Abstract: The seismic while driving technology is one of the most urgently needed geological guarantee technologies for the intelligent rapid driving development of roadways. Due to the large difference between the seismic while driving source and active sources such as explosives, conventional cognitive and data processing technologies have failed to achieve results. In view of the lack of cognitive of the source of seismic while driving and the lack of effective forward simulation methods for seismic while driving data, the research on the generation mechanism analysis and source time function construction of seismic while driving source were carried out. It is found that: the seismic while driving signal is a complex, variable frequency, continuous signal with a certain duration. Its relative low-frequency signal is mainly related to the mechanical rotation of the cutting head, the action of the gangue raking machine, the random coal falling, and the operation of the loader and belt conveyor, its high frequency and strong energy signals are determined by cutting coal, generally characterized by a pseudo random signal superimposed by multiple sources. Secondly, under the linear array data acquisition mode, the seismic while driving source can be approximated to a comprehensive virtual point source formed by the superposition of multiple sources that are continuously fired at different time delays. The seismic source is formed by the combined action of the vibration when the fully mechanized driving machine is cutting the coal seam and the coal body fracture, random coal falling, refuse raking, drilling, belt transportation, etc., in which the coal seam cutting plays a major role. Then, based on the analysis of source generation mechanism, the time function of the seismic while driving source is constructed, and the numerical model experiment of the seismic while driving is carried out with the time function as the source loading item. The simulated seismic while driving data obtained has the same time domain, frequency domain and time-frequency domain characteristics as the measured seismic while driving data, which show that the source time function constructed can effectively carry out forward modeling of seismic data while driving, and improve the cognition of the source while driving. In addition, in order to verify the effectiveness of the impulse algorithm based on spike deconvolution and cross-correlation in seismic while driving data processing, taking two common seismic interference technologies as comparison methods, effective signal extraction experiments were conducted using the simulation seismic while driving data. The experimental results show that: the impulse processing results based on spike deconvolution and cross-correlation are more similar to the wave field characteristics of seismic records obtained by using conventional Ricker wavelet simulation under the same simulation conditions, which verifies the effectiveness of the impulse algorithm. The research in this paper provides a basis for the further research and application of seismic while driving technology.