Guest post from WePower: Artificial Intelligence and the future of energy
Using AI in renewable energy production prediction, energy grid balancing and next generation understanding of energy consumers.
About the author. Dr. Tadas Jucikas is a co-founder and CEO at Genus AI — a next generation artificial intelligence platform which enables businesses to interact with their customers in an emotionally intelligent way. Tadas advises WePower, a revolutionary blockchain-based green energy trading platform, on how to harness the power of data science and artificial intelligence. Tadas has completed a PhD in Computational Neuroscience at the University of Cambridge, UK.
Artificial Intelligence will change the future of energy. I am often asked how Artificial Intelligence (AI) can be used to help interpret the past, optimise the present and predict the future. Having helped build data science and machine learning solutions in both private and public sectors I’m always pleasantly surprised by the multitude of applications and opportunities that AI technology can offer.
There are only three limitations to building successful AI systems — computing power, availability of data and imagination. More than often the latter is the hardest to realise.
Even though the successes are just starting to emerge AI has proven that it can revolutionise energy as well. The sector already depends on optimisation and predictions: energy production, energy grid balancing and consumption habits.
AI offers a unique solution to these challenges and due to its capacity to evolve and learn it will undoubtedly become a critical component of the energy industry.
AI plays a critical role in the WePower platform as well. The WePower platform is the next generation utility company and has in-built capabilities to use AI technology for renewable energy forecasting, grid balancing and in-depth consumer understanding.
Early Traction. Emerging technology team lead Dan Walker at the British Petroleum’s (BP) Technology Group says [1]:
“AI is enabling the fourth industrial revolution, and it has the potential to help deliver the next level of performance.”
AI provides an inspiring area for talented individuals as a career path. Solving energy problems directly relates to improving living conditions for generations to come. Bill Gates, founder of Microsoft, wrote an online essay to college students graduating worldwide in 2017 where he stated [2]:
“If I were starting out today… I would consider three fields. One is artificial intelligence. We have only begun to tap into all the ways it will make people’s lives more productive and creative. The second is energy, because making it clean, affordable, and reliable will be essential for fighting poverty and climate change.”
The last one that he mentioned was biological sciences.
It has prompted students across the world to enrol in these subjects and study as well as discuss challenges. Harvard University’s Franklin Wolfe, a graduate student in the Earth and Planetary Science programme writes a great overview of the challenges of the grid and how a new and different ‘smart grid’ could be enabled by AI [3].
However, has there already been a success story where AI helped the energy industry? The answer to this question requires a bit of research. A lot of pilot programmes are not publicised and are still in their early stages. However, a mounting body of evidence indicates a bright future for AI in the energy sector.
A great example of early traction is Google’s DeepMind technology which became famous for teaching itself to play the ancient game GO through technology called reinforcement learning and becoming the World’s number one player [4]. The team behind the technology announced that its machine learning algorithms could cut electricity usage at Google’s data centres by 15% [5]. The predictions centred around anticipating a higher load on the data centres’ cooling systems and controlling equipment more efficiently. This decreased the energy usage by 40% percent and translated into saving hundreds of millions of dollars for Google over several years [6].
This announcement prompted discussions on how such an approach could be used elsewhere. One of such applications is with the National Grid in the United Kingdom.
“We are in the very early stages of looking at the potential of working with DeepMind and exploring what opportunities they could offer for us,” said National Grid. “We are always excited to look at how the latest advances in technology can bring improvements in our performance, ensure we are making the best use of renewable energy, and help save money for bill payers.” [5].
The focus of this partnership would be to use AI technology to balance energy supplies to the National Grid. The expected savings are substantial — DeepMind aims to cut the national energy bill by up to 10% [7]. The abundance of historical data supports such advanced predictive capabilities.
Several other big players are already active in this space.
Originally, IBM had everyone guessing why it had acquired The Weather Company. Many joked that they misinterpreted what it means to compute on the “cloud”.
