System and method for unsupervised anomaly detection
Some embodiments are associated with a system and method for deep learning unsupervised anomaly detection in Internet of Things (IoT) sensor networks or manufacturing execution systems. The system and method use an ensemble of a plurality of generative adversarial networks for anomaly detection.
1. A computer-implemented unsupervised anomaly detection (UAD) system, comprising:
a sever configured to receive real-time data from a plurality Internet of Things (IoT) sensors or manufacturing execution systems and convert real-time data to a data stream;
a processor configured to receive the data stream from the server and process the data stream using a UAD artificial intelligent (AI) core to generate an anomaly score, wherein the UAD AI core comprises an ensemble of a plurality of generative adversarial network (GAN) architectures; and
a web dashboard configured to present telemetry real-time data to a user and deliver the anomaly score alerting about potential operational anomaly in the IoT sensor networks or manufacturing execution system;
wherein the ensemble of a plurality of GAN architectures comprises long-short term memory (LSTM)-LSTM GAN, gated recurrent unit (GRU)-LSTM GAN, GRU-GRU GAN, LSTM-Multi-layer perceptron (MLP) GAN, and GRU-MLP GAN.
2. The system of claim 1 , wherein the system sets a threshold on a combined score of a discriminator score and a generator score of each GAN using Youden's statistics.
3. The system of claim 1 , wherein a length of a rolling window of a multivariate sequence for LSTM GAN or GRU GAN is equal to one.
4. The system of claim 1 , wherein the UAD system is configured to deliver alerts of the potential operational anomaly to the user's mobile app or email.
5. The system of claim 1 , wherein at least one of the plurality of GAN architectures comprises a long-short term memory (LSTM) GAN or gated recurrent unit (GRU) GAN.
6. A computer-implemented method for unsupervised anomaly detection (UAD), comprising:
receiving, via a sever, real-time data from a plurality Internet of Things (IoT) sensors or manufacturing execution systems;
converting, via the sever, real-time data to a data stream;
receiving, via a processor, the data stream from the server;
processing, via the processor, the data stream using a UAD artificial intelligent (AI) core to generate an anomaly score, wherein the UAD AI core comprises an ensemble of a plurality of generative adversarial network (GAN) architectures; and
presenting, via a web dashboard, telemetry real-time data to a user; and
delivering, via the web dashboard, the anomaly score alerting about potential operational anomaly in the IoT sensor networks or manufacturing execution system to the user;
wherein the ensemble of a plurality of GAN architectures comprises long-short term memory (LSTM)-LSTM GAN, gated recurrent unit (GRU)-LSTM GAN, GRU-GRU GAN, LSTM-Multi-layer perceptron (MLP) GAN, and GRU-MLP GAN.
7. The method of claim 6 , further comprising setting a threshold on a combined score of a discriminator score and a generator score of each GAN using Youden's statistics.
8. The method of claim 6 , wherein a length of a rolling window of a multivariate sequence for LSTM GAN or GRU GAN is equal to one.
9. The method of claim 6 , further comprising delivering alerts of the potential operational anomaly to the user's mobile app or email.
10. The method of claim 6 , wherein at least one of the plurality of GAN architectures comprises a long-short term memory (LSTM) GAN or gated recurrent unit (GRU) GAN.