Classifying Rooftops Through Neural Networks to Eliminate Energy Waste of Facilities
Project completed! Results attached!
The challenge partner is a Techstars Energy tech startup with the mission to provide a digital map of the largest commercial and industrial energy users in North America.
In this two-month Omdena Challenge, 50 AI changemakers collaborated to build AI solutions, which can help to significantly improve energy efficiency and sustainability of facilities.
Access to customer electricity data is extremely important for society and energy companies who are looking to prioritize efficiency, sustainability, and make operational decisions about facilities (think solar, energy retrofit, and energy consulting companies). It allows them to understand how these facilities consume power, and where the best opportunities exist. However, electricity data is difficult to obtain and often requires significant engagement with the end-user to understand if they are qualified.
This lag in time is inefficient and slows down go-to-market teams and efficient society impact. The Partner Platform allows users (energy sales teams) to easily search any territory, and view Energy Assessments of the covered facilities right away. The Energy Assessments include the estimated kilowatt-hour consumption for each of the 250,000 facilities that the platform covers. The Platform engages in predictive analytics to build these estimations and help energy sales teams go to the needy market quickly and effectively. One indicator of energy consumption could be rooftop type and rooftop equipment.
The Project Outcomes
The AI solution built has the intent to gain a better understanding of the energy intensity of a facility by assessing rooftop characteristics. The models classify rooftops into types, using a convolutional neural network (“CNN”) and Edge Detection. With the access to rooftop images for over 200,000 facilities across the US (soon to be Canada, Europe, and Asia). These images were classified into 1 of 5 categories:
Flat and Clear
Flat and Cluttered
The project partner provided image examples of each category to help train the CNN. In short “cluttered” (see point 2) is a reflection of rooftop equipment such as HVAC systems. “Industrial” is a characteristic of the roof which tends to have more cylindrical or rectangular-shaped equipment. Existing solar would classify rooftops that already have solar panels installed.
Even generating more classifications, and that there are more facility types that the partner hasn’t thought of.
Below is some initial research on how edge detection might be helpful in building this AI:
Edge Detection Neural Net
This may be able to help distinguish industrially complex rooftops from simple or flat ones. Below is a screenshot from Google Maps which compares two facilities somewhere in Texas. You will get the code for this in an HTML file.