However, it was revealed that IBM planned to launch a new product called Deep Thunder which will offer precision weather predictions at a 0.2 mile to 1.2 mile resolution. The focus of the Watson product will be on how minor changes of weather can affect consumer behaviour and help businesses to react more effectively [8].
IBM has also worked extensively in solar energy prediction. Even as early as 2013 IBM’s research division partnered with the Department of Energy in the United States on leveraging machine learning for clean power. IBM has over 200 partners that use their solar and wind forecasting technology [9]. The technology is built by combining dozens of forecasting models and then integrating a multitude of data sources about the weather, the environment, atmospheric conditions and how solar plants and the power grids operate. The predictions range in availability anywhere between every fifteen minutes up to thirty days in advance. IBM’s product manager Hendrik Hamann claims that the self-learning weather model and renewable forecasting technology is 50% more accurate than the next best solar forecasting model [9].
Early signs of AI changing the industry sector are also present in the technology community, not only large companies. A data scientist Evan Baker posting on a Medium shares exciting results:
“Using a random forest model, I was able to predict expected annual savings to around $15.00, a 75% increase in accuracy over predictions generated by the National Renewable Energy Laboratory (NREL)” [10]
Evan outlines his approach with the support of a homemade pipeline that uses publicly available data and open source tools. His machine learning models give highly accurate predictions mapping out the expected return on energy generated by a prospective solar panel. He made the predictions available on his website; http://solarcalculator.xyz.
AI powered tools such as this make information regarding the ability to switch to solar more accessible and readily available for everyday use. It has the potential to bring down soft costs associated with installation and may accelerate the transition to renewable energy and micro-production in the future.
AI technology background. With the rise of cloud computing and the ever-decreasing costs associated with computations, now and in the future this technology will be more and more widely available. One of the most process heavy steps in AI systems is model training and validation. Being able to pay per minute or even second for the use of computing power removes the need for large upfront investment and data centre maintenance costs. With Google Cloud, IBM Bluemix and Amazon Cloud the power to perform highly complex computations is readily available for everyone today [11].
The systems architecture for machine learning which underpins artificial intelligence is also seamlessly provided by cloud solutions. Some of the computations require highly parallel processing capabilities and are best performed by hardware specifically built for such purposes. Cloud solutions which offer native architecture for such computations are already available. A good example is Google’s Tensor Flow platform that enables next generation deep learning [12].
One of the unexpected beneficiaries of this trend are graphic processing unit (GPU) manufacturers who tried for decades to create highly paralleled processing capabilities for the gaming industry. Nvidia is a great example — their stock price has risen dramatically because such processing units are ideal for artificial intelligence systems. Some even say that Nvidia’s lead in enabling AI computing is nearly impossible to replicate [13].
Data is the next big resource that has grown in capacity. It has been estimated that in the year 2017 alone we will produce more data than in the last 5,000 years of humanity combined [14]. Data for artificial intelligence is like air, food, and water for us humans. The first fields which started to generate significant amounts of data were physics and biology — LHC in CERN, Switzerland and DNA sequencing followed by the explosion in the digitisation of life phenomenon.
One other area where data is continually generated at a large scale is in the energy sector.
Predicting renewable energy. The production of energy from renewable sources is growing rapidly. With the advancement of technology development harnessing energy from wind, sun, hydro, amongst others it is becoming more popular and economically accessible. Negative effects on the environment from energy sources such as natural gas, oil and coal have further accelerated this shift.
The United Kingdom’s move to renewable energy reached a new milestone in 2016. Grant Wilson, teaching and research fellow at the University of Sheffield details the shift in a widely circulated piece on The Conversation [15]. According to Wilson, in 2016, just 9.3% of British (not UK — as Northern Ireland is calculated separately) electricity was generated from coal, down from more than 40% in 2012. That would be the lowest percentage of coal that has even been provided in the system’s 100 year history and the lowest absolute quantity burnt since the start of World War II [15].
