A novel building flexibility potential assessment method based on hybrid CNN-GRU-CALDA framework considering consumer psychology

Buildings are rapidly getting on top of the agendas of various initiatives aimed at enhancing the operational flexibility and resource utilization. Accurately and comprehensively assessing the flexibility potential of buildings is pivotal for unlocking and efficiently harnessing demand-side flexibility, thereby bolstering the overall development of the energy system. In this study, a novel transfer learning strategy based building flexibility potential assessment methodology is proposed. Firstly, the consumption profiles of buildings from heterogeneous industries (e.g., residential, commercial, and industrial) are physically simulated through building energy simulation tool. To overcome the performance degradation occurred in single-source transfer learning, a hybrid convolutional neural network-gated recurrent unit (CNN-GRU) framework integrated contrastive adversarial learning for multi-source time-series domain adaptation (CALDA), termed as CNN-GRU-CALDA is proposed. In this way, the source and target feature representations can be better aligned and thereby improving the prediction performance of target building energy consumption by leveraging multi-source building data (domain). Furthermore, an HVAC load response model based on consumer psychology is introduced in the assessment process to realize the comprehensive consideration of consumers with different preferences. Finally, experiments are conducted to validate the effectiveness and applicability of the proposed methodology. Results demonstrate that the proposed CNN-GRU-CALDA model performs better than other models with respect to extracting domaininvariant features and prediction accuracy. Besides, considering consumer psychology is more reliable and suitable to optimally exploit the building energy flexibility in practice.