关键词: 中长期径流预报；自适应噪声完备集合经验模态分解；原子搜索算法；时间卷积网络；岷江流城

Abstract: Accurate prediction of monthly runoff is crucial for water resource management in a watershed. In order to enhance the accuracy of medium and long - term runoff prediction, CEEMDAN - IASO - TCN combined model is proposed, which is constructed by combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), improved atomic search algorithm (IASO), and temporal convolutional network (TCN). The model firstly uses CEEMDAN to decompose the monthly runoff sequence, and then uses IASO to optimise the batch size, learning rate, and discard factor of the TCN model to obtain the optimal time convolution network structure and predict the components using the optimal ASO - TCN, and finally reconstructs the component prediction results to obtain the final monthly runoff prediction results. The monthly runoff data from 1957 to 2019 at Zhenjiangguan Hydrological Station in Minjiang River Basin are taken as the study object, and the proposed model is compared with other models. The results show that the CEEMDAN - IASO - TCN model has the highest prediction accuracy, with Nash coefficients of 0.9191 and 0.8691 in the training and testing stages, respectively. The research results can provide a reliable basis for the sustainable use of water resources.

Key words: medium and long - term runoff forecasting; complete ensemble empirical mode decomposition with adaptive noise; atomic search algorithm; temporal convolutional network; Minjiang River Basin