Automatic meter reading through artificial intelligence framework
Approximately 28 million consumers are being facilitated by 10 power distribution companies of all Pakistan (excluding Karachi city and its surrounding areas). Water and Power Development Authority (WAPDA), being a central entity, is responsible for the current billing system mainly relying on the m
2025-06-28 16:30:27 - Adil Khan
Automatic meter reading through artificial intelligence framework
Project Area of Specialization Artificial IntelligenceProject SummaryApproximately 28 million consumers are being facilitated by 10 power distribution companies of all Pakistan (excluding Karachi city and its surrounding areas). Water and Power Development Authority (WAPDA), being a central entity, is responsible for the current billing system mainly relying on the manual meter reading process, whereby a meter reader notes down readings on a register and takes an image of the meter as a proof-of-reading. Subsequently, the written reading is compared with the meter-image for veri?cation purposes, followed by the issuance of consumption bill. This whole process is repeated every month for each meter, thus requiring a lots of human e?orts and time. In line with the Punjab Information Technology Company (PITC)’s initiative of mobile meter reading system, this project aims to develop and design an arti?cial intelligence based framework which could assist humans with automatic meter reading. With the purpose to reduce wrong meter reading complaints and to reduce billing cycle time, the key objective of this research is to replace manual meter reading process with an automatic meter reading protocol. To achieve this, an image-based automatic meter reading scheme will be developed employing state-of-the-art deep machine learning architectures.
Project ObjectivesWith the purpose to reduce wrong meter reading complaints and to reduce billing cycle time, the key objective of this research is to replace manual meter reading process with an automatic meter reading protocol. To achieve this, an image-based automatic meter reading scheme will be developed employing state-of-the-art deep machine learning architectures.
Project Implementation MethodAn image-based automatic meter reading (AMR) approach incorporates three stages, to be speci?c as shown in Figure 3.1:
• Counter detection
• Digit segmentation
• Digit recognition
Counter detection is central to the entire AMR framework as it largely decides its accuracy and processing speed.To start with, the initial step will be to collect raw images, of all types of meters installed in Pakistan, which will be used to make training, validation and test datasets with 90/10/10 cross-validation scheme. In the next stage, a well-known you only look once (YOLO) algorithm by Zhao et al. [5] will be utilized as an object detector to detect the counter region of a meter.Next, counter recognition scheme will be designed to detect the meter reading.The entire pipeline will be trained, validated and tested on a big dataset.The trained model will be deployed in circle-level central computer. A meter reader will take a snap of the meter from a mobile app, which will be automatically sent to the central computer (or server) for extraction of necessary information by simulating the trained model. The entire framework will be evaluated in a sequential order. First, the manual annotation of images will be performed along with their quality check. Only those images will be considered which will be readable by the human eye. The multiclass classi?er will be tuned on the available data with the aim to achieve acceptable accuracy. Once the multiclass classi?er has been developed with acceptable accuracy, di?erent variants of you only look once (YOLO) algorithm will be explored and optimized for counter detection. Being central to the whole framework, this stage will be ensured having maximum accuracy. Lastly, di?erent digital recognition schemes will be evaluated to extract the relevant information by Laroca et al [6] e.g., meter readings. The performance of all stages will be evaluated in quantitative terms e.g., accuracy. In addition to all these, the feedback from meter reading sta? will also be considered for as a tool to further re?ne the proposed pipeline.
•The electricity consumers in Pakistan are mostly unsatisfied due to overbilling complaints because of human errors made by meter readers. By automatic meter reading the reading will be accurately recognized and errors will be reduced.
•Moreover, the consumer will not suffer. The utility companies will not face the problems like payment delays and additional processing.
•The manual meter reading process will turn into automatic process after completion of that project.
•In automatic meter reading process, the cost on reading process and bill generation will be low as compared to the manual meter reading process.
•It will reduce the billing cycle time.
•It will uplift the image of overall power sector.
Technical Details of Final Deliverable- Comprehensive literature review of deep learning architectures.
- Programming frameworks including Python.
- Software implementation of counter detection, digit segmentation and digit recognition.
- Development of a software routine which will tell the energy consumed by pulling out necessary information from Energy Meter.
| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 70000 | |||
| Computer system that supports nividia gpu | Equipment | 1 | 70000 | 70000 |