In this paper, a single user multiple input single output downlink wireless communication system is investigated, in which multiple reconfigurable intelligent surfaces (RISs) are deployed to improve the propagation condition. Our objective is to optimize the phase shift matrices of all the RISs by exploiting the statistical channel state information (CSI). In particular, two model-free algorithms are proposed, which are applicable for any channel statistical assumptions. Numerical results show that the proposed algorithms significantly outperform the random phase shift scheme, especially when the channel randomness is low.