Machine-learning-integrated load scheduling for reduced peak power demand

Load scheduling over cyclic electrical devices can reduce the peak power demand. In this paper, we propose a machine-learning-integrated load control (MILC) scheme for improved performance and reliability. By dynamic capacity adjustment and interactive load heuristic, MILC tries to reduce the power deviation while keeping the temperature violation ratio and switching counts within an acceptable range. A prototype of the proposed scheme has been implemented and, through experiments using load traces from a real home, we evaluate the performance of MILC. The results show that MILC reduces the peak demand from 4993 W to 4236 W and successfully decreases the power deviation by 12.1% on average.