Authors: Rafael Gonçalves, Diogo Magalhães, Rafael Teixeira, Diogo Gomes, Mário Antunes, and Rui Aguiar
Abstract: Residential energy forecasting is a challenging task due to the non-stationary nature of the data. Therefore, model weights need to be often updated to adapt to changes in the data distribution and avoid performance degradation. However, it is not always feasible to frequently retrain regression models with lookback and forecast windows large enough to capture energy patterns because increasing the number of features leads to longer training times. To address this problem, we propose an approach that guarantees fast convergence through dimensionality reduction. We first focus on achieving good performance on a synthetic neighbourhood dataset with an Artificial Neural Network (ANN), a 1D Convolutional Neural Network (1D-CNN) and a Long Short-Term Memory (LSTM) network. Then, we mitigate the observed increase in training time using Principal Component Analysis (PCA) and a Variational AutoEncoder (VAE). To ensure the suitability of the proposed models for a residential context, we also explore the concepts of locality and globality and discuss the trade-offs between low error and training speed. In this regard, we test three different scenarios: a global model, a local model for each building, and a global model that is fine-tuned for each building. The results showed that by combining the best reduction technique with the best model architecture, it is possible to decrease the Mean Squared Error (MSE) by up to 63% and accelerate training by up to 80%.