Applying Machine Learning Methods and Outlier Detection to Process and Analyse Incomplete Heat Meter Data
Abstract
Digitalization plays an important role to achieve optimized operation of renewable-based energy supply systems that couple the sectors heat, electricity, and mobility. The benefit derived from optimization methods, however, also depends on the quality of the used data. Therefore, the following study proposes methods for the processing of low-quality heat meter data. Machine learning algorithms are used to fill the gaps of incomplete load and return temperature multi-year profiles retrieved from a dataset of five heat meters installed in a heating grid of an industrial consumer. On average, the coefficient of determination R² for the black box regression models is 0.93. Additionally, four statistics-based methods are developed to detect outliers in continuous profiles to see if this can improve the accuracy of the models.