Design and simulation of an off-grid photovoltaic system with duty cycle prediction using neural network controller

Research output: Contribution to journalArticlepeer-review


Global concerns over the inappropriate utilization of abundant renewable energy sources, the damages due to instability of fuel prices, and fossil fuels' effect on the environment have led to an increased interest in green energy (natural power generation) from renewable sources. In renewable energy, photovoltaic is relatively the dominant technique and exhibits non-linearities, leading to inefficiencies. Maximum Power Point is required to be tracked rapidly and improve the power output levels. The target is to use a Neural network controller by training historical data of ambient irradiance and temperature levels as inputs and voltage levels as output for the photovoltaic module to predict duty cycles across the DC-DC converter. The DC-DC converter is the electrical power conditioner at the Botswana International University of Science and Technology, Palapye Off-Grid photovoltaic system. Perturb and Observe algorithm on PSIM environment is only implemented to acquire the historical data for the training and Matlab for the modeling of the network. Relatively long period ambient irradiance and temperature data of Palapye were acquired from the Southern African Science Service Centre for Climate Change and Adaptive Land Management (SASSCAL) WeatherNet in Botswana. Matlab environment was used for the simulation of the backpropagation algorithm for training. The Neural network's feedforward to optimize the non-linear nature of the PV module input and output relationship with relatively fewer processes is required. The results show promising, and the Mean Errors appear to be typically about 0.1 V, and the best performance is 193.5812 at Epoch 13, while the regression delivered a relatively low measured error. The maximum power delivered by the duty cycles from the model with 90 % prediction accuracy. The article demonstrates Neural Network controller is more efficient than the conventional Perturb and Observe Maximum Power Point algorithm.

Original languageEnglish
Pages (from-to)181-210
Number of pages30
JournalInternational Journal of Engineering Research in Africa
Publication statusPublished - Nov 2021

All Science Journal Classification (ASJC) codes

  • General Engineering


Dive into the research topics of 'Design and simulation of an off-grid photovoltaic system with duty cycle prediction using neural network controller'. Together they form a unique fingerprint.

Cite this