Multi-objective UAV trajectory planning algorithm based on the hybrid MS-INSGA2 strategy

To address the critical challenges in 3D UAV trajectory planning—specifically the severe tendency to fall into local optima and the inherent dilemma of balancing convergence speed with population diversity in multi-objective optimization—this paper proposes a unified optimization framework integrating multiple mechanisms, namely the Multi-Strategy Improved Non-dominated Sorting Genetic Algorithm II (MS-INSGA2). This framework is founded on a linear interpolation mechanism with safety verification, effectively guaranteeing the safety of initial solutions and the quality of exploration starting points. During the evolutionary process, dynamic guidance and adaptive crossover and mutation mechanisms are deeply coupled, enabling real-time adaptive parameter adjustments based on the population status. This successfully breaks the deadlock between convergence and diversity. Subsequently, a local search strategy is introduced to meticulously refine the Pareto front, comprehensively enhancing trajectory smoothness and overall quality. Comprehensive evaluations on the ZDT, DTLZ, and WFG benchmark suites, supported by the Wilcoxon signed-rank test, confirm the superior underlying optimization capabilities of the proposed framework. Furthermore, in two complex 3D trajectory planning tasks, MS-INSGA2 demonstrates overwhelming superiority over 11 state-of-the-art algorithms, including MOBA, MOGOA, MOGWO, IMOABC, MOSSA, MOFA, MOEA-2DE, MPSOD, NSGA2, INSGA2, and DN-NSGA2. Specifically, it significantly reduces the Average Fitness by 27.87% and 32.06%, and decreases the Best Fitness by 6.91% and 14.18%, respectively. The research findings indicate that this unified framework not only substantially shortens the flight distance but also drastically mitigates the spatial collision risk, providing robust algorithmic support for highly reliable replanning in complex airspaces.