The new record capacity of electricity comes from renewable energy, mainly from wind and solar power.
However, the biggest challenge with renewables is that energy production is intermittent. The production depends on weather conditions, such as the wind blowing or sun shining. Unlike conventional power, this means such sources cannot necessarily meet surges in demand.
Valentin Robu, a lecturer in Smart Grids at the university of Heriot-Watt discusses how AI can provide a solution and ‘future proof the grid’ [16].
There has been a lot of research studying accuracy and prediction capabilities. A paper from the office of Energy Efficiency & Renewable Energy discusses a multi-scale, multi-model machine learning solar forecasting technology [17].
Solar is not the only forecasting that is tackled by researchers. A talk by Andy Clifton from NREL National Wind Technology Centre in the US discusses machine learning applications in modelling power output inside the wind turbines shows promising results [18]. The methods described in the talk are regression tree based algorithms, however, the results present a compelling case for further explorations in wind energy forecasting at both an individual rotor and subsequently for the whole plant.
A paper by Gul M Khan from University of Engineering and Technology in Peshawar describes neural network approaches in creating power generation predictions of wind based power plants. The results show the predictions from a single hour up to a year with mean absolute percentage error as low as 1.049% for a single day hourly prediction [19].
In 2015, IBM was able to show an improvement of 30% for solar forecasting while working with the U.S. Department of Energy SunShot Initiative [20]. The self-learning weather model and renewable forecasting technology integrated large data-sets of historical data and real-time measurement from local weather stations, sensor networks, satellites, and sky image cameras. The platform is exploring how to address forecasting challenges in wind and hydro-power plants.
Nils Treiber and his colleagues from the University of Oldenburg discuss how machine learning can be used to predict wind power [21]. Their study focuses on predictions for individual turbines and then how entire wind parks can predict production from a matter of seconds to hours. They compare their results to a persistence model and show an increase in accuracy of over 24% [21].
A survey by Kasun S. Pereral and colleagues from the Technical University Dresden and the Masda Institute of Science and Technology in Abu Dhabi discuss the need for accurate forecasting and its implications on balancing the grid [22]. The creation of a ‘smart grid’ is discussed which touches upon the need to identify renewable energy plant locations, and integration points and sizes. Machine learning has become a tool for strategic planning and policy making for renewable energy.
The need for a smart grid. The first grid was created by Thomas Edison in 1882 as the Pearl Street Station plant in lower Manhattan which powered 59 customers. The customer base has since increased to hundreds of millions of users, but its overall structure and approach still has not fundamentally changed. The grid consists of a vast network of transmission lines, distribution centres and power plants.
In order to adapt to the intermittent nature of the renewable energy generation there has been a world-wide effort to modernise the grid. The U.S. Department of Energy has made supporting the ‘smart grid’ a national policy goal, which includes a ‘fully automated power delivery network that monitors and controls every consumer and node, ensuring a two-way flow of electricity and information’ [23]. It has been reported that in the last seven years the department has invested over $4.5 billion into the smart grid infrastructure. Part of the investment is focused on installing 15 million smart metres and monitoring energy usage per device in order to alert utilities of local blackouts. This programme is estimated to limit the rise in peak electricity loads on the grid to only 1%. This is especially important knowing that the total U.S. energy demand is expected to increase by 25% in 2050 [23].
Michael Bironneau, the technical director at the UK’s Open Energi, a company that gives energy users the power to participate in the energy market has been exploring the future of the grid:
“In the UK alone, we estimate there is 6 gigawatts of demand-side flexibility which can be shifted during the evening peak without affecting end users. Put into context, this is equivalent to roughly 10% of peak winter demand and larger than the expected output of the planned Hinkley Point C — the UK’s first new nuclear power station to be built in generations. Artificial Intelligence can help us to unlock this demand-side flexibility and build an electricity system fit for the future; one which cuts consumer bills, integrates renewable energy efficiently, and secures our energy supplies for generations to come.” [24].
