Devices and processes for data sample selection for power consumption management

A device for selecting data samples related to power consumption so as to proceed with power consumption management, said device comprising:

• at least one input adapted to receive available data samples related to power consumption and associated with respective sample times, said available data samples being distributed into a sliding time window including at least one new batch of data samples, said available data samples being then called current samples, and into a past period preceding said time window, said available data samples being then called past samples;
• at least one processor configured for determining in said past period, selected past samples to be kept for said power consumption management, by keeping among said past samples:
• a first share of said past samples, consisting in most recent ones of said past samples, irrespective of said at least one similarity criterion;
• and
• a second share of said past samples through eliminating among said past samples deprived from said first share, called a complementary share of said past samples, part of said past samples in function of said at least some of said current samples, and of said elimination conditions depending on said at least one similarity criterion applied to at least the first share and the complementary share of said past samples;
• wherein said at least one processor is configured for determining said first share and said second share of said past samples so that said first share and said second share have relative proportions given by a hyper-parameter;
• at least one output adapted to provide said selected past samples as a complement to said current samples for said power consumption management;
• and
• detect, based on the selected past samples and after an elimination step of eliminating the part of said past samples, a predicted anomaly in a power system based on received available data samples;
• and
• provide, to a user, in advance of a predicted timing of the predicted anomaly, a notification comprising the predicted anomaly and decisional guidance relating to the predicted anomaly;
• wherein:
• said at least one input is adapted to receive evaluations of relevance of said selected past samples, said evaluations of relevance being respectively associated with successive time positions of said time window and determined with respect to machine learning processing for said power consumption management corresponding to next time positions of said time window;
• and
• said at least one processor is configured for determining said hyper-parameter by selecting among at least two candidate values of said hyper-parameter a best candidate value providing smallest averaged errors associated with said evaluations of relevance over said successive time positions of said time window.