A Q-learning guided search for developing a hybrid of mixed redundancy strategies to improve system reliability

Mixed redundancy strategies (MRSs) are based on leveraging a diverse combination of active and cold-standby components to improve system reliability. This study integrates three designed MRSs and employs them to develop a hybrid model to solve redundancy allocation problems (RAPs) and reliability-RAPs (RRAPs). The MRS best suited for a specific subsystem is difficult to determine. To resolve this issue, in the process of system reliability optimization, the state of limited resource utilization (e.g., cost, weight, and volume) in each sub-system facilitates the adoption of MRSs and is used as a learnable factor. To realize this learning process, Q-learning is used in this study to build a knowledge library (i.e., a Q-table) of MRS usage, where the Q-table guides the main optimization technique, the artificial bee colony algorithm (ABC), to expedite the convergence in searching for near-optimal solutions. For convenience, the Q-learning-guided ABC search method is abbreviated as QABCS, and the new MRSs obtained by QABCS are called Q-mixed. The experimental results show that Q-mixed not only improves system reliability but also reveals the preferences of each subsystem for the MRSs.

Automated summary

This study integrates three designed Mixed Redundancy Strategies (MRSs) to develop a hybrid model for solving redundancy allocation problems. Q-learning is used to build a knowledge library of MRS usage, which guides the artificial bee colony algorithm in searching for near-optimal solutions. The experimental results demonstrate that the proposed Q-mixed approach improves system reliability and reveals subsystem preferences for MRSs.