Answers are grounded in this paper and cite their sources.
The accurate, extended period prediction of individual customer energy consumption is critical for utility providers. Machine learning techniques, particularly neural networks, have proven effective in predicting household energy consumption by identifying correlations and patterns. However, these predictions often generalize across the entire dataset, neglecting the distinct behaviors of specific sub-groups. This paper presents an innovative transformation architecture aimed at enhancing the prediction of gas consumption for multiple households or population subgroups concurrently. The adaptability of the transformation layer to various neural network frameworks allows for broader applicability. The model's performance is assessed based on prediction accuracy and efficiency. Furthermore, as the transformation layer may introduce private information during training, we also evaluate the robustness of the model against inference attacks and its resilience to Additive White Gaussian Noise (AWGN) and adversarial examples. Our results demonstrate that the proposed approach not only achieves parallel prediction with high accuracy but also maintains the ability to forecast consumption over an extended period without the need for recent meter readings.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: