Abstract
Reliable carbon price prediction can help to stabilize the carbon market, minimize financial risks for investors, and encourage innovation in green industries. The forecasts have a crucial role in reaching advanced goals for reducing emissions, aiding the shift toward an economy with reduced carbon emissions, and lessening the adverse effects of climate change overall. This paper proposes applications of LSTM with Temporal Pattern Attention (TPA-LSTM) to predict carbon price fluctuations. The prediction of carbon price fluctuations not only draws on its own historical information but also from its main predictors, including fossil fuel prices from 2018 to 2023. The TPA-LSTM method is a combined method that uses the LSTM layer as the initial input of the model. Furthermore, the output generated by the LSTM layer serves as the input to the TPA layer, which is then used to predict the carbon price for the following day. The model is tested by predicting the test data and calculating the evaluation results. The experimental results indicate that TPA-LSTM has surpassed other models in accuracy by showing better performance based on MSE, RMSE, MAE, and MAPE metrics.