Solar enerji kullanımında makine öğrenmesi

In this dissertation, the solar energy utilization processes were studied via machine learning algorithms. First, a bibliometric analysis of the topic was performed; it was observed that photovoltaics and hydrogen-related applications have been studied more extensively while machine learning, mostly neural network and deep learning algorithms, has been mainly used for forecasting and optimization. Desalination, dye-sensitized solar cells and photo(electro)catalysis, which are the most commonly studied solar technologies, were selected for further analysis in this dissertation. The data were collected using Web of Science database search with related keywords while R language was used for the analysis. Tree-based ensemble methods gave higher predictive power, apparently due to the presence of a large number of categorical variables; the decision trees were used to deduce heuristic rules for high performance, especially for the cases in which the predictive models were not successful. Although the band gaps of semiconductors were generally predicted successfully, the accuracy in the prediction of gas production rate and photocurrent density was lower; hence, the classification with a decision tree was employed to identify the range of input variables for high performance. Results in desalination analysis suggested that the inlet water and air temperatures and flow rates have a higher influence on the performance. The dye type was found to be the most important variable for the performance of both synthetic and natural dye solar cells; the type of counter electrode also becomes important for synthetic dyes whereas electrolyte is more influential in natural dye counterparts. In photocatalysis, it was also found that the band gap mostly depended on the preparation method of the semiconductor while the photocurrent density was affected by bias and electrolyte. For $CO_2$ reduction, the method used for co-catalyst deposition and reaction temperature were important for the gas and liquid phase systems, respectively.