The Blessings of AI in Making Renewable Energy Sources More Reliable

The Blessings of AI in Making Renewable Energy Sources More Reliable

As global efforts towards transitioning to a low-carbon economy proliferate, AI can change the way renewable energy sources are presently operated and maintained.

With rising carbon emissions globally, there hasn’t been any better time to switch to renewable energy sources. In line with the United Nation’s Sustainable Development Goal 13  and the Paris Agreement  aimed at reducing emissions and building climate resilience,     global citizens need to be responsible, compassionate and empathetic towards their natural environment. Of the multitude sources of energy, the most popular and prevalent types of renewable energy include solar energy, wind power, hydro energy, geothermal and biomass energy.  There have been significant technological advances in the last few decades which have all made renewable energy a reliable and clean alternative to conventional sources of energy. However, with the growing capacity and complexity of such energy forms, the challenges in operations and maintenance (O&M) have continued to rise continuously.  

Consider, for instance, solar power plants which regularly experience challenges in preventive (routine inspections to prevent unexpected breakdowns and downtimes) and corrective (activities to fix faults during any anomalous event and restore normal operation) maintenance. Similar challenges are faced by wind energy sources, wherein, as wind turbines continue to be deployed in challenging and complex environments (such as offshore), they become more prone to unexpected failures and downtimes, which can lead to significant costs for O&M to turbine operators. To ensure sustainable (and profitable) energy production for the operators (such as solar power companies and wind farm operators) as well as minimise downtimes and usage costs for organisations (households, industries etc.) switching to renewables, it is vital to bring trust and reliability to how these energy are operated and maintained. While Artificial Intelligence (AI) techniques have played an intrinsic role in shaping the future of many disciplines like e-commerce, healthcare, finance etc. with wide adoption across the core areas of Natural Language Processing (NLP)) and Computer Vision, it is interesting to note that the renewable energy sector has witnessed little attention in this area over the last decade.

Better late than never, the AI community for social good has started to realize the benefits which machine learning techniques (and more recently, deep learning) can play in tackling climate change and helping improve the reliability of renewable energy sources such as solar power and wind energy. A good reflection of this change is the growing number of events (workshops and conferences etc.) which are now being organized in the domain of AI for Tackling Climate Change and the rise in publications in this discipline. Notable amongst these is, for instance, the Climate Change AI Workshop which has been regularly held at leading AI conferences such as NeurIPS, ICML and ICLR since 2019. Such events gather a community of interdisciplinary researchers from academia and industry, all working towards a common goal- utilizing advances in AI for tackling climate change  Interestingly, a consensus which generally prevails at such events is the discussion on why the renewables sector is not progressing as much as other disciplines in applying AI. The leading reasons for this are the limited resources which are available to  researchers (including sufficient amounts of quality data) to develop AI models, which generally require large amounts of data for optimal training for making these renewable sources of energy more reliable, especially supporting their predictive and corrective

maintenance. And another reason, is most likely the lack of a clear perspective on the benefits which AI can provide in this aspect, which we aim to outline in this article.  

AI models are generally extremely competent in generating useful (and reflective) decisions for diverse situations, such as in expert systems. Imagine the everyday usage of AI in our lives: in chatbots, recommendation systems in Netflix and Youtube and so on, not to forget the likely bright future of self-driving vehicles and robots. In all such situations, AI models do one common thing- they generate optimal decisions for given environmental conditions. This is the very aspect which can help make AI a boon for the renewable energy sources, by supporting their O&M and helping e.g. diagnose operational anomalies and irregularities before they lead to unexpected outcomes. 

There has been some early (but very promising) demonstration of this premise in the real-world. Energy companies have utilised AI models, specifically using Microsoft Azure and Apache Spark to develop efficient decision support systems which can predict in advance the solar energy production over future time, improve profits and thereby ensure a stable energy future. Similar encouraging results have been demonstrated by DeepMind and Google in their efforts to utilise AI models, especially neural networks trained on historical weather forecasts to predict wind power output in wind turbines up to 36 hours in advance. Such applications are useful as they can help recommend optimal commitments to the grid operators for hourly delivery of energy, making wind farm operators better aware of the role which smarter, faster and significantly data-driven applications for decision support can play in meeting electricity demand. 

So what is the exact idea behind using AI models? This is most likely their diverse nature which span a family of algorithms ranging from simple regression models and classification techniques, to more sophisticated algorithms for reinforcement learning and natural language processing. All these models can play a vital role in:

  1. Time-series forecasting of vital operational parameters (e.g. solar power output, turbine power output, wind speed forecasts etc.), by adopting the task as a regression problem.
  2. Predicting operational anomalies in the system as a whole, or in its various sub-components, such as solar power controllers and solar panels, wind turbines and their many sub-components (gearbox, pitch system etc.), generally a classification problem.
  3. Generating explainable and transparent predictions in the form of human-intelligible messages through natural language generation. Here, AI models can be used to provide natural language descriptions of various causes which affect normal operation of the renewable sources of energy, and in some cases, also the appropriate techniques to fix ongoing/upcoming issues in O&M.

The best part of this is that the approaches are not necessarily limited to more popular sources of energy at present (wind, solar etc.), but can easily be adopted in other cases e.g. for tidal energy and hydro-power. In fact, in late 2019, there have been some interesting developments in initiation of projects on applying AI for tidal energy, relating to monitor and forecast waves (vital parameters such as wave heights), which can directly reduce the lifetime cost of energy significantly.

Hydro-energy is also not far behind in the endeavour to apply AI for tackling climate change. Some companies have started demonstrating the brilliant promise which AI and big data analytics hold in identify potential machine damages well in time through preventive maintenance of hydropower plants. There has also been a special focus in such projects on intelligent noise analysis, wherein, noise patterns (which vary under the conditions of anomalies) can help hydropower plant operators identify maintenance and future repair actions. 

In another domain, AI has been shown to improve biomass processing, with algorithms based on fuzzy logic for instance, being applied successfully for data-driven decision making process in utilizing sensor data (such as moisture, dirt etc. prevalent in biomass) to automatically control and adjust the Process Development Unit (PDU), a complex control system which converts biomass into useful energy/biopower. The idea behind showing all these successful examples is to arrive at an integral conclusion: AI can be a game-changer for the renewable energy sector, by learning optimally from historical conditions of operating such energy sources, specifically based on past successes and failures, and utilizing these to make highly accurate and useful predictions for O&M of such energy sources.

While we mention about all the positives here, it is important to not forget the challenges in applying AI to the renewables sector. AI needs to be adopted in a manner which is transparent, accurate and scalable, and ethically responsible. This may seem to be a topic of little relevance at present (given the limited application of AI In the renewables domain), but as these sources of energy continue to develop, become more complex and generate greater power, they would demand AI models which are not only accurate, but also explainable. Companies and operators of such energy sources would only adopt such models in practice, if the decisions made by the models (which are generally black-box natured) are clearly explained, including potential reasons for why and how the model generates certain predictions during decision support. And as always, we should never forget, “Good AI is not just accurate: it is transparent, ethical and scalable”.


Read more on The Good AI

You may also like

Comments are closed.