A new approach for CMS RPC current monitoring using Machine Learning techniques

Monitoring the stability of the RPC current is a tedious job where more than a thousand individual high voltage (HV) channels have to be analyzed. The current depends on several parameters (applied voltage, luminosity, environmental parameters, etc.), and sometimes it is not evident if it changes due to variation of external parameters or if it is due to a malfunction of the chamber. A Machine Learning approach is introduced to monitor and detect possible HV problems. A Generalized Linear Regression algorithm is trained to recognize the HV current behavior of a given chamber. The algorithm is then used to predict the HV current under certain data-taking conditions and environmental parameters. The divergence between the predicted HV current and the measurement is an indication of a problem. Results from several chambers are displayed. The algorithm is trained and tested with data from 2017 and 2018. The software development is at the ``proof of concept'' level and the results are encouraging.