Blog | April 14, 2017
How machine learning improves energy consumption
At the intersection of machine learning and energy consumption stands an incredibly powerful force with the potential to transform the way we globally produce and consume energy. So powerful in fact, that the concept of merging machine learning and renewable resources has been named the “energy internet” by economic theorist and author Jeremy Rifkin or “digital efficiency” by Intel and GE.
Going green with machine learning solutions can drastically improve the way we consume energy, in terms of lower operational costs, more efficient production, better use of natural resources and lower environmental impacts.
Last year, Google, with the help of its U.K.-based subsidiary DeepMind, reduced the amount of energy used to cool its data centers by 40%. By introducing machine learning to compensate for the nonlinear interactions between equipment and environment, and using the unique architecture and environment of each data center, this decrease saves Google millions of dollars each year.
In many parts of the world, energy is created and dispersed on a grid via distributors, leading to environmental and economic waste. With machine learning, the grid can be transformed into a model more closely resembling a neural network, with two-way channels of communication and increased storage capabilities.
Grid operators will be able to use machine learning to model individual storage units through meters and sensors. In addition to large-scale grid transformation, the millions of smaller buildings that produce tiny amounts of energy and store it in hydrogen can be added to the network to send excess electricity across continents for others’ use, similar to the way media formerly had to be stored digitally and can now be shared across the cloud.
Collection and consumption
Before energy reaches the grid, machine learning has the ability to revolutionize the way it is collected. Detailed predictions of passive solar capacity and wind speed could improve the efficiency and reliability of the systems used to collect solar and wind energy. Systems in place at wind farms, for example, are able to account for wind’s speed and direction and adjust the angle and speed of the blades and rotors accordingly to maximize the farm’s intake while driving costs down.
At the opposite edge of the grid, it can be leveraged to accurately predict demand, based on variables like holidays and weather conditions, to understand when people are likely to consume more energy. By having a sense of both supply and demand before and after the energy is collected and transported, waste and cost can be significantly reduced.
In energy and in other industries, machine learning is touted as a maintenance godsend, alerting users to potential issues long before they require action. For instance, one of our partners uses machine learning to understand engine behaviors in heavy machinery, understanding when a machine is operating sub-optimally and assessing issues before they happen. In energy, predictive maintenance can prevent shortages and blackouts, saving providers and users millions of dollars and infinite headaches.
GE developed the digital twin, an exact digital replica of, in its example, a turbine, equipped with algorithms and a voice that will actually call the user anytime its physical counterpart may experience an issue. The twin can assess historical data, other turbines in the fleet and experiential knowledge of the stress on its counterpart, then converse with the user to find the often simple, preventative solution that will go on to save GE and its clients millions in down time.
Optimization through collaboration
GE, other energy and heavy industry leaders, and machine learning research groups have been working on addressing these challenges, yet implementing changes across highly centralized networks is no small task. Collaboration between startups and these traditional organizations enables both groups to better serve the market with innovative machine learning technology and spark a new network that optimizes legacy grids.
The Global e-Sustainability Initiative (GeSI) found that information and communication technologies can reduce global carbon dioxide emissions 20% by 2030 through machine learning-enabled solutions and reduce electricity expenditures by $1.2 trillion and fuel expenses by $1.1 trillion over the same time period. This landscape serves as an incredible opportunity for early investors and startups in this space to not only shake up an industry in need of disruption, but have a positive global impact, socially, environmentally and economically.