IMPROVING PREDICTION OF OPTIMAL TIME FOR INTERFERENCE NORMALIZATION OFFICE NETWORK SERVICES AND SCADA BY USING ARTIFICIAL INTELLIGENCE (AI)

• Main Author: Suherianto Sinaga, Eidelbert
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86680
• Institution: Institut Teknologi Bandung

Computer network services are a crucial part in supporting the reliable business processes of the electrical system. Operational office and service processes at PT PLN (Persero) such as billing management, sales, human resources, procurement, disturbance services, and remote devices are currently mostly accessed using Web or Desktop applications that require an Office network connection (Intranet/Internet) as well as SCADA (Supervisory Control and Data Acquisition) networks. Therefore, if network disturbances occur, they must be addressed immediately to prevent widespread impacts on business processes and services. Currently, targeted resolution times for disturbances must not exceed the predetermined performance targets, which are 1,85 hours per month for Office disturbances and 2,16 hours per month for non-redundant SCADA disturbances. However, when network disturbances occur, the ongoing recovery estimation lacks a standard or firm basis, making current solutions ineffective in providing accurate recovery time estimates. Accurate and efficient prediction of optimal recovery times for Office and SCADA network service disturbances is vital for operational continuity. Furthermore, teams working to resolve disturbances in the field need to have time targets based on the history of past network disturbance resolutions. If disturbances are resolved before the predetermined targets, these can serve as benchmarks for network disturbance recovery. However, if recovery exceeds the set targets, strategies need to be reassessed to prevent network service recovery from exceeding the allocated time. Addressing these issues, researchers aim to apply a method that can improve prediction accuracy in determining the optimal recovery time class for Office and SCADA network service disturbances. While several methods exist for determining fault duration, few studies utilize Machine Learning (ML) methods to determine the fault recovery duration class (Joyokusumo et al., 2020). Machine Learning can be described as a method to enable programs to learn from experience and improve task execution through gaining more experience (Ray, 2019). In the conceptual framework of Artificial Intelligence (AI), Machine Learning serves as models that can be implemented as pre-trained models within intelligent agents without the capability to learn additional insights from the environment (Kühl et al., 2019). Based on these methods, the researchers use several Machine Learning-based methods, namely Random Forest Classifier and XGBoost (Extreme Gradient Boosting), to predict the fault recovery duration class for network services. Comparisons were also made between the estimation using Naïve Bayes Classifier and Support Vector Machine. The dataset used in this research comes from SCADA and Office network disturbance data from January 2021 to October 2024 and currently consists of 2591 disturbance records. The variables (input features) used in this research include 5 variables: SID, Stop Clock (Duration), Ticket Open, Action, Interference Detail, and one target variable, Fault Duration. The benefits obtained from predicting the fault recovery duration class include contributing to the development of more effective and efficient disturbance recovery management strategies when the estimated recovery time is known. Additionally, disturbance recovery strategies can be reassessed to minimize the risk of prolonged recovery times. With more accurate predictions, recovery teams can work more efficiently and accurately, ensuring that disturbances do not significantly impact the company's business operations. Machine Learning technology in predicting fault recovery duration classes is expected to maintain the continuity of critical services for companies like PT PLN (Persero). Thus, the developed prediction model can become a vital tool in network disturbance recovery management, providing necessary information for better decision-making and more effective handling strategies.