Research on Deep Learning Based-carbon Measurement Model in UHVDC System

With the increasing application of the Ultra-High Voltage Direct Current (UHVDC) system, carbon emissions produced by these systems are growing in proportion within the power industry. However, several challenges exist in reducing carbon emissions for the UHVDC system, including difficulties in processing high-dimensional and strongly coupled data and establishing electrical-to-carbon conversion models. Considering this, this paper investigates the carbon measurement issue in the UHVDC system as follows: First, the loss mechanism is analyzed to determine its distribution. Then, considering the characteristics of loss, a factor accounting algorithm is selected to calculate the electrical-to-carbon conversion for the system. Finally, a carbon measurement model is constructed, combining deep learning models. To address the challenges of feature extraction and fusion in the UHVDC system, a model fusion strategy based on Residual Network and Long Short-Term Memory (LSTM) is proposed. Experimental results demonstrate that the evaluation indexes of the fusion carbon measurement model are superior to other deep learning models. Compared to the LSTM, the fusion algorithm model reduces mean squared error, root mean squared error, mean absolute error, and mean absolute percentage error by 30.51%, 20%, 16.67%, and 2.17% respectively.