运用人工神经网络的防空系统威胁评估模型_21页_1mb

Summary of the Document: Threat Assessment Model in Air Defense Systems Using Artificial Neural Networks

Core Content

This document presents a novel threat assessment model for air defense systems that leverages artificial neural networks (ANNs) to improve decision-making processes. The model integrates missing data completion, multi-criteria analysis, and ANN-based learning to dynamically update threat scores in real-time, adapting to changing conditions in the battlefield.

The study emphasizes the importance of automating threat assessment and target assignment in combat environments, where traditional methods often fail to provide the necessary speed and accuracy. The proposed model is designed to be dynamic, multi-criteria, and resilient to incomplete data, making it a significant improvement over existing literature.

Main Points

• Objective: Automate threat assessment and target assignment in air defense systems using a dynamic AI-based model.
• Key Innovations:
  • Integration of missing data completion.
  • Use of multi-criteria decision-making with 26 criteria.
  • Introduction of the Combined Geometric Threat Score (CGTS) to provide a balanced and verifiable threat assessment.
• Data Sources:
  • Data was collected from 56 studies and expert opinions.
  • A total of 223 target situations were analyzed, with 5,798 data points compiled.
• Data Processing:
  • Criteria values were standardized and normalized.
  • Missing data was filled using nearest cluster-based estimation and statistical methods (average, median).
• Model Structure:
  • The model consists of three layers: input, hidden, and output.
  • The input layer includes normalized values of 26 threat criteria.
  • The output layer produces the CGTS values.
  • The hidden layer uses a variable number of neurons and the ReLU activation function.
• Training Process:
  • The Levenberg-Marquardt backpropagation algorithm was used for training.
  • This is a hybrid algorithm that combines least squares, gradient descent, and Gauss-Newton methods.
  • The model's performance was evaluated using MSE and R-squared values, with MSE ranging from 0.0005 to 0.0072 and R above 95%.
• Model Evaluation:
  • The CGTS was designed to reduce the bias caused by different sources of threat data.
  • The Mean Absolute Error (MAE) between the original threat scores from the literature and the model's calculated scores was 0.039, indicating high reliability.
  • The model shows good adaptability to varying threat conditions and provides a more accurate and consistent assessment than traditional static models.

Key Information

• Number of Criteria: 26, selected based on frequency of use, importance, and data availability.
• Data Points: 5,798 total, with 1,552 available and 4,246 imputed.
• Performance Metrics:
  • MSE: 0.0005–0.0072
  • R-squared: Above 95%
  • MAE: 0.039
• Training and Testing:
  • The model was trained on a comprehensive dataset that includes both real and imputed data.
  • It is designed to respond adaptively to new and evolving threats.
• Comparative Analysis:
  • Most existing studies use fewer criteria and are static.
  • This model is more scalable and versatile, especially for network-centric air defense systems.
• Geographical and National Perspectives:
  • Different countries have varying priorities in threat assessment.
  • The USA and China focus on big data and AI, while Türkiye and Israel emphasize operational speed and precision.
  • Germany and Sweden prioritize modular systems and scenario-based training.
  • China also develops countermeasures against cyberattacks and uses quantum radar.
• Limitations:
  • The lack of a universal model that can integrate all methods and adapt quickly to new threats remains a challenge.
  • Some studies face issues such as high costs, limited data, and expert knowledge dependency.

Conclusion

The study concludes that the proposed AI-based threat assessment model significantly improves decision-making speed and accuracy in air defense systems. It reduces human influence, enhances system effectiveness, and provides a scalable and adaptable framework for threat prioritization. The CGTS and ANN-based approach offer a more robust and dynamic solution, capable of handling incomplete data and providing verifiable results. The model's high performance in terms of MSE and R-squared validates its reliability and accuracy in real-world applications.