A Reinforcement Learning-Based Pantograph Control Strategy for Improving Current Collection Quality in High-Speed...

Area of Science:

• Railway engineering
• Control systems
• Artificial intelligence

Background:

• The pantograph-catenary system (PCS) is crucial for high-speed train power supply.
• Double-PCS (DPCS) operation leads to trailing pantograph (TP) contact force fluctuations, impacting power collection.
• Active pantograph control is key to mitigating these fluctuations.

Purpose of the Study:

• To propose a novel multiagent reinforcement learning (MARL) algorithm for active pantograph control.
• To address the coordinated control challenges in double pantograph systems (DPCS).
• To improve the quality of current collection in electric multiple units (EMUs).

Main Methods:

• Developed a cooperative proximity policy optimization (Coo-PPO) algorithm based on the Nash equilibrium framework.
• Implemented heterogeneous agents within a cooperative environment guided by a global value function.
• Introduced a novel reward propagation channel and curriculum learning approach.

Main Results:

• The Coo-PPO algorithm achieved higher rewards compared to existing methods.
• Demonstrated significant suppression of pantograph-catenary contact force (PCCF) fluctuations (up to 41.55%).
• Achieved a substantial decrease in the trailing pantograph's offline rate (up to 10.77%).

Conclusions:

• Coo-PPO is the first MARL-based approach for coordinated control of double pantographs in DPCS.
• The proposed method effectively reduces PCCF fluctuations and improves power collection quality.
• MARL technology offers a promising solution for enhancing high-speed railway power supply systems.