AI in Renewables: A Practical Guide for Digital Transformation

In 2023, we are looking towards a future where algorithms dictate much of what happens within renewable energy operations. Find out if your organisation is ready for the robots – and ready for success in the coming decades. 

Artificial Intelligence – Hype or Real? 

The world of technology is a hype machine. Once it acquires a target the buzz can be incessant – whether justified or not. It has been hard to avoid the explosion of “web3”/crypto news for the past few years – even for those of us in renewables, where the space of applications is vanishingly small. The crypto hype seems to have crashed and burned for now, and the hype machine has trained its focus on AI and Machine Learning. 

Some reasons for the recent buzz include well-publicised releases of new “AIs" such as ChatGPT – a large language model (LLM) that can do a convincing job of writing language like a human – and generative image models like Stable Diffusion. The excitement seems to be reaching an apex after building over the last decade, with vast improvements in speech recognition and translation, advances from Google’s DeepMind in areas such as protein folding and Tesla’s self-driving capability (though you can’t always believe what you see).

Since writing the first draft of this article we’ve also seen Microsoft release their new GPT-enabled bing, and Google responded to this with a preview of their new LaMDA-enabled Bard chat-based search product. These releases indicate a significant platform shift - we can no longer rely on outdated assumptions about technology.

As we observe this shift from the renewables industry, it’s reasonable to ask: is this technology useful for us? How can we determine if this is real progress, or just a fancy algorithm desperately looking for an application (looking at you, blockchain…)? There are several reasons to believe that progress here is real and will have wide benefits. Firstly, the broad range of domains that have seen advancements thanks to AI progress indicates that there is something general underlying this progress. Secondly, the reason that breakthroughs are happening now is easy to understand – in general, ML gets better the more data and compute it has access to. The amount of data available for these applications is growing exponentially, and compute is decreasing in cost exponentially. Add to this a virtuous cycle of capital injection giving more access to compute giving better models driving capital injection… it’s easy to see how real progress can be made in any domain with the right approach. 

Finally and most importantly for us, we are already seeing valuable uses of ML in renewables. Bitbloom data scientist Harley recently introduced some of the machine learning work we’re doing, and our recent whitepaper dived into improving wind power performance with machine learning. We use ML to model component temperatures, wind speed, and more in our monitoring work. There has also been an explosion of solutions in the renewables space promising ML approaches. 

That is not to say that ML in renewables is not suffering from the dazzling overpromises generated by the hype machine. The challenges with ML are not often discussed, and we have seen that many solutions overpromise and under-deliver. At Bitbloom we remain optimistic about the long-term and realistic about the short-term. We know that our clients need reliable solutions to deliver value now – from raw data to action – but we believe that in the future that value will increasingly be informed and ultimately delivered by ML. 

Capitalising on the promise of ML is a daunting challenge facing many companies in the renewables space. Given the challenges and the long-term opportunities, what steps should your organisation be taking? Bitbloom can draw on its experience of working with renewables data to give some recommendations. 

Hungry for Data 

There is no silver bullet for succeeding with ML, but it is certain that ML will not be able to deliver on its promise for your organisation without sufficient data. One of the first steps to take on the road to our ML future is to collect as much data as possible (and practical). There are many challenges in doing so, but one at least should not be a barrier – storage costs have fallen so substantially over the past decade that physically storing data is no longer an expensive consideration. 

Renewable asset data connectivity is another key challenge to overcome. 5G networks and satellite internet hold some promise that in the future even the most remote of wind farms will ultimately be capable of delivering at least moderately high-frequency data to the cloud. We see some interesting companies operating in this space. 

Once assets are connected, the remaining challenge is data organisation. The specific solution depends heavily on the organisational requirements of the data – both immediate and long-term. There are numerous routes one could go down: from data warehousing, where data is stored in a structured, queryable system such as a database; to data lakes, which impose less structure and defer the organisation for some point in the future; to a lakehouse architecture, which combines a data lake approach with queryable structures and staged data improvement. 

