To solve the problems of low efficiency,poor accuracy and easy to fall into local optimal in road disease prediction,a multi-particle swarm zero sample neural architecture search method is proposed to automatically explore the optimal neural architecture for road disease prediction.Firstly,the multi-particle swarm strategy is used to initialize the high-quality architecture in the scale-adaptive search space.Then particle swarm dynamic adaptive update architecture is used to prevent local optimization.Finally,zero-sample learning,parameter and floating-point operation are combined for multi-objective optimization to achieve lightweight and improve prediction accuracy.The results show that:1)Scale adaptive search space effectively captures multi-scale road disease information;2)PSO dynamic adaptive updating effectively prevents the search process from falling into local optimization;3)Multi-objective optimization improves the classification accuracy,F1 score,Kappa coefficient,AUC,exponential balance and search efficiency by 19.34%,23.37%,23.77%,4.28%,20.26%and 91.30%,respectively.
维普
万方数据