Author: Marcel Horstmann.

What is it that self-driving cars and smart home energy use have in common?

They are both enabled by a technology called ‘deep learning’. Deep learning allows computer algorithms to learn very complex rules and patterns, just by giving the algorithm a set of example data to learn from.

Why is it relevant?

Worldwide, utilities are trying to ‘go digital’ with the aim of delivering a much better customer experience. Current electricity customers are rightfully upset when receiving an expensive bill, that just states cryptic numbers: 3682kWh, $920.50!

“Where did that money go?”

“What can I do about it?”

“Is my utility overcharging me?”

These are the questions on the minds of millions of utility customers worldwide. Thousands of utility call-center service workers know them all too well. The problem is: Without a detailed insight into where that electricity actually went, the question is very hard to answer, and it is hard to keep the customer happy.

This is where load disaggregation comes into play: This technology, applied to ever more available smart meter data, breaks the utility bill down into its components (consumption of the fridge, the washing machine, the EV, ….), allowing detailed insights into where the electricity (and the money!) actually went. Was it that innocent looking space heater, which has been running for months on end? Or that old inefficient fridge in the garage, that we forgot to turn off after the last BBQ?

blog image

Energy insights driving smart and sustainable home energy use.

Our algorithms deliver these answers. They are based on deep learning allowing these insights to be delivered with both unprecedented accuracy and scale.

Deep learning explained

For the self-driving car, think of: “That pedestrian over there, she’s going to cross the road right in front of us! Brake, now!”

This is intuitive for human drivers with years of experience. As long as we are paying attention (and are neither drunk nor staring on our iPhone…), we will consistently apply it.

Conventional approaches of programming a computer to act in the same way are cumbersome to implement. They require incredible amounts of engineering efforts, as all possible rules, driving scenarios and obstacles have to be precisely written in computer language. A Herculean effort, impossible to execute given the ever changing nature of our roads and cities.

blog image

A self-driving car finding its way in our complex world.

With deep learning, the computer learns from sets of examples that we provide.

Like a student at driving school, the computer metaphorically looks over our shoulder. The computer can watch what what we do and what is the desired behaviour in any given situation. After a while the computer is allowed to drive, and only corrected when an action taken is not appropriate. These corrections are fed back to the deep learning algorithms, which re-adapt their internal configuration, learning from their mistake. It will not happen again!

California-based electric car company Tesla Motors is running a very large data collection effort in this space: All their recent cars are equipped with sensors and cameras that monitor a 360° view around the car. The self-driving algorithms learn from the collective wisdom of tens of thousands of drivers worldwide.

Deep Learning@ONZO — Load Disaggregation

We here at ONZO are doing a very similar thing for load disaggregation. Since our inception in 2009, we have been collecting huge amounts of data about different appliances worldwide: How and when they are used. What their power consumption patterns look like.

Our deep learning algorithms are becoming better and better every day at breaking utility bills down into their components. From “3682kWh, $920.50!”, we make “$50 for your washing machine, $100 for the fridge, $80 for cooking — and yes, all the rest was due to that smallish space heater!”

blog image

Bill breakdown allowing customers to understand their energy usage.

The result is helpful for both the end customer, who gains a much better understanding of their electricity consumption. How they can reduce it, contributing to energy sustainability in their own home. And for the utility, who can now understand the unique needs of their customers:

  • Why is this group of customers about to leave?
  • Who is most in need of advice about energy efficiency?
  • Who would be most interested in having a solar panel installation on their rooftop?