Artificial Intelligence and Photovoltaics

May 1, 2019

Artificial intelligence is not a fashion that disappears in a few months. The use of artificial intelligence changes the entire industries. The question is which applications are possible, useful and profitable. At the Intersolar Europe on 15 May at 3 p.m. we would like to discuss with the physicist Nicolas Hoyer the meaning of artificial intelligence for the photovoltaic sector. In the following blog post, Nicolas Hoyer explains as a guest author the characteristics of artificial intelligence and how its use in photovoltaics can provide an additional value.

 

What is artificial intelligence?

The term artificial intelligence is already part of the active language use of our society. Media, trade unions, governments, househusbands and schoolchildren speak of AI, AI or Deep Learning. The IT industry also speaks of machine learning. How are these terms related? Broadly speaking, machine learning is the application of artificial intelligence. In machine learning, artificial intelligence is an algorithm that learns a certain behaviour from data. A good example of machine learning is when a machine learns to recognise a person’s handwriting. This machine learning works by supervised learning. The machine is taught letters in image form and their meaning as letters in pure form. The opposite of supervised learning is unsupervised learning. Thereby, machines can detect anomalies or clusters by “observation” in large data sets and do not have to be trained explicitly with prepared data.

In addition, there are other techniques. Reinforcement learning is widely known. The Alpha Go and Open AI systems, which regularly appear in the media, use an algorithm that learns by reinforcement. The machine experiences this reinforcement, for example, by winning a Go game or by taking the first parts of a covered distance. All these machine learning methods have one thing in common: Firstly, they are not based on strict if-then-rule rules. Secondly, in machine learning, the individual software modules are combined to form networks, which enables the system to learn. The use of deep and larger neural networks is defined as deep learning. For such a system to work sufficiently, it is necessary to learn from data. The more, the better.

 

Applications of Artificial Intelligence in Photovoltaics

How do machine learning methods support the construction of a solar park? And what kind of data can solar park operators use in the future? Generally speaking, “unsupervised” anomalies can be detected from the data already available in the park. These could be further analysed and named. To give an example, a detected anomaly would be the work of a cleaning team while cleaning the modules of a tracking system. The machine would recognise and name the pattern of a cleaning based on the approach angle which is characteristic for the cleaning process. The reduced energy yield could be explained automatically.

This basic procedure can easily be applied to more complex processes, such as the evaluation of recorded energy from electric motors or the performance of strings. After a technical analysis, these anomalies can be clustered and automatically detected after another supervised learning process. If this recognition works, anomalies can not only be logged as “something is conspicuous”, but also with their learned classification. A prolonged loss of performance of just one string may not be explained by cloud shadowing. In the near future, drones can produce thermographic images to identify pollutions or property security. Using supervised learning, pedestrians can be easily distinguished from birds or rolling bushes. Or people who illegally dismantle solar panels.

 

What does this mean for project planners?

Some of the tasks outlined may seem trivial. And whether a monitoring or a monitoring by an AI or a human is necessary or simply not necessary at all, finally the costs decide. Especially during operation. AIs offer the advantage of being able to monitor continuously and then possibly trigger alarms automatically. Certainly, an advantage which should not to be neglected. A necessary precondition is the availability of data in sufficient quantity and quality. In order not to obstruct future evaluations and the use of AI technologies or to have to install relatively cost-intensive sensor technology retrospectively, it makes sense to plan systematic data collection directly for each new system. Low additional development and hardware costs enable efficient monitoring at a later stage, which minimises error-related costs, downtimes and loss of performance.

 

Key data of the event

AI-Talk at the Intersolar Europe 2019
Mounting Systems
Hall A3, Booth 340 (A3.340)
15 May, 3 p.m.
No registration required

 

About the author

Nicolas Hoyer is a physicist and founder of the AI start-up PERCEPTX that teaches deep learning machines to detect anomalies.
Nicolas lives and works in Hamburg.
Visit: www.percept-x.com

 

About the fair

Find here our topics at Intersolar
Here you can find the events’ main topics of this year’s Intersolar

 

Back to news Next post