Quirin Schiermeier in her Nature column last year explores the efforts in Germany to modernise the grid [25]. She quotes Malte Siefert, a physicist at the Fraunhofer Institute for Wind Energy and Energy System Technology in Kassel, Germany, and a leader on the projectcalled EWeLiNE:
“To operate the grid more efficiently and keep fossil reserves at a minimum, operators need to have a better idea of how much wind and solar power to expect at any given time”.
The German government has promised that by 2050 at least 80% of the country’s electricity will come from renewables. The challenge is that on calm and cloud days grid operators still need to use conventional power stations to meet the expected demand. The reverse applies on windy and sunny days. The grid operators must swiftly order coal and gas-fired power stations to reduce their output.
Quirin notes that such requests, called re-dispatches, cost German customers more than €500 million (US$553 million) a year because grid operators must compensate utility firms for adjustments to their inputs [25]. In addition, this ends up generating needless carbon dioxide emissions if the operators generate extra power that is not used. Renate Hagedorn, a meteorologist with the German weather service in Offenbach comments that:
“It is quite a concern that renewable energy here is expanding so fast without a proper database for an accurate power forecast” [25].
In the EWeLiNE project machine learning models are used to predict power generation over the next 48 hours. The team checks these powerforecasts against what actually materialises, and machine learning then improves the predictive models. This closes the learning loop and enables AI to be highly effective [25].
The National Centre for Atmospheric Research (NCAR) in Boulder, Colorado, USA has embarked on a similar project to in the one in Germany. It was started in 2009, and now is operational in eight US states [25].
Drake Bartlett, a renewable-energy analyst with Xcel Energy, the utility firm with the highest total wind capacity in the United States says that:
“The number of forecasting errors has dropped since 2009, saving customers some US$60 million and reducing annual CO2 emissions from fossil-reserve power generation by more than a quarter of a million tonnes per year” [25].
Understanding consumers. The third pillar for creating a stable, scalable, and smart energy system is understanding energy consumers.
Energy, like any other product has seen a rise in differentiation in terms of brands, usage plans and sources of energy.
Customers are more vocal about their preferences in terms of environmental impact that energy producers have. Consumer opinion and choices have a tremendous impact on the energy sector.
Consumers also produce continuous stream of data that comes through the power grid itself. There has been a significant push by utility providers to install smart meters. The meters are able to send the information to the utility providers on sometimes even hourly basis. It not only helps to predict the network load, but also predict consumption habits. Someone who is more of a ‘night owl’ and tends to work at night will have a completely different energy usage pattern when compared to someone who enjoys their 6am run in the morning.
There are over 7.4 billion people on the planet and no two individuals are the same. However, grid managers, utilities companies and governments still see people through a simplistic understanding of geography and demographic based segments.
Understanding the consumer’s habits, values, motivations, and personality helps to further bolster the balancing and effectiveness of a smart grid. It also allows for creating policies more effectively and enables an understanding of the human motivations associated with renewable energy adoption.
Genus Artificial Intelligence focuses on analysing first and third-party data about consumers to help organisations understand people. This understanding allows engaging with audiences in an emotionally intelligent way at scale. Genus AI strives to make human level emotional intelligence a reality and helps to deploy to inform and have a positive impact on the real world.
Appreciating individual level consumer differences in the context of energy platforms will unlock the next phase of optimisation and forecasting.
Genus AI is proud to work with WePower, an energy trading market powered by blockchain technology to achieve this goal [27]. With ongoing utilisation, the system will become more and more accurate, and will enable the further evolution, success and growth of the WePower platform.
The use of AI in the smart energy network will enable the long-awaited transition to fully decarbonised energy production and consumption.
Originally published at medium.com on November 2, 2017.
