TS-DP: An Efficient Data Processing Algorithm for Distribution Digital Twin Grid for Industry 5.0
Lijun Xiao, Dezhi Han, Ce Yang, Jiahong Cai, Wei Liang, Kuan‐Ching Li
Abstract
As known, the smart Grid is an essential scenario for Industry 5.0. With its rapid development, the huge number of sensors and smart devices widely used in the industrial field generate significant amounts of data that sharply increase. Facing the power grid environment with a high amount of data, it is easy to cause abnormal conditions in the power grid system and even cause the system to collapse abruptly. To tackle this problem, we simulate a layered digital twin power grid model, using the comparative learning method to process the time series in the power grid and select the positive and negative time series samples using a new slider. In the extended Convolution Neural Network (CNN), we use the Atrous Convolution model, limiting the receptive field and concentrating more data close to the dimension of the model. The time series and process classification tasks are predicted on Electricity Transformer Temperature (ETT), electricity, and University of California, Riverside (UCR) data sets, and experimental results show that the proposed method reduces the error rate of 4.38% and 5.21% in the prediction task and improves the accuracy of 8.39% and 12.73% in the classification task, indications to control the power grid model more accurately, predicting the power grid operation in the future, and take corresponding measures in time.