
Traditional electricity systems are designed to provide unidirectional power flow from generators to consumers via the transmission and distribution networks. However, with the increased use of weather-sensitive renewables, energy storage, and demand side response, electricity networks are becoming increasingly complex. Understanding the changing behavior of electricity consumers requires new analytical solutions that account for this complexity, including the increased sensitivity of transmission and distribution systems to weather events.
Established long-term demand forecasts rely on a combination of historical records of demand and socio-economic models broadly describing changes in population, consumer behavior, and the adoption of new technologies (e.g. rooftop PV installations, or electric cars). Underpinning these future scenarios is a baseline extrapolation of current weather-corrected demand trends.
Traditionally, this analysis is conducted on a network wide level, which does not allow for geographical variations in the weather, nor the varied behavior of consumers at different network connection points. This makes it difficult to determine the long-term baseline demand at a local level, and thus the level of investment required in different parts of the network.
Problem:
The capacity of each piece of infrastructure in an electric utility’s transmission or distribution network depends on peak level of consumer demand. It is therefore very important to have the most accurate consumer demand data when planning both the size of the future network and the upgrade and improvements program.

One factor that skews the “raw” consumer demand data is the influence of the weather. For example, consumer demand is higher when the weather is cold than it is in milder conditions.
At present, many utilities correct for these weather effects at a regional level (i.e. national or state), assuming that weather patterns are similar over large distances. This is not true, but lack of credible weather data means it is the only option. Typically, only the effects of temperature are included, but other weather variables such as wind chill and cloud cover can have a big impact. Both of these factors can lead to incorrect assumptions being made about the peak level of consumer demand at a given substation. This, in turn, can mislead decision makers in terms of future planning and upgrade schedules. Figure 1 demonstrates the diverse projections of observed versus weather-corrected annual peak demand at a single substation, which could potentially lead to very different decisions.
Solution:
Digital Engineering (DE) has developed a demand model which considers the effect of different weather parameters on consumer demand at a substation level. This significantly improves the quality of data and improves planning and maintenance operations, ensuring budgets are spent in the areas that need them most, leading to a better network with fewer faults and less customer downtime.
The project removed the effect of weather using long histories of weather data, a process known as ‘normalizing’ demand, exposing the underlaying changes in customer demand behavior.
The first step of the project was to cluster the substations based on consumer behavior, because different types of customers respond differently to the same weather conditions. Residential areas are quite sensitive to the weather, while commercial areas show little seasonal changes.
State-of-the-art weather models were applied to recreate ten years of historical weather data at high spatial and temporal resolution across Scotland. Weather varies significantly across an electricity network, because of ground elevation, weather patterns, coastal effects or urban heat-islands. Figure 2 demonstrates how annual average temperature can vary across a 150 mile (240 km) section of land.
A suite of consumer demand response models was trained using weather variables such as temperature, wind speed and solar irradiance, as well as non-weather variables such as day of the week. These models used advanced machine learning algorithms to determine the response of consumers to these different environmental and non-environmental conditions. The demand response of the consumer types was identified using a clustering algorithm, including rural villages to inner-city commercial districts and everything in between.
The substation models can be used on their own for granular data about the networks, or they can be combined to provide accurate regional or network level data. These models are data driven and adaptable to any geographical location worldwide.
Results:
The project processed and filtered the demand data from primary substations, it calculated ‘weather correction factors’ by comparing the peak demand in any one year to peak demand values predicted using ten years of weather data. It helped analyze historical trends in annual peak demand at each substation and predicted the impacts of extreme weather on different substations to allow the asset owners to see how their network will respond to future weather events.

This analysis is helping the client improve prioritization of investments, assess the potential impact of extreme weather scenarios on different parts of their network and improve baseline projections for future demand scenarios. It is also being used to enable early warning of substation-level shifts in demand, to cleanse SCADA data, provide more accurate load forecasting and transition to the role of distribution system operator.
June 22, 2020