Abstract: Accurately predicting power load demand is crucial for optimizing power generation and scheduling plans, improving economic efficiency, and ensuring the safe operation of the power grid. A short-term power load forecasting model based on the Spotted Kingfisher Optimization Algorithm(PKO) optimized TCN-GRU Attention is proposed for this purpose. The model first performs hyperparameter optimization on the TCN-GRU network through PKO, then extracts features from the data through a temporal convolutional network(TCN), and finally captures and remembers long-term dependency information of the data through a gated recurrent unit(GRU) and self attention mechanism(Self-Attention). Through simulation verification analysis of the total load dataset of a certain location in southern China in 2023, the prediction model PKO-TCN-GRU Attention outputs root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination of 25.419, 18.236, 1.064%, and 0.994, respectively, proving the generalization and reliability of the model in practical environmental applications.