Abstract
Prediction of solar output power is a valuable research work for
analyzing photovoltaic(PV) power. This study develops a futuristic
deep-learning algorithm that predicts solar power output. The solar
output data is collected in real-time for a series-parallel combination
of PV systems with a 1 KW capacity that is available in our laboratory.
The collected data is pre-processed via, initialization, normalization,
and validation for accurate prediction. The normalization process is
used to create the data set needed to fill in the missing values. The
k-nearest neighbor (KNN) algorithm and the interpolation method are used
to fill in the missing value. Then, the data is validated using a newly
proposed deep long short-term memory (DLSTM) algorithm for solar output
power prediction. Also as a new approach, the DLSTM algorithm and a
recurrent neural network (RNN) are combined with the capture of
time-series data in the validation process to improve the prediction
accuracy. To prove its superiority, the proposed DLSTM-RNN model is
compared with other exciting models, like the artificial neural network
(ANN), long-short-term memory (LSTM), and recurrent neural network
(RNN). All the models are trained and tested using three different
activation functions viz Sigmoid, ReLU, and tanh with different epoch
values. Finally, the accuracy is evaluated in terms of different
performance error indexes, such as the basic error index (BEI) and the
promoting percentage error index (PPEI).