Abstract:

The SF₆ gas volume data from the PMS system is incomplete and contains significant errors, rendering it inadequate for power grid enterprises to calculate carbon reserves and formulate carbon planning for future substations. To address this problem, this paper studies the carbon storage accounting method for substations, taking into account buses and circuit breakers. The input parameters of the neural network are selected using the MIC method based on the field measured data in Ningxia grid, and three neural network models (GA-BP, PSO-BP, and HPO-BP) with six input parameters are developed. The verification results indicate that the HPO-BP neural network model outperforms the other two models in evaluation index and prediction results with a relative error of 6.28%, and it can accurately calculate the SF₆ gas volume of circuit breakers. Additionally, considering the parameter uncertainties, the linear relationship between different parameters is analyzed using the PCCs method, and a three-input-parameter HPO-BP neural network model is constructed, yielding a relative error of 7.69% for the prediction results. Concurrently, the ergodic output mode is examined to generate estimated SF₆ gas volume data for multiple groups of circuit breakers under uncertain parameter conditions. The total SF₆ gas volume of the substation is determined using the cumulative sum method, consequently quantifying the total carbon storage of the substation, thereby offering data support for power grid enterprises to accomplish the objective of "double carbon".

Key words: SF₆ gas volume, carbon storage, neural network model, PSO, uncertain parameters