Tackling climate change with machine learning [part 1] – Electricity systems

By loading the video, you agree to YouTube's privacy policy.
Learn more

Load video

Always unblock YouTube

On 10th of June, 2019, twenty-two AI researchers, including Andrew Ng, David Rolnick and Yoshua Bengio, published a paper on how climate change can be tackled with machine learning. I really enjoyed reading it and I am convinced that the paper as well as the climatechange.ai initative, which emerged from it, deserve more attention. For that reason i created a series of blog posts and videos which provide a dense summary, listing many of the proposed solutions and linking research work as well as ongoing projects. In the big picture, all solutions aim to reduce greenhouse gas emissions.

As my contribution to the global #ClimateStrike week from September 20th to 27th, I will post one chapter (video and blog post) on every working day. You can subscribe to our YouTube channel or follow me on Twitter to be notified when new content is published.

This is part one of a six-part series:

1. Electricity Systems
2. Transportation
3. Buildings & Cities
4. Farms & Forests
5. Industry & Carbon Dioxide Removal
6. Datasets & further resources

Notation

Some solutions are highlighted by the following flags:

• [high leverage]: bottlenecks that could be particularly well solved / improved with ML
• [long-term]: solutions that might have their primary impact after 2040
• [high risk]: solutions where either the technology proposed is uncertain, or it might not lower greenhouse gases as expected or negative side effects might occur

Furthermore, most solutions include brackets with a link to corresponding research work and the technique / algorithm used (if i figured it out), for example:

[Paper – Reinforcement Learning]

So let’s get started!

Electricity systems

Electricity systems are responsible for a quarter of human-caused greenhouse gas emissions.

Machine Learning can help reduce the carbon footprint of electricity systems by

• ↓ forecasting power generation & demand
• ↓ accelerating materials science
• ↓ optimizing the capture of ambient energy
• ↓ increasing the safety of nuclear power plants
• ↓ advancing research of nuclear fusion
• ↓ reducing fossil fuel and electricity loss during transport
• ↓ modeling live emissions of electricity
• ↓ collecting data in regions with scarce information about the local grid

[high leverage]

Since renewable energy production varies depending on wind conditions and solar radiation, energy grid operators use polluting standby plants, which can feed electricity into the grid when unexpected production shortages occur. By applying Machine Learning to better forecast how much power will be generated by renewable sources and also how much power demand there is, the reliance on polluting standby plants can be reduced. Also, long-term forecasts help operators understand where and how much new renewable energy plants should be built.

ML can help

• forecast energy produced by solar
  [Paper – Long-Term Recurrent Convolutional Network] [Paper ] [Paper – SVM, Regression Trees, Random Forests] [Paper – CNN] [Paper – Recurrent Neural Networks]
• forecast energy produced by wind
  [Paper – Extreme learning machine] [Paper – SVM, Genetic algorithm]
• forecast energy produced by water [Paper ]
• make better long-term forecasts which can inform decisions how much variable plants should be built.
• to better predict power demand [Paper – Regression, Neural Networks]

[high leverage] [long-term] [high risk]

The process of discovering new materials can be slow and imprecise, the physics behind materials are not well understood so experts apply heuristics. ML can accelerate the discovery by automating parts of the process by combining existing heuristics with experimental data, physics and reasoning.

ML can help

• accelerate the material discovery for alternative fuels, e.g. solar fuels by understanding a proposed material’s structure [Paper ] [Paper] [Paper ]
• improve battery storage technologies with support-vector regression to design conducting solids for lithium-ion batteries [Paper ]
• design alternatives to cement or create better CO2 sorbents
• improve chemical experiments by identifying promising catalysts [Paper – Neural Networks] [Paper – Graph Convolutional Networks]

Solar panel detection [Malof  et al., 2016 ]

For wind turbines and solar panels to capture as much ambient energy as possible poses a challenge.

