Aiming at the problem that the number of clustering classes and fuzziness index of the traditional fuzzy C-means（ FCM） algorithm are determined by prior knowledge so affect the precision of clustering results

an improved FCM method for repairing traffic flow missing data is proposed. Firstly

according to the fuzzy decision theory

the FCM objective function and the partition entropy are used together to determine the optimal fuzziness index. Secondly

the simulated annealing algorithm is used to optimize the number of FCM clusters. Finally

the spatial-temporal correlation of traffic flow data is combined to improve the repair effect. The highway traffic data provided by the California Transportation Bureau’s Operation Monitoring System is taken as the research object

and the experimental results show that the method overcomes the problem of insufficient global search ability of the traditional FCM algorithm. Compared with the traditional FCM algorithm

the root mean square error is reduced by about 30 %.