Model-Agnostic Meta-Learning Based Short-Term Carbon Emission Price Forecasting

Considering that the available training data are insufficient in building a conventional machine-learning based carbon price forecasting model for the national carbon emission market in China, an alternative machine-learning scheme, i.e. the model-agnostic meta-learning (MAML), is employed for short-term carbon price forecasting in this work. Specifically, a pre- trained global meta-model in MAML is generated with training data from all carbon emission trading centers, and then customized to personalized forecasting models for each carbon emission trading center, and then the carbon price forecasting accuracy of the carbon emission trading center with insufficient training data could be enhanced. Numerical results show that the proposed method could attain accurate carbon emission price forecasting results.