Machine Learning for Improved Efficiency Analysis & Asset Information

Machine Learning for Improved Efficiency Analysis and Asset Information

Eis Icon

Introduction

The use of unstructured data combined with machine learning and data fusion techniques, has the potential to enhance today’s energy analysis tools and provide asset information for public and private sector building portfolios.  This project will develop automated approaches to determine building characteristics, and retrofit and operational efficiency opportunities. 

State of the art analytics software and modeling tools can provide valuable insights into efficiency opportunities. However prior research has shown that key barriers include relatively limited data sources (smart meters and weather being most common in commercial tools), or reliance upon user-provided inputs for which default values may be the fallback. There is great opportunity to apply techniques based on multi-stream data fusion and machine learning to overcome these challenges. Secondarily, these techniques can be applied to provide enhanced asset information for public and private sector building portfolios.  

Project Summary Machine learning

This project is developing automated approaches to determine building characteristics and efficiency opportunities using unstructured data from a subset of the following types:

1) Drone-based RGB images

2) Drone-based thermal images

3) Satellite or aerial images

4) Building footprint GIS data

Research Outcomes

Current Research Outcomes are summarized below. The work is also documented in the following papers:

Aerial 3D Building Reconstruction from RGB Drone Imagery

Machine Learning for Automated Extraction of Building Geometry

1) Opportunity analysis to apply machine learning to enhance building energy-efficiency measure identification. 

 We gathered information regarding data availability, providers, and methods of collection for ~27 novel data sources (public, proprietary, private, or researcher datasets) and assessed how these data can be used to enhance identification of  ~18 building energy use and asset features (such as equipment, envelope, interior, context, occupancy, utility cost etc.), and ~70 building energy efficiency measures.  

The analysis surfaced satellite/aerial data, building footprints GIS data, and drone-based visible and thermal images as the most promising data sources to pursue via machine learning methods.

2) Replicable, open solutions for ML-based feature extraction from satellite/aerial images

The open source Automated Building Footprint Extractor codebase is available from Github with three modules:

  • Module to generate training data for deep reinforcement learning algorithms to determine building footprints from satellite data
  • Module for deep learning semantic segmentation 
  • Module to post-process building footprints into GIS-Format (i.e., GeoJson)

3) Open solution for drone-based generation of 3D geometry

We designed a drone system, defined the data acquisition methodology, and developed a workflow to automatically extract building 3D geometry. The open source Aerial 3D Building Reconstruction from Drone Imagery is available from Github with two modules:

  • Module that uses aerial drone images to construct a building footprint with height information to create a 3D model of the building using the steps line processing, polygonization, and polygon merging
  • Module to compute the window-to-wall ratio of the building facades

4) Thermal anomaly and rooftop unit (RTU) detection in buildings through machine learning

We developed algorithms to detect 1] thermal anomalies and 2] rooftop units [RTU] from UAV-based thermal and RGB images, respectively. The thermal anomaly detection algorithm uses a clustering algorithm while the RTU detection uses an object detection algorithm. The open source code is available from Github.

Partners

NYSERDA, Signetron, TRC

Team

Mechanical Staff Scientist/Engineer
Statistician Research Scientist/Engineer
Program Manager
Program Manager