The characteristic of energy systems

in which the source varies according to the load

dictates that the load side bears responsibility for the system's carbon emissions. Jointly predicting multi-energy loads and carbon emissions forms the fundamental basis for the low-carbon operation of dynamic integrated energy systems. Given the challenge of analyzing correlation relationships within massive heterogeneous data from the time domain perspective

this paper employs an enhanced data decomposition technique to uncover data relationships from a frequency domain perspective. Addressing the computational demands associated with learning heterogeneous data characteristics

this study establishes a joint prediction model for multi-energy loads and carbon emissions using a teacher-student network architecture based on knowledge distillation. The teacher network extracts crucial information from extensive low-density data to train the student network

ensuring model prediction performance with fewer computational resources. Subsequently

the measured data are used for verification. Compared with the existing prediction models

the prediction model using the improved model decomposition technology exhibits higher prediction accuracy. The joint prediction model adopting the knowledge distillation method not only improves the prediction accuracy of the model

but also reduces the calculation time.