Global warming is one of the biggest problems humanity is currently facing. Despite governmental efforts to reduce the output of greenhouse gasses, the emissions still rise every day. More and more, technology seems to be a promising tool to battle this issue.
In 2019, climatechange.ai published a paper highlighting some high-impact problems where existing gaps can be filled by machine learning. As mentioned in the paper, machine learning is not a silver bullet solution and will only have an impact when well-matched to specific problems.
As in many machine learning areas, there is a big gap between research and putting solutions into action. That is why we at ML6, decided to dive deep into some real-life use cases from four different domains where machine learning can be applied to help the planet.
The amount of produced and consumed energy in the grid always has to be equal, or else a black-out would occur. The increased reliance on renewable sources has immensely complicated the task of balancing the grid because the production of these sources varies based on external conditions such as the amount of wind or solar radiation.
Machine learning can be used to forecast the short-term production of wind and solar plants, as well as predict electricity demand to come closer to a grid that relies completely on green energy. These new machine learning algorithms will have to require domain-specific insights, and they will have to be more complex than most current methods in use. Additionally, long-term forecasts give operators insights into where and when new wind and solar farms should be built.
Currently, ML6 has a couple of solutions in production in the renewable energy domain. At Otary, for example, a machine learning model is used to give insights into what factors can lead to underperformance in wind turbines. This allows to remediate these effects and increase the overall power output.
The schedules for equipment systems such as HVAC (heating, ventilation, and air conditioning) in big offices or buildings, are often only set once or are operated by people on site. These schedules usually remain very static and imprecise. This might lead to very unwanted situations such as lighting and heating activating on a day where no one is present at the office.
Smart buildings and the humans using them generate a lot of data. Machine learning algorithms can learn from this data to automate HVAC tasks to anticipate occupant needs, reduce expenses (by using electricity when it’s cheap), and increase efficiency. A large variety of variables can be taken into account, increasing the overall comfort for the occupants. For example in a company setting, a system utilizing machine learning will use information about air quality, time of year, humidity, temperature, number of occupants, and even audio-visual needs of meeting attendees to create optimum conditions for that specific group of people. By making these decisions based on generated data (reflecting the occupant's needs) machine learning will improve energy management and therefore reduce a building’s carbon footprint.
Some remarkable results have been made in this domain such as Google’s DeepMind, which was able to reduce the Google data center cooling bill by 40%. Another example is DeltaQ, which reduced Sodexo’s energy costs by 30%, while significantly reducing time spent on operating the building’s energy systems.
In essence, smart buildings are quite a big and complex use case to tackle where various methods such as supervised learning, unsupervised learning, and reinforcement learning can be adopted. Using these methods will result in fewer expenses concerning electricity, lower carbon footprint, and higher occupant satisfaction and production.
It is estimated that more than half of the logging that takes place globally is illegal. Illegal loggers mainly use a technique called selective logging, which is in contrast to clear-cutting. With clear-cutting a large patch of forest is cut down, leaving little behind except wood debris and a denuded landscape. Selective logging is a practice where one or two trees are cut while leaving the rest intact before moving to another area. This technique makes it a lot harder to spot illegal loggers on satellite imagery.
Recent machine learning methods, however, have become increasingly powerful and they are accurate enough to spot selectively cut down trees from satellite images (which have also experienced great improvements in resolution).
Thus by using machine learning on satellite images to discover illegal logging activity, it is possible to maintain the indispensable rainforest trees that counter climate change. Additionally, it allows to fine those involved in illegal operations.
Products are frequently overproduced and overstocked. Excess products not only waste resources through their production but also cause greenhouse gas emissions when shipped and stored in climate-controlled warehouses.
Industries collect more data than ever and cloud storage and computing is becoming more affordable. By implementing demand forecasting based on this available data, supply and demand will be better balanced, lowering the produced waste. This avoids sunk costs for the company while at the same time reducing their carbon footprint. Additionally, demand forecasting gives a company valuable information about the potential of their products in certain regions. This allows managers to make more informed decisions about budgeting, pricing, and growth strategy.
In short, better forecasting means maximizing sales and results with more revenue and profit for retailers while reducing greenhouse gas emissions.
Machine learning is a promising tool to push back greenhouse gas emissions caused by human activity. This technique can deliver substantial added value in multiple domains such as electrical systems, smart buildings, deforestation, and industry. Often, the ecological benefits are also complimented with business value such as a reduction of expenses, automation of processes, etc.
Are you interested in starting up a CSR (corporate social responsibility) project with ML6? Please let us know!