Deep learning-based estimation and correction algorithm for synchronisation time error in smart grid

As the core infrastructure of power supply in modern society, the robustness and efficiency of smart grids are crucial for daily operations. Accurate estimation and correction of such errors are crucial. However, traditional methods lack accuracy and robustness. A deep learning model based on a combination of two-dimensional convolutional neural networks and long short-term memory networks (2D-CNN-LSTM) is innovatively proposed for the first time in the estimation and correction of synchronisation time error. This model utilises the spatial feature extraction of 2D-CNN and the time series learning ability of LSTM to improve the accuracy and robustness of synchronisation time error estimation. The results showed that the model achieved the highest accuracy with a 50-length sliding window. The model exhibited resilience, with over 90% accuracy within 5-metre distance. Smart grid's actual time error overlapped with correction results, validating the model's effectiveness. The root-mean-square error of the neural networks exhibited a stabilisation without a significant decrease at a lag time step of 6, which is a crucial finding for the stability of the smart grid. The proposed algorithm significantly improves accuracy and aids operational efficiency and safety.