Abstract: The traditional offset optimization method is based on adjacent intersections while it ignores the internal correlation of offset among multiple continuous intersections. In order to solve this problem, several offsets between multiple continuous intersection are taken as research objects, and through neural network an optimization model is established to describe the relationship between offset and arterial vehicle delay, and a genetic algorithm is used to gain the optimal offset scheme. Firstly, according to the survey data the simulation environment is built for different vehicle delay data corresponding to multiple offsets, and based on this, a neural network is used to fit the relationship between offset and vehicle delay. Secondly, to get the optimal offset scheme based on the neural network, a genetic algorithm is introduced. Finally, a simulation experiment is carried out to prove the model efficiency, and the model is compared with Synchro. The result shows that the proposed model can effectively improve the signal scheme, and the arterial vehicle delay can be decreased by 22.27%.