The invention discloses a self-learning control system and method for a transverse movement process of a cutter suction dredger based on deep learning. The system comprises an off-line learning moduleand an on-line learning module, wherein the off-line learning module is used for constructing a virtual transverse movement process simulation environment by utilizing a deep neural network model according to data acquired by actual construction, and constructing a rough prior control model empirical database by adopting a depth deterministic strategy gradient model; and the on-line learning module is used for directly carrying out self-learning control on the real transverse movement process of the cutter suction dredger by utilizing the depth deterministic strategy gradient model, and perfecting and updating the prior control model empirical database constructed by the off-line learning module. According to the system and the method, the control of the transverse movement process is more intelligent, the operation process is simpler, the flexibility is good, and the rapid optimization can be realized.