Pre-stack seismic inversion is a high-precision method used for lithology and fluid identification in the exploration of reservoirs. However, the conventional inversion method is limited in the case of producing inversion results of high-frequency and being applied to reservoirs that have strong randomness of lateral changes. We proposed a multi-parameter similarity pre-stack seismic inversion method, in which a seismic waveform is employed to generate high-resolution results. The proposed method focuses on characterizing the spatial variability, which depends on seismic data and the reservoir structure. Firstly, we implement multi-parameter variable optimization based on the gather waveform similarity, Amplitude Variation with Offset (AVO) features, and spatial distance-extracted wells with structures that are similar to those of spatial estimation samples; secondly, we propose initial models of the gathers to be discriminated and use statistical elastic impedance as prior information; thirdly, we employ Markov Chain Monte Carlo (MCMC) stochastic simulation algorithm to implement the pre-stacked seismic waveform inversion; and finally, high-precision inversion results are obtained using the estimated elastic parameters. This method is applied to real seismic data acquired over shale oil reservoirs to verify the reliability, and we conclude that the proposed method is useful for producing reliable results that are used for identifying geological sweet spots.