CCPS Talks: Thesis Defenses

The following presentations will take place on the above date:

14:00: Patrick Brumund (Master Thesis)
“Stochastic Nonlinear Model Predictive Control of Offshore Energy Systems using Gaussian Processes” Presentation language: English
Supervisor: Maik Pfefferkorn, M.Sc.

Guests are very welcome.

Room S3|10 406A

or via Zoom: https://ovgu.zoom-x.de/j/68502552967

Passcode: 222031

Abstract:
For the model predictive control of an offshore hybrid power system consisting of wind turbines, a gas turbine battery storage, and power consumers, predictions of the expected wind power are required. As numerical predictions of the wind speed are subject to uncertainty in wind power, a probabilistic model to estimate the uncertainty is needed. For this, two different models based on Gaussian processes are tested, which estimate the prediction error based on measurements and numerical weather predictions. One of these models presents an approach to modeling the wind speed as a non-stationary time series using two nested Gaussian processes, while the other model utilizes separate Gaussian processes for different horizons. Both models are able to improve the accuracy of the forecast and provide intervals to estimate the uncertainty. The models are integrated into the stochastic nonlinear model predictive control of the power system, which controls the output of the gas turbine and the battery to satisfy a given power demand. Chance constraints are set up to include the uncertainty, for comparison a scenario-tree based approach is implemented as well. Through simulations, it is shown that both controllers are able to significantly improve the capability of the controller to satisfy the power demand, and using the Gaussian process to predict the uncertainty leads to better results than assuming a static Gaussian distribution of the uncertainty. Finally, flexible loads are considered by allowing the generated energy to vary within a given band, increasing the flexibility of the system.