CatBoost-Enhanced Convolutional Neural Network Framework with Explainable Artificial Intelligence for Smart-Grid Stability Forecasting

Stephanie Ness^1,2*

• ¹University of Vienna, Vienna, Austria
• ²Diplomatic Academy of Vienna, Vienna, Vienna, Austria

Mobile robots, such as drones, rovers, and autonomous ground units, are becoming more crucial for inspecting, monitoring, and maintaining smart grid infrastructure. These robotic agents are sent to assess electricity lines, substations, offshore wind farms, and other critical components in risky or remote places after severe weather occurrences or system breakdowns. This study proposes a hybrid diagnostic technique designed to optimize robotic operations by providing accurate, interpretable projections of smart grid stability. To employ structured sensor and spatial data simultaneously, the model integrates Deep Convolutional Neural Networks (Deep CNNs) with CatBoost gradientboosted decision trees. The system, which includes Local Interpretable Model-Agnostic Explanations (LIME), is critical for controlling autonomous robot behavior since it ensures openness and trust in predictions. This CatBoost + DeepCNN architecture outperforms conventional neural network models on test datasets, scoring 98.23%. This backend artificial intelligence system may guide robotic deployment plans, enabling mobile platforms to prioritize at-risk sites, adjust mission goals in response to projected grid instability, and assess their own sensor data in connection to system-wide energy patterns. Our results demonstrate the model's considerable potential for enabling intelligent, resilient, and real-time robotic responses inside modern energy networks.