A carregar...
Publicação
Hybrid Knowledge Representation for Temporal Forecasting
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.11 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Accurate electricity demand forecasting increasingly relies on complex machine learning models that lack explicit interpretability. This paper presents a hybrid framework that integrates statistical time-series forecasting with symbolic knowledge extraction to produce human-interpretable representations of climatic demand drivers. Using open-access meteorological and consumption data from multiple Portuguese cities, we train SARIMAX and ensemble models with exogenous inputs and extract logical IF–THEN rules that describe local demand regimes. These rules are then formalized as knowledge graphs encoding variable-level and rule-level causal structures. A case study on Coimbra shows that temperature and day length define compact, physically meaningful consumption regimes, demonstrating that high forecasting accuracy can coexist with explicit symbolic knowledge. The proposed approach provides a reproducible pathway from numerical prediction to formal knowledge representation for energy systems.
Descrição
Catarro, P. I., & de Campos Souza, P. V. (2026). Hybrid Knowledge Representation for Temporal Forecasting: Interpretable Symbolic Extraction from ARIMAX and Rule-Based Learning. Paper presented at International Joint Conference on Neural Networks (IJCNN) 2026, Maastricht, Netherlands. https://linklings.s3.amazonaws.com/organizations/WCCI/wcci2026/submissions/stype114/iuy8g-ijcnn_pap4667s2.pdf