People

Distributionally Robust Model Predictive Control for Virtual Power Plants

Renewable Energy Sources (RES), such as solar and wind, increasingly contribute to the global energy production and mark an important pillar of the modern energy system. RES come with numerous advantages such as the provision of clean as well as broadly available energy production capacities. However, one major factor limiting the widespread adoption and penetration is the inherent underlying uncertainty of RES. Here, the concept of virtual power plants (VPP) can assist. A VPP is a collection of small-scale energy resources that, aggregated together and coordinated with grid operations, can provide the same kind of reliability and economic value as traditional power plants. To further increase the penetration and adoption of solar energy plants, intelligent strategies for VPPs are required to deal with the uncertainty of RES.

This PhD thesis addresses the uncertainty within the decision-making process of virtual power plants. In this context, uncertainties can derive from various sources such as the sunlight-dependent energy production, unpredictable (stochastic) market prices and variable energy demands. To tackle this problem, probabilistic forecasting is combined with strategies from control engineering. These control engineering strategies, particularly (Stochastic) Model Predictive Control and Distributionally Robust Optimization provide a promising approach in the context of VPPs as they allow i) to make informed decisions under uncertainty, ii) to hedge against quantifiable risks and iii) the end user to trade-off between conservatism and performance. By integrating these strategies, this research aims to enhance the decision-making capabilities of virtual power plants, ultimately leading to a more reliable and efficient integration of RES into the power grid.

Key Words: Virtual Power Plants, Distributionally Robust Optimization, Model Predictive Control, Uncertainty Quantification

Connection on terms as a facilitator for colocalization of power production and consumption.

The power grid in Norway faces numerous challenges regarding power distribution. This master's thesis will examine the experiences of distribution system operators (DSOs) and customers with grid connections under specific terms. Furthermore, it will analyze how a power line in the grid will react when both a solar plant and a data center are connected under these terms in the same area. The aim is to provide insights into how local production and consumption can potentially reduce the need to wait for the DSO to expand the grid.

Mitigating overvoltage in the Norwegian distribution grid due to high PV penetration

This thesis analyzes how active power curtailment and battery storage can mitigate overvoltage in the Norwegian distribution network due to high PV penetration. Power flow simulations in Pandapower are conducted on a reference grid to evaluate different mitigation strategies and their impact on voltage regulation

Modelling of temperature for mono- and bifacial photovoltaics in Nordic conditions

The efficiency of a photovoltaic (PV) module is highly dependent on the operating temperature of the PV cells. Measurements of the operating temperature is not available in the design-phase of a PV system, and in some cases it is not even measured for operating systems. In these cases, we must rely on models which estimates the cell temperature from measurements of ambient temperature, irradiance and wind speed to assess the performance of the system. These models typically rely on a set of empirical coefficients which have been determined for locations that experience very different weather conditions than what is common in the Nordics. This master’s thesis is focused on assessing the performance of existing temperature models and carry out further model development to enhance the accuracy of temperature modelling of PV systems operating in Nordic conditions

Wave-induced loss for floating photovoltaics

Floating photovoltaics (FPV) is a promising solution for renewable energy generation in areas where land is limited or costly. For FPV systems, waves will, if present, induce movements in the floating structures that the PV panels are mounted on. These movements will cause changes in the incident irradiance on the panels, giving rise to wave-induced losses (WIL). This master’s thesis will expand on previous efforts to simulate WIL by including time series of wave, wind and irradiance data and wave models to improve the model and account for potential correlations between power production and sea states.

Improving Solar PV Forecasting for New Installations: A Comparative Study of Physical vs. Transfer Learning Methods

This thesis investigates how to accurately forecast the energy yield of newly installed solar PV systems that lack historical data. First, a physical model using meteorological inputs provides a baseline forecast without needing operational history. Then, a transfer-learning approach is explored to see if data-driven methods can improve forecast accuracy under data-scarce conditions. By comparing these two approaches, the study aims to highlight trade-offs—including complexity, data requirements, and computational cost—and provide guidance for practitioners seeking reliable PV forecasting solutions for new installations.

Machine Learning Correction of Horizontal Irradiance for Solar Production Forecasting

This thesis examines the relationship between global horizontal irradiance forecasts and actual
solar production from tilted photovoltaic installations, focusing on the dynamic self-shading of
solar roof tiles. Existing approaches for solar roof tiles rely on manual adjustments that fail to
capture real-time variations in shading. By integrating numerical weather prediction data with
environmental, geometric, and overhead mapping information, this research systematically
incorporates self-shading eƯects to enhance forecasting accuracy and support optimized
energy management.