In the project 36152.1IP-EE - NILMplus, a cost-effective system, which does not require any intervention in the cabling, for power measurement of industrial loads was researched This was to be realized by means of low-cost current sensors and self-learning methods of Non-Intrusive Load Monitoring (NILM).
As part of the project, the relevant requirements for the current sensors were worked out and three NILM approaches were developed. Although the mean measurement error in many tests was below the targeted 3%, the number of unacceptable measured values was too high for a universal application. It was shown that, among other things, the consumption of the machine in relation to that of the total input is very crucial for the accuracy.
Since this ratio is unknown without a reference measurement, the uncertainty of the method cannot be estimated in advance.
Thus, the method is not suitable for economic implementation. The project team concluded that self-learning NILM approaches, even with the help of additional sensors on the cable, are unsuitable for the application at hand.
Nevertheless, the results obtained in the project are considered valuable: The limits of self-learning NILM in the industrial environment have been more precisely defined, the data set collected will drive the further development of NILM approaches for the industrial sector, and the intensive engagement with the topic of "current measurement on the multiphase cable" raised further research questions and resulted in the submission of a bridge discovery application.
