Abstract: Power load forecasting is a key task in the dispatching and management of power systems. To improve the forecasting accuracy and stability, this paper proposes a power load forecasting method based on the LSTM-GRU-KAN combined model. This method combines the characteristics of long short-term memory networks (LSTM) and gated recurrent units (GRU), and further optimizes the model performance through the kolmogorv-arnold networks (KAN). The LSTM, GRU and LSTM-GRU models were compared on the power load dataset. The results show that the LSTM-GRU-KAN model performs better than the traditional models in multiple evaluation indicators.