The Integration Modeling of YOLOv8 Algorithm and Reinforcement Learning: Intelligent Tracking and Optimization of Target Objects in Power Systems

Abstract

The complexity and real-time nature of modern power systems pose serious challenges to traditional anomaly detection methods, especially with the increasingly intricate grid structure and the widespread integration of renewable energy sources. Traditional approaches often fail to capture rapid dynamic changes and abnormal events, creating an urgent need for more advanced, adaptive, and data-driven anomaly detection techniques to ensure system stability. This study addresses these challenges by introducing a novel hybrid deep learning framework, YGPPNet, which enhances anomaly detection accuracy and real-time response in next-generation smart grids. The YGPPNet model is built upon the YOLOv8 architecture and integrates three key components: YOLOv8, Generative Adversarial Networks (GANs), and Proximal Policy Optimization (PPO). YOLOv8 is responsible for real-time detection of target objects in power system environments, especially those associated with abnormal operational patterns. GANs are employed to model normal system behavior and generate more distinguishable representations for detecting deviations. Based on the abnormalities detected, PPO further optimizes decision-making and adaptive response strategies through reinforcement learning, thereby improving the efficiency and stability of the overall system.

Experimental evaluations conducted on multiple benchmark power system datasets demonstrate that YGPPNet surpasses traditional machine learning approaches and recent deep learning models in terms of sensitivity, detection accuracy, and reliability across diverse operational scenarios. The proposed framework offers a comprehensive end-to-end solution for anomaly detection and adaptive control, contributing a significant methodological advancement to intelligent power system monitoring. The YGPPNet model provides a strong foundation for improving the safety, sustainability, and reliability of future power systems and shows promising potential for real-world grid deployment.