DCCMAPI/CCET

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
DISCENTE: ARTUR FELIPE DA SILVA VELOSO
DATA: 08/08/2023
HORA: 14:00
LOCAL: Sala virtual - Modalidade II
TÍTULO: Deep-clustering Based Load Profiling Microservice for Demand Side Energy Management as a Service
PALAVRAS-CHAVES: Internet of Things; Deep-clustering; Load Profile; Smart Grid; Microsservice; HyDSMaaS; Demand Side Management.
PÁGINAS: 28
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
SUBÁREA: Sistemas de Computação
ESPECIALIDADE: Teleinformática
RESUMO: The lack of consumer awareness regarding energy consumption, coupled with the need to improve energy efficiency and reduce electricity costs are problems that can be attributed to the lack of availability of energy consumption data by electric utilities (EPC) to their consumers. Another problem is the generation of Load Profile (LP) for suggesting customized energy consumption services. The LP is a record of a customer’s electricity consumption over time, which is critical for understanding each customer’s energy consumption behavior and providing personalized suggestions for energy consumption services. Additionally, LP generation is a challenging task due to the complexity of electricity consumption data and the need to analyze and process large amounts of data. Moreover, often the available power consumption information is insufficient, which may limit the accuracy of personalized power consumption service suggestions. With that in mind, this work proposes the implementation of an algorithm based on the time series model for LP generation, which can reduce the size of the database processed during the execution of services that could be offered on the end-consumer side. Among these services are Demand Response, energy consumption management, prediction and classification of consumption profile to avoid peak hours and reduce electricity costs. Considering the mentioned challenges, this work proposes a LP management micro-service based on deep-clustering algorithms. This service aims to manage electric energy from the demand side, offering personalized suggestions of electric energy consumption services for each customer, and overcoming the limitations of LP generation. To achieve this goal, the time series model for LP generation was implemented and six different clustering algorithms were applied: K-means, Gaussian Mixture Model (GMM) and Spectral (considered classical algorithms) and Deep Embedded Clustering (DEC), Hierarchical and Autoencoder with K-means (considered deep clustering algorithms). These algorithms were compared to determine which one would be the most suitable for the proposal of this work. As results, more than 30 hours of processing were reduced in the general microservice with more than 5500 consumers in a database of one and a half years, the database was reduced from more than 160 million data to a little more than 5500 records, thus making the database even more accurate with a smaller amount of records. In clustering, the DEC algorithm obtained the best result, allowing the identification of patterns in the electricity consumption data and grouping consumers into only three distinct groups, which can be very useful for developing customized electricity consumption management strategies.
MEMBROS DA BANCA:
Presidente - 005.693.753-98 - JOSE VALDEMIR DOS REIS JUNIOR
Externo à Instituição - FILLIPE MATOS DE VASCONCELOS - UFMT
Externo à Instituição - ROGÉRIO ANDRADE FLAUZINO - USP