Aplicação de Gêmeos Digitais Baseados em KNN para Previsão e Controle de Pressão em Sistemas de Abastecimento de Água

Abstract

Predicting pressure in water supply systems is fundamentally important for optimizing operations and reducing waste. In this work, an offline digital twin was developed to predict pressure in an experimental hydraulic network using machine learning algorithms. The studied system consists of a test bench composed of a motor-pump set, pressure sensors, and automated valves, allowing the simulation of various possible operational scenarios. Machine learning models such as KNN were tested for pressure prediction in an offline environment, where their performance was analyzed in comparison to a model using a MultiLayer Perceptron. The results show that the model is capable of predicting the system's pressure with satisfactory accuracy, enabling a future implementation of an online version connected to the experimental hydraulic network. This reinforces the application of digital twins for the monitoring and control of hydraulic systems, contributing to operational efficiency and waste reduction.