Carbon Emission Prediction and Sensitivity Evaluation of Virtual Power Plants Based on Big Data and Multiscale Analysis

Abstract

In order to address the issue of increased prediction errors in the peak carbon emissions of virtual power plants due to various influencing factors of electricity carbon emissions, the authors propose a study on the prediction and sensitivity evaluation of virtual power plant carbon emissions based on big data and multi-scale analysis. Firstly, it analyzes the original data sequence and cumulative sequence, use grey BP neural network to construct a carbon emission peak prediction model, then it analyzes the factors affecting electricity carbon emissions, and use recursive calculation method to calculate electricity carbon emissions. Then, it compress the model coefficients to zero through a penalty function and filter out significant variables. Based on the adjacency characteristics of carbon emission flow, the node carbon potential is calculated through finite recursion, and iterative training is carried out within the allowable error range to solve the model and obtain the predicted peak carbon emissions of electricity. The experimental results indicate that the prediction results of the designed method under three scenarios of benchmark setting, low-carbon, and enhanced low-carbon are 40 million tons, 390 million tons, and 40 million tons, respectively, which are consistent with the actual results, indicating that the prediction error of this method is lower and the prediction results are more accurate. The method studied by the authors can provide technical support for carbon emission control and improve prediction accuracy.