Digital Twin Technology Comes to Solar

What is a Digital Twin?

A Digital copy or (Twin) of a physical object or business process has become an important tool as companies try to become more efficient. Originally used for complex systems like aircraft manufacturing, (as the cost of collecting data and sensors has fallen, while the ability of AI to be more capable of creating insights has risen), Digital Twin technology is spreading to other industries including now – solar energy.

The Digital “Twin” uses both historic and current behavior using sensors, and data combined with modeling and simulation. For a solar power plant, it allows us to go from reacting to the state of a module, to predicting the performance we need to model.   Traditional software models are limited as they don’t take into account factors like precise location, environmental factors, aging of parts, damage history, and individual tolerances of components or processes used in their manufacture.

Digital Twins however, use a combination of sensor data and machine learning to provide individualized algorithms, which are then, able to model actual performance of a module. Next, Machine learning, enables the Digital “twin” to model history and individual tolerances.  Added to that, sensor data enables the digital twin to model actual operating conditions for the module, creating an intelligent algorithm which keeps improving with time.

Quadrical’s Proprietary Digital Twin Technology

Quadrical’s expertise in the areas of Prediction AI and Anomaly Detection AI, underly the technology. This allows us to create accurate models which predict the behaviour of individual modules (panel strings, SCBs, or inverters) in a solar plant. In particular, our predictive digital twin models can predict the actual output of an operating solar plant, and our anomaly focused Digital Twin models can detect and predict when an individual module may be about to fail. Combined, these can make a 4% difference, in the operating cost of a plant whether small or big.

The trend to current large-scale plants, add additional complexity. A 20MW solar plant contains about 100,000 panels typically organized into about 2,000 strings. With larger plants like a 2GW one, we’re looking at 100X.  At this scale, even RealTime monitoring makes careful management of data streams extremely complicated.  Combining them into an AI-ready Data Lake, (necessary for RealTime insight) also becomes a significant engineering exercise. Lastly,  maintaining individual digital twins for all panels and strings is quite a challenging computing exercise.

Again, being able to build and scale these systems requires real experience and expertise. Quadrical RealTime, AI-ready data lake technology is behind Quadrical Ai’s 5-in1, Management Platform for solar plants. The combination of this Data Lake providing RealTime data, and our extensive expertise, in designing and managing large scale data engineering platforms combine to provide a solution which will scale with the growth in the size and complexity of each string going up to the totality of all the solar plants in your portfolio.

With AI insights from Dr. Hugh Hind, CTO at Quadrical Ai

Author: Kitty Chachra, CRO at Quadrical Ai

Originally published on September 6, 2020