Research and Implementation of a Carbon Emission Prediction Method Based on Electricity Data-Driven Approach

In the context of low-carbon development, the establishment of a reliable carbon emission prediction model is of great significance for the accurate monitoring of carbon emissions and the formulation and implementation of future carbon reduction goals. This paper proposes a carbon emission prediction method based on information entropy feature selection and tree-structured Parzen estimator optimized neural network. Firstly, the raw data is normalized and the mutual information analysis is conducted on the feature variables to obtain the information entropy between carbon emissions and each feature. Secondly, strong features are selected, and an extreme learning machine network structure is constructed. Thirdly, the tree-structured Parzen estimator is used to optimize the hyperparameters of the network, and the optimal hyperparameters are fed into the extreme learning machine. Additionally, a regularization term is added to the model to prevent overfitting and obtain the prediction of carbon emissions. Finally, the effectiveness of the proposed model is validated on a carbon emission dataset from 120 steel enterprises. The results show that the proposed model, which optimizes the mutual information extreme learning machine with the tree-structured estimator, outperforms the control group in all evaluation indicators, demonstrating its superiority in carbon emission prediction in the steel industry. The fast optimization capability of the TPE algorithm and the high accuracy and robustness of ensemble learning improve the accuracy and stability of the prediction model.