“Load forecasting is crucial for the planning and operation of power systems. The high proportion of new energy integration increases the uncertainty of source load, leading to increased volatility of new power system loads and difficulty in accurately predicting multi-user loads. The research team proposes a multi-user short-term load forecasting method based on frequency domain enhanced Transformer, which uses an encoder to capture feature information, obtains frequency domain information through discrete cosine transform, and enhances data through channel attention mechanism. The decoder integrates and outputs the prediction results, reducing errors by 21.8% compared to traditional Transformer models, providing a high-precision solution for short-term load forecasting in new power systems.”