Hossein Panamtash

Solar energy is one of the alternative resources for fossil fuels. Many countries have invested in the growth of solar power generation, particularly the Photovoltaic (PV) technology, with the top installed capacity being from China, Japan, Germany, and USA. As of 2016, there were 40 gigawatts of PV capacity installed in the United States, with 2016 installed capacity twice the capacity from the previous year. Around 39% of all new generation capacity in the United States came from solar energy, more than any other sources. As the solar energy gains a noticeable portion of the grid, operators face new challenges maintaining power system reliability. Achieving an accurate forecast for solar power generation is critical for system operation.

Solar power forecasting is unique in the energy forecasting domain in terms of exogenous data usage, such as sky images and Numerical Weather Prediction. These exogenous data are also employed in generation of spatio-temporal forecasts. In contrast to the common practice of using exogenous data as inputs to the forecast model, we propose a Bayesian forecasting approach where exogenous inputs are treated as random variables. We selected ambient temperature as the exogenous input, but the proposed method can be generalized to include other available inputs.

A copula-based Bayesian method is implemented to improve probabilistic solar power forecasting by capturing the joint distribution between solar power and ambient temperature. A prior forecast distribution is first obtained using different underlying point forecasting models. Parametric and empirical copulas of solar power and temperature are then developed to update the prior distribution to the posterior forecast distribution. A public solar power database is used to demonstrate effectiveness of the proposed method. Numerical results show that the copula-based Bayesian method outperforms the forecasting method that directly uses temperature as a feature. The Bayesian method is also compared with persistent models and shows improved performance.

Considering the coherency among multiple PV sites, a reconciliation is applied using a copula-based bottom-up method or proportion-based top-down method. Numerical results show that the proposed methods efficiently produce accurate and coherent probabilistic solar power forecasts.

Dr. Inalvis Alvarez Fernandez

With the high penetration of inverter-based distributed energy resources (DERs), network situational awareness is a challenge for both transmission and distribution system operators and planners. It is becoming increasingly important to equip transmission planners with the visibility of DER dynamic performance in distribution systems. Not having visibility of the disconnection of DERs with the occurrence of transmission events, could result in an erroneous view of the stability of bulk power system. This presentation will introduce the parameterization of the aggregated distributed energy resource (DER_A) model (able to represent the aggregated dynamic response of inverter-based DERs in distribution circuits), and machine learning application leveraged from the data generated during the parameterization process.