Using Machine Learning to Speed Up Electrical Grid Mapping

Ad:

Lack of accurate documentation and the laborious nature of pouring over imagery to located high voltage towers can dramatically slow down mapping of electrical systems.  Development Seed, working for the World Bank Group, created a methodology for integrating machine learning with manual mapping in order to speed up the mapping of  high-voltage (HV) grids in developing countries.  The project used this method for mapping projects in Pakistan, Nigeria, and Zambia, and found that it increased the speed of mapping by 33-fold compared to manual only mapping.

The lack of a comprehensive and accurate of a county’s high-voltage (HV) infrastructure can make decisions about where to invest or expand electrical grids challenging.  To add to this dilemma, existing schematics tend to be outdated, lacking in areas, and split among a multitude of agencies.  Manual mapping of these grids typically involves laboriously scanning  imagery to pinpoint high-voltage towers which is a slow and time intensive process.  Development Seed came up with a two-step system.  The first involves using machine learning to flag possible HV towers on satellite image tiles.  This process allowed for thousands of tiles to be scanned per hour.  The second step involves a team of mappers that is then able to visually located HV towers to develop a map of HV infrastructure.

In this imagery section, machine learning was able to flag 3 out of 4 HV towers. A team of mappers then manually traced connections between the towers. Source: Development Seed.

In this imagery section, machine learning was able to flag 3 out of 4 HV towers. A team of mappers then manually traced connections between the towers. Source: Development Seed.

A detailed description of the project is available here:

See Also


Advertising



Like this article and want more?

Enter your email to receive the weekly GIS Lounge newsletter:

Advertising