Prophet-LSTM-BP Ensemble Carbon Trading Price Prediction Model.

Accurately identifying changes in carbon trading prices can provide reasonable reference indicators for a government's macrocontrol and can also help companies more effectively avoid risks brought by carbon emissions and increase the income of carbon assets. Based on the Prophet model, LSTM neural network model, and backpropagation (BP), this paper proposes a method to predict carbon trading prices using the ensemble learning model and uses the Hubei carbon trading market data to predict carbon trading prices. Results show that in terms of accuracy, the Prophet-LSTM-BP ensemble learning model achieves better predictive ability than existing models; its RMSE, MAE, and MAPE are 1.479, 0.951, and 2.135, respectively, which are markedly smaller than the Prophet model's 5.631, 4.471, and 9.661, and the LSTM model's 3.352, 3.105, and 6.880, respectively. Compared with the traditional time series ARIMA model, the MAPE of ARIMA reaches 12.933, which is nearly 1.5 times that of the Prophet model, nearly 2 times that of the LSTM model, and nearly 7 times that of the ensemble learning model. In terms of applicability, when the model is applied to the national carbon trading market, the difference in MAPE compared with the Hubei carbon trading market is only 0.6%, and the other parameters are not more than 16%. The model improves the relevant research on carbon trading price predicting, and concurrently, this method provides ideas for carbon trading price predicting in other carbon trading markets and promotes the sustainable development of the national carbon trading market. [ABSTRACT FROM AUTHOR]