【摘要】 虽然时间序列预测问题已形成较为完整的理论体系,但是在复杂情况下的预测准确度和灵活性仍有提高的余地。针对复杂时间序列预测问题提出一种基于深度学习的集成模型。首先分别使用时钟驱动循环神经网络(CW-RNN)和向量自回归(VAR)模型进行预测,然后采用stacking的方式集成两者预测结果。实验结果表明,相比单一模型和传统模型,集成模型对时间序列的预测准确度和灵活性均有显著提高。更多还原

【Abstract】 Theoretical time series methods are already helpful for forecasting,but accuracy and flexibility of models are still can be improved. To make better forecasts for complex time series,a new deep learning based stacking algorithm is proposed. At first,it uses clockwork recurrent neural network(CW-RNN)and Vector Auto-regression(VAR)model to forecast independently.Then two estimated results are integrated by using stacking algorithm. The experimental results show that compared with single models and traditional methods,the improved stacking model has higher accuracy and more flexible.更多还原