Aiming at the nonlinear, nonstationary, and time series characteristics of power load, this study proposes a load forecasting method based on empirical mode decomposition and particle swarm optimization of the gated recurrent unit neural network. First, the original power load data are decomposed into a limited number of modal components and a residual component by using empirical modal decomposition to reduce the nonstationarity and complexity of the load sequence and decrease the association between different IMFs. The subsequences build prediction models based on the gated recurrent unit neural network, respectively, and use the particle swarm algorithm to optimize the network-related hyperparameters to increase the parameter accuracy of the model; finally, superimpose the prediction results of each subsequence to obtain the final load prediction value. The results of the case study show that compared with the traditional forecasting algorithm, the proposed EMD-PSO-GRU forecasting model method can better dig the trend information of forecasting, fit the load curve better, and have higher forecasting accuracy.