基于人工神经网络的集成预报方法研究和比较

STUDY AND COMPARISON OF ENSEMBLE FORECASTING BASED ON ARTIFICIAL NEURAL NETWORK

中文关键词:  神经网络  预报集成  长期预报  比较分析

英文关键词:Neural network  Consensus for ecast  Long-term forecasting  Comparative analysis

中文摘要:

      用人工神经网络方法对同一预报量的各个子预报方程进行集成预报研究,并以同样的子预报方程进行回归、平均和加权预报集成。对神经网络集成预报模型与各个子预报方程及其它集成预报方法进行了对比分析研究。结果表明,人工神经网络方法所构造的集成预报模型不仅对历史样本的拟合精度比各个子预报方法及其它集成预报方法更好,独立样本的试验预报结果也显示出更好的预报准确性。并且,采用神经网络方法进行预报集成,可以避免以往集成预报方法难以确定权重系数的困难

英文摘要:

      In term so fanartificial neural net work(ANN), an ensemble forecasting for a number of submodels of the same predict and is established, and consensus forecast expressions of the regressing, average and weighted meanare formulated with the aid of the same submodels. Results show the ANN is superior in fittings and predictions compared to the submodels and other consensus forecast due to its self-adaptive learning and non-linear mapping. The ANN'sensemble fo recasting is easy application in such a way to ascertain weighting coefficient, thus providing a new-line for the research of prediction integrated on long-term forecasting of flood and drought.