5 min read

Last week, a team of machine learning experts published a paper titled “Tackling Climate Change with Machine Learning”. The paper highlights how machine learning can be used to reduce greenhouse gas emissions and help society adapt to a changing climate.

Climate change and its consequences are becoming more apparent to us day by day. And, one of the most significant ones is global warming, which is mainly caused by the emission of greenhouse gases. The report suggests that we can mitigate this problem by making changes to the existing electricity systems, transportation, buildings, industry, and land use. For adapting to the changing climate we need climate modeling, risk prediction, and planning for resilience and disaster management. This 54-page report lists various steps involving machine learning that can help us adapt and mitigate the problem of greenhouse gas emissions.

In this article, we look at how machine learning and deep learning can be used for reducing greenhouse gas emissions from electricity systems:

Electricity systems

A quarter of human-caused greenhouse gas emissions come from electricity systems. To minimize this we need to switch to low-carbon electricity sources. Additionally, we should also take steps to reduce emissions from existing carbon-emitting power plants.

There are two types of low-carbon electricity sources: variable and controllable:

Variable sources

Variable sources are those that fluctuate based on external factors, for instance, the energy produced by solar panels depend on the sunlight.

Power generation and demand forecasting

Though ML and deep learning methods have been applied to power generation and demand forecasting previously, it was done using domain-agnostic techniques. For instance, using clustering techniques on households or game theory, optimization, regression, or online learning to predict disaggregated quantities from aggregate electricity signals.

This study suggests that future ML algorithms should incorporate domain-specific insights. They should be created using the innovations in climate modeling and weather forecasting and in hybrid-plus-ML modeling techniques. These techniques will help in improving both short and medium-term forecasts. ML models can be used to directly optimize for system goals.

Improving scheduling and flexible demand

ML can play an important role in improving the existing centralized process of scheduling and dispatching by speeding up power system optimization problems. It can be used to fit fast function approximators to existing optimization problems or provide good starting points for optimization. Dynamic scheduling and safe reinforcement learning can also be used to balance the electric grid in real time to accommodate variable generation or demand.

ML or other simpler techniques can enable flexible demand by making storage and smart devices automatically respond to electricity prices. To provide appropriate signals for flexible demand, system operators can design electricity prices based on, for example, forecasts of variable electricity or grid emissions.

Accelerated science for materials

Many scientists are working to introduce new materials that are capable of storing or harnessing energy from variable natural resources more efficiently. For instance, solar fuels are synthetic fuels produced from sunlight or solar heat. It can capture solar energy when the sun is up and then store this energy for later use. However, coming up with new materials can prove to be very slow and imprecise.

There are times when human experts do not understand the physics behind these materials and have to manually apply heuristics to understand a proposed material’s physical properties. ML techniques can prove to be helpful in such cases. They can be used to automate this process by combining “heuristics with experimental data, physics, and reasoning to apply and even extend existing physical knowledge.

Controllable sources

Controllable sources can be turned on and off, for instance, nuclear or geothermal plants.

Nuclear power plants

Nuclear power plants are very important to meet climate change goals. However, they do pose some really significant challenges including public safety, waste disposal, slow technological learning, and high costs. ML, specifically deep networks can be used to reduce maintenance costs. They can speed up inspections by detecting cracks and anomalies from image and video data or by preemptively detecting faults from high-dimensional sensor and simulation data.

Nuclear fusion reactors

Nuclear fusion reactors are capable of producing safe and carbon-free electricity with the help of virtually limitless hydrogen fuel supply. But, right now they consume more energy that they produce. A lot of scientific and engineering research is still needed to be done before we can use nuclear fusion reactors to facilitate users.

ML can be used to accelerate this research by guiding experimental design and monitoring physical processes. As nuclear fusion reactors have a large number of tunable parameters, ML can help prioritize which parameter configurations should be explored during physical experiments.

Reducing the current electricity system climate impacts

Reducing life-cycle fossil fuel emissions

While we work towards bringing low-carbon electricity systems to society, it is important to reduce emissions from the current fossil fuel power generation. ML can be used to prevent the leakage of methane from natural gas pipelines and compressor stations. Earlier, people have used sensor and satellite data to proactively suggest pipeline maintenance or detect existing leaks. ML can be used to improve and scale the existing solutions.

Reducing system waste

As electricity is supplied to the consumers, some of it gets lost as resistive heat on electricity lines. While we cannot eliminate these losses completely, it can be significantly mitigated to reduce waste and emissions. ML can help prevent avoidable losses through predictive maintenance by suggesting proactive electricity grid upgrades.

To know more in detail about how machine learning can help reduce the impact of climate change, check out the report.

Read Next

Deep learning models have massive carbon footprints, can photonic chips help reduce power consumption?

Now there’s a CycleGAN to visualize the effects of climate change. But is this enough to mobilize action?

ICLR 2019 Highlights: Algorithmic fairness, AI for social good, climate change, protein structures, GAN magic, adversarial ML and much more