AI vs. Power Thieves: St. Petersburg State University Trains Smart Meters to Detect Manipulated Readings
Researchers at St. Petersburg State University have developed a specialized detector that uses smart-meter data to identify illegal electricity consumption. The intelligent monitoring approach could help utilities combat one of the most common forms of electricity theft.
Smart meters alone can no longer guarantee that electricity theft will be absent from a smart city. Fraudsters have learned to exploit features of these systems for financial gain. They use various methods to falsify data, manipulate time-of-use tariff periods and understate meter readings while preserving the shape of consumption profiles. Meanwhile, monitoring systems that track electricity use in real time cannot always distinguish such manipulation from normal changes in load.
Researchers at St. Petersburg State University have developed a physics-guided detector that targets one of the most common forms of theft, known as multiplicative under-registration. The actual meter reading is multiplied by a factor below one, so that consumption of 100 kWh, for example, is recorded as 70 kWh. Yet the daily profile continues to look normal, retaining its typical daily or seasonal peaks and troughs.
The St. Petersburg researchers' approach uses weighted regression. A regression model built from a large body of previously collected data determines what the actual daily profile of a particular customer should look like. Different points in that profile carry different weights: electricity use is low and more stable at night, while evening peaks provide a clearer picture of consumption patterns.
The detector's key feature is its ability to virtually eliminate false alarms by accounting for legitimate changes caused by other smart-city technologies, including heat pumps, electric-vehicle charging and other systems. According to the researchers, the detector achieved a demonstrated precision of 91% based on an analysis of electricity consumption from more than 1,000 homes.
The researchers argue that specialized detectors designed to identify specific types of anomalies perform better than general-purpose systems. Combining multiple types of fraud-detection algorithms with smart meters, they say, could deliver the greatest benefit.
From Power Retail to Exports: The Potential of Russian AI Systems
As smart-city systems become more widely deployed, demand for this kind of technology is expected to keep growing. Algorithms capable of detecting different forms of fraud could help prevent resource theft while making metering systems as transparent as possible.
Power retailers and grid operators, along with smart-metering system operators and manufacturers, have already become the main users of the technology. Over time, development is expected to shift toward analytical platforms capable of working with multiple types of detectors and identifying electricity-consumption anomalies at facilities of any scale.
Electricity theft is a problem in many countries, which means Russian technologies could also find demand in international markets once adapted to national requirements.
From Field Inspections to Neural Networks
For many years, electricity theft was most often detected through on-site inspections aimed at finding illegal connections. As smart metering systems became more widespread, digital algorithms increasingly took over that task.
In 2020, PJSC MTS and Bashkirenergo LLC began deploying the EnergyTool platform to identify non-technical electricity losses using geospatial analytics. The system analyzes large datasets with AI technologies. It can identify meters transmitting understated readings and pinpoint specific sections of the grid where electricity theft is occurring.
St. Petersburg State University researchers have spent several years developing different methods for detecting electricity theft. In 2024, working with JSC Electrotechnical Plants Energomera, they unveiled an electricity meter with integrated AI algorithms designed to detect and prevent unauthorized consumption, including illegal cryptocurrency mining.
The university's latest invention illustrates the shift from automated metering toward intelligent infrastructure management. A combination of different detectors and analytical systems could minimize electricity theft. Data on customers' actual consumption profiles could also help optimize electricity use and forecast the future development of smart-city energy infrastructure.
