Time sequence missing value filling method based on bidirectional cyclic codec neural network

The invention provides a time sequence missing value filling method based on a bidirectional cyclic codec neural network. According to the method, an auto-encoder and a recurrent neural network are combined, and modeling of a time sequence containing missing values can be achieved; according to the method, the difference between a filling sequence and a label sequence is measured through two training losses, and an encoder and a decoder are reversely updated in an asynchronous mode; according to the method, the response of the gating unit amplification network to missing data is coordinated. According to the method, the problems that the space-time relationship of the time sequence containing the missing value cannot be correctly modeled by a common method and the filling effect is sensitive to the change of the missing rate are solved. 本发明提供一种基于双向循环编解码器神经网络的时间序列缺失值填充方法。该方法结合了自编码器、循环神经网络，可实现对含有缺失值的时间序列进行建模；该方法通过两种训练损失衡量填充序列和标签序列的差异，并通过异步的方式反向更新编码器和解码器；该方法通过协调门控单元放大网络对缺失数据的反应。本发明方法克服了一般方法无法正确建模含缺失值的时间序列的时空关系、填