Real-time optimization of submersible pump performance using artificial neural networks and shiny interface

Abstract

Submersible pumps are essential components in irrigation, potable water supply, and drainage systems, yet they often operate inefficiently under fixed-speed control, leading to excessive energy consumption. Addressing this challenge, the present study introduces a real-time optimization framework that integrates artificial neural networks (ANNs) with a Shiny-based decision support interface. Experimental data were collected from a 3-inch submersible pump operating across seven discrete pressure levels (0.8–2.0 bar) and frequencies ranging from 35 to 50 Hz. Frequency (F) and outlet pressure (P) were used as model inputs, while volumetric flow rate (Q) and system efficiency (η) served as outputs. Two ANN models were developed and optimized through grid search, achieving high predictive accuracy with coefficients of determination of 0.997 for Q and 0.991 for η. During model development, performance on the train–test split was evaluated using R², RMSE and MAPE, while the agreement between ANN predictions and an independent set of 20 additional measurements was summarized using R² and MAE. Root mean squared errors were 1.00 m³ h⁻¹ for flow rate and 1.27% for efficiency, while validation against 20 real-world measurements yielded mean absolute errors of 3.24% and 3.14%, respectively. A synthetic dataset comprising 949 operational scenarios further demonstrated the models’ generalizability. The trained ANNs were embedded into a Shiny application that enables users to input target flow and pressure requirements, dynamically filter results, and receive recommended operating frequencies that maximize predicted efficiency. Case studies confirmed that the interface operates in real time (<0.5 s per query), handles infeasible inputs safely, and supports robust decision-making. Overall, the findings demonstrate that the proposed ANN Shiny framework can reduce energy losses, improve operational water management, and provide a scalable foundation for intelligent and sustainable pump control.