基于长短期记忆网络的时间序列预测研究

Time Series Prediction Based on Long Short-Term Memory Network

针对递归神经网络(RNN)模型难以训练和梯度消失等问题,引入长短期记忆网络算法(LSTM)。介绍了LSTM的基本原理,并将其应用于时间序列预测领域。以Wiener退化过程为例进行分析,针对传统预测方法无法兼顾退化数据的非线性及时序性特点,利用LSTM方法对Wiener退化过程时间序列进行预测。该预测算法与传统的预测算法进行了比较,研究结果表明,所构建的模型具有更高的预测模型精度,达到了预测要求。

In order to solve the gradient extinction of recurrent neural network(RNN), a long short-term memory network(LSTM) algorithm was proposed. Its basic principle was introduced and applied in the time series prediction. Taking the Wiener degradation process as an example, a new prediction algorithm based on LSTM is put forward in this paper, which solves the non-stationary feature of degraded data. Results show that the LSTM algorithm is effective compared with the traditional method.