Application of LSTM Model Considering Temporal Dependency in Electric Vehicle Power Load Forecasting

Power load forecasting is crucial for ensuring the safe and economical operation of modern power systems. Nowadays, a large number of electric vehicles (EVs) have been integrated into power systems, causing certain impacts on them. To address the issues of nonlinearity and temporal dependency in electric vehicle load data, this study proposes a short-term power load forecasting model based on Long Short-Term Memory (LSTM) neural networks. The model first performs data preprocessing, including missing value imputation and min-max normalization, and constructs 96-time-step sequences through feature engineering. It adopts a two-layer LSTM network combined with fully connected layers, with Mean Squared Error (MSE) as the loss function, and is trained for 30 epochs. Experimental results based on 2024 electric vehicle load data from a certain region in China show that the MSE of the training set and validation set drops to 0.00013 and 0.00009, respectively. The Mean Absolute Percentage Error (MAPE) between the predicted values and the actual values is only 0.91%, with a high degree of curve overlap. This indicates that the LSTM model can effectively capture the temporal dependency in power loads and provide an efficient and reliable solution for power systems.