中文摘要：

针对极限学习机易陷入过度学习和隐层节点数目难确定的问题,提出基于双层差分进化算法的极限学习机（双层DE-ELM）预测模型,把极限学习机（ELM）的隐层节点数目和节点参数作为差分进化（DE）算法的外层和内层进化个体,利用DE算法通过自然选择淘汰机制对其进行学习和完善。将该模型应用于上证综合指数和标准普尔500指数的短、中期预测,并与DE-ELM等模型的预测结果进行对比分析。实证结果表明：双层DE-ELM预测模型能够有效地选择隐层节点数目和参数,具有较强的预测能力和较高的稳定性。

英文摘要：

Aiming at the problems that neural networks are easy to fall into over learning and the number of nodes are difficult to determine,we establish a double-layer DE-ELM model. This model sets the number of nodes and the hidden layer node parameters of ELM as the individuals in outer and inner layer of DE algorithm,and then uses natural selection mechanism to select the best parameters. We make a short-term and medium-term prediction with Shanghai Composite Index and SP 500 Index by double-layer DE-ELM prediction model,and compare with DEELM,ELM and BP neural network model. The empirical results show that this prediction model can select the parameters of ELM model more effectively,and it has better prediction ability and better stability than comparison models.