Dynamic liquid level modeling of sucker-rod pumping systems based on Gaussian process regression

In practical oil production it is very important to precisely detect the real-time dynamic liquid level of the sucker-rod pumping system. This contributes to laying down the reasonable working institution of the pumping system in order to improve the efficiency of the pumping system and reduce the fault rate of the pumping equipment. In the paper the dynamic liquid level modeling method based on Gaussian process regression (GPR) is proposed. By introducing the simulated annealing algorithm to traditional GPR such an approach can nicely solve the drawbacks of common optimization methods in traditional GPR. The above drawbacks are that the optimal values of hyperparameters in the model are easily affected by the initial values and fall into the local optimal solution. The simulation analysis and practical running data show that Gaussian process regression is very appropriate for dynamic liquid level modeling. The model entirely meets the precision and reliability requirement.