Proceedings of the

The 33rd European Safety and Reliability Conference (ESREL 2023)
3 – 8 September 2023, Southampton, UK

Deep Reinforcement Learning for Space Power Source Regulation

Tingyu Zhanga, Ying Zengb, Yan-Feng Lic, Xin Huangd and Hong-Zhong Huange

School of Mechanical and Electrical Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan, China.


As one of the key subsystems of space equipment, the main task of space power supply system is to ensure that it can provide continuous and stable electric energy during orbital operation as well as the bus regulation function of power supply system. For the space power supply control system represented by the S4R type, this paper proposes a dynamic analysis model of network cascade fault based on multiple charge-discharge adjustment tests which on the basis of using complex network theory to evaluate the structural reliability of the main error amplification system. Furthermore, it models the actual operation characteristics such as photovoltaic conversion and power regulation in the dynamic analysis of cascading faults, and analyzes the impact of the real-time change process on the overall reliability of the system. In this research, the regulation problem of the space power supply system is modeled as a Markov decision process model, and a power regulation algorithm based on deep reinforcement learning is further proposed to achieve intelligent monitoring and diagnosis of power supply network faults through different ways of bus regulation and filtering technology, so as to reduce the overall cascading fault risk of the space power supply distribution and supply network, while maintaining a reasonable utilization rate of stored power.

Keywords: Space power system, Complex network theory, Power control, Deep reinforcement learning.

Download PDF