Methodology

Data Sources

Metered Electricity Demand

30 minute metered electricity demand from January 2019 - January 2022. Data was accessed from the Western Power Distribution (WPD) data portal over 3 different Electrical Supply Levels (ESAs):

• 81 Grid Supply Points (GSPs)
• 150 Bulk Supply Point (BSPs)
• 924 Primary Substations

Verisk (Geomni)

Information about the buildings within an ESA:

• number of premises
• mean bedroom count
• proportion_premise_use_* where the * could be residential or industry or health etc.

EXPERIAN - MOSAIC

Consumer classification system. The population is grouped into 15 groups based on consumer behaviour. The features we currently include are the proportions of people within the ESA that belong to each group.

PV

Estimation of the amount of PV embedded energy generated within an ESA as a function of time.

Weather

Rainfall, Temperature and Radiation measurements

Model Training

The diagram below describes the workflow associated with training and then using the model.

Figure 1: Training and Inference workflow

Model Training

Firstly, we split the ESA's into three different sets:

• Train: ESA's used for training the model. Important that this data is not used to evaluate the model.
• Validation : ESA's for monitoring the models performance during training and is used to determine when to stop training the model.
• Test: Unseen dataset used for evaluating the models performance. Data is kept completely seperate from training.