Abstract: Aiming at the problem that the different values of the weight of the input layer and the bias of the hidden layer of the extreme learning machine（ELM） have a great influence on the prediction results and the existing prediction model has low prediction accuracy for the power demand prediction of Guangdong Province, a method based on the tuna swarm optimization（TSO） algorithm is proposed to optimize the ELM to obtain the optimal value and construct the TSO-ELM prediction model. After normalizing the six influencing factors and power demand data of Guangdong Province from 2008 to 2018, a prediction model is constructed to predict the power demand of Guangdong Province from 2019 to 2021. The simulation results show that compared with the four prediction models of SVM, BP, ELM and GWO-ELM, the TSO-ELM prediction model has higher prediction accuracy.