Short-term power load forecasting of a city in Henan Province using Attention based LTSM

In order to ensure the security and stability of power grids and the efficient operation of power markets, grid dispatchers and power market participants have high requirements on the accuracy of power load forecasting. Power load forecasting based on engineering computation and deep learning is an important foundation for power optimization and smart grid dispatching. Addressing the challenge that traditional load forecasting methods cannot simultaneously model the changing patterns of the load itself and its influencing factors, a load forecasting method based on the self-attention mechanism of Long Short-Term Memory units (LSTM-Self Attention, abbreviated as LSTM-SA) is proposed. This method utilizes historical electricity load data and relevant influencing factor data to predict the load for the upcoming day. The critical influencing factors of short-term load forecasting are analyzed first. Using Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) neural networks as examples, the principles of neural network models and their alignment with load forecasting tasks are explained. Subsequently, the self-attention mechanism is introduced to automatically allocate appropriate weights to LSTM hidden layer states, distinguishing the importance of different time load sequences. This effectively reduces the loss of historical information and highlights the information of crucial historical time points. Finally, the ultimate load prediction results are output through a fully connected layer. Experimental analysis is conducted using actual power load data from a city in Henan Province as an example. Based on real load data, the performance of load prediction for different historical time windows and network architectures is validated. The results, compared with other load prediction models, indicate that the proposed model can improve prediction accuracy and is more effective in short-term load forecasting. This model can serve as a reliable basis for power system planning and stable operation.