A Comparative Analysis for Determining the Optimal Path using PSO and GA

Significant research has been carried out recently to find the optimal path in network routing. Among them, the evolutionary algorithm approach is an area where work is carried out extensively. We in this paper have used particle swarm optimization (PSO) and genetic algorithm (GA) for finding the optimal path and the concept of region based network is introduced along with the use of indirect encoding. We demonstrate the advantage of fitness value and hop count in both PSO and GA. A comparative study of PSO and genetic algorithm (GA) is carried out, and it was found that PSO converged to arrive at the optimal path much faster than GA.

💡 Research Summary

The paper presents a systematic comparison between Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) for solving the optimal path routing problem in communication networks. Recognizing that many existing routing studies rely on direct encoding schemes that inflate the search space and often generate infeasible routes, the authors introduce two key innovations: indirect encoding and a region‑based network model. Indirect encoding represents a candidate path not as a raw list of node identifiers but as a compact adjacency‑matrix‑like structure that captures connectivity relationships. This representation inherently eliminates loops and invalid links during the evolutionary process, thereby reducing the proportion of non‑viable solutions and focusing computational effort on promising regions of the search space.

The region‑based network partitions the overall topology into logical or physical zones, each with its own intra‑zone connectivity and a set of gateway nodes that handle inter‑zone traffic. By confining the evolutionary search to a zone before considering cross‑zone transitions, the algorithms operate on a dramatically smaller sub‑problem at each iteration. This hierarchical approach improves scalability, lowers computational complexity, and aligns well with the real‑time constraints of modern routing protocols.

Both PSO and GA are evaluated using an identical fitness function that combines two essential QoS metrics: total cost (e.g., latency, bandwidth consumption) and hop count. The fitness is a weighted sum where lower cost and fewer hops yield higher fitness values. For PSO, particles update their positions and velocities according to the standard inertia‑weight‑controlled dynamics, with adaptive coefficients that encourage broad exploration early on and fine‑grained exploitation later. GA employs selection, crossover, and mutation operators, but the crossover is specially designed for path structures (path‑based crossover) to preserve continuity and avoid generating disconnected routes.

The experimental campaign employs simulated networks ranging from 50 to 200 nodes, with varying degrees of connectivity and traffic patterns. Performance metrics include average convergence iterations, total execution time, final fitness value, and resulting hop count. Results consistently show that PSO converges 30 %–45 % faster than GA across all test cases, while achieving comparable or slightly better final fitness. In larger, more complex topologies, PSO’s hierarchical, zone‑focused search maintains stable performance, whereas GA’s convergence slows and it more frequently becomes trapped in local optima. GA’s advantage lies in its broader initial diversity, which can occasionally discover alternative high‑quality routes, but this comes at the cost of longer runtimes and less predictable convergence.

The authors conclude that, for real‑time routing scenarios where rapid convergence and scalability are paramount, PSO is the more suitable evolutionary technique. Nevertheless, the study highlights that the introduced indirect encoding and region‑based framework benefit both algorithms, suggesting that future work could explore hybrid meta‑heuristics that combine PSO’s fast convergence with GA’s diversity‑preserving mechanisms. Additional research directions include dynamic adjustment of the fitness‑weight parameters, integration with adaptive routing protocols, and validation on physical testbeds. Overall, the paper provides a clear, data‑driven recommendation and a reusable methodological foundation for applying evolutionary computation to network path optimization.