本文为英国皇家理工学院（作者：AKASH SINGH）的硕士论文，共61页。

我们探讨了长短期记忆（LSTM）在时间数据异常检测中的应用。由于获取标记异常数据集的困难，因此采用了无监督的方法。我们用LSTM单元训练递归神经网络（RNN），以学习正常时间序列模式并预测未来值。通过对预测误差进行建模，给出异常值评分。我们研究了维持LSTM状态的不同方法，以及使用固定时间步长对LSTM预测和检测性能的影响。还将LSTM与具有固定输入时间窗的前馈神经网络进行了比较。我们对三个真实数据集的实验表明，尽管LSTM RNN适用于通用时间序列建模和异常检测，但维持LSTM状态对于获得预期结果至关重要。此外，对于简单的时间序列，可能根本不需要LSTM。

We explore the use of Long short-termmemory (LSTM) for anomaly detection in temporal data. Due to the challenges inobtaining labeled anomaly datasets, an unsupervised approach is employed. Wetrain recurrent neural networks (RNNs) with LSTM units to learn the normal timeseries patterns and predict future values. The resulting prediction errors aremodeled to give anomaly scores. We investigate different ways of maintainingLSTM state, and the effect of using a fixed number of time steps on LSTMprediction and detection performance. LSTMs are also compared to feed-forwardneural networks with fixed size time windows over inputs. Our experiments, withthree real-world datasets, show that while LSTM RNNs are suitable for generalpurpose time series modeling and anomaly detection, maintaining LSTM state iscrucial for getting desired results. Moreover, LSTMs may not be required at allfor simple time series.

1 引言

2 相关理论

3 方法与数据集

4 实验与结果

5 讨论

6 结论

更多精彩文章请关注公众号：