Machine Learning and Renewable Energy
Today, ML and AI represent a valuable opportunity for energy corporations and investors to implement more efficient and productive processes for greater investment returns and to support the energy transition. Whilst the varied applications of ML are already considerable; it is the potential to accelerate positive change in the energy sector and sector decarbonization that is now becoming a larger focus.
Machine Learning For The Energy Sector
Machine learning refers to technology capable of sorting through algorithms and data in such a way that it learns and improves its methods through increased experience. It represents a subset of artificial intelligence (AI), with AI intended to be a network capable of mimicking human thinking. Analysts and Experts across the industry are now predicting that AI will play a pivotal role in the future of energy. This is particularly true for sustainable energy, and while the applications of ML in this field are already plentiful, its potential to make a monumental impact with regard to change and innovation has still not been fully realized. Investment in AI within sustainable energy is expected to be over $7.78 billion by the year 2024. To put this into a broader perspective, the potential to create value from AI across 19 energy industries ranges from $3.5 trillion to $5.8 trillion. Needless to say, it is hard to ignore the opportunities for growth ML offers. Major industry players are now incorporating ML into their strategy with a view to capturing the promise it offers.
The energy sector represents about 2% of funding for AI in Europe, which while significant, indicates that there is still a lot of room for growth. The ML industry is home to a multitude of start-ups with plenty to offer, and corporations and governments are starting to take note.
This year, BP’s investment of $3.6 million into R&B AI technologies represented their first foray into the Chinese AI landscape, which is aiming to be a $300 billion industry by 2030. Of the $250 million invested from NGPs venture fund, $105 million of it was reserved for emerging technologies in the energy and software crossroads. Government policies are increasingly incorporating ML now as well. For example, the UK government announced it is investing $119 million this year into new technologies and systems that can be deployed in extreme environments. Following this, in April 2020 the U.S energy department announced $30 Million of funding for ML and AI Research. It is clear that there are opportunities for investment, and these are expected to increase. The number of applications for ML and AI in the sector is almost countless, and it is now more a matter of finding which applications will be the most useful, sustainable and profitable.
ML represents an elegant fit for some of the problems it is beginning to solve for energy companies. This is particularly true for solar and wind power, which have been historically burdened by weather patterns that are difficult to predict and have many variables to consider.
Machine Learning for Renewable Energy Applications
Solar Energy
Solar production has already benefited from AI in several ways, notably in weather prediction. By having more accurate weather predictions, grid supply can be predicted more accurately. This is highly valued by grid operators. In America, the department of energy partnered have with IBM to develop its ML technology – Watt-Sun – which sorts through data gathered from a substantial bank of weather reports. The goal was to reduce the uncertainty of the variable energy output from solar energy with the investment into the project being justified by the ability to significantly reduce costs related to excess energy storage and deficit energy production. This technology was able to increase the accuracy of forecasting by up to 30%. The variability of sunlight is unavoidable, but its unpredictability can be diminished.
AI integration with micro-grids are proving to be another exciting area in which ML is being implemented in. The challenges of variable supplies of electricity and variable load are now being tackled through ML. BluWave AI and Sustainable Power Systems recently collaborated in an attempt to take on some of these challenges. By analyzing current flows, operators can work to mitigate bottlenecks in the grid and decide whether energy should be stored in a battery, distributed, or sold in an electricity exchange at a particular location and moment of time resulting in greater energy consumption efficiency. The struggles that have accompanied the Microgrid through the years of its inception for how to best distribute energy now seem to be reliant on developments in AI and ML to once again be feasible. The benefits offered by such programs are tantalizing, and they are not limited to solar energy.
Wind Energy
Weather predictions – wind energy faces many of the same problems as solar in regard to variable weather conditions. Google acquired the AI based company DeepMind in 2014 for $500 million. When applying ML algorithms to widely available weather reports, they were able to predict output 36 hours ahead of time. This helped them achieve their goal of boosting the value of their energy, increasing the value of it by up to 20%. Initial talks were held to see if this could be incorporated into the UK national grid, and while these talks ended, it is possible they will have future talks and collaboration.
Condition monitoring is another application of ML that is proving to be capable of significantly reducing costs for energy producers, especially when applied to wind turbines. Recent models have been able to monitor blade faults or the generator temperature, allowing for rapid repairs and maintenance to be performed using ML technology and data. Given the fluctuating demand and production for energy, ensuring operations run smoothly at offshore windfarms is crucial. Having monitoring and regulation that is both reliable and precise is a key aspect of this. Smoothing out the bumps in the energy supply chain and increasing efficiency is an effect that cannot be understated; however, this alone is not enough to accurately predict the power output of a given turbine or even farm.
There have been improvements on the traditional power curve methodology in wind farms, namely the regression tree model which incorporates ML into its predictions. Without any new data being required than would be typically be necessary for a wind resource assessment, the regression tree model has three times the accuracy of the traditional power curve methodology. This would once again benefit the farm operators as well as the grid operators due to the importance of accurately predicting the output of a given site.
Countries Focusing on ML
Countries are now realising the potential ML and AI have to offer, and are incorporating them into their policy.
Australia Investing $25 million into AI and aiming to target the energy sector among others.
|
China Taking steps to adopt a wide-spread smart grid across the country, utilizing its existing advanced capabilities in AI.
|
Italy Installing 41 million smart meters as part of a project starting in 2001. Italy now has the most advanced smart grid in the world as a result
|
UK Investing $119 million into the robotics and AI in extreme environments programme, for research of robotics and AI technologies used in extreme environments such as offshore oil and gas locations.
|
U.S.A The department of energy announced funding of $30 million directed toward ML and AI.
|
|
Notable ML Projects
Exxon Mobil
Working with M.I.T to integrate self-learning robots that will scan ocean floors to detect areas which are rich in oil.
US Department of Energy
Invested over $4.5 billion in establishing smart grid infrastructure. This involves installing over 15 million smart meters, which monitor energy demand and supply. Additionally, they are investing in ‘synchrophasers’, which provide richer data for grid optimization.
Google DeepMind
Google and DeepMind have been analyzing 700MW worth of wind energy to see how it could best be optimized. The ML networked utilized data from historic weather forecasts and turbine data, allowing them to now predict wind energy output 36 hours ahead of time.