Carbon trading price prediction based on a two-stage heterogeneous ensemble method

Several countries have formulated carbon–neutral plans in dealing with global warming, which have also derived various carbon trading markets. All parties involved in carbon trading aim to obtain the maximum benefit from it, and this requires participants to accurately judge the carbon trading price. This study then proposes a two-stage heterogeneous ensemble method for predicting carbon trading prices. To accurately capture the characteristics of the time series data, we extracted four feature sets based on the lag length, moving average, variational mode decomposition, and empirical mode decomposition methods. Subsequently, four algorithms, linear regression, neural network, random forest, and XGBoost, constructed the first-layer model. We used a neural network algorithm to build the second-layer model to enhance the predictive model fit. Moreover, we used the particle swarm optimization algorithm to optimize the crucial parameters involved in the model. Extensive numerical experiments were conducted on carbon trading data from the Beijing carbon trading market in the past five years (2016–2021), and showed that our proposed method is superior to other popular methods such as LightGBM, support vector machine, and k-nearest neighbor.