You will develop short-term PV forecasting / nowcasting algorithms to improve operational stability of power systems with a high share of renewables
We will soon reach a point where solar energy will become an abundant source of clean energy in modern power systems, driven by decreasing costs and increasing performance of modern, integrated PV technology. However, due to the intermittent and non-dispatchable nature of PV generation, this will present additional challenges in terms of system stability. Integrating and operating distributed PV in a system-efficient way will be the next big challenge to evolve to a (near) 100% renewable energy system. The solution lies in a more flexible and decentralized system, where local energy management will be key. A key building block to improve the operational stability of power systems dominated by renewables will be the availability of (very) short-term and fine-grain forecasts (also known as nowcasts), able to predict high ramp-rates caused by rapid changes in weather conditions due to cloud movement. The use of such advanced forecasting techniques for proactive power plant control will allow to accommodate more PV in next generation grid and power system architectures.
The Ph.D. candidate will develop and validate a coupled irradiance and PV generation forecast model based on a combination of physics-based PV system models and artificial intelligence algorithms. Together with global horizontal (GHI) and direct normal irradiance (DNI) data, sky images from (low-cost) sky cameras form a valuable source of input data for constructing the angular distribution of sky radiance which is needed to calculate the irradiance in the plane of the module. When transposing irradiance onto the plane of a solar module solely based on GHI and DNI, the effect of clouds on the anisotropy of sky radiance cannot be captured. Sky images, in turn, can be used to correct the standard distributions used in irradiance transposition. Under frequently cloudy conditions typical of, e.g., temperate oceanic sites, the sky radiance distribution often varies widely and rapidly. Therefore, the in-plane irradiance can change notably more than the changes in GHI or in DNI might suggest. Therefore, one of the objectives of the Ph.D. project should be to develop, test, and implement computationally efficient methods to correct the standard sky irradiance distributions using the sky images and basic irradiance parameters. Image processing and machine learning techniques will be required to detect and classify cloud patterns and motion trajectories from the sky camera input, and translate this into usable information for the forecasting model. Furthermore, accurate but fast transient thermal models will be needed to capture the impact of temperature fluctuations on power generation during variable weather conditions. The candidate will develop, test, and implement computationally efficient methods for fast execution of the developed algorithms, in order to deliver highly accurate PV forecasts in a real-time operational context.
The candidate will have the unique opportunity to test the developed algorithms in real life conditions with our industrial partners and use the extensive Energyville infrastructure.
Required background: Physicist, electrical or software engineer with a strong background in modelling and simulation of energy systems in Matlab and/or Python. Knowledge of machine learning, AI or image processing is an asset.
Type of work: 40% modelling, 30% implementation of models and simulations, 30% experimental validation
Supervisor: Jef Poortmans
Daily advisor: Joris Lemmens, Arttu Tuomiranta