Développement d'une méthodologie d'analyse comparative des performances basée sur les données pour un grand nombre de climatiseurs d'autobus

Bus air conditioners (ACs) are responsible for providing a comfortable cabin environment for passengers. Identifying the bus ACs with degraded performance from a large number of city buses is a critical and challenging task in the development of smart cities. This study developed a data-driven benchmarking methodology to detect anomalous operations with degraded energy performance from a large number of bus ACs. For each target AC to be benchmarked, its similar operation data in other ACs, termed comparable peer samples, are first identified by a Long-Short-Term-Memory (LSTM) autoencoder-based similarity measurement method. The comparable peer samples are then used to develop a LSTM network-based reference model for predicting the power consumption of the target AC. A key energy performance indicator termed power consumption ratio (PCR) is defined for the target AC as the ratio of its measured power to the predicted power. Statistical analysis-based trend and change detection algorithms are designed to identify a trend or change of PCR over a few days for anomalous detection. To validate the benchmarking methodology, two fault experiments were conducted in field-operating bus ACs, and the results show encouraging potentials of the proposed methodology for health monitoring of a large number of ACs serving the city bus fleet.