AI and Efficiency
Artificial intelligence without data is much like a red ribboned box without a present: it seems promising on the outside but it’s rather empty on the inside. Our computers are still not smart enough to learn without the help of large datasets, and haven’t really progressed to a point where they are able to solve things we, as humans, aren’t able to by ourselves. So, why AI at all then?
One word. Efficiency.
Efficiency is the one thing human brains aren’t good at; they are inferior to computers when it comes to processing information fast and taking into account all possible outcomes simultaneously. That’s why we started losing to them at chess and Go: ML-driven programs nowadays are so smart that, given the right computational capabilities, they can process even incredibly large datasets within seconds.
What makes this trivial detail much more fascinating is the fact that progress often happens when we uncover something new exactly by processing large datasets during a period of years, decades, and even centuries. Not one of humanity’s revolutionary inventions came out of nowhere: it was the pinnacle of a research process started by curious people many years before the final discovery.
Now think about electric power systems. Even though it may not seem that obvious to you at first sight, power systems are dependent on data. Lots of data. The more effectively it is managed, the lesser the costs for energy generation, transmission, and distribution.
Hence, the justification of this article. And the reason why you should go on reading.
The Story Behind
Ever since 2006, Google has worked dedicatedly on designing the most efficient data centres in the world, enabling the company to use the absolute minimum of energy to run them. The company has been carbon neutral since 2007. And in 2015, Google committed to the goal of purchasing more renewable energy than any other company in the world.
The strategies have not been merely eco-friendly: despite a tremendous growth in energy demand during the past five years, Google’s consumption has remained virtually flat. In fact, last year, Google proudly announced that it has reduced the amount of energy used for cooling the data centres by almost half (40 percent) by applying DeepMind’s machine learning algorithms to optimize the cooling processes.
The success didn’t go unnoticed. In March 2017, the news broke that Google’s DeepMind started discussing with the UK’s National Grid to use AI to balance the energy supply and demand across Britain. The discussions are still in the early stages, but the promising outcome already excite many.
“It would be amazing if you could save 10 per cent of the country’s energy usage without any new infrastructure, just from optimization,” said Demis Hassabis, DeepMind’s CEO.
Key Problems to Be Solved by ML and AI
Power systems are currently undergoing significant changes. There is an apparent shift of focus to renewable energy resources and distributed (decentralized) generation. However, despite all the buzz caused by electric cars and household solar panels, and despite the obvious advancements in the field, it wouldn’t be an exaggeration to say that humanity is still largely amateurish in the field.
The main problem is the broad range of factors and conditions and the need for sophisticated predictions. And mistakes cost lots of money – while impacting end users as well. And this is where smart systems appear in the picture.
Capable of concurrently “taking into account” plentiful parameters, AI-driven systems are able to make much more accurate predictions contributing to the development of an extremely efficient system and reducing the number of uncertainties to minimum. And ML and AI are efficient in more than one field:
- Energy trading
- Planning and predicting
- Diagnostics and restoration
To sum up: almost every aspect of power systems may benefit from the application of AI. Even though many of the applications are still at a trial stage, the results are already visible. Meaning: paying attention is essential if you don’t want to miss the next revolutionary discovery.
AI can be utilized to optimize power systems in many different ways, varying from conventional pseudo-AI techniques to advanced and complex networks which are able to outperform humans in few relevant areas. Thorough analysis of each should show the benefits much clearer, but for now, even a brief overview of the leading approaches may suffice.
Artificial Neural Networks (ANNs)
ANNs can be employed whenever there’s a need to make predictions and calculate possible outcomes. ANNs are trained on large data sets of system parameters, following which they are able to take into consideration nuances undistinguishable by humans. System planning and security checks can benefit immeasurably from the use of ANNs.
Expert Systems (XPSs)
Expert systems are computer programs which – much like the human experts from which they obtain their knowledge – are trained in a narrow domain and possess specific competence. Or, as described better in Artificial Intelligence Techniques in Power Systems:
“An expert system captures the knowledge of a human expert in a narrow-specified domain in a machine implementable form. It utilizes this (knowledge) to provide decision support at a level comparable to the human expert and is capable of justifying its reasoning.”
Additionally called knowledge-based or rule-based systems, expert systems are trained with the help of knowledge structures such as:
- production rules (translated into “if-then” logic),
- frames (data structures which divide knowledge into substrates by representing “stereotyped situations”),
- semantic nets (graphs which express relations within a concept), etc.
Despite its somewhat cartoonish name, fuzzy logic is a form of many-valued logic (any real number between 0 and 1 can be a logical value) intended to resemble human reasoning. Fuzzy logic has four main parts: 1) fuzzification module which converts system inputs (crisp numbers) into fuzzy sets; 2) knowledge base which stores “if-then” rules provided by experts; 3) inference engine which simulates human reasoning by juxtaposing the inputs against the knowledge base and producing “fuzzy set” conclusions; and 4) defuzzification module which converts these fuzzy sets into crisp numbers.
When perfected, fuzzy logic will mean that machines may be able to perform somewhat like humans; or, in other words, they should be able to plan and predict outcomes by themselves, and not follow strict rules when the opposite is essential.
As is the case with many other human endeavours, data mining has come to be vital when it comes to energy as well. In power systems, data mining is commonly used to solve energy market issues, such as gathering and structuring information, i.e. the most energy- and time-consuming aspect of data handling in general. This is why developing new data mining techniques is essential for maximizing efficiency whichever the field. Power systems are no exception.
Planning and predicting outcomes, developing and maintaining secure systems, diagnosing and restoring problems, handling the energy market – these are only few of the areas where power systems will greatly benefit from implementing AI. Even though most of the novel AI approaches are currently at a trial stage, each of them – whether it is data mining, ANNs, expert systems, or fuzzy logic – has already found some use in power systems.
And the effects are evident already.