Application of a Deep Hybrid Model Based on Transformer-LSTM and GCN-GAT in Load Disaggregation with Sparse Data

A method for power load forecasting and decomposition. This method is based on a hybrid model and combines several techniques such as T-LSTM, GCN-GAT and ANFIS. The role of T-LSTM is to capture temporal characteristics and use it for daily load forecasting. With it, the characteristics in terms of time can be well grasped. Predict the daily load situation. GCN-GAT is used to optimize the spatial relationship among power equipment and make the spatial connection among power equipment more reasonable. ANFIS can precisely decompose the daily load data into hourly load data, that is, refine the load data measured in days to each hour. The results obtained from the experiment show that This method has achieved relatively high accuracy rates in both the tasks of load forecasting and decomposition. It can effectively improve the precision of load forecasting and decomposition, making the results of forecasting and decomposition more accurate.