Abstract
Integration of wind and solar power generation into power systems has grown significantly over the past decade. While system operators have managed the variable and non-dispatchable nature of these resources at current levels, their large-scale integration would pose new challenges in power systems operation procedures. In particular, net load, which is the conventional load minus the non-dispatchable generation, would significantly deviate from load as the penetration level increases. The main non-dispatchable sources of electricity generation are utility-scale and small- scale behind-the-meter wind and solar power.
This thesis focuses on characteristics of the net load in power systems when a large amount of wind and solar power generation is integrated into the grid. Historical and simulated net load scenarios are analyzed from a variety of perspectives. It also evaluates the effect of wind integration level on the net load forecasting accuracy. Additionally, the thesis proposes two methodologies to estimate invisible solar power generation using the data from a limited number of sites. The first approach uses data mining tools to identify the critical sites for continuous monitoring. The second approach models the uncertainties of the invisible solar power production using fuzzy arithmetic applied to publicly available production data. This is the first study using public data in the field.
Numerical simulations are provided based on California, Alberta, and Ireland power systems. The results show the importance of understanding the changes related to significant wind and solar power generation. New morning downward and an increased level of afternoon upward net load ramps were found compared to the conventional load. The net load was also found to be more volatile compared to the load.
In addition, numerical results prove the efficiency and accuracy of the proposed methodologies for the invisible solar power generation estimation. The results showed that continuous monitoring of a small number of sites is enough for accurate estimations. Moreover, the fuzzy model is capable of producing accurate estimations by using public data of only 20 sites per subregion.
