I-LAD: Innovating Losses Analysis and Detection (SIF Round 4)

Project Overview

Description

Electrical losses are an unavoidable consequence of transferring energy across electricity networks, where they have financial and environmental impacts, and are forecast to increase significantly as we accelerate electrification of heat and transport.

Technical Losses, while inevitable, are directly dependent on demand at the time. Current technologies only allow for estimations of losses, making it challenging for DNOs to target losses reduction technologies effectively. While Non-Technical Losses are somewhat preventable, current identification methods are piecemeal, rely on 3rd party intelligence, use of tools like google maps and manual information exchange between all parties.

Additionally, no tools exist to monitor all losses in real/near-real time. Improvement in losses management requires a radical new approach rather than refinement of existing processes. I-LAD proposes to address this by:

1. Developing novel approaches to automating and modernising losses data collection, identification and modelling: e.g., currently energy theft is difficult to detect, and even when it is, by the very nature of bypassing the meter, DNOs won’t have a load profile dataset from a smart meter which can be used train a model. To address this, we will use techniques like synthetic data (where very small amount of “real” data available is used to generate “fake” examples of NTL to train the model). Those techniques have been applied in other industries (e.g. fraud detection) but are untested in this context.
2. Fully understanding total losses landscape using model developed in item 1 create a new digital service allowing DNOs to automatically identify, model and record losses at a granular level through innovative use of Machine Vision, AI and Machine Learning and traditional analysis of network data.
3. Novel approach to losses governance: develop new data sharing options to address data privacy barriers (GDPR) and drive improved cross-sector collaboration. Improved visibility of losses will give better clarity on their sources and who is best placed to address them.
4. Automating losses monitoring and measurement: to evidence effectiveness of losses mitigation. This will be unique and will help reflect the dynamically changing network we expect during the energy transition.

Expected Benefits

This project will deliver tools to identify, classify and monitor electricity losses more efficiently using novel data and modelling techniques, resulting actions and coordinated losses interventions, leading to measurable losses reduction at a higher level than is currently achieved.

It’s estimated that losses account for 5% – 8% of the total distributed units, costing a typical household around £100pa and accounting for around 90% of a DNO’s total Greenhouse Gas emissions. Whilst already high, losses are expected to significantly increase in coming years from electrification of heat and transport, increasing volumes of low carbon technology altering and increasing power flows and heightened cost of living pressures impacting the levels of Non-Technical Losses.

A reduction in losses will provide two main streams of benefits:

Financial: the greater the losses, the greater the costs to customers through their electricity bills. This is due to having to generate more electricity to cover losses. Therefore, losses reduction will directly contribute to a relative reduction on customers’ bills. E.g., It is estimated that total annual losses due to theft across the networks are approximately 2.2TWh. The largest source of theft is believed to be cannabis farms, accounting for around 0.75TWh. A methodology which improved their detection allowing even 10% of them to be removed would save a staggering 75GWh. Using standard values in Ofgem’s ED2 CBA this equates to a saving of over £4m pa, and nearly 20,000 tCO2

Environmental: The greater the losses, the greater the carbon emissions and environmental impact to society. This is due to losses representing fuel consumed and emissions produced in the process of electricity generation. Other: Reducing different sources of losses can have widely varying social benefits. E.g., theft can be associated with significant health and safety risks. Illegal modification to the network is often linked to other illegal activity, therefore better identification of these may support identification.

Progress

This project started in February 2025.

Discovery phase started with reviewing what is known about the different types of distribution losses, with key findings listed below:

• The main source of losses are “technical losses” which are intrinsic to the physics of the power system. The largest proportion of these are variable losses (9 TWh annually) which are higher the more power is being consumed – for example due to wires and transformers heating up. There are also various sources of fixed losses (making up 4 TWh annually) which are always present such as from the cores of transformers.
• Network companies can estimate technical losses based on the characteristics of their network – although even at this high level, there is substantial uncertainty as to the exact size of these losses. Based on previous research, they’re likely to be about 13TWh.
• The remaining 1.5TWh or so of losses are “non-technical losses. These include:
  • Energy theft
  • Unknown unmetered supplies. Unmetered supplies are things like lampposts which don’t have individual meters. DNOs have records of how many unmetered supplies there are and the energy they should be consuming. However, if these records are inaccurate then this will appear as a loss.
  • Meter inaccuracies will also appear as non-technical losses.

By their nature these forms of loss are hidden from the DNO and suppliers so they’re extremely difficult to quantify. One of the benefits of the losses service from the I-LAD project we’ll go on to describe would be a much more systematic way of putting together different data sources to understand where the greatest losses are and where taking action can really make a difference.

Main outputs of Discovery: 

• The Project showed that clustering, time series, classification, and AI techniques can successfully analyse these data layers to reveal where losses occur, their likely source, and confidence levels.
• The Project designed a set of principles by which any such model will be governed. The NTL Governance Model will be:
  • Fair: Customer protections and identification of vulnerable households (via connection to SSEN’s VERIFY project)
  • Secure: GDPR-compliant data sharing and user-based access.
  • Accurate: based on confidence levels; reduced risk of false positives; ground truthing for model learning.
• The Project explored intervention options and established a basic rule set (e.g., type of loss, value, cost, confidence) for selecting the appropriate. The Project developed a wireframe of a proposed losses service.

Funding

Total Cost – £175,005
SIF Funding – £149,167

Start/End Date

3rd February to 30th April 2025