An unmanned intelligent inspection technology based on improved reinforcement learning algorithm for power large-area...

Area of Science:

• Robotics and Automation
• Artificial Intelligence
• Power Systems Engineering

Background:

• Efficient power system surveillance is crucial for grid reliability.
• Unmanned intelligent patrols offer a promising solution for large-area monitoring.
• Current patrol path planning methods face challenges in efficiency and safety.

Purpose of the Study:

• To develop an improved reinforcement learning algorithm for multi-scene unmanned intelligent patrol path planning in large power areas.
• To optimize patrol efficiency and ensure comprehensive surveillance coverage.
• To enhance the safety and reliability of power system inspections.

Main Methods:

• Utilized an improved Q-learning algorithm, a reinforcement learning subset, for path planning.
• Incorporated prior environmental knowledge as heuristic information to enhance Q-learning efficiency.
• Developed a novel reward function for collision-free traversal checking.
• Modeled the system with patrol UAVs, wireless charging, and multiple inspection targets.

Main Results:

• Generated patrol paths that ensure UAV safety and minimize travel distance.
• Achieved a coverage rate exceeding 98.5% for all designated patrol targets.
• Demonstrated enhanced search efficiency and optimized reinforcement learning performance.

Conclusions:

• The proposed multi-scene unmanned intelligent patrol technology effectively addresses power system surveillance needs.
• The improved reinforcement learning approach optimizes path planning for safety and efficiency.
• This technology significantly enhances the quality and coverage of power grid inspections.