Short-Term Building Electrical Load Prediction by Peak Data Clustering and Transfer Learning Strategy

Energies, Journal Year: 2025, Volume and Issue: 18(3), P. 686 - 686

Published: Feb. 2, 2025

With the gradual penetration of new energy generation and storage to building side, short-term prediction power demand plays an increasingly important role in peak response supply/demand balance. The low occurring frequency electrical loads buildings leads insufficient data sampling for model training, which is currently factor affecting performance load prediction. To address this issue, by using clustering knowledge transfer from similar buildings, a forecasting method proposed. First, building’s are clustered through peak/valley analysis K-nearest neighbors categorization method, thereby addressing challenge data-sparse scenarios. Second, clusters, instance-based learning (IBTL) strategy used multi-source domains enhance target prediction’s accuracy. During process, two-stage selection applied based on Wasserstein distance locality sensitive hashing. An IBTL strategy, iTrAdaboost-Elman, designed construct predictive model. proposed validated public dataset. Results show that reduces error 49.22% (MAE) compared Elman Compared same without clustering, approach also achieves higher accuracy (1.96% vs. 2.63%, MAPE). forecast hourly/daily demands two real campus USA China, respectively. effects both analyzed detail.

Language: Английский

---

Day-Ahead Demand Response Potential Prediction in Residential Buildings with HITSKAN: A Fusion of Kolmogorov-Arnold Networks and N-HiTS

Energy and Buildings, Journal Year: 2025, Volume and Issue: unknown, P. 115455 - 115455

Published: Feb. 1, 2025

Language: Английский

---

Modeling the complex spatio-temporal dynamics of ocean wave parameters: A hybrid PINN-LSTM approach for accurate wave forecasting

Measurement, Journal Year: 2025, Volume and Issue: unknown, P. 117383 - 117383

Published: March 1, 2025

Language: Английский

---

Towards sustainable architecture: Enhancing green building energy consumption prediction with integrated variational autoencoders and self-attentive gated recurrent units from multifaceted datasets

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(4), P. e0317514 - e0317514

Published: April 25, 2025

Global awareness of sustainable development has heightened interest in green buildings as a key strategy for reducing energy consumption and carbon emissions. Accurate prediction plays vital role developing effective management conservation strategies. This study addresses these challenges by proposing an advanced deep learning framework that integrates Time-Dependent Variational Autoencoder (TD-VAE) with Adaptive Gated Self-Attention GRU (AGSA-GRU). The incorporates self-attention mechanisms Multi-Task Learning (MTL) strategies to capture long-term dependencies complex patterns time series data, while simultaneously optimizing accuracy anomaly detection. Experiments on two public building datasets validate the effectiveness our proposed approach. Our method achieves 93.2%, significantly outperforming traditional methods existing techniques. ROC curve analysis demonstrates model’s robustness, achieving Area Under Curve (AUC) 0.91 maintaining low false positive rate (FPR) high true (TPR). presents efficient solution prediction, contributing conservation, emission reduction, construction industry.

Language: Английский

---

Power supply quality prediction method based on LSTM and self-attention mechanism

Journal of Computational Methods in Sciences and Engineering, Journal Year: 2025, Volume and Issue: unknown

Published: April 26, 2025

Existing LSTM-based power quality (PQ) prediction models primarily rely on historical information, which limits their ability to fully capture contextual dependencies. Furthermore, these process inputs sequentially without accounting for the varying importance of different time steps, leading significant inaccuracies. To address limitations, this study proposes an enhanced PQ model that integrates Bidirectional Long Short-Term Memory (BiLSTM) with a Self-Attention (SA) mechanism. The BiLSTM module is introduced both forward and backward temporal dependencies, enabling more comprehensive long-term patterns in series data. SA mechanism dynamically adjusts steps through weighted summation, enhancing model’s focus critical features improving its capacity nonlinear relationships. from layer are then mapped connected generate final outputs. Experiments were conducted using data Nanchang as primary dataset, additional datasets Nanjing, Wuhan, Changsha, Beijing used generalization testing. results demonstrate BiLSTM-SA outperforms traditional LSTM across all metrics, achieving mean absolute error (MAE) 0.09 voltage deviation, 0.05 improvement over single-layer LSTM. Notably, maintains robust performance complex supply scenarios, generalized MAE only 0.2 Beijing. These findings highlight effectiveness combining reducing errors ensuring stability quality, offering advancement methodologies.

Language: Английский

---