Valuable Data: Importance of analytics in detecting power theft and curbing losses

Importance of analytics in detecting power theft and curbing losses

The primary objective of discoms is to meet the electricity demands of consumers in a reliable and efficient manner. However, they face myriad challenges in realising this objective. The major challenges associated with supplying round-the-clock power are breakdown responses, high losses, consumer non-engagement and input power mismanagement. These issues need to be addressed in an objective manner to ensure uninterrupted power supply.

Analytics holds the key in this regard as it helps utilise the data collected by discoms through meters and other sensors. This information can be used in combination with other applications for a range of purposes. It can help discoms reduce costs, enhance supply quality, increase reliability and provide better customer services.

Analytics can aid in curtailing energy and revenue losses, detecting theft, and deciphering tamper messages/patterns to help distribution utilities prioritise their actions.

Power quality can be significantly enhanced through analytics. The analysis of meter data can assist in quantifying losses, monitoring delivery voltage and outages, evaluating voltages and improving reactive power. Analytics can assist utilities in replacing reactive processes with proactive management and ensure optimal system performance.

With regard to condition monitoring, analytics can provide up-to-date details of load; identify unanticipated load increases and overutilised, underutilised and at-risk transformers and other assets; and help in the evaluation of transformer sizing, using loading history and peak seasonal load levels.

The growing integration of renewable energy in the grid is increasing the importance of forecasting due to the variable nature of resources. Analytics can play an important role by helping forecast the total consumer usage at the feeder or substation level; net usage reduced for distributed generation (in the case of rooftop solar, for example); and the usage of new technologies like plug-in electric vehicles. It can also be deployed to estimate demand response availability at the facility level.


There is an urgent need for discoms to revamp the way they carry out analytics activities. It has been observed in countries across the world that utilities generate data without being clear about its objectives. Hence, data collection needs to be preceded by identifying the applications that a discom intends to develop. This will also provide discoms with information about the frequency and quantum of required data, whether there is a need for online or offline data, communication methods, and server capacity.


There are two ways of performing analytics in discoms (especially for theft detection and network management purposes): on the basis of domain knowledge or on the basis of scientific analytics. Each has its advantages and disadvantages.

Engineers prefer analytics based on domain knowledge, which attempts to derive the logic between possible causes and symptoms. This method is effective and provides an immediate outcome but its expertise is limited only up to a point as it is based on existing knowledge.

Scientific analysis relates to pattern recognition, trend recognition, machine learning and artificial intelligence, and is generally used by IT companies and professional analysts. However, this methodology has huge requirements of historical data and software. Moreover, the output in the first three or four years is almost zero as there are no past records. But this improves with time as a repository of historical data is created.

Case study: BSES Rajdhani Power Limited, Delhi

BSES Delhi serves the electricity needs of almost two-thirds of the city with its two discoms: BSES Rajdhani Power Limited (BRPL) and BSES Yamuna Power Limited. Realising the importance of data analytics, in 2006-07 BRPL planned to collect all data, including single- phase meter data, using computerised meter reading instruments and then perform analytics on the same. The utility’s focus was to undertake object-oriented analytics, irrespective of the source of data (online or offline) without being constrained by the absence of a smart grid. Since 2007, BRPL has been using data generated by meters deployed at various nodes of the power supply chain to conduct a variety of analytical exercises. These include grid meter analytics, distribution transformer (DT) meter analytics, and electricity theft analytics.

Grid meter analytics

In terms of grid meter analytics, BRPL conducts circuit-wise bill verification of Delhi Transco Limited’s (DTL) bills and compares this with the quantum of energy purchased by the discom as predicted by their analytics techniques. With this exercise, BRPL has claimed about 198 MUs from DTL because of billing errors. However, over the past couple of years, the gap between the two values has reduced to almost zero, and if there is a deviation, the gap is not more than ±1 MU. This procedure also involves express feeder analytics, DTL feeder meter analytics and extra high voltage consumer meter analytics. The discom also prepares other reports pertaining to feeder power outage, voltage optimisation and the simultaneous peak load of feeders for network planning as part of grid meter analytics.

DT meter analytics

BRPL engages in performing various kinds of analytics on DTs, pertaining to the health of transformers, overloading, voltage unbalancing, low tension feeder-wise analyses and high tension loss computation. A DT health report is generated, which serves as an important tool for loss-reduction activities. Over the years, this analysis has helped the discom significantly reduce distribution failures.

Electricity theft analytics

Analysing data has helped the discom identify reasons for a host of abnormalities related to voltage principle, load, consumption patterns, meter parameters, etc.; power theft being one of them. BRPL has formulated various maturity levels for identifying electricity theft on the basis of the overall consumption pattern of an individual, the consumption patterns of similar consumers, tamper events and instantaneous parameters.

Level 1: This is the preliminary level of theft analysis based on consumption patterns. However, its strike rate is generally low as consumption trends alone cannot be treated as evidence of theft.

Level 2: Data pertaining to consumption is collected through surveys and a vast amount of information is collated through secondary sources. After data collection, the consumption patterns of similar consumers are compared to detect abnormalities. However, benchmarking in this regard requires a high level of experience.

Level 3: Several newly designed meters directly identify abnormalities and log events caused by theft. Such leads generally have a high strike rate (above 90 per cent) but can only identify theft as per pre-defined methods.

Level 4: Apart from energy consumption and tamper event logging, meters also log instantaneous parameters like voltage, current, power factor and power on/off data. The patterns of instantaneous parameters and the variations in the same are analysed to identify power theft. However, logic is very important in this regard as it helps formulate a relationship between the theft method and its impact on the parameters. At present, BRPL has developed 104 logic modules. This method has a high strike rate and wide acceptance, but it requires extensive knowledge of metering and electrical engineering.

Level 5: The biggest drawback of the first four maturity levels is the generation of filter logics that require extensive knowledge. Moreover, there is a need to continuously develop new logic to improve the strike rate. The fifth level is based on artificial intelligence and works on two basic principles. One principle is that the outcome of two events with similar trends is expected to be similar, while the other says that utilities hold huge banks of meter data, consumption data and secondary data information through the four levels of analytics.

The working principle of this maturity level involves identifying cases where theft has been observed in the past, plotting data to identify the trends, filtering consumers showing similar trends for a given set of parameters, recording outcomes (success/failure) and using employee metering knowledge to refine the logic.

The biggest advantage of artificial intelligence is the self-improving nature of the software. It aims to develop a system from experience and continued analysis. BRPL’s various modules for theft analytics have helped it reduce thefts from 53 per cent to 12.5 per cent in 2015-16 (estimated).

The way ahead

Discoms need to focus on using the information generated from meters in a more productive manner. They also need to integrate information from all other data sources to realise the full value of analytics. Over the coming years, they can even leverage cloud infrastructure for this purpose as it can overcome the various bottlenecks associated with traditional hardware and software systems.

Based on presentations by Rajesh Bansal, Senior Vice-President, BRPL; and Mitul Thapliyal, Senior Principal and Practice Leader, Energy Utilities and Smart Cities, Infosys