PREDICTION OF LIGHTNING STRIKES AND ANALYSISOFLIGHTNING DISRUPTIONS ON POWER TRANSMISSIONLINESUSING A MACHINE LEARNING

• Main Author: Adiaksa, Aditya
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86684
• Institution: Institut Teknologi Bandung

Frequent lightning strikes in Sumatra pose significant challenges for the powertransmission network of PT PLN UIP3B Sumatra, causing transmission disruptionsand substantial economic losses. This study aimLightning strikes are a major causeof disturbances in electrical transmission systems, particularly in the operational areaof PT PLN UIP3B Sumatra. These disturbances often result in widespread power outages, damage to infrastructure such as towers, cables, and transformers, and significant economic losses. Moreover, lightning-induced outages reduce the reliability of powersystems, disrupting industrial, commercial, and residential activities. Therefore, aproactive and data-driven mitigation approach is necessary to ensure systemreliability.This study aims to develop an integrated predictive framework that combinesforecasting methods to predict the number of lightning strikes with classificationmethods to assess the risk of lightning-induced disturbances. In the forecasting phase, the Transformer model was employed for its superior ability to capture complextemporal patterns in time-series data from 2018 to 2024. Evaluation resultsdemonstrate that the Transformer model outperformed traditional models suchasRecurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM). WithanR- squared (R²) value of 0.9543, a Mean Squared Error (MSE) of 0.3957, and aRoot Mean Squared Error (RMSE) of 0.6291, the Transformer model provided highlyaccurate predictions even for volatile data patterns. The forecasting results wereintegrated into the classification phase as key features to detect disturbance risks.Intheclassification phase, XGBoost with SMOTE (Synthetic Minority OversamplingTechnique) was applied to address the significant class imbalance between disturbance(1) and non-disturbance (0) cases. Data imbalance often leads to models being biasedtoward the majority class, making it dif icult to detect minority cases. By using SMOTE, XGBoost ef ectively learned patterns from both classes more equitably. The evaluationrevealed that this method significantly outperformed other models such as Naive Bayes, Support Vector Machine (SVM), and Random Forest. XGBoost with SMOTEachieveda Test Accuracy of 89.15%, a Precision of 88.98%, a Recall of 89.16%, and anF1- Score of 89.05%. The model not only provided binary predictions (disturbance or nodisturbance) but also probabilistic risk assessments for each tower, categorizedintothree risk levels: Safe (probability < 50%), Alert (probability 50–80%), and Critical (probability ? 80%). Additionally, correlation analysis showed that key features suchas Count (+), Count (-), and Density had significant relationships with disturbancerisks, making them primary indicators in risk management. Supporting features likeTower Height + Elevation and Grounding Resistance also contributed significantly The negative correlation between Grounding Resistance and disturbance probabilityemphasized that better grounding quality reduces lightning-induced risks.This researchmakes a significant contribution to managing lightning disturbances in electrical transmission systems. By integrating lightning strike forecasting using the Transformermodel and disturbance classification using XGBoost, this study provides acomprehensive data-driven solution. The approach allows PLN to prioritize mitigationresources for towers in the Alert and Critical categories, enhancing the reliabilityof the transmission system. The probability-based framework supports more proactive andef icient mitigation planning, making power systems more resilient to lightningdisturbances. This study of ers a substantial contribution to data-driven mitigationmethods in transmission systems and highlights the transformative potential of machinelearning techniques in improving power system reliability.