Abstract: Due to factors such as cost,site conditions,construction period and weather,the sampling of crude oil for PVT （pressure,volume and temperature） in offshore oilfields is relatively limited. The conventional empirical formula method for predicting formation crude oil viscosity considers limited factors,and the error is relatively large. To address these issues,a new method is proposed：a BP neural network-based prediction model for formation crude oil viscosity is established by comprehensively considering parameters such as sampling depth,formation temperature,formation pressure,gas-oil ratio,surface crude oil density,and surface crude oil viscosity of Oilfield P as input parameters. The research results show that the model exhibits extremely high prediction accuracy. When the coefficient of determination （R²） of the formation crude oil viscosity model reaches 0.94 for the training set and 0.95 for the prediction set,the dominant controlling factors of formation crude oil viscosity in Oilfield P are surface crude oil density and surface crude oil viscosity.

Key words: formation crude oil viscosity, machine learning, BP neural network, surface crude oil density