Semisupervised refrigeration plant cooling disaggregation by means of deep neural network ensemble

The awareness of the energy usage has become a recurrent topic during the last decades. Identifying the end-use energy of each individual device can lead to a substantial improvement in efficiency and fault detection. The cost of instrumentation and especially the ones which involve fluids, makes the monitoring unfeasible. Hereby, the necessity of Non-Intrusive Load Monitoring (NILM) techniques has increased in order to avoid the aforementioned associated costs. In this paper, the cooling power of a refrigeration plant is disaggregated to identify the cooling power spent in each compartment. A data-driven methodology based on a semisupervised deep neural network ensemble is presented, which takes advantage of the data acquired from the typical installed sensors in a refrigeration plant. The proposed strategy is able to disaggregate accurately the cooling power without the necessity of introducing any additional sensing device in the installation. The proposed methodology is validated with a test bench simulation and also with real refrigeration plant data.