The present invention describes a system and method for selecting and optimizing a sound stimulus using a deep neural network to regulate and improve human sleep quality. The deep neural network has the capability of characterizing processing of human brain cortical neurons for external stimulus (images, sounds, etc.) information. By inputting massive sound stimuli into the deep neural network, a sound mode which causes model-estimated sleep electroencephalograph to be optimal can be found, the sound mode is applied to a real human body, and the intensity of corresponding sleep waves of the human body in different sleep stages is enhanced through closed-loop optimization so as to realize the purpose of regulating sleep. The present invention mainly aims at solving the technical problem of how to select and optimize, when a sound stimulus means (music, speech, natural sounds, white/colored noise, etc.) is used to assist in human sleep.