Application Of Supervised Learning Algorithms In Predicting Solar Energy Production

Abstract:There are the lowest forms of environmental pollution and the lowest level of carbon footprint in solar power categorized as renewable energy. It is high future demand to use solar energy in order to avoid energy crisis. So estimating the energy generated by solar plants is of great significance. In recent years, machine learning techniques starts to prevail in energy production analysis and prediction in solar farms. This research shows that application of supervised learning algorithms can significantly increase predictive power thus providing more flexible solutions for decision-makers and supporting their decisions with tested statistical models. Preprocessing and prediction methods of solar production data have been studied aiming at the characteristics of solar data in this thesis.First, the status of supervised learning algorithms is reviewed in this thesis. Data preprocessing methods are analyzed and researched. The use of mean value substitution method is proposed for missing data processing combined with the characteristics of solar data. The method based on K-fold cross-validation is put forward for obtaining and processing the data set for the training data set and validation data set acquisition. Visualization techniques are used to show the characteristics of the data set from different angles. Finally, Gradient Boosting, Random Forests and Extremely Randomized Trees three supervised learning algorithms are applied on processed data sets for forecast and analysis of solar energy data. The researcher has got a conclusion that Extremely Randomized Trees algorithm is the most optimal supervised learning algorithm for solar energy prediction. UCI((University of CaliforniaIrvine) datasets are used to conduct a comparative analysis of these three algorithms. The experiments have shown Extremely Randomized Trees algorithm that has better accuracy and stability.Software tools used in this research are Python and R. Python is mainly used for data visualization. Besides, the implementation of the three mentioned algorithms uses Python’s machine-learning library to make solar prediction. R is employed for performance above-mentioned algorithms comparison of the results obtained from Python.