METHOD FOR DISAGGREGATION OF CUMULATIVE ENERGY CONSUMPTION SIGNAL Russian patent published in 2023 - IPC G01R31/00 G06N3/985 

FIELD: computer technology.

SUBSTANCE: method identifies natural classes of devices stored in the database that do not contain the same or similar signals in the resulting classes, by performing a clustering method for each individual class of devices, in which parameters are selected at which the clustering method determines the boundaries of the classes in such a way that their minimum number of crossings is achieved; generating a database of aggregate energy consumption signals from the energy consumption signals of individual devices stored in the database, with assigned labels of natural classes of devices; training a machine learning model based on the database generated in the previous stage, where the input data are signals of aggregate energy consumption, and the output data are signals of energy consumption and/or participation share and/or an indicator of presence in aggregate energy consumption for labels of natural classes of devices; receiving the aggregate energy consumption signal from the distribution network and disaggregating it using the trained machine learning model; wherein additionally performing post-processing of the labels of natural classes of devices obtained at the output of the machine learning model by comparing them with the corresponding data about devices from the database of energy consumption signals of individual devices and labels, as well as by comparing the result of disaggregation at a given point in time with the history of disaggregation up to a given point in time and determination, based on post-processing of the disaggregation result, of the operating modes of individual devices.

EFFECT: increased accuracy and interpretability of energy disaggregation results.

2 cl, 3 dwg, 3 tbl