Three-dimensional UAV path planning in complex environments poses substantial challenges owing to high-dimensional search spaces, nonlinear operational constraints, and multimodal solution landscapes. Although particle swarm optimization (PSO) has demonstrated considerable promise for this task, conventional PSO variants exploit only a limited subset of the historical positional data generated during the search process, thereby constraining their overall optimization capability. To overcome this limitation, this paper proposes a neural-guided multi-strategy particle swarm optimization algorithm (NGMS-PSO), which systematically mines evolutionary process data to enhance algorithmic search performance. Specifically, a neural network is trained on collected evolutionary data to construct a neural-fusion evolutionary operator that provides knowledge-driven directional guidance throughout the search. An evolutionary state indicator is further introduced to stratify the swarm into elite, normal, and exploratory subpopulations, each governed by tailored evolutionary strategies that collectively maintain a dynamic balance between exploitation and exploration. To mitigate premature convergence, an adaptive escape mechanism is additionally incorporated, which applies targeted perturbations to particles exhibiting evolutionary stagnation. Extensive comparative experiments conducted across test scenarios of varying complexity against 11 competitive algorithms demonstrate that NGMS-PSO consistently generates safe, smooth, and cost-efficient UAV trajectories, achieving statistically significant performance improvements in the majority of cases.
