Banca de DEFESA: TIAGO TARGINO SEPULVEDA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : TIAGO TARGINO SEPULVEDA
DATA : 03/12/2020
HORA: 09:00
LOCAL: https://conferenciaweb.rnp.br/events/defesa-de-dissertacao-de-mestrado-de-tiago-targino-sepulveda
TÍTULO:

Application of neural networks for tracking maximum power in photovoltaic panels with shadowing

PALAVRAS-CHAVES:

Photovoltaic systems; Maximum power point tracking; Incremental conductance; Articial neural networks; DC-DC converter.

PÁGINAS: 88
GRANDE ÁREA: Engenharias
ÁREA: Engenharia Elétrica
SUBÁREA: Sistemas Elétricos de Potência
ESPECIALIDADE: Geração da Energia Elétrica
RESUMO:

This work presents a Maximum Power Point Tracking (MPPT) method in photovoltaic systems combining an Articial Neural Network (ANN) with a classic tracking technique known as Incremental Conductance (InC). Despite being widely applied, the InC technique generally fails to track the Maximum Power Point (MPP) under partial shading conditions due to the existence of local maximum points on the P-V characteristic curve. For this reason, an ANN has been trained in order to provide an initial reference voltage to the system that ensures that the InC tracking starts in a region which will converge to the MPP. To help in simulations, a solar cell is modeled and, based on parameters from a commercial solar module, the model of a photovoltaic panel is built using MATLAB software. The main methods of MPPT are described, pointing out the positive and negative sides as well as the respective implementation algorithms. The average small signal model is applied to the Boost converter, which considers small variations around an operating point, to obtain a linear model. With the delevoped model, it is possible to calculate the transfer function which re ects the in uence of the switching duty cycle of the DC-DC converter on the bus voltage that a photovoltaic arrangement is connected and thus insert a compensator to improve the efciency of the system. The main concepts of an ANN are presented and neural network training is performed using the backpropagation algorithm, which is based on the error calculated in the output layer to adapt the synaptic weights of the neuron layers. Then, the Constant Voltage, Perturb and Observe and Incremental Conductance methods are simulated. The results are discussed and compared taking into account the efciency of the search under variation of radiation and temperature and also under the efect of partial shading. The results of the proposed technique are presented considering dierent cases of arrangements and partial shading in order to demonstrate the efciency of the technique.

MEMBROS DA BANCA:
Interno - 1666309 - FABIANO FRAGOSO COSTA
Interno - 1504041 - LUCIANA MARTINEZ
Externo à Instituição - DURVAL DE ALMEIDA SOUZA - IFBA