Objectives of the service
The main objective of the Roof Solar Panel Calculator (a.k.a. THOR - Tellurian Holistic Object Recognition) is to automate the process of selecting a building’s roof as a candidate for the installation of solar panels and to estimate associated costs.
Before making an investment decision for the installation of roof-mounted solar panels, homeowners are normally interested in knowing the cost and return on investment. Until now this process has been cumbersome, slow, and not user friendly. The process normally involves the following steps:
-
Measure the roof
-
Calculate the pitch of the roof (where applicable)
-
Find directions of roof faces
-
Calculate solar power production based on roof data
Since the process is manual, it is expensive and time consuming. This results in customers discontinuing the process and the manufacturer losing a potential sale.
THOR mitigates this problem by making roof information available to homeowners and solar roof providers with just a few clicks in a web application.
THOR offers a way to automatically generate information about:
-
Roof size – roof geometry and measurements
-
Roof pitch – solar panel production depends on panels’ angle to sun
-
Roof orientation – solar panel production depends on the ideal panel direction
-
Power – a calculator estimates approximate solar power production for a roof
With quick and timely information provided by THOR, it is easy to calculate both investment size and return on investment.
Users and their needs
There are two main groups who can benefit most from THOR:
-
Solar roof providers and manufacturers
-
THOR simplifies their sales process
-
THOR provides roof information faster
-
-
Homeowners
-
THOR instantly provides a rough estimate of the investment required
-
In principle, THOR is scalable for worldwide use since we use publicly available satellite imagery as a basis for analysis and calculations. Initially, the service and resulting data is available for Estonia, Latvia, Lithuania, Poland and Scandinavia. Data may be prepared and calculated for other regions upon request.
Service/ system concept
THOR processes satellite imagery, along with other geographical data, to identify individual buildings and each roof segment of the building. For each roof face, the rough shape, area, azimuth, and pitch are calculated using dedicated machine learning models. The system also predicts the solar power production potential for each roof using defined parameters.
Since satellite imagery is updated quite frequently and has global coverage, THOR stays up to date and scales better than solutions which use data sources that are less frequently updated or gathered in a standardised way.
As a deployed service, THOR provides users with a simple and intuitive GUI that offers the following functionality:
-
Searching buildings by address
-
Visualising roof faces
-
Manual adjustment (if needed)
-
Estimating which solar power producing roof solution is the most reasonable based on:
-
Power generation
-
Yearly production
-
Installation cost
-
Return on investment
-
Payback period
-
For solar roof manufacturers and installers, THOR may be integrated into existing sales pipelines such as e-commerce platforms, manufacturing, scheduling and logistics information systems.
Space Added Value
THOR uses satellite images to identify buildings and calculate roof face size. By using satellite images, the system can be applied quickly to any country or region in the world. With other techniques (e.g., Lidar), accessibility of base data varies between countries and regions, meaning data is not universally available. Satellite images are universally available for entire world and are available at relatively low cost. Depending on the level of desired accuracy, it is possible to use cheaper- and lower-quality images to reduce the costs of data sets.
Current Status
Machine learning processes are complete and web applications fully developed, making THOR ready for production. The first client is currently integrating the Roof Solar Panel calculator into their systems and processes.
The project was challenging in many aspects since it required the following:
-
Building a multistep machine learning pipeline where the next step takes input from the previous step
-
Very careful planning and fine tuning of each step’s input and output
-
Combining multiple source data types into one
-
Generation of 3D models and shapes from 2D images
Prime Contractor(s)
