For Govinda Upadhyay, CEO of AI-based predictive analytics firm SmartHelio, a self-proclaimed passion for renewable energy and the potential it offers is more than just a professional interest. Growing up in rural India, he says he didn’t see electricity “for a pretty long time”. This formative experience led him to set up his first company, one dedicated to teaching children in unelectrified villages how to make simple solar lamps to help with their studies.
It was during this time that Upadhyay noticed the rapid growth of large solar plants being installed in India and around the world and, more importantly, that many of them were delivering far less power than expected.
Unlock unlimited access for 12 whole months of distinctive global analysis
Photovoltaics International is now included.
- Regular insight and analysis of the industry’s biggest developments
- In-depth interviews with the industry’s leading figures
- Unlimited digital access to the PV Tech Power journal catalogue
- Unlimited digital access to the Photovoltaics International journal catalogue
- Access to more than 1,000 technical papers
- Discounts on Solar Media’s portfolio of events, in-person and virtual
Or continue reading this article for free
“These solar plants were not working very well,” Upadhyay says. “All the companies [that built them] were talking about how hard it is to know what the problem is: I just know that solar is working or not working, but even though I know it’s underperforming, it’s so hard to identify what the problem is exactly, until I spend money on that.”
This realisation led Upadhyay to formulate what would become the core mission of SmartHelio, a company he went on to found in Switzerland towards the end of the last decade: to improve the performance of existing solar plants and make them cheaper.
“My objective became: I don’t care about the new technology anymore. I want to make sure that existing technology works efficiently. That’s the way to go because already money has been spent on them, we don’t have to reinvent the wheel. So that’s the question I took: how I can make solar plants more efficient by extending their lifetime.”
Underperformance guaranteed
According to Upadhyay, it’s almost a universal fact that after five years, “any PV plant” will underperform by 5-10%. “That’s an average; some are really doing well, but almost 70% of those solar plants will be beyond 10% [underperformance] after five years.”
And underperformance is self-compounding; issues that emerge in the first few years of a plant’s operational life that go unaddressed carry on getting worse, leading to “unrecoverable losses”. Upadhyay uses the analogy of human beings who doesn’t exercise, thinking they can reverse the ageing process by going into “full-on exercise mode” at age 50.
“No! It doesn’t work like that. You have to start when you’re young, right? So I am pushing this idea that whether you like it or not, you will have that 10% of losses [in your PV plant] after five years but either you’re going to recover all those 10% of losses if you start this at the beginning of the installation, or you just going to recover only 3-4% after that, because it’s permanent damage and it will cost you more.”
Saving costs, no hardware required
That objective was the starting point for the suite of AI-enabled asset management tools that SmartHelio has developed and is now operating across a growing portfolio of PV projects worldwide, all designed to enable plant owners to spot faults before they happen and take early action. The data required for such a system to work was extensive, and Upadhyay explains how it was first gathered using an internet of things-based device the company produced called HelioHealth.
“We realised that if we want to build a model, we would have to understand exactly what was happening in these solar plants,” he says. “So at the very beginning of SmartHelio, we built a device to collect very high-frequency data and started doing experiments on panels in Asia, in the US, in Europe, in Africa … We collected millions of data points.”
The exercise involved deliberately damaging the panels so that the gathered data contained the “signatures” of different faults commonly seen in the field. “We got people to put some soiling on the panels, break the panels, break a diode, emulate some problems … We recreated almost 50 kinds of problems and collected those data points,” Upadhyay says.
This mass data-gathering process enabled SmartHelio’s engineers to build a model that could overlay these fault signatures onto customers’ own plant data to detect problems as they emerged, without the need for any special monitoring hardware to be installed.
“No one wants to install hardware extra—it’s more expense, more devices, more problems,” Upadhyay explains. “So we built an algorithm which can work because we already have the [fault] signatures from our devices. You’re able to translate those signatures into the existing data points our customers collect [through] inverters … and find underperformance.”
In addition to the fault detection algorithm, SmartHelio has also built an automated workflow platform that raises tickets when faults are detected and quantifies the financial gains from acting on that ticket.
