Through machine learning, companies can turn their data into insight and advanced analytics into foresight, in order to improve decision making. This enables them to achieve real and measurable improvements compared to traditional methods. Many sectors have already embraced machine learning, but the energy sector is lagging behind. However, it now has the opportunity to leap frog other sectors, with the latest available methodologies. This is illustrated in how swhere and Powel have implemented machine learning, successfully proving that it can be an important step in making intermittent renewable energy production both more predictable and more profitable.
Put simply, machine learning is the science of getting computers to act without being explicitly programmed. Although it may sound futuristic, it is already so pervasive in society that you probably come into contact with it several times a day without even realising it. Machine learning captures a lot of complex information, learns from it and then applies the knowledge to better estimate or predict future events.
Developments within machine learning is what is behind a wide range of today’s technologies, from something as commonplace as effective web search to far more complex and futuristic inventions, such as self-driving cars. Inciting both scepticism and optimism, fears and hope, one thing is undebatable, machine learning is here to stay and it is only going to get more prevalent and important.
Powel, innovation and world-class expertise
Powel prides itself on being at the forefront of the latest technological innovations and providing this to our customers. Recently we partnered up with machine learning experts swhere to look at how we can implement machine learning in our solutions to benefit the energy sector.
“We have always had the philosophy that we cannot be experts in everything from day one,” says Tom Røtting, VP of Business Development in Powel. “This means that if we do not have the expertise in-house, we search for it elsewhere. In this instance we found swhere, who are real heavy-weights when it comes to machine learning.”
Established in 2014, swhere uses state-of-the-art predictive analytics techniques to extract full value from data. Through this, they enable their clients to make the transition from management by knowledge to management by inquisition.
Dr. Ernst van Duijn, an internationally renowned expert within the field and founder of swhere, explains that the increasing complexity of the utility market requires management to ask the right questions rather than provide the right answers. To answer these complex questions with a high degree of confidence, machine learning has proven its value: It allows for complex predictions to be made with high accuracy, thereby improving senior management’s decision making. The question now is, how can this capability to make better predictions and to drive a successful transition from deterministic to probabilistic management benefit Powel's customers?
Machine learning possibilities in the energy sector
“Many people ask whether robots and machines are going to take over,” says Van Duijn, “but I firmly believe that the future lies in the symbiosis between people and machines. A machine predicts and a human optimises and takes action.”
“When it comes to the energy sector, I think machine learning will take some of the jobs that are currently being done by people, as with many white-collar jobs in several sectors. With some roles eventually being substituted by machines, it is a nice challenge to find the right balance between where we need people and where we need machines. The trick is to find the balance and knowing how to apply human expertise where needed.”
“The energy sector lags behind when it comes to embracing these new opportunities. The telecommunications and finance industries are years ahead in that respect. However, the advantage for the energy sector is that those looking to machine learning now will be using the latest technology and the science has developed immensely in the last five to ten years,” says Van Duijn. “The speed, lower cost of data storage and availability of data thanks to sensors and smart devices all combine to make this a very good time for the energy sector to embrace machine learning.”
Greener energy through data analytics
European markets are moving towards an asset-centric future, margins are tighter and profits are harder to come by. Although photovoltaic and wind power are on the rise, there are inherently challenging as the sun does not always shine and the wind is equally unpredictable. Until better solutions are found for energy storage the balancing act continues to be a challenge for renewable power producers everywhere.
In Germany, hailed as the poster child for transition to renewable energy, they are struggling with grids that are unable to cope with the erratic nature of wind and solar power. The consequences of production being hard to predict is that nominations becomes more difficult to place accurately, leading to high penalties. This is one of the areas in the energy sector where Van Duijn is confident machine learning can be utilised.
Powel, together with swhere and a large Norwegian utility, put machine learning for wind forecasting to the test. Combining our domain knowledge with swhere’s machine learning expertise, we applied predictive analytics and machine learning techniques to see how we could optimise production further and make improvements for our customers.
“The key here is the data quality,” says Van Duijn. “Are the data actually good enough to start machine learning? What often happens is that the client think they are not, yet the data is better than they think. Machine learning techniques can fill in missing data with a high degree of accuracy. It needs to be at granular level, which allows businesses to make better decisions and gaining the most from it. It also needs to be at a level where those in charge can actually do something with the information.”
The trouble with wind power
“Planning wind production in Norway is difficult partly because the weather prognosis they work from is only updated every six hours. Wind farms are typically out at sea or in open areas, where the weather can change in an instant, especially as it travels across the North Sea,” explains Røtting
As a result, intraday and day ahead rarely has accurate enough predictions, resulting in loss of revenue. Although many companies have been working with power production from wind for some time now and are managing increasingly more accuracy in predictions, there is still a long way to go.
“The basic idea behind this project was that if we are able to find a pattern locally, we can create a mini weather prognosis centre linked directly to the wind parks. Hopefully, by doing localised analysis we will end up with forecasts that are much more accurate,” says Røtting.
So what exactly is it machine learning can do in terms of predicting and forecasting, that goes beyond what human expertise and experience can achieve?
“The process started with specifying in detail which targets to be predicted by looking at the data we had to hand. Based on the very specific target and available data, our scientists then made a list of methodologies best suited for the purpose,” says Van Duijn. “Data scientists are constantly developing new methodologies, and we need to always find the one that will deliver the highest quality and accuracy for the project.”
When planning wind power production, three key questions need answering. When is the wind, how powerful is it, and in what direction is it blowing. We wanted to see if there was a pattern in the wind that we had not seen before and used new algorithms to find these connections.
In this case, swhere were given a lot of data, both weather-, production- and forecast. They started looking at it and questioning the meaning of the information, before we started testing methodology. The results, at the end of six weeks, were very uplifting and showed great potential for the use of machine learning techniques in such cases.
“This was a good case for us at swhere, not just because of the results, but also for the methodology we used,” says Van Duijn. “By applying niche predictive analytics techniques, we managed to reduce uncertainty in wind power production by more than 45% resulting in the imbalance costs being more than halved.
Enabling utilities to change
“In addition to specific tasks, such as production planning, I also believe machine learning has great potential in enabling utilities to change at a faster pace along the entire value chain. The energy sector is becoming more complex and just in a few years it has seen a vast amount of unprecedented changes in regulations, customer behaviour, technology and innovation. It is impossible for a few people to know it all,” says Van Duijn.
In a competitive market, it is imperative for anyone who wants to survive to find their competitive edge. In traditional analytics you work on data, whereas in machine learning, the data work for you. Machine learning often has the flexibility to change the original model by just a few parameters.
“My prediction is that successful managers in future will be the ones who ask the correct questions, and this is where machine learning can help by dealing with many different parameters. People can only cope with thinking of a set amount of parameters in a day. A computer can do infinitely more than a person.”