Energy Theft Reduction Strategy
Under the REC, we are required to progress a Theft Reduction Strategy. While the theft reduction strategy will be further developed as part of our overall strategy and forward work plan, some of the elements are already in place and have been migrated from legacy code arrangements:
We are also expected to develop a replacement for the former Theft Risk Assessment Service (TRAS), subject to industry consultation informed by a robust business case and project plan. Earlier this year we commissioned an independent report on the options that might be available for such a service, intended to inform any subsequent business case we develop. The results of that initial ‘discovery phase’ report we presented to a Theft Strategy workshop in March, the slides from which are available below. The ‘discovery phase’ report set out a spectrum of five options (which could each be further sub-divided) that may be available to RECCo based on the desired degree of service and technology transformation. The anticipated cost of the options ranged from <£1m to >£4m per annum, the latter of which was comparable with the cost of the former TRAS service.
Before we take further any commitments on the replacement of the TRAS, we believe it is in best interest of REC Parties to first establish a methodology for understanding the scale of the energy theft within the market. It is also expected that this Theft Estimation Methodology would provide a means of periodic evaluation, helping us to assess the efficacy of any measures taken to mitigate the problem of theft.
Theft Estimation Methodology Project
We issued a Request for Proposal earlier this summer to competitively procuring a service provider to support the development of the ‘Theft Estimation Methodology’. Four submissions were received, which were taken through to evaluation by a panel made up of RECCo Executive team members and subject matter experts, who concluded that Capgemini offered the strongest proposal.
We are now progressing with the project working alongside Capgemini, the project will use four initial methods:
Method 1 - WHOLE SYSTEM ENERGY BALANCE (AGGREGATE)
Aim: to understand the order of magnitude for potential volume of theft as a baseline
Sum of all energy fed into the distribution/transportation systems minus the sum of all reported consumption, the also subtract modelled ‘Technical losses’.
Techniques: data aggregation, confidence estimation, uncertainty propagation.
- Gas and electricity settlement data;
- Performance assurance data;
- Smart meter data.
Method 2 - NETWORK CENTRIC ENERGY BALANCE (GRANULAR)
Aim: to understand the order of magnitude for potential volume of theft in geographical segments. Compares the energy entering and exiting a section of network, working at a more granular level than “whole system” above.
Sample several network segments for which the input and sufficient endpoint energy metering data is available today
Techniques: subgraph sampling, Bayesian inference, uncertainty estimation.
- DNO/GT substation, feeder or similar meter data;
- Supplier meter data flow for those networks;
- Property data;
- Company categorisation;
- Former TRAS data;
- ETTOS data;
- Technical loss/leakage date.
Method 3 - PRIOR THEFT INCIDENTS EXTRAPOLATION (SEGMENTED)
Aim: to identify patterns and drivers of theft to extrapolate for the entire network. Uses the historical records of theft detection (and suspicion) to extrapolate the likely volume of similar theft across the population. The key to this will be meaningful characterisation and segmentation of the theft incidents.
Techniques: feature engineering, feature importance analysis, ML regression, hypothesis testing.
- Former TRAS data;
- Cannabis market trend data, etc;
- Other specific theft;
- Deprivation data;
- Energy poverty trends;
- Energy consumptions trends;
- Theft data from other sectors (e.g. water).
Method 4 - TREND EXTRAPOLATION
Aim: to understand time-based effects in theft and propose forecasting methods for theft volumes. Modelling hypotheses around the driving factors for energy theft and project how they are likely to affect the volumes of theft
Supports the periodic re-estimation requirement and will benefit from feedback of detection results in the future.
Techniques: temporal feature engineering, trend/seasonality analysis, autoregressive forecasting with exogenous variables.
Primary data: As per method three.
The progress of this project is heavily dependent upon the provision of timely data. Therefore, although an agile approach is being taken, timelines may need to be adapted in light of progress. The project is being structured around a small number of short sprints as outlined below:
|0 – Kick-off||
Hypothesis Development – the refinement and documentation of the hypotheses outlined above, with a more detailed articulation of the required data sets for each Data Availability Assessment – assurance activity that the delivery team has access to the required data sets outlined as part of Hypothesis Development
Data Requirement Documentation – articulation of missing data requirements and preparation for issuing RFIs
|1 – Convergence on the Ideal Method||
Data Acquisition & Transformation – the acquisition, quality assessment, cleaning, aggregation and transformation of the available data from disparate sources Data Analysis & Method Convergence – exploratory analysis of data, testing of available methods based on the available data sets and convergence on an ideal method
Data Requests Issued – any requirements for additional data requested through RECCo
|Take stock and if necessary re-plan|
|2 – Method-specific in-depth analysis||
Data Acquisition & Transformation – the acquisition, quality assessment, cleaning, aggregation and transformation of newly available data from requests
Data Analysis & Model Development – exploratory analysis of data, testing of available methods based on the available data sets
|3 – Model development||Model Refinement – implementation and validation of the final TEM|
|Close – Business Case Development||Documentation – documentation of the final estimation model including data sources, modelling methodologies, assumptions and user guides|