Home

Awesome

Energywise: understanding building energy use

<img src="http://dssg.io/img/partners/lbnl.png" width="200"> Agentis

An energy analytics tool to make commercial building more energy efficient. Energywise profiles a building's energy consumption and gives building managers insights on how to mke their buildings more energy efficient.

This project is a rewrite and extension of Lawrence Berkeley National Laboratory's open source fingerprint tool.

This project is part of the 2013 Data Science for Social Good fellowship, in partnership with Lawrence Berkeley National Laboratory and Agentis Energy.

The Problem: uncertain energy savings for building retrofits

Buildings use too much energy, and nobody knows what to do about it.

Normally, the people that can take action on what goes on in a building are building managers. To reduce a building's energy consumption, they need to know how their building uses energy to begin with.

Even though they would like to spend less in their energy bills, it is too risky to decide what changes to make in their building to become more energy efficient because accurately quantify savings is hard. As a result, building energy efficiency goes unexploited.

Read more about the problem in the wiki

The Solution: energy analytics tool

Energywise is a tool that helps buildings managers identify ways to reduce their building's energy consumption. The tool consumes energy use data collected by a building's smart meter, analyzes the data, and generates a report that iluminates the connection between building behaviors and electric consumption.

The analysis includes:

  1. Phantom load: each building has systems that operate all the time. We estimate the building's baseline energy consumption produced by its systems.

  2. Periodicity: buildings have regular, predictable cycles in their energy consumption. Examples include:

    • Weekend/Weekday
    • Time of the day
    • Holidays
    • Schedule
  3. Anomaly detection: Classifying days according to their energy consumption patterns. Days can be grouped according to their cycles, periodicity and variability.

    <img src="http://dssg.io/img/posts/anomaly_detection.png" width="600">
  4. Outlier detection: within each group of days, particularly extreme days can be recognized. This helps identify erratic energy behavior within a building.

    <img src="http://dssg.io/img/posts/outlier_detection.png" width="600">
  5. Temperature sensitivity classification of a building:

    • High sensitivity: energy consumption is extremely reactive to outside temperature both when it is high and low.
    • Medium sensitivity: energy consumption is very reactive but only when temperature is high. This can be due to alternative heating systems.
    • Low sensitivity: the consumption of energy is not highly correlated with outside temperature.
  6. Peak prediction: Modeling and prediction of energy consumption yields an estimate on the savings due to the elimination of peaks in consumption.

    <img src="http://dssg.io/img/posts/peak_prediction.png" width="600">

Read more about our analysis in the wiki

The data: building energy interval data

Agentis energy, a company that builds software for energy utilities, supplied us with anonymized, hourly meter data - electricity usage in kWh - for roughly 7,000 buildings. A year's worth of data for one building looks like this:

interval

They also gave us the NAICS code and business type for each building. Because the data was anonymized, and we didn't know the location of these buildings, Agentis provided us with the associated temperatures for the same time series.

Read more about the data in the wiki

Project Layout

Installation Guide

git clone https://github.com/dssg/energywise.git
cd energywise
python setup.py install

Team

team

Contributing to the Project

License

MIT license, see LICENSE.txt