ML can help

• • • detect rooftop solar panels and inform electricity system operators about estimated solar capacities [Paper – Random Forest]
    • detect faults in rooftop solar panels [Paper – Regression]
    • optimally place wind turbines in wind farms [Patent – Genetic Algorithm]
    • control wind turbine blades and movable solar panels [Paper – Reinforcement Learning, Bayesian Optimization]

Nuclear power plants might be essential to meeting climate change goals, although they face significant challenges regarding public safety and waste disposal.

ML can help

• speed up inspections by detecting cracks and anomalies from image and video data [Paper – Convolutional Neural Networks, Naive Bayes]
• detect faults from high-dimensional sensor and simulation data [Paper – Convolutional Neural Networks, Denoising Autoencoders, k-means Clustering]
• design next-generation nuclear reactors or simulate nuclear waste disposal options (speculation)

[high leverage] [long-term] [high risk]

Disruption prediction workflow of nuclear fusion reactor [Kates-Harbeck et al., 2019 ]

Nuclear fusion reactors have the potential to produce safe and carbon-free electricity using hydrogen fuel but today have a negative energy balance: they consume more energy than they produce. Research in this field can be accelerated by ML.

ML can help

• • • guide experimental reactor design [Paper ]
    • monitor physical processes [Paper – stochastic perturbation method (actually not machine learning)]
    • detect disruptions in tokamak reactors [Paper – CNN, RNN] [Paper ] [Paper – SVM] [Paper – Neural Network] [Paper – Neural Network]
    • steer plasma into safe states through reactor control by simulating how plasma’s state evolves over time (speculation)

[high leverage]

Some of the fossil fuel gets lost during transport, for example due to leaks in pipelines. In the case of Methane, its release into the atmosphere should be avoided since it is roughly 30 times more potent as a heat-trapping gas than CO2.

Detecting pipeline leaks with sensors [Wu et al., 2012 ]

ML can help

• • • prevent leakage of methane from natural gas pipelines and compressor stations by detecting pipeline damages and predicting pipeline maintenance with sensor & satellite data [Paper  – SVM] [Paper ] [Project Bluefield Technologies ] [SLED Project ]

As electricity gets transported to consumers, some of it gets lost as resistive heat on electricity lines.

ML can help

• • • lower avoidable losses by suggesting electricity grid upgrades with predictive maintenance [Paper – LSTM] [Paper – Bipartite ranking] [Paper – Neural Network & Clustering]

realtime information on electricity composition [electricitymap.org ]

Without knowing how much emissions are caused by consuming electricity at every moment, industry and consumers cannot make climate-friendly decisions regarding when machines with high energy consumptions should be run.

ML can help

• • • model current and future emissions of electricity and thereby informing people when the consumption of electricity would cause lower emissions. Have a look at these projects:
      • [WattTime project (USA) – Regression techniques]
      • [electricityMap project (Europe) – Ensemble models]
      • [Carbon Intensity project (Great Britain) – Ensemble models]
    • improve the planning of electricity grids in developing countries and thereby reduce the need for diesel generators, wood-burning stoves etc. used by isolated communities
    • generate data which aids energy access policy

• • • better manage energy access solutions by balancing electricity load in rural microgrids [Paper – LSTM, Neural Network]

[high leverage]

Many countries do not share or even collect data about their electricity system.

ML can help

• gather information through satellite imagery or cellular network data.
• translate insights from data-abundant to data-scarce regions using transfer learning

Potential pitfall:

Innovations that reduce greenhouse gas emissions in the oil and gas industries could instead increase emissions by making these sources cheaper and thereby making them more competitive to renewable energy sources.

Credits

Many thanks to all researchers of the paper:

David Rolnick, Andrew Y. Ng, Priya L. Donti, Lynn H. Kaack, Kelly Kochanski, Alexandre Lacoste,  Kris Sankaran, Andrew Slavin Ross, Nikola Milojevic-Dupont, Natasha Jaques, Anna Waldman-Brown, Yoshua Bengio, Alexandra Luccioni,  Tegan Maharaj, Evan D. Sherwin, S. Karthik Mukkavilli, Konrad P. Kording, Carla Gomes, Demis Hassabis, John C. Platt, Felix Creutzig and Jennifer Chayes.

→ Continue to Part 2: Transportation