Strengthening SCADA System Security through a Novel Intrusion Detection Method Using artificial intelligence Algorithm

Abstract

Background Intrusion detection in SCADA systems is essential for ensuring operational security and preventing unauthorized access. Traditional methods often face challenges related to high dimensionality and inefficiencies in accurately identifying threats.

Materials& Methods: This paper introduces a novel methodology that optimizes intrusion detection performance through a series of systematic steps, beginning with data preprocessing to eliminate redundant features, reducing the initial dataset from 35 to 17 features while maintaining critical information integrity. The approach utilizes Principal Component Analysis (PCA) to reduce dimensionality, transforming variables that are related into unrelated main components and retaining 99% of the original data variance. For classification, a Radial Basis Network (RBN) is employed, with parameters such as spread value and the number of hidden layer neurons carefully selected to enhance model performance. To avoid the problem of vanishing gradient, Emperor Penguin Optimization is applied to the training process. It should be noted that the population size and maximum iterations are carefully adjusted for an optimized training procedure. The obtained performance is 99.22% for accuracy, 99.31% for precision, 99.09% for recall, and 99.20% for F1 score.

Conclusion: These results confirm the model's appropriate performance for the detection of intrusions while avoiding any false alarms and validate it as a very efficient solution for strengthening security on SCADA systems.