The most important thing for ML algorithms is that they can read and understand the data. A large corpus of historical data will make for better models. Specific requirements, for example streaming/live data insights, can be met when there is a tangible business need. The first step is to collect the data. 

Data Is Not Born Ready 

There is a well-known concept in ML – Garbage In, Garbage Out (GIGO). This refers to the idea that if one feeds an algorithm with garbage (noisy and/or erroneous) data, then the output is likely to be nonsense.  Another way to look at this is that data is not born ready – we need to make it ready. 

Data improvement is the next step on the way to capitalizing on the promise of ML. Bitbloom has been using messy SCADA data signals for their analytics for a long time, so we’re good at this. There are many steps that one can take to make data ready for ML – signal flagging, filtering, binning, rebuilding, and more. The key requirement for an organisation that wants to make the most of ML is that these steps should be automated and standardised. 

Automation is key, but it can also be a challenge. In the case of wind turbines, each turbine is unique – it has a unique history, including installation, operating conditions, and servicing, not to mention large differences between models, components, and sensors. The right automation processes and tools are necessary to prevent a snowball of complexity when approaching a whole portfolio of renewable assets. 

We have the data, and it’s ready – let’s build some ML! 

The Right Mix 

People are at the core of succeeding with ML. As a result of some of the challenges enumerated above, we believe that we are still some distance from a future when predictive models can be rolled out across global assets with ease. That future will come – but until then we’ve got some building to do, and we need talented people. 

At its core, succeeding with analytics and ML is about understanding how to translate business questions into algorithms. To do that reliably, your organisation will need people who understand the domain (renewables, wind, solar), as well as your organisation’s goals, challenges, and opportunities. At Bitbloom, we believe in the value of domain experts who can understand the business as well as solve the hard technical problems presented by operating assets in the field. 

Harnessing The Power Of ML 

If finding insights using ML can feel like splitting the atom, generating long-term reliable insights is like building a power station. It is not enough to have the right data, prepare the data, and put in place the right people to derive insights. All of this takes substantial investment. To justify this investment an organisation needs to continuously harness this power, and they need it to get more and more powerful as the underlying technology improves and evolves. That means we need to address some final missing pieces: we need to turn our insights into action, and we need to make more and better insights over time. 

Though turning insights into action and value is not easy, the good news is that organisations can start doing this now. Generating value from insights requires the right processes and tools. By putting these in place early, organisations can build the muscle memory they will require to capitalise on the advances that ML will bring in the future. 

One obstacle to adopting ML more widely in the enterprise is the ability of stakeholders to understand ML models. Many advances in the field can appear too much like an algorithmic black box into which some numbers go and out of which come some more numbers – it is up to the stakeholders to decide how much to trust these numbers. When taking any action has a cost, the threshold of explainability is high for stakeholder buy-in. 

There is no silver bullet here, and trust must be built through a gradual move rather than a sudden dash to ML. At Bitbloom, we take the low-hanging fruit offered by traditional, explainable analytics first; when we use ML, it is to push the envelope of what is possible and provide insight that cannot be found through standard approaches. Continuous improvement of the analytics processes and the coordination of stakeholders is the key to being ready for ML. 

AI Is Coming – Will You Be Ready? 

As Bitbloom and others are currently showing, the value that machine learning can deliver to renewable operations is real, substantial, and increasing all the time. It should be considered at the core of every organisation’s digital strategy. 

Using good data collection practices, automation of analytics, and starting on the journey of helping your stakeholders get more value from your data, you can ensure your organisation is ready to make the most of machine learning. 

As we look to a possible – and we believe inevitable – future where algorithms dictate much of what happens within renewable energy operations, we can trace a path back to the present and identify the steps we need to take to make that future happen. Those organisations that start that journey now will be well-placed to be amongst the most successful in the coming decades.