“Once you know there’s underperformance, that’s basically money on the table,” says Upadhyay. “To grab that money, you have to act. Before and after reporting really creates the impact matrix for the customer, saying that, look, we detected ten faults, which has an impact of US$1,000 per unit, let’s say. And now you’ve acted upon that, you’ve grabbed US$800 already. Of course, that proves the USP of our technology, but also it drives our mission, because clients really see how much money they’re able to grab using such advanced technology and push their people.”
Upadhyay says SmartHelio has a growing client base in Europe, Australia, Japan, India and America. Africa and the Middle East are also on the horizon as promising markets. “Our aim has been to work with clients who have very large portfolios. We love it because we know the impact is much higher,” he says.
On the ground, Upadhyay claims the use of the SmartHelio platform is having a dramatic impact, notably reducing project operation and maintenance costs—by as much as 30-50%—and enabling a job previously done by five people to now be done by two.
‘Solar GPT’
Upadhyay is naturally excited about AI’s role in helping accelerate the clean energy transition and believes we’re “just at the starting point” of what it has to offer. To illustrate his vision of the future, he gave me a “sneak peek” of something SmartHelio is developing called ‘Solar GPT’. It’s a large language model akin to ChatGPT that interacts with his company’s own databases to provide instant plant due diligence reports on a range of metrics that in the past would have taken numerous people a lot of work to pull together.
“These answers would have taken months to get for a 100MW plant. Months!” Upadhyay exclaims.
Upadhyay describes how such a system could also be used to inform PV plant repowering strategies, something many developers are grappling with as they seek ways to improve the performance of ageing assets.
“Many developers have plants that are, let’s say, ten years old. They have great power purchase agreements (PPAs) because ten years back, pricing was very different,” he explains. “So it means that they’re looking into repowering their system now because the plants are not working the way they want, and they’re wondering what should be the strategy to repower. Should I just keep it like it is? Should I introduce new technology, like the new solar panels that are much more efficient? What about batteries?”
The answers to these questions would usually take a large team of people across various disciplines “a quarter“ to complete, says Upadhyay. But, using some simple prompts, he shows me how Solar GPT can produce a repowering strategy complete with suggestions on technology choices and return-on-investment estimates in a matter of seconds.
It’s an impressive feat and a glimpse of what Upadhyay believes will be a “paradigm shift” in the way PV power plants and fleets are managed. In his view of an AI-enabled future, gone will be the days of plant performance dashboards and software. In their place will be a single AI agent, one that can provide detailed, data-rich responses to all manner of questions, whether asked by a CEO, asset manager or operations and maintenance (O&M) operative.
“You can ask this Solar GPT model that we’re building: show me the underperformance of the plant in monetary terms—that matters to the asset manager. But it will also show me the places where I can act upon and which days—that matters to the O&M guy on the ground. The CEO will say, ‘Tell me the best strategy to maximise my portfolio performance.’
“It’s the same large language model giving answers to everyone. So I do believe in a world where we don’t need dashboards, we don’t need so much software. It’s only one chatbot; we just talk to the AI agent.”
‘An AI job’
Upadhyay reveals SmartHelio is also now turning its attention to energy storage and already working with a number of research institutes to develop AI-based tools to meet the needs of this fast-developing market. Among its initial objectives are questions such as how to use AI to improve the reliability and safety of battery energy storage systems and to inform strategies for the optimal dispatch of the energy from storage systems.
As the energy transition gathers pace and the penetration of renewables and storage technologies increases, Upadhyay believes the management of such a complex picture will be “truly an AI job”. “I don’t imagine that we are in any state right now where we can provide reliable power using renewables to our society,” Upadhyay says.
But through its ability to optimise, AI offers the opportunity of anticipating what generation is coming down the line from a variety of renewables sources and what storage is available to allow that power to be dispatched optimally. “The unlocking will truly happen when these renewables—solar, wind, heat pumps, storages or hydrogen—all these guys are able to provide me with a flat base curve,” Upadhyay says. “Once we achieve that, we are truly into the energy transition phase.”