How to get actionable insights from energy consumption data streams in minutes?
As energy prices keep going up and more environmental regulations are issued, many B2B and B2C firms are committed to optimizing their energy consumption. From coaching occupants to detecting malfunctioning devices and identifying consumption drifts, there is much to learn from consumption data: it carries valuable, yet personal information about individual habits. But monitoring it is hard, and understanding it even harder.
craft ai energy: The AI building blocks for energy performance optimization
“ craft ai energy brings craft ai’s individual-level automated machine learning solution to the world of energy.”
Predict, Alert, Explain
Because energy consumption is inherently related to human activity it reflects its variability. Your habits are not those of your neighbors, and designing general rules that efficiently describe them all means losing valuable insights at the individual level. Moreover, extracting rules from personal data is hard. Habits depend on many intertwined factors, and so does the data: it’s hard to reduce that complexity to simple rules.
craft ai can continuously extract meaningful patterns from energy consumption . By learning individual behaviors, it can output personalized predictions on future consumptions and detect anomalies. Every decision taken by the model is expressed in simple rules that speak your language, so you don’t have to trust a black box. But let’s see how this all works in practice.
Saturday Night Fever
Let us consider a fictional household in Canada where Eva lives with her family. From 2012 to 2014, Eva measured her instantaneous power consumption every half-hour. On the graph below, click on the menu on top right to chose a week of data. Let’s compare it with the week before by selecting Last week’s load. As expected, great variations are observed from one week to another. Saturday night of week 96 for example, Eva consumed more than twice the amount of power she had needed the Saturday before. Is it possible to detect a pattern?
By feeding craft ai energy with Eva’s data, rules are extracted about her and her household’s consumption. These rules can then be used to predict future consumption. Every Sunday, craft ai energy uses Eva’s past data and future weather reports to predict her consumption for the coming week. You can select craft ai predictions to see how they compare to Eva’s real consumption. The green area defines the confidence interval for every prediction made: that is, every measure that falls within its boundaries is an observation of a normal behavior. This allows to detect anomalies in Eva’s consumption and alert her when needed! In this case, Eva’s energy-expensive saturday evening falls within the expected load values: it is nothing to be worried about! The same cannot be said about her Monday afternoon: select the Anomalies checkbox to display all the anomalies. Because every decision of the model is explainable, it is easy to backtrack along the model’s decision process and extract the rules it followed. Click on a red dot, and look at the corresponding rules below.
Energy data can bring you valuable insight to understand, predict or optimize your consumption. But they often display both a great complexity and a variability that make these tasks very hard. craft ai energy provides an easy way to extract meaningful patterns from your data and detect anomalies. As a result, craft ai energy empowers energy companies to quickly deploy & run explainable AIs that improve energy management performance and deliver hyper-personalized experience. Those AIs are used by craft ai client to enable process automation, predictive maintenance, coaching and smart push.
If you want to learn more about craft ai energy solution download the official presentation below and/or contact us !