Artificial Neural Networks Predict Discharge Pressures of ESPs

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 216598, “Application of Artificial Neural Networks in Predicting Discharge Pressures of Electrical Submersible Pumps for Performance Optimization and Failure Prevention,” by M.A. Mahmoud, M. AbuObida, and O. Mohammed, University of Khartoum, et al. The paper has not been peer reviewed. _ This paper presents a novel approach using artificial neural networks (ANNs) to predict the discharge pressure of electrical submersible pumps (ESPs). A data set of more than 12,000 data points collected from 40 different wells was used to train and test various ANN models with different input parameters. The ability to predict discharge pressure accurately can lead to the early detection of possible anomalies. Methodology Data Collection. The data-collection process for ESP sensor data involved collecting data from 40 wells operational during 2019–2023 with starting dates from 2019 through 2022. The data included key parameters provided in the complete paper. Data Cleaning and Integration. The data-cleaning process was performed to ensure that the accuracy and reliability of the data were maintained. Scatter plots were used to visualize the distribution of each variable in the data set, and outliers were identified and removed or smoothed. Missing values were addressed by fill-in using statistical methods or by removing them from the data set if the percentage of missing values was significant. Inconsistencies in the data were dealt with by removing duplicates, fixing errors, and ensuring that the data was consistent across all sources. The data-integration process involved combining data from different sources to create a single data set for analysis. Data were labeled and formatted to facilitate analysis, including assigning meaningful names to the variables and ensuring that data were in the correct format. Data Analysis and Physical Relationship. The MATLAB program was used to analyze the data statically. A code was made to generate charts and statistical calculations to gain insights into characteristics such as minimum, maximum, mean, variation, standard deviation, skewness, and covariance. Understanding the physical relationship between the various operating parameters and the pump-discharge pressure is crucial. A comprehensive analysis was conducted to examine the correlation between 18 key parameters and the pump-discharge pressure. Model Development and Optimization. The 18 input parameters were used to build an ANN using MATLAB. The neural network was trained using a set of known input/output pairs and tested on a separate set of data. The performance of the neural network was evaluated using various metrics such as mean squared error and correlation coefficient. After the ANN model was built, its performance was evaluated to determine the best configuration of the model. This was achieved by varying the number of input parameters, hidden layers, and neurons in each layer. The best model configuration was then identified that provided the most-accurate predictions and minimized the errors in the model.