Scientists at the King Fahd University of Petroleum and Minerals in Saudi Arabia have developed an innovative approach to identify optimal locations for utility-scale wind and solar power projects. By incorporating distributionally robust optimization (DRO) and deep learning features for solar radiation forecast, this new model offers a more accurate and realistic depiction of uncertain parameters compared to conventional stochastic methods.
The DRO technique takes into account partial distributional information and evaluates solutions based on the worst-case scenario among a range of probability distributions of uncertain variables. By doing so, it provides a less conservative estimation of forecast uncertainties for wind turbines and PV system production when compared to robust optimization models.
The model also leverages a simplified version of the long short-term memory (LSTM) model known as the Gated Recurrent Unit (GRU). This recurrent neural network is capable of learning order dependence in sequence prediction problems, making it ideal for time series and sequence data analysis.
In their study, the researchers applied their methodology to potential wind and PV project locations in Saudi Arabia. Using historical weather data spanning from 2010 to 2020, they evaluated 28 sites based on various data entries such as temperature, wind speed, solar irradiation, and more. The results showcased that the DRO model was able to accurately forecast trends closer to the actual recorded data compared to an older robust optimization model.
By embracing this advanced approach, the scientists believe they have achieved a more realistic modeling of uncertain variables, surpassing the limitations of deterministic or robust models. Their findings also highlighted the importance of avoiding candidate sites with low wind speeds or those located in close proximity to each other, based on the map of mean wind speed and global horizontal solar irradiation in Saudi Arabia.
This groundbreaking research was documented in the paper titled “Multi-objective distributionally robust approach for optimal location of renewable energy sources,” which can be found in the Alexandria Engineering Journal.
FAQ
What is distributionally robust optimization (DRO)?
Distributionally robust optimization is an optimization technique that considers the worst-case scenario among a family of probability distributions of uncertain variables to evaluate optimal solutions.
What is deep learning and how does it relate to this study?
Deep learning refers to a subset of machine learning techniques that involve the use of neural networks with multiple layers. In this study, deep learning was employed to forecast solar radiation, using a simplified version of the long short-term memory (LSTM) model called the Gated Recurrent Unit (GRU).
Why is this new method significant?
This new method offers a more realistic modeling of uncertain parameters for wind and solar project siting compared to conventional stochastic approaches. By incorporating DRO and deep learning features, it provides more accurate forecasts and optimal locations for renewable energy sources.
How did the researchers apply their methodology?
The researchers analyzed 28 potential locations in Saudi Arabia using historical weather data collected between 2010 and 2020. They considered various data entries such as temperature, wind speed, solar irradiation, and more to identify sites with maximum power output and minimal production variance.