a rule-based control, an optimized control and simulation results including a forecast. The results show that the operational costs can be reduced by 7.3% for the chosen test-period using the optimized algorithm in the laboratory compared to the same system with a rule-based control. The results also indicate that even under perfect forecast conditions the hardware, metering and energy management cause latencies and inaccuracies leading to deviations, which are not accounted for in simulations. Hence, the accuracy of the forecast methods need not be higher than the deviations introduced by the hardware. These deviations often lead to unwanted charging and discharging events of the battery. A faster way of processing data and a second order or low level control is needed for short term reaction and higher efficiency. [ABSTRACT FROM AUTHOR]