Comparative analysis of methods for simulating the well operation with electric submersible pump installations

Abstract

This article presents the research results of parametric and non-parametric identification methods of the technological models of well operation using electric submersible pump installations. The use of a hybrid approach is proposed, combining parametric and non-parametric models to obtain accurate models that allow the prediction of well performance parameters. Studies of simulation methods under conditions of interference effect of different level, which are typical for signaling channels of real data management, control systems, and measuring instruments, have been conducted. The combined models proposed have been constructed with the help of the Rosenblatt-Parzen non-parametric regression, parametric models with automatic adaptation of parameters and artificial neural networks. Such combined models have been shown to possess essential generalizing possibilities, allowing for smoothing of parametrical data and the restoration of initial dependences with a significantly smaller error in relation to the disturbing interference. The developed methods and models were implemented for research purposes in the software system, which allows a complex simulation of changes in parameters during well operation using the electric submersible pump installations. To evaluate the results' statistical significance, methods of statistical processing have been applied using ANOVA. The results demonstrate that for an effective solution to the problem of the process simulation of well operation and to ensure high adaptability of the models, the combined approach is the most effective method. Models on the basis of artificial neural networks after adjustment allow us to improve the efficiency of the solution to the prediction problem and at the same time have necessary flexibility for adaptation of the computational structure under the conditions of changing performance parameters. The parametric block of models allows us to use a priori information about dependences of performance parameters and to identify reasonably accurate the drift of parameters under the conditions of instability of the process under study.