A simulation-to-forecast framework with correlation-based feature selection and multiple machine-learning models for PCM-integrated buildings energy prediction
Energy, cilt.360, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 360
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.energy.2026.141615
- Dergi Adı: Energy
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, Environment Index, Geobase, INSPEC, Public Affairs Index, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Anahtar Kelimeler: Building energy, Feature selection, Influential building and environmental parameters, Machine learning methods, Phase change materials (PCM)
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
The building sector accounts for a substantial share of global energy consumption, making energy-efficient building design a key component of sustainable development. Phase change materials (PCMs) can reduce building heating and cooling demand; however, optimizing PCM thermo-physical properties and envelope configurations using repeated dynamic simulations is computationally intensive. To address this shortcoming, this research proposes a novel correlation-based feature-selection workflow coupled with gene expression programming (GEP), multi-expression programming (MEP), and Support vector machine (SVM) to forecast monthly heating-and-cooling energy consumption of PCM-integrated residential buildings in a hot semi-arid (BSh) climate. EnergyPlus was used to generate a dataset of 194,400 cases for nine residential building models, considering building characteristics, environmental conditions, and commercially available PCM properties. Pearson, Spearman, and Kendall correlation analyses were applied to identify influential variables and remove redundant inputs, reducing database-generation requirements by approximately 17%. The best GEP model achieved high predictive accuracy (R2 > 0.97) and required less training time than the best SVM model. Furthermore, the derived symbolic expression enabled interpretable PCM selection, identifying an optimal melting temperature near 26 °C and predicting energy reductions of up to 16%.