The training process is the process of determining the
connection weights of the neural network. A set of input
output pairs of the training patterns or training data are
needed to accomplish the training process. The input data set
is PV generator maximum output power Pmax and the
rotational speed of the motor ωm, whereas the output data set
is the corresponding duty ratio of the buck chopper. The
training data has been obtained by varying the solar radiation
level from 50W/m2 to 1200W/m2 with incremental value of
10W/m2. While the PV surface temperature is incremented by
2 oC per step from 0 to 70 oC, and the corresponding duty
cycle is calculated for each case.
A neural network toolbox in MATLABSIMULINK has been
used to achieve the training process. During the network
training, all the computations are offline. The backpropagation
algorithm is used in this work; this type is
considered one of the most widely used algorithms because of
the ease of implementation, robustness and stability. The
learning stage of the network is performed by updating the
weights and biases using a backpropagation algorithm with
the gradient descent method in order to minimize a mean
squared error performance index E by using Equation (15).
The smaller the mean square error is the better performance
and the accuracy of the network in real life will achieve.