Message ID Sequence GAN Training for UAV Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unmanned moving objects face challenges in symptom detection due to insufficient training data, leading to biased information and inadequate performance enhancement in deep learning models, which is exacerbated by the high cost of securing large datasets.
Innovation Solution
A generative adversarial neural network model is employed to generate message ID sequences for training an anomaly detection model, involving data preprocessing, generation of similar data through a first neural network, and training a second neural network to predict symptoms using a hierarchical structure with 1D convolutional, GRU, and dense layers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of training data is increased to enhance model performance, then the model performance is improved, but the cost for securing data increases
Solution Approach 1:
The patent uses a generative adversarial network (GAN) to create synthetic copies of training data. The generator network creates artificial message ID sequences that mimic real communication patterns, while the discriminator network distinguishes between real and synthetic data. This copying approach enables the system to expand the training dataset without collecting additional real data, thereby improving model performance while avoiding the high cost of securing more actual training data.
Solution Approach 2:
The patent employs synthetic data generated by the GAN as a cost-effective alternative to real training data. The synthetic message ID sequences are created through computational processes rather than requiring expensive data collection efforts. These disposable synthetic data points can be generated in large quantities at minimal cost, providing sufficient training material to improve model performance without the financial burden of securing extensive real datasets.
2Quantity of substance
If geometric conversion is applied to image data to increase data amount, then the amount of data is increased, but the method can only be applied to specific domain data
Solution Approach 1:
The patent applies the GAN copying principle to message ID sequence data rather than relying on geometric transformations. The generator network learns the statistical patterns and distributions of real message ID sequences and creates synthetic sequences that follow the same patterns. This approach is domain-agnostic and can be applied to any data type that has identifiable patterns, not just image data, thereby increasing both data quantity and adaptability across different domains.
Solution Approach 2:
The patent transforms the data through parameter changes in the GAN training process. The generator network adjusts parameters such as message ID sequences, timestamps, and communication protocols to create variations of the training data. This parameter transformation approach allows the system to generate diverse data samples from a single domain, expanding data quantity while maintaining domain-specific characteristics and improving versatility.
3Reliability
If additional actual data is secured to improve training, then the model performance is enhanced, but the cost and complexity increase
Solution Approach 1:
The patent uses the GAN copying mechanism to generate synthetic training data that replicates the characteristics of real communication data. The generator network is trained on a small initial dataset and then creates numerous synthetic message ID sequences that mimic real communication patterns. This copying approach provides a simple and efficient way to expand the training dataset without the complexity of implementing data collection systems, data cleaning pipelines, or data management infrastructure.
Solution Approach 2:
The system performs self-service by using the GAN to automatically generate training data from a small initial dataset. The generator network learns from the initial data and then autonomously creates synthetic data samples without requiring external data collection efforts. This self-service capability eliminates the need for complex data securing operations such as data acquisition, data validation, and data storage management, thereby reducing overall system complexity while improving model performance.
Data Source
AI summary
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.