Aiming at the problem of low prediction accuracy of power load data with large volatility and high complexity, a hybrid power load forecasting model combining variational mode decomposition (VMD), improved osprey optimization algorithm (IOOA), bidirectional gated recurrent unit (BiGRU), and error compensation was proposed, called VMD-IOOA-BiGRU-EC. First, VMD was used to perform the initial decomposition of the load sequence to extract multiple modes and residuals. Then, three strategies circle chaos mapping, dynamic elite guidance mechanism, and "optimal-random mean" mutation were used to improve the relevant hyperparameters of the OOA optimized BiGRU model to enhance the prediction effect of the initially decomposed modes. Simultaneously, a VMD-BiGRU model was constructed for the residuals generated by the initial decomposition; the residuals are decomposed twice, and BiGRU was used to predict the decomposed residuals, constituting error compensation. Finally, the prediction results of each mode after the initial decomposition and the secondary decomposition were superimposed to obtain the final load prediction value. Experimental verification was conducted on a real power load dataset from Zhuzhou City, Hunan Province. The results show that the proposed method has lower average absolute error, root mean square error, and average absolute percentage error than other comparison models, demonstrating its effectiveness in processing complex load data.