Smarter Homes, Greener Future: unveiling the power of causal Machine Learning in energy consumption

As part of the SENDER project’s research into energy consumption, an important step has been taken in forecasting electricity demand for residential buildings. These structures make a significant contribution to energy consumption, with the residential sector accounting for a large share in both the European Union (26.1%) and the USA (21%).

In the context of energy consumption, residential buildings play an important role, accounting for a substantial share of energy consumption in the European Union and the USA. Focusing on the optimisation of heating, ventilation and air-conditioning (HVAC) systems, CRS4 researchers are developing key predictive models for demand-side management (DSM) and demand-response (DR) strategies.

In a recent scientific paper (see left), entitled “Total and thermal load forecasting in residential communities through probabilistic methods and causal machine learning”, an innovative approach (causal machine learning) was introduced, with the aim of predicting electricity demand not only at the aggregate level, but also by dissecting components such as base loads and heating, ventilation and air conditioning (HVAC) consumption. Such innovations are essential for improving energy efficiency and facilitating informed decision-making in the growing field of energy management and sustainability.

The SENDER project, carried out at three pilot sites in Europe, involves the development of a home automation system to regulate thermal loads, demonstrating the practical application of our research.

The novelty of our research lies in the use of causal machine learning methods, a pioneering approach that distinguishes our work from conventional forecasting techniques.

Unlike traditional methods, which often focus solely on correlations between variables, causal machine learning enables us to understand cause-and-effect relationships in the complex dynamics of energy consumption. By incorporating physical knowledge of phenomena, our methodology provides a more insightful and interpretable means of predicting the components of energy demand in residential communities.

This innovation not only improves the accuracy of our predictions, but also sets a precedent in the use of advanced machine learning techniques to untangle complex relationships, contributing to the development of more robust and transparent predictive models in the field of energy forecasting.

The main innovation is to offer a short-term, one-day-ahead probabilistic forecast of electricity demand, distinguishing between total aggregate demand, base load and HVAC consumption. Using conformal quantile regression and causal machine learning techniques, we obtained accurate forecasts without the need for direct measurements of heating system electrical loads.

This breakthrough is essential for energy communities, as it facilitates optimal coordination of energy production, distribution and consumption. The next steps will be to refine the methods, address limitations and explore broader applications beyond heating and cooling systems. The ultimate goal is to contribute to the advancement of energy forecasting tools for more effective management of distributed networks and energy communities.