Data processing in the diapir fuzzy zone of the Yinggehai Basin faces challenges, specifically manifested as a small effective offset for shallow seismic data, a low signal-to-noise ratio for deep seismic data, and significant spatial velocity variations. These factors render it challenging for purely data-driven tomographic inversion to effectively invert the formation velocity, resulting in unclear delineation of shallow faults and chaotic imaging of deep formations in seismic images.In response to the above issues, this paper proposes a velocity modeling method based on multi-information constraints. Firstly, a deep neural network is trained by integrating multiple aspects of information which includes geological horizons, seismic velocities, seismic attributes, the trend laws of velocity changes inside and outside the fuzzy zone, and original migration velocities. After that, we used the deep neural network to construct a seismic velocity model with multi-information constraints. Then, the velocity in the fuzzy zone is updated based on high-precision grid tomography velocity inversion technology, and finally, an optimized velocity model is obtained.The application of practical data demonstrates that this method not only enhances the consistency and accuracy of velocity and seismic imaging outcomes, but also markedly improves the imaging of shallow tomography and deep fuzzy regions. Consequently, this approach exhibits high practicality and effectiveness.