Anomaly detection, interpolation and real-time forecasting of time series for reservoir operation and water distribution

Author(s)

Leonardo Fonseca Larrubia

• Document type: Master's Dissertation
• Press: São Paulo.
• Institution: Universidade de São Paulo (USP). Instituto de Matemática e Estatística (IME/SBI)
• Defense date: 2021-04-08

Examining board members

Chang Chiann; Valderio Anselmo Reisen; Marcelo Magalhães Taddeo

• Advisor: Chang Chiann

Abstract

We developed a time series analysis solution for a recurring problem in Operational Control Centers of water distribution systems: anomalies in real time data received from telemetry stations. The developed solution consists of detecting outliers, reconstructing missing values and forecasting. For this purpose, we analyzed time series generated by measuring equipment: level, input and output flows of reservoir and upstream and downstream pressures of valves that control the flow of water. The data, referring to the water distribution system of the city of Peruíbe, were provided by Sabesp located in Baixada Santista and it is sampled every hour from 1:00 a.m. on January 1, 2017 to 11:00 p.m. on December 31, 2018. We propose procedures that use three main techniques for the detection of outliers and imputation of missing values: adjustment of curves by regression, classical decomposition combined with regression and STL decomposition. For forecasting, we used rolling analysis techniques in combination with SARIMA models, regression models with autocorrelated errors and BATS and TBATS models. The results showed that the proposed methods, both for detecting outliers and imputing of missing values, as well as for forecasting, perform very well for most time series. (AU)

• FAPESP's process: 18/26592-5 - Development of a real-time forecasting model for demand and levels of reservoir with big data of water industry
• Grantee: Leonardo Fonseca Larrubia
• Support Opportunities: Scholarships in Brazil - Master