A validated method for estimating regional electricity demand from national time-series

The share of renewable energy in the electrical grid will likely increase as mitigation measures for future climate change are put into effect. Renewable energy production, such as wind and solar, is largely dependent on the weather and is thus subject to variability on (sub-)daily, weekly and yearly timescales. To facilitate a smooth transition of the energy system, it is necessary to have a thorough knowledge of such a future energy system on hourly level for multiple years. 

Information on the optimal spatial distribution of renewable energy sources, on the supply patterns of different renewable energy sources, and on the cost-effective operation of such a system can be obtained through a variety of methods. For each of these methods intricate knowledge of the regional electricity demand is essential, as without it you have no insights into the required installed capacity, the regional size and sign of the weather-induced impact, or to what degree units have to be committed. 

While in general the knowledge on (renewable) energy systems has steadily increased, the knowledge on the demand for electricity is still basically the same as 20 years ago. The reason for this is simple: sub-national demand data has not been measured until recently and is still not readily available to the public and thus to researchers. Therefore the possibility to study the sub-national demand for electricity in more detail is limited.

The aim of the presented work is to show that national demand time-series can be spatially disaggregated by taking the population distribution and the spatio-temporal variation of temperature into account. Similar approaches have been used in the past, but they never tested this assumption due to a lack of historical regional data. In our work we use 5-minute measurement data of all transformers of the Dutch transmission grid for the period 2012 until 2019. As there exist a plethora of methods for modelling the national demand for electricity, based on socio-economic data and climate variables, the method presented here focuses only on the spatial distribution of demand. 

Using the data on national demand (EN_t) [ENTSO-E], population (P_x,t) [NASA GPWv4.11] and temperature (T_x,t) [ERA5], a variety of linear regression models were constructed for the regional demand (E_x,t). Each of these models allows a researcher to disaggregate the national demand time-series to the regional time-series in a simple, but effective manner. Based on our data the model with the highest accuracy out of the sample is of the following form: 

E_x,t = EN_t  ( α₁ P_x,t + α₂ T_x,t )

By using this method for the regionalization of electricity demand, a whole new range of research becomes possible. For instance, electricity transmission between regions can be explicitly modelled, enabling the identification of future congestion problems in the network.