基于动态长短期记忆网络的设备性能退化预测方法

针对目前基于数据驱动的旋转机械退化状态预测中时序列信息考虑不充分、寿命标签制定不合理、退化模型累计误差大等问题，提出一种融合趋势滤波、模糊信息粒化、动态长短期记忆网络（LSTM）的旋转机械退化趋势与退化区间预测方法。以振动信号为例，首先提取表达设备退化信息的特征指标，然后通过趋势滤波与模糊信息粒化提取主要退化趋势与模糊退化边界，其次利用动态LSTM进行综合性能退化预测；最后，利用网络公开的轴承训练数据集验证了本文方法的可行性与有效性。

An equipment performance degradation prediction method based on a dynamic long-short-term memory network

In an attempt to tackle associated with the problems current data-driven degradation predictions for rotating machinery-such as insufficient consideration of time series information, unreasonable life labeling, and large cumulative error of degradation models-a method involving fusion trend filtering, fuzzy information granulation, and a dynamic long-short-term memory network (LSTM) has been proposed for predicting the degradation trends and degradation intervals of rotating machinery. Taking the vibration signal as an example, the characteristic index of the degradation information of the equipment is first extracted, and then the main degradation trend and the fuzzy degradation boundaries are extracted through trend filtering and fuzzy information granulation, and finally the comprehensive performance degradation is predicted using dynamic LSTM. The feasibility and effectiveness of the method were verified using a bearing training data set published on the internet.