Prediction interval models have been proposed to characterize uncertain phenomena, which provide expected values and the upper and lower bounds of the interval given a desired coverage probability. However, existing methods of generating prediction intervals mainly use simple models such as recurrent neural network (RNN), long short-term memory (LSTM), or predominant models such as transformers as the backbone network, which have limitations in modeling long-term multivariate sequences. To address the issue, the paper proposes LMDCPI (long-term multivariate dual channel prediction interval), which employs dual branches to capture both intra-variable temporal dependencies and inter-variable dependencies from global and local perspectives. Experimental results show that LMDCPI achieves state-of-the-art performance on several public data sets.
Li et al. (Tue,) studied this question.