Abstract: A historical path routing algorithm of opportunistic network is proposed on the basis of clustering algorithm in order to optimize the traditional historical path algorithm. The algorithm uses the k-means++ algorithm in unsupervised learning to encode the nodes and uses the coding method to update the historical path algorithm. It is characterized by low cache space occupation, high node search speed and strong adaptability in the environment with varied topology. The experimental results show that the RACA algorithm has better performance in many aspects, especially in terms of delivery ratio and overhead ratio. The good network performance enables RACA algorithm to be used in scenarios of limited resources and changeable network environment, for example, the in-vehicle network environments.