摘要
结合Kalman滤波与回声状态网络,将在线回声状态网络算法应用于变形数据预测。回声状态网络的输出权值通过Kalman滤波训练,直接对网络的输出权值进行在线更新,克服了传统递归网络需要收集大量样本后才能进行拟合预测的缺陷,同时也保证了预测精度。实例计算验证了该方法的有效性。
A new kind of on-line predictor is constructed by combining Kalman filtering with the echo state network. The method of Kalman filtering is applied to the echo state network output weights training, directly on-line updating the network output weights, overcoming the defects in traditional recurrent neural network(RNN) which is needed to collect a large number of samples. The examples demonstrate the effectiveness of the proposed method.
出处
《大地测量与地球动力学》
CSCD
北大核心
2016年第7期617-619,629,共4页
Journal of Geodesy and Geodynamics
基金
国家自然科学基金(61572015)~~