The trend towards automation is noticeable in the management of solar systems. Automatic fault detection is part of the standard repertoire of monitoring systems today. Almost all the systems in our overview can do this, according to the information provided by the manufacturers, and 48 of the 77 systems should also be able to classify fault patterns. However, digital fault detection also brings difficulties. Every solar PV system is unique, and different types of fault occur depending on the location and system design.
In order to prevent false alarms, operators can adjust the values for fault detection in many systems. With 40 systems in our overview, operators should be able to adjust the limit values themselves.
But it is not enough for an operator to know that an error has occurred. To derive a need for action, they must also clarify what exactly the error is, why it occurred and how it can best be rectified – if it is even worthwhile to take action. For this, usually an expert is still needed, even if 24 of the 77 systems in our overview are capable of independent advanced error analysis. The question is: How well does automatic analysis work? And how much human action is still needed?
PV system operator Enovos believes that highly qualified technical personnel can be better employed than for monotonous reading and interpretation of endless data. Richard Rath, Head of Operations & Maintenance at Enovos, hopes for improvements in automatic fault analysis. “There is currently no truly satisfactory software on the market,” says Rath. In most cases, the data are already well presented, he finds, but an intelligent evaluation does not usually take place. Either there are more false alarms than real alarms, or real alarms are not identified quickly enough. “Then human monitoring and analysis are still necessary.”
AI does analysis
In order to further improve and automate error analysis, some providers rely on self-learning algorithms. The solar PV system configures itself, and after configuration triggers fewer false alarms than other solutions. In our overview, 10 companies state that they already use “self-learning error detection” via algorithms. A further 10 say that they are working on this technology and want to bring corresponding functions to the market. One company currently developing such an artificial intelligence system is Solytic. The german monitoring provider is relatively new to the market. The company’s monitoring and analysis software has been available since the end of 2017, followed by its own data logger in 2018. Solytic wants to turn the monitoring market upside down with its own AI. Chief Communications Officer and cofounder Alwin Nagel explains how they want to bring intelligence into their monitoring system. Solytic is currently collecting data from around 12,500 PV plants with a total output of around 800 MW in 42 countries. The company uses the data to develop an algorithm that performs certain analysis functions.
Recommendations and predictive maintenance
“The first step is a good analysis of the monitoring processes in order to optimize the workflow,” says Alwin Nagel. The next step would be to reduce manual work for O&M providers. According to Nagel, this raises questions such as: “Where can you save time? How can data be better compared? Where can you make the system more intelligent?” The third step is smart alerts. In other words, alarm messages that not only indicate that something is wrong, but give a precise description of the problem and offer recommendations to solve it. This should reduce the current flood of irrelevant messages.
Once this works well, says Nagel, the recommended action could also be forwarded directly to the service technician, skipping the human inspection and saving further costs. According to Nagel, the goal is to get as close as possible to “predictive maintenance.” This means that the artificial intelligence should find errors before they occur and lead to loss of revenue. If the AI is programmed correctly, Nagel considers this possible. Solytic wants to go this way and received prominent support for this. In March 2018, the Vattenfall Group invested €3 million in the Berlin based startup. However, development is still in its infancy. The first “intelligent” functions are to be integrated into the system this year.
Richard Rath of Enovos also thinks that such plans are very promising. “Companies like Solytic will fundamentally change the way photovoltaic systems are monitored and managed.” While other providers focus on accurate visualization, the focus of this new approach is on automation – from the exact data presentation to the “invisible control room” in which all processes run automatically. “This makes O&M services scalable,” says Rath.
Are the machines winning?
Markus Zerer, Head of SCADA and Monitoring at system provider Zebotec, is also convinced that the trend is and should be towards automation. However, he is skeptical about the noble goals of some competitors. “If software is to automate everything and replace human analysis, then it will have to be comprehensive and complicated software.” Adapting it to individual systems without constantly triggering false alarms would be very difficult and time-consuming.
“The question is: Who is more intelligent?” Zerer says. “Often a person can evaluate complex situations much faster and better in their head than software can.” Of course, people would need the right information. In some cases, this also includes knowledge of the local conditions to go into the details of the respective operator and processes.
Besides, most self learning systems are still in development. “Often it is not yet clear where the journey can lead at the end.” Zerer therefore worries about misleading promises of marketing terms like “artificial intelligence” that promises more than is actually feasible today. From Zerer’s point of view, many providers who say ‘Our AI can learn all this by itself’ are still not ready yet.
Zebotec therefore takes a more moderate approach when it comes to automation. “We rely on the integration of human capabilities in an expert system. This system is trained by the operator and then provides an automated and plant-specific evaluation,” says Zerer.
The number of monitorable systems per person could also be increased in this way. “From our point of view, there are processes that can be handled faster and better by a human being for a long time to come,” says Zerer. “And these processes should therefore remain a human responsibility for now.”
So, it looks as if human and machine will have to get along with each other for some time to come.
It will probably be a few years before no human intervention at all is necessary for error analysis. Nevertheless, even smaller steps in automation can make the work of O&M service providers easier and more cost efficient. And many providers of monitoring systems have now begun to take the first steps in this direction.
By Mirco Sieg