Modeling of adversarial artificial intelligence in blind false data injection against AC state estimation in smart grid security, safety and reliability
Computer-implemented methods and systems for training an adversarial neural network to simulate a stealthy blind false data injection attack on a cyber physical system are provided. Supervised learning is used to generate an initial attack vector, by an adversarial attack generation model, based on inferred grid topology and historical measurements. A final attack vector is generated, by an adversarial verification model, based on a filtered subset of the initial attack vector utilizing a substitute bad data detection threshold, wherein the final attack vector enables creation of a counter measure.
1 . A computer-implemented method for training an adversarial neural network to simulate a stealthy blind false data injection attack on a cyber physical system, the method comprising:
using supervised learning to generate an initial attack vector, by an adversarial attack generation model, based on inferred grid topology and historical measurements, the adversarial attack generation model generating a mapping representing the inferred grid topology and the adversarial neural network being trained by the adversarial attack generation model, with a subset of historical measurements as input and all historical measurements as output;
determining, by an adversarial verification model, a substitute bad data detection threshold by obtaining pseudo-estimated measurements from the adversarial neural network utilizing the subset of the historical measurements as input;
generating a final attack vector, by the adversarial verification model, based on a filtered subset of the initial attack vector utilizing the substitute bad data detection threshold, the final attack vector enabling of a counter measure, the final attack vector including malicious data and the counter measure including storing final attack vectors in a security catalog for security purposes;
classifying, by the adversarial verification model, data in the initial attack vector into abnormal data and malicious data; and
generating and sending updated security parameters to the cyber physical system based on the counter measure.
2 . The method of claim 1 , wherein the historical measurements include one or more of voltage angle, magnitudes, active power flows, reactive power flows, and generated power.
3 . The method of claim 1 , further comprising selecting input data, by the adversarial attack generation model, for training the adversarial neural network, the selected input data including a subset of the historical measurements based on a state to be estimated for the cyber physical system.
4 . The method of claim 1 , further comprising selecting input data, by the adversarial attack generation model, for training the adversarial neural network, the selected input data including a subset of the historical measurements, wherein the subset includes voltage angles and magnitudes.
5 . The method of claim 1 , further comprising generating, by the adversarial attack generation model, intended state deviation using a random gaussian distribution.
6 . The method of claim 1 , further comprising using supervised learning, by the adversarial attack generation model, to generate the initial attack vector, wherein input data for training the adversarial neural network includes a subset of the historical data and an intended state deviation.
7 . The method of claim 1 , wherein the adversarial neural network comprises four layers.
8 . The method of claim 1 , wherein the adversarial neural network utilizes a Tan-Sigmoid activation function to infer grid topology.
9 . The method of claim 1 , wherein determining the substitute bad data detection comprises:
determining the filtered subset of the initial attack vector by filtering the initial attack vector utilizing the substitute bad data detection threshold.
10 . The method of claim 1 , wherein determining the substitute bad data detection threshold comprises obtaining pseudo-estimated measurements from the adversarial neural network utilizing a subset of the historical measurements as input.
11 . The method of claim 1 , wherein determining the substitute bad data detection threshold comprises obtaining an error vector determined by a difference between the historical measurements and the pseudo-estimated measurements.
12 . The method of claim 1 , wherein determining the substitute bad data detection threshold comprises setting a scalar factor for adjusting filtering strength.
13 . The method of claim 1 , further comprising filtering, by the adversarial verification model, the initial attack vector utilizing a substitute bad data detection threshold and a normalized residual vector.
14 . The method of claim 1 , wherein the counter measure is based on at least one of a signature detection mechanism, an anomaly detection mechanism, and an intrusion detection mechanism.
15 . A non-transitory computer readable medium or media containing instructions for executing a method for training an adversarial neural network to simulate a stealthy blind false data injection attack on a cyber physical system, the method comprising:
using supervised learning to generate an initial attack vector, by an adversarial attack generation model, based on inferred grid topology and historical measurements, the adversarial attack generation model generating a mapping representing the inferred grid topology and the adversarial neural network being trained by the adversarial attack generation model, with a subset of historical measurements as input and all historical measurements as output;
determining, by an adversarial verification model, a substitute bad data detection threshold by obtaining pseudo-estimated measurements from the adversarial neural network utilizing the subset of the historical measurements as input;
generating a final attack vector, by the adversarial verification model, based on a filtered subset of the initial attack vector utilizing the substitute bad data detection threshold, the final attack vector enabling creation of a counter measure, the final attack vector including malicious data and the counter measure including storing final attack vectors in a security cataloge for security purposes;
classifying, by the adversarial verification model, data in the initial attack vector into abnormal data and malicious data; and
generating and sending updated security parameters to the cyber physical system based on the counter measure.
16 . A system for training an adversarial neural network to simulate a stealthy blind false data injection attack on a cyber physical system, the system comprising:
a database connected to a network, configured for receiving and storing historical measurements and attack vectors; and
one or more processors and memory, the memory containing instructions executable by the one or more processors whereby the system is operative to:
use supervised learning to generate an initial attack vector, by an adversarial attack generation model, based on inferred grid topology and historical measurements, the adversarial attack generation model generating a mapping representing the inferred grid topology and the adversarial neural network being trained by the adversarial attack generation model, with a subset of historical measurements as input and all historical measurements as output;
determine, by the adversarial verification model, a substitute bad data detection threshold by obtaining pseudo-estimated measurements from the adversarial neural network utilizing the subset of the historical measurements as input; and
generate a final attack vector, by the adversarial verification model, based on a filtered subset of the initial attack vector utilizing the substitute bad data detection threshold, the final attack vector enabling creation of a counter measure, the final attack vector including malicious data and wherein the counter measure includes storing final attack vectors in a security catalog for security purposes;
classify, by the adversarial verification model, data in the initial attack vector into abnormal data and malicious data; and
generate and sending updated security parameters to the cyber physical system based on the counter measure.
17 . The system of claim 16 , wherein implementing the counter measure is based on at least one of a signature detection mechanism, an anomaly detection mechanism, and an intrusion detection mechanism.
18 . The system of claim 16 , further comprising generating and sending updates for the adversarial attack generation model.
19 . The system of claim 16 , further comprising generating and sending updates for the adversarial verification model.
20 . The system of claim 16 , further comprising training the adversarial neural network with the classified data.