基于粗糙集的GMDH神经网络的时间序列预测

Time series forecast of GMDH neural network based on rough set

针对随机时间序列的强不确定性和非线性特征,结合粗糙集理论和成组数据处理的神经网络技术建立了基于粗集的GMDH神经网络预测模型。同时就自然界大多数的随机时间序列数据维数较大的问题,为提高约简效率,提出了基于快速求核和集合近似质量的约简算法,并进行了仿真验证。结果表明,基于粗集的GMDH神经网络预测模型合理可行,约简算法快速有效。

In view of the random time sequence＇s strong uncertainty and the misalignment characteristic, GMDH neural network forecast model was established by combining rough set and group data handling. Meanwhile, most of the nature random time sequence has the problem of large data dimension, to improve the reduction efficiency. A reduction algorithm was proposed based on fast seeking nucleus and similar quality of sets, and simulation confirmation was given. The results indicate that the proposed GMDH neural network forecast model is reasonable and feasible, and the reduction algorithm is fast and efficient.