Artificial intelligence (AI) is no new concept - in fact, we’ve been talking about it and attempting to perfect it for decades.
Yet, for various reasons including lack of data, low computing power and the resulting absence of funding, the technology has evaded us - until now.
Hal Varian, Chief Economist at Google, cites data as a fundamental cause behind this shift in capability, noting that in all of the time before 2003, the world generated around five exabytes of data. Since 2017, we’ve generated that much every two days - a number only set to increase.
Likewise, the availability and performance of computing power now enables more powerful AI algorithms to be developed, enabling this technology to permeate the world markets in a way never possible before.
Now, with the price of data and its generation being so low, new technologies such as AI, deep learning and machine learning can finally be realised in industries the world over - including in utilities.
In an Engerati webinar featuring Gianluca Mauro, Founder of AI Academy and Ronnie Dibbaut, Subject Matter Expert MPO and Product Innovation, Ferranti Computer Systems, we learned more about the use cases for AI in utilities, and how exactly machine learning develops.
How to create machine learning
The ability to create a knowledge base within computers is a crucial way for utilities to improve services, reduce costs and provide new services that increase revenue.
Traditionally, to teach a computer to solve a problem, programmers create a scenario and solution and translate it into code so that the computer can replicate it.
While there is nothing wrong with this process, Mauro explained, it can be improved by allowing computers to extract knowledge from data, instead of being hardcoded.
One method of creating machine learning algorithms, supervised learning, drives inspiration from the way that we, as humans, learn - by experience.
The new approach, Mauro continued, replicates this: “First, you need to get the right data that describes the problem you are trying to solve. Notice, we are not starting with a solution, we are starting with the data that describes a problem.”
He continues: “Then, we feed this data into the computer and the computer uses machine learning algorithms to learn from the data. Once you’ve done this, the machine can finally solve the problem.”
Once the learning process has been completed, machine learning and AI is then able to make inferences and predictions.
Another methodology that is less used in certain fields but equally important option is unsupervised learning, which transforms unstructured data into data clusters.
Using machine learning algorithms, utilities can implement new measures to solve a variety of common issues within the energy industry. Mauro identified a few use cases as follows:
1] Demand-response management
Before renewables permeated the market, demand-response management was simpler - the volatile demand side could be matched sufficiently by plannable fossil production.
Now, however, renewable energy adds another volatile player to the chain, and as usage increases, so too does the risk of imbalanced demand-response management.
Mauro explained how AI can assist in mitigating this in demand side predictions: “We take some variables that we think are going to affect the demand, for instance occupancy of buildings, sun radiation, etc., as well as the previous building consumption. The machine learning can then predict the consumption and what the utilities need to match.”
Furthermore, the same systems can be used to ascertain what renewable generation may be achieved by evaluating external factors and their effects on generation to predict the energy production.
Finding the best location to build new renewable assets can be just as challenging as monitoring them - a problem Mauro has identified that could also be mitigated by AI.
He says: “You can use the same information to predict how much energy a farm can produce in a certain place, and therefore optimise the location of your investment.”
2] Energy savings and resources optimisation
By using a neural network, utilities can improve a data centre’s power usage effectiveness (PUE) - something for which Google has also used its advanced AI.
“What Google has done is build what is called a digital twin. They used information to model the PUE of the data centre, and once they’d done that, they optimised it without having to test it in reality.”
The results, Mauro says, were astonishing, with Google saving 40% on its energy usage.
The same approach can be utilised for predictive maintenance, where asset data including vibrations, temperature, speed and consumption can be fed into a machine learning algorithm that can then predict when or where a fault may occur.
This can save money in repairs and replacements and reduce the need for specialist attendance to machines thanks to the higher accuracy of machine learning.
3] Providing better energy insights
By using machine learning to manage and sort data more efficiently, Mauro explained how utilities can improve insights and get greater market intelligence.
He demonstrated this through the example of Opower, which, following an acquisition by Oracle, used unsupervised machine learning in order to finesse its data capturing and analysis.
By changing the data from what Mauro identified as a ‘hairball’ graph with 1,000 inputs of daily consumption rates into separate data clusters, Opower was able to identify five key types of consumer patterns.
This new wealth of comprehensible data enables the utility to provide more tailored services, to ensure optimal return and set up demand-response programmes.
Watch the webinar on-demand
To learn more about key takeaways and learnings from the development of AI technology and machine learning, watch our Engerati webinar “Applying artificial intelligence to energy challenges” on-demand now!