| Summary: | Environmental issues are among the critical problems plaguing the world. Since the
Solar energy has been considered as one of the fastest growing among the renewable
energy sources that addresses the energy demand in producing clean energy. The
continuous development in solar-based harvesting technology contributes to increase
reliability of the solar power technology. One of the key issues for solar energy is the
dependency to the weather conditions, which generates energy fluctuations in the
amount of energy generated, and furthermore leads to electric grids to be unstable.
Therefore, there is a need to have an accurate prediction of solar energy generation. The
main objective is to predict the amount of energy generated by solar panels in different
weather conditions using supervised machine learning. The model of prediction is
developed based on 2000 dataset collected using sensors through Internet of Things
platform. The solar energy dataset collected represents parameters of weather
conditions, namely: outlook; temperature· and humidity. The model of prediction
algorithms such as Artificial Neural Network (ANN), K-Nearest Neighbour (KNN), and
Random Forest (RF) are selected because these three algorithms are supervised machine
learning techniques. The three algorithms were analysed and compared using
correlation coefficient (r) and error ratio mean absolute error (MAE), to measure their
performance. The use of r evaluates the ratio of accuracy and strength of the relationship
between weather condition parameters and solar energy produced. The r used to
measure how strong a relationship is between two variables, on a scale that varies from
+ 1 through 0 to - 1, the higher r-value represents a higher accuracy of the prediction.
MAE measures the average magnitude of the errors in the test sample of the absolute
differences between prediction and actual value, in a set of predictions. Where lower
MAE value indicates that the algorithm has higher accuracy and less error rate. The
result shows that RF has the best r of voltage compared to KNN and ANN algorithms
with 97.26%, 97.04%, and 95.11 % respectively. RF has the lowest error rate compared
to KNN and ANN algorithms with 11.54%, 11.60% and 13.61% respectively. Higher r
and lower MAE represent higher prediction accuracy. The study was able to create
accurate prediction model using supervised machine learning techniques for the solar
panel voltage output. Using the prediction model, electric grid operators can predict
accurately the voltage that the solar power plant will produce based on weather
conditions.
|
|---|