About the Customer
The client is one of the utility-scale facilities that generate solar power and feeds it into the grid, supplying a utility with energy. Today, nearly 10,000 solar projects over 1 MW are in operation or development across the U.S.
Developing utility-scale solar power is one of the fastest ways to reduce carbon emissions, and digital transformation is the way forward for utility-scale facilities.
The Challenge
With solar becoming the fastest-growing technology in the world, the challenge is to create an ML model that can accurately forecast solar irradiance a day ahead of the bidding power and manage the end-to-end development process to meet the target, timeline, and Levelized cost of electricity (LCOE).
Predicting solar irradiance and controlling battery charge and discharge rates accordingly. Using computer vision technology and Machine learning algorithms for accurate solar irradiance prediction.
The Solution
Although machine learning algorithms and their uses are not new, these algorithms have been successfully used for solar forecasting and predicting incident solar radiation. However, the result may vary depending on the particularity of the studied location as a result of the meteorological parameters’ natural variability, such as sunshine duration, land surface temperature, and visibility. This case study presents three ML algorithms for solar forecasting.
The machine learning algorithm we used for solar irradiance included a Support Vector Machine (SVM), Artificial Neural Network (ANN), and Extreme Learning Machine (ELM).
How did we use these algorithms for solar forecasting?
Support Vector Machine: We used the SVM algorithm to predict daily and mean monthly solar radiation in an arid climate. To do this, we measured local temperatures and calculated maximum sunshine duration and extraterrestrial solar radiation to train multiple models and predict solar radiation daily. To predict monthly solar radiation, we needed fewer parameters to train the best predicting models than daily prediction.
Artificial Neural Network: We used the artificial neural network for hourly solar radiation. ANN consists of different topologies. We used a Generalized Regression Neural Network (GRNN) and a Cascade-Forward Backpropagation Network (CFBN) among different ANN topologies. We trained the models considering the previous hours of the predicted day and the days having the same number of sunshine hours in the dataset. GRNN had a higher prediction efficacy compared to other networks.
Extreme Learning Machine: We used a number of hidden neurons, ambient temperatures, and historical solar radiations to train ELM models. Then it was used to predict the solar PV power predictions online. ELM models delivered slightly more accurate 24-hour-ahead solar irradiance prediction than ANN models.
After training several ML models and using them for solar forecasting, we were able to predict solar irradiance accurately. Also, we used computer simulations to analyze the hour-by-hour performance of solar irradiance.
Benefits and Outcomes
We were able to obtain accurate results showing a good agreement between measured and predicted solar radiation data.
The SVM-based models showed good accuracy and required simple parameters compared to other models.
Accurate prediction improved the planning and operation of photovoltaic systems and yielded economic advantages.
Having access to solar radiance data for local locations, engineers and architects can evaluate the effects of window size and orientation on the energy consumption of a particular building and determine the size of equipment needed for heating and air conditioning. They will be able to use this information, combined with desired levels of natural lighting and building aesthetics, to develop the final design of the building.
Utility engineers will be able to use solar irradiance data to identify whether the output of a solar electric power plant could economically and reliably help meet their expected electric demand.
We used data driven from solar irradiance prediction to identify whether or not the average daily radiation for the day or month will be sufficient to prevent the batteries from discharging for several days.
The Impact
Increased solar panels efficiency by 25%
Developed strategies to leverage the supply chain
Optimized operations and maintenance using data and analytics
Digital transformation for utility-scale solar design and construction
Why Transpire?
Transpire is a global IT consultancy company with over 15 years of experience in digital solutions, providing businesses with a roadmap to digital transformation. Utility-scale facilities can benefit from our services and embrace digital transformation. We offer several solutions for digital construction, including robotics, supply chain, and digital worker. We help manage asset operations and maintenance to optimize architecture.
