Researcher at HSE University in Perm Predicts Electricity Consumption in Residential Buildings
Aleksey Kychkin, Associate Professor in the Department of Information Technologies in Business at HSE University in Perm, together with Georgios Chasparis, a scientist at the Software Competence Center Hagenberg (SCCH, Austria), built models to predict energy consumption in residential buildings for the day ahead. The electricity consumption profile of a group of residential buildings, which is determined for the day ahead, will allow electricity demand to be effectively managed. The results of the research were published in ‘Energy and Buildings’ journal.
The need for accurate balancing in electricity markets and a larger integration of renewable sources of electricity require accurate forecasts of electricity loads in residential buildings.
Researchers at HSE University in Perm and their Austrian colleagues have been reviewing various forecasting methodologies that have been used in the electric power industry in Russia and Europe for several years to build load profiles of a residential building for the day ahead. Initially, the researchers reviewed standard forecasting methodologies.
The researchers have discovered that electricity loads for groups of buildings connected to the same substation or located in the same energy district change greatly over time, which means that the accuracy of forecasting using persistence models systematically decreases and does not achieve the project’s goal.
The researchers tried to create the best combination of persistence models. But any set of forecasting models always worked a bit worse than the best model in terms of accuracy at a specific time period. The scientists then decided to develop a series of machine learning models that carry out selection of a forecasting strategy taking into account an adaptive strategy. As a result, the research presents three forecasting models: i) the Persistence-based Auto-regressive (PAR) model, ii) the Seasonal Persistence-based Regressive (SPR) model, and iii) the Seasonal Persistence-based Neural Network (SPNN) model. With the help of modelling, the researchers demonstrated the accuracy of forecasting for all considered models based on real energy consumption data of a large number of buildings.
As a result of the experimental calculations, it turned out that the proposed models increase the quality of forecasting for energy consumption. Such models are more robust in relation to persistence models for long time periods, they are prepared for changes in the behaviour of single energy consumers who form a load profile, changes in user behaviour, and seasonal climatic changes.
The Seasonal Persistence-based Regressive (SPR) model with the training sample up to 1 month showed the best accuracy and adaptability. It can be used in the early stages of forecasting buildings’ energy consumption.
As the training sample grows, the researchers recommend switching to models that use neural networks as a tool for determining nonlinear dependencies in the selection mode.
The forecasting strategy presented in the article makes it possible to guarantee an increase in the accuracy of persistence models by at least 5%, which can be critically important for large power systems. The energy consumption profile of a group of residential buildings, which is determined for the day ahead, allows researchers to decide whether it is reasonable to manage price-dependent electricity demand.
The ability to accurately forecast building’s energy consumption also helps researchers to create scenarios for controlling energy storage, which along with pricing models can be used to balance generation and consumption, including the use of renewable energy sources.