Abstract: Aiming to the non-intrusive load monitoring（NILM）of residential households，a method of load

profile decomposition based on the appliance switching events and the long short-term memory networks（LSTM）

was proposed. The difference-summation algorithm was performed on the total active power data of a residential

household，and the hard threshold function was used to filter the noise of the obtained difference-summation data，so as to accurately detect the switching events with extracting the abrupt change of the difference-summation data.Furthermore，the LSTM method was adopted to forecast the data of the power profile，and the power profile of each individual appliance was extracted from the total power profile of the residential household to realize the appliancelevel profile decomposition. The validity of the proposed method was verified by the experimental data and the reference energy disaggregation dataset（REDD）.

      Key words: non-intrusive load monitoring（NILM）；load profile decomposition；long sort-term memory

networks（LSTM）；difference-summation；load switching event