Leonardo Di Caprio did an amazing job with “Before the flood”: he provided a great overview of the dramatic situation of our environment, but most importantly he raised the awareness of many people that wouldn’t have cared otherwise.
As an energy engineer, I dedicated 5 years of my life to study how we produce and consume energy, and I’ve always felt this was a great choice to contribute to one of the biggest problems we are facing as a species. Later, I got passionate about Artificial Intelligence, which many people may think is unrelated to my main field of study. But is it?
“Before the flood” had such an effect on a friend of mine that he criticized me for investing in AI rather than focusing on my original field of study: he believed that one automatically excludes the other. An interesting discussion started where I convinced him that not only AI can be applied to solve energy problems, but it’s the best tool we have in our hands, right now. Now I want to share my main arguments with you.
#1: One of the biggest obstacles to renewables is a data problem
For those who don’t know, the energy you use in a certain moment to turn on your lights is produced exactly in that moment in some powerplant of your country. This is mainly due to the fact that energy can’t be efficiently stored in big quantities. This implies that if the grid demands X kW of power now, you need to answer producing X kW of power at the same moment. Produce significantly less, you’ll have a blackout. Produce significantly more, you’ll have an overload, and then probably a blackout. That’s not nice.
Until the rise of renewable energy you could match production and consumption simply by forecasting the demand. You forecast how much the grid will want tomorrow, and you plan the production on time: there’s only one random variable in play. It’s like having guests for dinner and knowing that they won’t bring any food along to help you: you need to correctly forecast just how much food to buy in order to satisfy them.
But what happens when you start producing energy also using renewable resources? You add a stochastic variable to the equation: now you need to forecast the energy demand from the grid, and the energy production from renewable resources at the same time. If you correctly forecast that the grid tomorrow will demand X kW of power at a certain moment, you need to know what power Y will be provided by renewable resources, and then set your thermoelectric power plants to produce X-Y. It’s like organizing a dinner, and knowing that your guests will bring some food to help you, but not how much: now you need to predict both how much food they’ll bring, and how much to add to make them feel satisfied. Two random variables: double the probability of making a mistake.
As the price of solar energy drops, energy markets get deregulated, and people start producing power on their own rooftops and selling it to the grid, the problem becomes nastier. Now not only the production is unpredictable, but spread all over thousands of micro-generation sites, a opposed as the old paradigm of a few powerplants to power the whole country.
Forecasting correctly the amount of energy to produce is crucial to a smooth transition to renewable energy. Not unlike financial predictions or Netflix movie recommendations, this is a data problem, not an energy issue: AI can help big time.
#2: In a cash-strapped economy, optimizing is better than renovating.
Imagine that some crazy scientist invents an amazing new technology that leveraging new crazy materials cuts heating bills in half. Who would afford the investment to deploy it at scale? New technologies are often expensive for a while until economies of scale and industrialization kick in, but if you had some of the sense of urgency I felt with Di Caprio’s documentary, you’ll understand that we don’t have much time.
What if you could increase the efficiency of your building’s heating system without changing anything, but simply optimizing what you already have and enjoy a +30% in efficiency, from today. That’s a job for AI, and that’s amazing because:
- We don’t need to wait for the amazing discovery of the crazy scientist
- It amplifies the potential of what you already have available, meaning that the upfront installation cost can be close to zero. This is particularly important when capital is scarce, and the big investments needed for renovations are painful.
I know that increasing the efficiency of a system without changing anything but the way it’s controlled may sound unrealistic, but it’s true. Very true. Energy systems are often governed by complex non-linear relations that traditional formulas and human intuition fail to find solutions that can be instead found with AI. Take a look at what Google did in their datacenters, where a simple AI algorithm achieved an energy consumption reduction of 40%.
Switching to a clean energy produced from clean sources can’t be the only solution to the energy problem, it has to go together with increasing efficiency both in production and utilization.
#3: Energy systems evolve with time.
Your car doesn’t have the same performances now that when you bought it. It also doesn’t behave in the same way in summer and winter. The same is true for more energy-consuming systems like a thermoelectric plant, or the cooling system of a massive mall. Wear of components and external factors such as temperature and humidity affect its behavior. Why don’t we update the control logics when the behavior of a system changes?
That used to be hard, once. Keeping track of the decrease of performances of a machine due to components’ wear for instance, is not a simple task for traditional formulas. But guess what? Is perfect food for AI.
Not only an algorithm can update its model of an energy system with time, exactly like Facebook’s news feed algorithm updates when you start clicking on contents, but it’s getting cheaper and cheaper (and easier) with the spread of cloud computing solutions, that allow constant real time processing of data streams coming from a plant.
To the best of my knowledge, this kind of approach is unprecedented, and allows us to set up a system to always run at its best working conditions, increasing (again) its efficiency during its whole lifetime.
#4: We never had so much data.
AI is a simple recipe: take a powerful enough computer, find a learning algorithm that works well for your problem, and add as much data as you can. If you don’t have enough data, you’ll find yourself potentially with a Ferrari and no gasoline to run it.
Guess what? We never had so much data for energy systems. Buzzwords like “Industry 4.0”, “Internet of Things (IoT)”, are a real thing, and are making possible for energy engineers and data scientist to meet, applying powerful AI techniques to important energy efficiency optimization problems.
Those arguments seem to be convincing the industrial workforce to bet on Big Data Analytics. A very interesting report by Accenture together with General Electrics called “Industrial Internet Insights Report” presents some clear insights on this trend, like this one:
I hope those four arguments convinced you too of the potential of AI in helping us to solve the energy challenge. That’s a problem our planet’s future depends on, and the more tools we have to solve it, the brighter our future looks like.
If you found this article interesting, please recommend and share it, it would mean a lot to me :)