Dynamic Fault-Tolerant three-dimensional cellular genetic algorithms

This paper proposes a new dynamic and algorithm-based approach to achieve fault tolerance using 3D cellular genetic algorithms (Dynamic Fault-Tolerant 3D-cGA). The proposed algorithm is an improved version of our previous algorithm (Fault-Tolerant 3D-cGA) that introduces and utilizes a dynamic adaptation feature to achieve further improvement. In Dynamic Fault-Tolerant 3D-cGA, faulty individuals are isolated and the maximum number of fitness evaluations is recalculated to adapt to faults encountered. To improve the performance of the algorithm, a mitigation technique is integrated into our algorithm by introducing an explicit migration operator. A benchmark of well-known real-world and test problems is used to test the effectiveness of the algorithm in order to investigate the influence of adaptation schemes and migration on algorithm performance. Faulty critical system data is tackled in conjunction with various fault ratios. To illustrate the improvement achieved, Dynamic Fault-Tolerant 3D-cGA is compared with Fault-Tolerant 3D-cGA, the previously proposed algorithm. The overall results demonstrate the ability of Dynamic Fault-Tolerant 3D-cGA to maintain system's functionality despite an increasing number of faults with up to 40% of processing elements (PEs), and clearly illustrate the importance of migration. Significant improvements in the performance of the algorithm, measured as efficiency, efficacy, and speed, are achieved, especially when migration is employed.