Abstract: As one of the main sources of potash fertilizer, polyhalite resource is in huge demand. Polyhalite deposits are well developed in the Triassic marine beds of Northeast Sichuan. However, these polyhalite deposits are very difficult to identify since they are deeply buried and scattered. To cut down the cost, seismic data are extremely useful in identifying well locations in pre-drilling exploration. Considering the complex relation between polyhalite deposits and seismic reflection in this area, it is impossible to directly predict potassium-rich salt layers solely by traditional impedance inversion method, herein an artificial intelligence (AI) approach is introduced. First, sensitive attributes suitable for characterizing potassium-rich layers are extracted using a fuzzy rough optimization algorithm. Then, on the basis of nonlinear mapping of the extracted sensitive seismic attributes onto the potassium-content well log, a polyhalite layer prediction method is established by using extreme learning machine. Using this method the polyhalite distribution in Northeast Sichuan was successfully identified. The AI-based seismic inversion methodology provides an alternative approach for the exploration of polyhalite layers in this area.

Key words: polyhalite deposits, artificial intelligence, extreme learning machine, sensitive attributes, potassium⁃rich layer