Integração avançada de análise de séries temporais, clusterização, previsão e criação de aplicação desktop para otimização da eficiência energética e gestão de tarifas

Abstract

This doctoral thesis investigates advanced analysis of electrical demand and energy consumption, with a focus on anomaly detection and the identification of consumption patterns in consumer units. By employing time series analysis techniques and specialized algorithms, anomalous behaviors were identified in both demand and consumption records. Additionally, the research

encompassed the identification of customers with anomalous consumption profiles using cluste- ring techniques based on average monthly consumption and contracted demand. The clustering

approach provides a broader perspective on consumer groups, revealing distinct characteristics in terms of consumption patterns—whether normal or irregular—as well as trend and seasonality behaviors. The research also incorporates consumption forecasting. Highly correlated time series within the same group are used to predict the behavior of new time series or those with limited historical data. This approach offers a deeper understanding of the behavioral patterns of older and more stable series, which is essential for estimating the annual consumption levels of new

consumer units and forecasting their behavioral trends. Forecasting also contributes to the deter- mination of the most suitable contracted demand, optimizing energy consumption efficiency. To

enhance the practical utility of these analyses, a specialized desktop application was developed. This tool enables the analysis of demand and energy consumption data, supporting the selection of the most appropriate tariff option for Group A energy consumers. The application highlights relevant statistics such as mean and standard deviation and compares current costs with different tariff alternatives. It also includes an anomaly detection tab that allows both graphical and textual visualization of values identified as anomalies. Furthermore, it features a dedicated tab for observing trends and seasonality in decomposed time series. The results of this research not only

improve the understanding of energy consumption and electrical demand patterns but also intro- duce practical and innovative tools for professionals in the sector. The study demonstrated that

energy costs for the analyzed public units could be reduced by approximately BRL 2.7 million in a single year. This thesis represents a significant contribution to the intelligent management of energy demand, integrating advanced time series analysis, clustering, and forecasting techniques to optimize operational and economic efficiency in the This doctoral thesis investigates advanced analysis of electrical demand and energy consumption, with a focus on anomaly detection and the identification of consumption patterns in consumer units. By employing time series analysis techniques and specialized algorithms, anomalous behaviors were identified in both demand and consumption records. Additionally, the research encompassed the identification of customers

with anomalous consumption profiles using clustering techniques based on average monthly consumption and contracted demand. The clustering approach provides a broader perspective on consumer groups, revealing distinct characteristics in terms of consumption patterns—whether normal or irregular—as well as trend and seasonality behaviors. The research also incorporates consumption forecasting. Highly correlated time series within the same group are used to predict the behavior of new time series or those with limited historical data. This approach offers a deeper understanding of the behavioral patterns of older and more stable series, which is essential for estimating the annual consumption levels of new consumer units and forecasting their behavioral trends. Forecasting also contributes to the determination of the most suitable contracted demand, optimizing energy consumption efficiency. To enhance the practical utility of these analyses, a specialized desktop application was developed. This tool enables the analysis of demand and energy consumption data, supporting the selection of the most appropriate tariff option for Group A energy consumers. The application highlights relevant statistics such as mean and standard deviation and compares current costs with different tariff alternatives. It also includes an anomaly detection tab that allows both graphical and textual visualization of values identified as anomalies. Furthermore, it features a dedicated tab for observing trends and seasonality in decomposed time series. The results of this research not only improve the understanding of energy consumption and electrical demand patterns but also introduce practical and innovative tools for professionals in the sector. The study demonstrated that energy costs for the analyzed public units could be reduced by approximately BRL 2.7 million in a single year. This thesis represents a significant contribution to the intelligent management of energy demand, integrating advanced time series analysis, clustering, and forecasting techniques to optimize operational and economic efficiency in the power sector.