Generative adversarial network model and training method to generate message ID sequence on unmanned moving objects
Disclosed is a method for training an anomaly detection model of an unmanned moving object utilizing a message ID sequence, which includes: collecting packet data generated from the unmanned moving object; pre-processing the collected packet data; generating converted data for a message ID sequence of the packet data by inputting the preprocessed packet data into a language model; generating similar data similar to the message ID sequence of the packet data by inputting the preprocessed packet data into a first neural network model; training and evaluating the first neural network model by inputting the converted data of the language model and the similar data of the first neural network model into a training model; and predicting and training a symptom of the unmanned moving object by inputting the preprocessed packet data and the similar data of the first neural network model into a second neural network model.
1 . A method for training an anomaly detection model of an unmanned moving object, the method comprising:
collecting packet data generated from the unmanned moving object;
pre-processing the collected packet data;
generating a second message identification (“ID”) sequence having a message ID pattern corresponding to a first message ID sequence of the packet data by inputting the pre-processed packet data to a language model to produce converted data for the first message ID sequence;
generating a third message ID sequence having statistical characteristics corresponding to the first message ID sequence of the packet data by inputting the pre-processed packet data to a first neural network model to produce a fourth message ID sequence;
training and evaluating the first neural network model by inputting the converted data of the language model and the fourth message ID sequence of the first neural network model to a training model; and
training a second neural network model in a way of predicting a symptom of the unmanned moving object, the training comprising:
pre-processing packet data generated in a communication state of the unmanned moving object; and
training the second neural network model to predict a message ID following a current message ID sequence of the packet data by inputting the pre-processed packet data to the second neural network model.
2 . The method of claim 1 , wherein the pre-processing of the collected packet data includes:
extracting a message ID from the collected packet data;
assigning an integer value to the extracted message ID; and
normalizing the integer value assigned to the message ID.
3 . The method of claim 2 , wherein the assigning an integer value to the extracted message ID includes assigning a different integer value according to a type of the message ID.
4 . The method of claim 3 , wherein the normalizing the integer value assigned to the message ID includes normalizing the integer value assigned to the message ID to a range of 0 to 1 in order to scale a size of the message ID.
5 . The method of claim 1 , wherein the generating of the converted data includes dividing the first message ID sequence of the packet data into N units, and generating the converted data for the first message ID sequence by sequentially performing an N-gram calculation for the divided sequence.
6 . The method of claim 1 , wherein the language model includes a message ID N-gram model.
7 . The method of claim 1 , wherein the generating the fourth message ID sequence includes:
generating the third message ID sequence having statistical characteristics corresponding to the first message ID sequence of the packet data by inputting a random noise vector to a generator of the first neural network model:
training to discriminate between an actual message ID sequence and the third message ID sequence by inputting the actual message ID sequence of the pre-processed packet data and the generated third message ID sequence to a discriminator of the first neural network model; and
training the generator by inputting determination result to the generator of the first neural network model.
8 . The method of claim 1 , wherein the first neural network model includes a generative adversarial network (GAN) model.
9 . The method of claim 1 , wherein the training and evaluating the first neural network model includes:
evaluating the first neural network model by inputting the converted data of the language model and the fourth message ID sequence of the first neural network model in a training model; and
training the first neural network model based on evaluation result.
10 . The method of claim 7 , wherein the training the first neural network model includes training the discriminator and the generator of the first neural network model individually, and then training both the discriminator and the generator to adjust a training ratio.
11 . The method of claim 10 , wherein the training the first neural network model includes a first training process of training only the discriminator M times, wherein Mis a positive integer, a second training process of training only the generator N times, wherein N is a positive integer, and a third training process of training the discriminator and the generator O times, wherein O is a positive integer, and
orders and the numbers of times of the first training process, the second training process, and the third training processes are varied based on a loss function of the generator.
12 . The method of claim 9 , wherein the training the first neural network model includes:
training the discriminator of the first neural network model M times, wherein M is a positive integer;
training the generator of the first neural network model N times, wherein N is a positive integer; and
training the discriminator and the generator of the first neural network model O times, wherein O is a positive integer.
13 . A non-transitory computer readable medium storing a computer program, wherein the computer program performs the following method for training an anomaly detection model of an unmanned moving object when executed by one or more processors, the method comprising:
collecting packet data generated from the unmanned moving object;
pre-processing the collected packet data;
generating a second message identification (“ID”) sequence having a message ID pattern corresponding to a first message ID sequence of the packet data by inputting the pre-processed packet data to a language model to produce converted data for the first message ID sequence;
generating a third message ID sequence having statistical characteristics corresponding to the first message ID sequence of the packet data by inputting the pre-processed packet data to a first neural network model to produce a fourth message ID sequence;
training and evaluating the first neural network model by inputting the converted data of the language model and the fourth message ID sequence of the first neural network model to a training model; and
training a second neural network model in a way of predicting a symptom of the unmanned moving object, the training comprising:
pre-processing packet data generated in a communication state of the unmanned moving object; and
training the second neural network model to predict a message ID following a current message ID sequence of the packet data by inputting the pre-processed packet data to the second neural network model.
14 . A computing device for providing a method of training an anomaly detection model of an unmanned moving object, the computing device comprising a processor including one or more cores; and a memory, wherein the processor is configured to:
collect packet data generated from the unmanned moving object;
preprocess the collected packet data;
generate a second message identification (“ID”) sequence having a message ID pattern corresponding to a first message ID sequence of the packet data by inputting the pre-processed packet data to a language model to produce converted data for the first message ID sequence;
generate a third message ID sequence having statistical characteristics corresponding to the first message ID sequence of the packet data by inputting the pre-processed packet data to a first neural network model to produce a fourth message ID sequence;
train and evaluate the first neural network model by inputting the converted data of the language model and the fourth message ID sequence of the first neural network model to a training model; and
train a second neural network model in a way of predicting a symptom of the unmanned moving object, the training comprising:
pre-processing packet data generated in a communication state of the unmanned moving object; and
training the second neural network model to predict a message ID following a current message ID sequence of the packet data by inputting the pre-processed packet data to the second neural network model.