Study on Meta-Learning-Improved Operational Characteristic Model of Central Air-Conditioning Systems

Establishing accurate models for central air-conditioning systems is an indispensable part of energy-saving optimization research. This paper focuses on large commercial buildings and conducts research on improving the energy efficiency model of chillers in central air-conditioning systems based on meta-learning. Taking the Model-Agnostic Meta-Learning (MAML) framework as the core, the study systematically addresses the energy efficiency prediction problem of chillers under different operating conditions and across different equipment. It constructs a comprehensive research process including data preparation, meta-model training, fine-tuning and evaluation, cross-device transfer, and energy efficiency analysis. Through its bi-level optimization mechanism, MAML significantly enhances the model’s rapid adaptability to new tasks. Experimental validation demonstrates that: under varying operating conditions on the same device, only 5 data points are required to achieve a relative error (RE) within 3%; under similar operating conditions across different devices, 4 data points achieve a RE within 5%. This represents a reduction in data requirements by 50% and 73%, respectively, compared to standard Multi-Layer Perceptron (MLP) models. This method effectively addresses modeling challenges in complex operating scenarios and offers an efficient solution for intelligent management. It significantly enhances the model’s rapid adaptation capability to new tasks, particularly its generalization performance in data-scarce scenarios.