Loading...
Publication
Sistema inteligente para deteção de anomalias em consumos hoteleiros
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 2.52 MB | Adobe PDF |
Abstract(s)
Esta dissertação de mestrado tem como objetivo o desenvolvimento de um sistema inteligente
para deteção de consumos anómalos em hotéis. O sistema em si, detetará
precocemente anomalias de consumo, quer sejam de eletricidade, água ou gás, para
poderem ser analisados e eventualmente corrigidos, refletindo-se em menores consumos,
numa maior poupança e numa maior segurança.
Para ser possível o desenvolvimento deste sistema, como em qualquer aplicação
que envolva modelos de aprendizagem máquina, primeiramente efetuou-se uma análise
exploratória a um conjunto de dados de um hotel, que de alguma forma seriam
representativos dos dados que a plataforma irá processar futuramente. Em seguida,
é efetuado um estudo comparativo entre dois algoritmos de aprendizagem máquina,
Isolation forest e Variational autoencoder, para decidir qual deles a ser implementado no
sistema. Para tomar esta decisão, desenvolveu-se uma métrica de desempenho de algoritmos
para deteção de anomalias em séries temporais não supervisionadas. Depois
é apresentada a arquitetura do sistema e alguns resultados de operação em condições
reais. Para concluir, apresentam-se as conclusões do trabalho e são feitas algumas sugestões
para possíveis trabalhos futuros.
This master’s dissertation aims to develop an intelligent system for anomalous consumption detections in hotels. The system itself will early detect consumption defects, whether electricity, water, or gas, so that they can be analyzed and eventually corrected, resulting in lower consumption, greater savings and higher security. In order to develop the system, like in other applications that involve machine learning models, an initial exploratory data analysis was performed on hotel datasets, that would somehow be representative of the data that will be processed in the future by the system. Then, a comparative study was carried out between two machine learning algorithms, Isolation forest and Variational autoencoder, to decide which one to implement in the system. To make this decision, a performance metric for anomaly detection algorithms in unsupervised time series was developed. Afterward, the system architecture and some operational results under real conditions are presented. To finalize, the conclusions of the work are presented and some suggestions are made for possible future works.
This master’s dissertation aims to develop an intelligent system for anomalous consumption detections in hotels. The system itself will early detect consumption defects, whether electricity, water, or gas, so that they can be analyzed and eventually corrected, resulting in lower consumption, greater savings and higher security. In order to develop the system, like in other applications that involve machine learning models, an initial exploratory data analysis was performed on hotel datasets, that would somehow be representative of the data that will be processed in the future by the system. Then, a comparative study was carried out between two machine learning algorithms, Isolation forest and Variational autoencoder, to decide which one to implement in the system. To make this decision, a performance metric for anomaly detection algorithms in unsupervised time series was developed. Afterward, the system architecture and some operational results under real conditions are presented. To finalize, the conclusions of the work are presented and some suggestions are made for possible future works.
Description
Keywords
Sistema inteligente Deteção de anomalias Aprendizagem máquina Isolation forest Variational autoencoder