Pumping motor with large electrical horsepower for small power-consuming equipment is commonly seen in beam pumping units

as it is difficult to understand exact relationship between motor working efficiency and complex dynamic loads. An online hybrid model for motor load torque under dynamic load changes was proposed by online sequential-extreme learning machine (OS-ELM) with a consideration of polished rod loads as the system's loads. First

mechanism models of each part in the pumping unit were separately built according to working principles of the system. Then

OS-ELM-based online soft sensor model was built to obtain value for a critical uncertain variable

the underground friction. Original structure of the soft sensor model was first set by offline training with historical production data and then online updated by a sliding window method. Therefore

the underground friction is no longer a constant value

which was given by any subjective experience in most other studies

but a dynamic value following system changes

which is more in line with actual operation conditions. The simulation results of proposed method on a normal oil well demonstrated validness and effectiveness.