Machine Learning-Based Anomaly Detection and Analysis in Control Systems

Introduction:
The increasing complexity of control systems and the growing volume of data require advanced approaches for anomaly detection. Ensuring the reliability and cybersecurity of systems managing critical production processes is becoming increasingly important. Traditional statistical methods often struggle with complex and high-dimensional data, which motivates the use of machine learning techniques.

Methods:
This study investigates the application of machine learning methods for anomaly detection in control systems, with a focus on autoencoders. The model is trained on normal operational data and detects anomalies based on reconstruction error. Additional approaches, including statistical methods and machine learning algorithms, are considered for comparison. The proposed methodology also includes analysis of reconstruction errors to identify parameters contributing to anomalies.

Results:
The results demonstrate the effectiveness of machine learning methods, particularly autoencoders, in detecting anomalies in control systems. The proposed approach enables not only the identification of anomalies but also the determination of key parameters influencing their occurrence. The method shows improved adaptability when working with complex and high-dimensional data compared to traditional approaches.

Conclusions:
Machine learning methods provide a flexible and effective solution for anomaly detection in control systems. The developed approach improves system reliability and enhances cybersecurity by enabling both detection and interpretation of anomalies. Future work will focus on improving model interpretability and extending the approach for real-world industrial applications.