An electricity utility provider ran a pilot to automate meter reading using CNN algorithm
The utility service provider was looking at multiple solutions to drive efficiencies in their operational processes. They turned towards image analytics to assess if an automated meter reading solution would help them reduce the human resource requirement, increase accuracy and efficiency and solve their problem of unavailable customers (locked houses) and help them generate bills on time – improving the efficiency of the billing process and push customer satisfaction.
Machine learning based automation improves efficiency
The utility provider was looking at image analytics based automated meter reading (AMR) solution that could solve a lot of problems .
- An accuracy of 84% on untrained sample of meter images
- Improved reading coverage of 4% in the pilot area
- Improved payment compliance of 2.7% in the pilot area
One of the key challenges that emerged out of the pilot was no reading in a significant number of cases due to poor lighting in the meter location and unreachable meter locations.
Inefficient, human resource driven process.
As a regular practice, the analog meters are used to collect data for the energy consumed. The meters could have a number dial or more recently, digital displays are used.
The utility provider’s employee visits each household with and notes down the reading at the end of every billing cycle. This process suffers from wastage of human labour, human error, manual process of reading the paper based notes every month to generate invoices.
This manual process is inefficient, time taking and cumbersome. Another big issue with the process is that it is not possible to take readings if no one is at home. This leads to delays in generating invoices, and in some cases clubbing bills for multiple months. The aim of the pilot was to showcase the current capabilities of machine learning based automated meter reading solution.
Latent benefits of Automated Meter Reading
Image analytics based automated meter reading solution allows customers to click the image of the meter through a web application. This process provides following benefits to the utility provider:
a. An automated solution generating efficiencies in the number of meters being read in each billing cycle
b. Enables timely and efficient generation of invoices, leading to a higher payment compliance
c. Freeing up human resources for other tasks and reducing human dependencies
The pilot though highlighted one of the key limitations of a machine learning based automated meter reading process. The best accuracy achieved was 84%, which was significantly lower than the manual process, which, despite all inefficiencies had an avg. error rate of ~2%.