Abstract: To solve the problems of existing carbon emission prediction models such as a limited number of indicators and slow data updates, this article proposes a substation carbon emission prediction model based on the improved firefly algorithm (IFA) optimized BP neural network. Firstly, in response to the slow convergence speed and tendency to fall into local optima in the firefly algorithm (FA), teaching and learning factors are introduced to modify the firefly position update process to improve population fitness. Secondly, IFA is introduced to perform hyper-parameter optimization on the BP neural network model, and an IFA-BP neural network prediction model is constructed. Then, based on the CRITIC method, select key carbon emission indicators for the input layer of the prediction model. Finally, the prediction model is trained using the training set data to predict the carbon emissions of the substation based on the trained model. The simulation results show that compared with the three comparison schemes, the root mean square error (RMSE) of the proposed IFA-BP neural network prediction model decreases by 59.61%, 15.77% and 26.65%, respectively. The coefficient of determination (R²) increases by 5.66%, 1.46% and 1.15%. The feasibility and superiority of the substation carbon emission prediction model proposed in this paper are fully verified.

Key words: carbon emissions, substation, IFA optimization algorithm, BP neural network, teaching and learning factors