Machine learning (ML) is touted as a technology on the verge of changing how we plan and optimize not only our businesses, but also our lives. The onset of the climate crisis leads us to ask questions about how we can use this technology to help fight ― and eventually prevent ― overall climate change over the next few decades. I’d like to take a few minutes to help frame the discussion for anyone interested in using ML to combat this threat. However, it is important to understand that, like many other efforts aimed at combating climate change, it won’t be a straight-forward and overnight process. Rather, it will require (re)thinking many of the ways we operate our businesses and ―even more ― how we operate as humans.
The nature of ML is to find an underlying consistency in the data provided by an underlying data-generating system. The increase in ways in which we can measure data-generating systems ― whether they are electrical power and delivery systems, the many forms of transportation systems, or the materials and systems used in construction of buildings ―allows us to identify underlying consistencies or patterns and use them to understand how we can optimize utilization and construction of the component systems. This is a long-winded way of saying that any system we’re able to measure could be subject to the application of ML to optimize operations, administration and maintenance of that system.
Let’s look at two specific areas of on-going research and application, and at how ML is changing system operations in ways that can positively impact our environment and, ultimately, the future of our planet.
Transportation greenhouse gas (GHG) emissions account for about a quarter of global energy-related CO2 emissions. Attacking this problem head-on will not only allow businesses to realize potential cost savings within their logistics spend, but may also lead to a larger overall environment net positive impact. It’s vital that we, as business leaders, pay close attention to how these technologies evolve and to how we can utilize them within our organizations.
Transportation is one of the hottest areas where ML is being applied today by way of autonomous driving. Many automotive industry leaders, even those who aren’t traditional automotive manufacturers, are investing heavily in pushing self-driving car technologies to the limits of what is possible. However, let’s take a moment to look beyond this example.
Whenever we produce a physical product that will be purchased by consumers, we’re required to source the sub-components of our product (if we don’t manufacture them) and to deliver the product to brick-and-mortar stores and/or to customers who purchase it. Wrapped up in many of the related business decisions is the opportunity to apply ML to a common problem. Specifically, how does one source and/or deliver products while minimizing the amount of resources required to make it happen?
This is sometimes referred to as the traveling salesman problem (TSP). However, there’s a more general problem hidden within TSP known as the vehicle routing problem (VRP).1 VRP is a framework we can use to describe and formulate many product sourcing and distribution problems that present themselves in the real world. While there are many ways to solve the TSP problem, and many algorithms can provide some form of provable guarantees, we can utilize methods from ML to push these methods even further and to make them amenable to the stochastic nature of the changing world of customer demand and product distributions.
We can also look to apply ML to things like increasing overall vehicle efficiency. With the race toward hybrid and electric vehicles, there are ways in which ML can be applied to power management methods.
- Work has been done to measure the operating envelope of internal combustion engines to understand when potential mis-fires may occur in a vehicle and to allow for control and planning.
- We can apply ML to modeling power use of hybrid electric vehicles for online model predictive control that allows for real-time optimal battery usage solutions while the car is being operated.2
Electricity is yet another resource that’s a cause of large amounts of GHG emissions generated by humans. It accounts for about one quarter of the GHG generated today.3 As industries and societies strive to move toward low(er)-carbon sources of electricity, it’s important to understand that there are many ways in which ML can be applied to the generation, transportation and consumption of electricity within many facets of both business and residential systems. Since we’re concentrating on technology applications within our businesses, I will restrict the discussion to how we can use ML to help alleviate our dependence on fossil fuel-powered electricity.
Many businesses are interested in leveraging solar or wind for power generation within their facilities. It’s important to understand that these two sources of power are considered “variable,” in that they do not provide a constant feed of electricity over time. For example, sometimes solar panels are operating at less than peak efficiency due to environmental conditions, such as debris or clouds. This can cause variance in the amount of electricity being provided. ML can be applied in these systems in a way that allows for forecasting, scheduling and controlling delivery rates from these inherently variable sources.4
We can also look to address storage of power being generated in excess of current consumption. When our systems are running at peak power, and we’re collecting more power than we’re using, we will need to store this energy for future use. Current materials research is leveraging solvers from the ML space to help identify how to structure future materials to store this energy.5
There are less-obvious ways in which we can utilize ML methods to identify potentially useful energy-related advances. For example, we can apply natural language processing (NLP) techniques to patent data to help us understand networks of innovation. Or we can apply NLP to look at how a patent knowledge graph is built over geographic locales.6
In this post, I’ve covered a small number of the many ways ML will impact businesses moving forward. However, when we think about future applications, we can do so in the context of anything we are able to measure in some meaningful way. We may be able to leverage ML to make a new discovery of an underlying structure present in a system and use the information to change the way systems operate to positively impact our environment! It’s up to everyone, regardless of affiliation, to help ease the burden we have on our planet, and I can think of no better way than using the advances being made in the field of ML.
To learn more about unlocking the value of data with artificial intelligence systems, explore Dell EMC AI Solutions and Dell EMC Ready Solutions for AI.
Ed Henry is a senior scientist and principal architect for machine learning at Dell EMC.
- Vehicle routing problem. Wikipedia, accessed July 9, 2019.
- Ahmed M. Ali and Dirk S¨offker. Towards optimal power management of hybrid electric vehicles in real-time: A review on methods, challenges, and state-of-the-art solutions. Energies, 11(3), 2018.
- Sources of Greenhouse Gas Emissions, U.S. Environmental Protection Agency, accessed July 9, 2019.
- C. Voyant et al., “Machine learning methods for solar radiation forecasting: A review,” Renew. Energy, vol. 105, pp. 569–582, 2017.
- J. Bai et al., “Phase Mapper: Accelerating Materials Discovery with AI,” AI Mag., vol. 39, no. 1, pp. 15–26, 2018.
- S. Venugopalan and V. Rai, “Topic based classification and pattern identification in patents,” Technol. Forecast. Soc. Change, vol. 94, pp. 236–250, 2015.
- D. Rolnick et al., “Tackling Climate Change with Machine Learning,” Jun. 2019.