Search

This study presents a new data mining strategy to discover the performance and operational patterns of a shared energy recovery (SER) system with a data centre and a district heating network. Multidimensional clustering incorporated with a composite performance metric was first used to evaluate the typical performance of the system and reveal the interactions among different performance indicators. Decision tree analysis was then used to identify distinct system performances under different external conditions. Temporal clustering analysis was lastly used to identify the impact of recovered waste heat on the variations in heat supply from the district heating substation. The strategy was evaluated through a case study SER system at a university campus located in Norway. It was found that the most frequent performance accounted for 34 % of the total operational period with the instantaneous waste heat recovery rate of 572.9 kW, the temperature of waste heat of 57.2 °C, and the coefficient of performance of the heat pumps of 2.0. The outdoor air temperature and supply water temperature from the main district heating substation to the campus buildings showed a significant impact on the SER system performance. Moreover, the results showed that the SER system can help reduce the energy use of the district heating networks while increasing the fluctuations of heat supply from the main district heating substation. • A two-step clustering strategy is proposed to study shared energy recovery systems. • A new performance metric considering multiple indicators is presented. • The performance under different external conditions is quantified. • Unsupervised and supervised data mining are combined for performance evaluation. [ABSTRACT FROM AUTHOR]