Article: Integrated power information operation and maintenance system based on D3QN algorithm with experience replay Journal: International Journal of Computational Science and Engineering (IJCSE) 2024 Vol.27 No.4 pp.377 - 384 Abstract: Due to the expanding scale of power system and the variety of equipment, the operation and maintenance tasks require more efficient scheduling algorithms to accommodate large-scale data. To solve the problems of delays and interruptions in scheduling, an integrated operation and maintenance task scheduling algorithm based on D3QN with experience replay is designed. The proposed method performs Markov modelling of the operation and maintenance scheduling scenario, which takes the work order queue and the current total amount of resources as the state, selects tasks in each discrete time period as the action, and minimises the average waiting time as the goal. Considering the different levels of urgency of faults, the reward function is designed according to the fault level. In addition, the five-tuple attributes of work orders and their constraints for different tasks such as execution time, maximum load and resource requirements are also taken into account. Finally, the action with the highest Q value in the current state is always selected as the optimal policy of the operation and maintenance task. Experiments on several tasks demonstrated that the proposed method significantly improves the efficiency of overhaul tasks. Inderscience Publishers - linking academia, business and industry through research

Title: Integrated power information operation and maintenance system based on D3QN algorithm with experience replay

Authors: Yu Yang; Heting Li; Dongsheng Jing

Addresses: State Grid Suzhou Power Supply Company, Suzhou Jiangsu 215004, China ' State Grid Suzhou Power Supply Company, Suzhou Jiangsu 215004, China ' State Grid Suzhou Power Supply Company, Suzhou Jiangsu 215004, China

Abstract: Due to the expanding scale of power system and the variety of equipment, the operation and maintenance tasks require more efficient scheduling algorithms to accommodate large-scale data. To solve the problems of delays and interruptions in scheduling, an integrated operation and maintenance task scheduling algorithm based on D3QN with experience replay is designed. The proposed method performs Markov modelling of the operation and maintenance scheduling scenario, which takes the work order queue and the current total amount of resources as the state, selects tasks in each discrete time period as the action, and minimises the average waiting time as the goal. Considering the different levels of urgency of faults, the reward function is designed according to the fault level. In addition, the five-tuple attributes of work orders and their constraints for different tasks such as execution time, maximum load and resource requirements are also taken into account. Finally, the action with the highest Q value in the current state is always selected as the optimal policy of the operation and maintenance task. Experiments on several tasks demonstrated that the proposed method significantly improves the efficiency of overhaul tasks.

Keywords: orthogonal architecture; reinforcement learning; maintenance; experience replay; integrated operation and maintenance.

DOI: 10.1504/IJCSE.2024.139685

International Journal of Computational Science and Engineering, 2024 Vol.27 No.4, pp.377 - 384

Received: 31 Jan 2023
Accepted: 06 May 2023
Published online: 05 Jul 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Integrated power information operation and maintenance system based on D3QN algorithm with experience replay

Keep up-to-date