International Journal of Science, Engineering and TechnologyAuthors: Anthony Vivian Onyinyechi, Umejuru Daniel, Vinani Nuka Precious
Abstract: In the present scenario of rapidly changing technology, the industrial control system (ICS) is the backbone of critical services like power generation, production units, and transport services. However, with the increasing interconnectivity of the components of the ICS, they are also increasingly exposed to various cyber-attacks, which may have varying effects from operational bugs to possible threats to public safety. The hybrid nature of the ICS network, along with the need for continuous real-time monitoring, creates a challenge in identifying possible threats to the ICS network. A cyber-attack on an industrial control system can lead to system unavailability, loss of production, economic loss, and even potential threats to public safety in extreme cases. This is a point of concern for a wide range of stakeholders who use ICS for their day-to-day business activities. Traditional security solutions are no match to the advanced nature of cyber-attacks, thus requiring the development of innovative solutions that can offer effective protection against cyber-attacks and unauthorized access. The proposed research work aims to make industrial control systems cyber-attack proof using the capabilities of artificial intelligence (AI) and deep learning (DL) models. The focus is on designing a Deep Learning-Based Intrusion Detection System (IDS) that can detect and mitigate port scanning and Distributed Denial of Service (DDoS) attacks in real-time on ICS networks, as the current IDS systems may offer some level of security but are not capable of dealing with the ever-changing nature of cyber-attacks. The proposed research work uses the Rapid Application Development (RAD) method, where the data from the ICS is collected and preprocessed to enable effective feature extraction and development of the model. The diagnostic parameters used in the proposed research work include the confusion matrix, accuracy, precision, recall, and F1-score. The proposed model was validated using a sample of the HAI 21.04 dataset and achieved an average accuracy of 98.58%, thus proving the effectiveness of the proposed model in detecting normal and abnormal patterns in the ICS data.
