Penetration depth acts as a crucial indicator reflecting laser welding quality, thus the control of its stability and the perception of its fluctuation state are increasingly garnering attention. This paper proposes a process information database-based control strategy for penetration depth, and the control validity is verified through penetration depth detection utilizing optical coherence tomography (OCT). The process information database stores diverse expected penetration depth knowledge formed by a substantial quantity of varying welding speeds with fixed other process parameters under undisturbed welding conditions. In the database, the stable average values inside the standard penetration depth information and the corresponding heat input (HI) values are connected and mapped via an artificial neural network (ANN). In response to abnormal variations in the penetration depth curve caused by interferences during welding, according to the HI gap predicted by the trained ANN from the penetration depth gap arising from the curve deviation, the control unit can calculate the new welding speed required to feed the penetration depth curve back to within the steady fluctuation range. Based on OCT, the keyhole depth signal is acquired, and a deep belief network is built to predict the penetration depth curve via the correlation between the reconstructed keyhole depth obtained by ensemble empirical mode decomposition and the penetration depth. This detection method demonstrates that the penetration depth curve can be controlled accurately. Finally, a closed-loop real-time feedback control system for penetration depth is established.