Autoencoders for Hourly Load Profile Reconstruction in Renewable Energy Communities

A Machine Learning-Based Electricity Consumption Forecast and Management System for Renewable Energy Communities

“…Intravaia et al [21] proposed a novel strategy for generating sensible hourly consumption profiles using information commonly found in energy bills. They adopted a machine learning approach based on autoencoders, with an input/output dimension of N = 168, allowing the network to generate hourly load profiles for an entire week.…”

“…Furthermore, the majority of the research is conducted using distinct and non-publicly available datasets, making it very difficult to properly compare the proposed solutions. The few identified works (e.g., [20,21]) that effectively take into consideration the concept of energy community and also employ publicly available datasets, lack an energy management system that makes use of the resulting predictions. Additionally, in the case of [20], no deep learning or non-linear shallow learning models were employed, only advanced statistical time series methods, limiting the analysis of the different available options for the forecast algorithms.…”

“…Additionally, in the case of [20], no deep learning or non-linear shallow learning models were employed, only advanced statistical time series methods, limiting the analysis of the different available options for the forecast algorithms. On the other hand, the work in [21] does not actually provide real-time forecasts of the REC buildings consumption, being instead more focused on the reconstruction of the typical user consumption profiles. As a result, a new approach based on the utilization of ML models for the short-term consumption forecast of different buildings inside an energy community is required, as well as an energy management system that takes advantage of these predictions for the optimization of the REC electricity usage.…”