Open AccessDesign of Automatic Switch System of Residential Load From Solar Cell and Power Plant Resources using Neural Network
Author(s) -
Desri Kristina Silalahi,
Bandiyah Sri Aprilia,
Wahmisari Priharti,
K. Kumillayly,
Sofia Saidah
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/771/1/012006
Subject(s) - artificial neural network , backpropagation , computer science , electric power system , power (physics) , process (computing) , electrical load , real time computing , voltage , simulation , engineering , artificial intelligence , electrical engineering , physics , quantum mechanics , operating system
Residential loads are electronic equipment that is often used at home. Residential loads are supplied from two sources, they are solar panel and the State Power plant. Selecting the supply load source, so an automatic switch system is needed from the load. Optimizing residential load resources and avoiding overloading, load balancing techniques are needed. The switch system uses a relay, determined by the solar panel power output. This research discusses the design of automatic load switch systems for residential loads from solar panels and State Power plant using Artificial Neural Networks (ANN). ANN control system that is arranged using ANN Backpropagation consists of 4 inputs, four hidden layers, each consisting of 4 neurons and one neuron in the output layer. In this research, to determine the network that has been formed to provide changes of load. The results of testing that the parameters used to get the smallest error rate in the process of setting an automatic power load switch is best to use a number of repetitions of 2000 times with an error percentage 5.3%. Artificial Neural Networks can experience convergent failure or not close to output because the initial guess is not good. Initial experiments with actual solar panel data with as many as 98 data produced non-convergent output values. The solution to improving the initial experiment is to simplify learning data 40 data produces convergent output.