The prediction accuracy of a single model will deteriorate because of load fluctuations when making ultra-short-term load forecasting. To solve this problem, this paper proposes a combinatorial prediction model based on a deep learning algorithm. First, variational mode decomposition is used to decompose the original load sequence to obtain a series of sub-sequences. Secondly, a bidirectional long short-term memory network and an optimized deep extreme learning machine are used to predict each sub-sequence. Thirdly, the improved Q-learning algorithm is used to weight the prediction results of the bidirectional long short-term memory network and of the deep extreme learning machine to obtain those of each sub-sequence. Finally, the prediction results of each subseries are summed to obtain the final load prediction results. The results show that the prediction model proposed in this paper performs better than other models in ultra-short-term load forecasting, with a prediction accuracy of more than 98%.