A novel anomaly detection method based on reinforcement ensemble learning and multi-source domain transfer for building energy consumption
Chen Chen, You Lu, Qiming Fu, Jianping Chen, Yunzhe Wang, Zhenping Chen · International Journal of Green Energy · 2025
The identification of anomalies in the energy consumption of buildings is imperative for the implementation of sustainable energy savings in smart buildings. The proposed methodology (REL-MSDT) integrates reinforcement learning and multi-source domain transfer in order to address the limitations of existing machine learning methods in new buildings with limited historical data. The transfer learning model is constructed through the integration of graph convolutional and long-short-term memory networks, with domain adaptation employed to ensure the model’s adaptability to target domains that are analogous to the source domain. Furthermore, it adjusts ensemble weights dynamically through the use of reinforcement learning, thereby ensuring the effective transfer of knowledge from multiple source domains to the target domain. The experimental results demonstrate that the proposed method exhibits superior performance in terms of accuracy, score, and recall when compared to alternative approaches.