In this paper, a chain-structure echo state network (CESN) is newly proposed to enhance the scalability, robustness and computational efficiency in spatio-temporal solar irradiance prediction. Firstly, the network structure, mathematical model, stability analysis, and training mechanism are studied for CESN. Secondly, according to the spatial correlation analysis, the number of ESN modules in CESN can be determined, such that the dynamics of different features can be fully fitted. In addition, the autocorrelation analysis is adopted in temporal information of each spatial variable to provide appropriate inputs for each ESN module. Finally, the spatio-temporal solar irradiance prediction model is established based on CESN. Simulation results illustrate that the CESN could achieve more accurate prediction, compared with backpropagation (BP) and Elman neural networks, classical ESNs.

本文提出了一种链结构回波状态网络（CESN），以提高时空太阳辐照度预测的可扩展性，鲁棒性和计算效率。首先，研究了CESN的网络结构，数学模型，稳定性分析和训练机制。其次，根据空间相关性分析，可以确定CESN中ESN模块的数量，从而可以充分拟合不同特征的动力学。另外，在每个空间变量的时间信息中采用自相关分析，以为每个ESN模块提供适当的输入。最后，基于CESN建立了时空太阳辐照度预测模型。仿真结果表明，CESN可以实现更准确的预测，