Anomaly detection in cyber-physical systems
AssigneesNEC Laboratories America, Inc., NEC Corporation
InventorsLuAn Tang, Wei Cheng, Haifeng Chen, Yuji Kobayashi
Life of this patent
Estimating the term…
Abstract
Methods and systems for training and deploying a neural network mode include training a modular encoder model using training data collected from heterogeneous system types. The modular encoder model includes layers of neural network blocks and a selectively enabled connections between neural network blocks of adjacent layers. Each neural network block includes neural network layers. The modular encoder model is deployed to a system corresponding to one of the heterogeneous system types.
Claims20
A method for training and deploying a neural network model, comprising:
- training a modular encoder model using training data collected from a plurality of heterogeneous system types, the modular encoder model including a plurality of layers of neural network blocks and a plurality of selectively enabled connections between neural network blocks of adjacent layers, wherein each neural network block includes a plurality of neural network layers;
- and
- deploying the modular encoder model to a system corresponding to one of the plurality of heterogeneous system types.
A method for performing anomaly detection, comprising:
- collecting local sensor data;
- encoding the local sensor data using a modular encoder model to generate an encoded representation, wherein the modular encoder model includes a plurality of layers of neural network blocks and selectively enabled connections between neural network blocks of adjacent layers, with the selectively enabled connections being controlled by a local system-specific policy model;
- decoding the encoded representation to generate reconstructed sensor data;
- generating an anomaly score based on a comparison between the local sensor data and the reconstructed sensor data;
- and
- performing a corrective action responsive to the anomaly score.
A system for performing anomaly detection, comprising:
- a hardware processor;
- and
- a memory that stores a computer program, which, when executed by the hardware processor, causes the hardware processor to:
- collect local sensor data;
- encode the local sensor data using to a modular encoder model to generate an encoded representation, wherein the modular encoder model includes a plurality of layers of neural network blocks and selectively enabled connections between neural network blocks of adjacent layers, with the selectively enabled connections being controlled by a local system-specific policy model;
- decode the encoded representation to generate reconstructed sensor data;
- generate an anomaly score based on a comparison between the local sensor data and the reconstructed sensor data;
- and
- perform a corrective action responsive to the anomaly score.
Dependencies are inferred from the claim text.