A Region-Based Approach for Missing Value Imputation of Cooling Technologies for the Global Thermal Power Plant Fleet Using a Decision Tree Classifier

Abstract

AbstractThe lack of information on the current water demand of individual thermal power plants is a problem for the planning of future energy systems, especially in regions with high water scarcity and an elevated power demand. This lack is linked to the limited availability of data on the type of cooling technology for these power plants. In this study, we propose a hybrid decision-tree based classification model to impute the missing values of the cooling technology for individual power plants globally. The proposed model is cross-validated on the GlobalData database and benchmarked against several approaches for missing value imputation of the cooling technology of individual power plants found in the scientific literature. The decision tree model (with the average test set accuracy of 75.02%) outperforms all alternative approaches in terms of accuracy, often by a considerable margin. In addition, for 103 out of the 137 minor regions in this study, the hybrid model yields the highest test set accuracy of all approaches. It is apparent that, in terms of accuracy, the proposed hybrid model seems to outperform more general models which are based on shares or the portfolio mix in a region/country. The proposed model can be replicated and used in future studies, which have different data sources at their disposal.KeywordsMachine learningSustainabilityWater-energy nexusPower generationSupervised learning