Against the backdrop of the carbon neutrality target, the low-carbon transformation of urban building energy systems faces significant challenges. Traditional energy station management modes rely heavily on manual inspections and isolated control strategies, leading to low energy efficiency and limited predictive capabilities. To address the dynamic optimization requirements of complex energy systems, this study, through integrating Internet of Things (IoT) and Digital Twin (DT) technologies, proposes a data-driven predictive model for energy stations. By constructing a full lifecycle DT model and incorporating advanced artificial intelligence (AI) algorithms, the framework enables three-dimensional visual monitoring, operational state prediction, and virtual interactive optimization of the system. The proposed Sensor Data Stream Prediction (SDSP) module, particularly the Empirical Mode Decomposition and Long Short-Term Memory for Sensor Data Stream Prediction (EMD-LSTM-SDSP) model, achieves enhanced prediction accuracy for chilled water supply and return temperatures, while also improving cooling load forecasting performance, by integrating physical insights from the digital twin. Validated using operational data from the Hankou Binjiang International Business District Water Source Heat Pump (WSHP) station in Wuhan, the method demonstrates superior reliability compared to three traditional time-series prediction approaches, with the EMD-LSTM-SDSP model attaining prediction accuracies of 98.49% for supply temperature and 98.52% for return temperature, and 86.55% for initial cooling load. Furthermore, the DT-enabled framework enables real-time forecasting of operational modes and cooling capacity variations, with the deduced cooling capacity prediction accuracy reaching 83.39%. Beyond improved prediction, the integration of DT reasoning facilitates closed-loop control by providing forecasted states as input signals to the physical system controller, bridging the gap between predictive analytics and real-time operational optimization. This approach overcomes the limitations of traditional energy management systems in data integration and intelligent decision-making, offering an innovative solution for the refined and intelligent operation and maintenance of large-scale energy facilities.
Zhang et al. (Mon,) studied this question.