Toward Enhanced Energy Forecasting for Smart Grid Integration in Net Zero Energy Buildings

Net zero energy buildings (NZEBs) represent a critical advancement in sustainable architecture, achieving equilibrium between annual energy consumption and on-site renewable energy production. Accurate short-term forecasting for energy generation and consumption is essential for efficient NZEB operation and smart grid management. This research introduces a novel hybrid deep learning (DL) framework by integrating temporal convolutional network (TCN) with bidirectional long short-term memory (BiLSTM) networks for precise energy forecasting in NZEBs. The proposed architecture employs multi-scale temporal features extraction using 1D convolutional layers with kernel sizes (1, 6, 12, 24) for capturing dependencies with different temporal resolutions. The residual TCN with dilated causal convolutions extracts complex temporal patterns while expanding receptive fields. The BiLSTM module processes sequences bidirectionally, capturing contextual information across entire time-window sequences. The key innovations include multi-head self-attention for dynamical temporal focus, the squeeze-and-excitation (SE) technique for adaptive feature refinement, and a custom dynamic time warping (DTW)-inspired loss function that preserves both magnitude accuracy and temporal structure. Comprehensive preprocessing incorporates data cleaning, normalization, environmental feature integration, curriculum learning for efficient sampling, and self-supervised pre-training for enhanced robustness. Experimental validation on two real-world datasets, solar PV energy generation from Alice Springs, Australia (26.5kw system), and four years of French buildings’ energy consumption (over 2 million samples), demonstrates superior performance over SOTA methods. The model achieves exceptional accuracy with MSE values of 0.0008 and 0.0007 or solar PV generation and building consumption, respectively, providing a scalable solution for NZEBs’ energy management and smart grid operations.