Research on Non-Intrusive Load Decomposition Method Based on s2p Model
The non-invasive monitoring of load decomposition can accurately obtain the electricity consumption information of household users and the power data of various electrical devices,providing a data foundation for demand side management and user electricity optimization.This paper predicts the active power data of the proposed five types of power-using devices based on the long short-term memory neural network(LSTM)model,and combines the im-proved differential double sliding window algorithm for event detection of the predicted data;a sequence-to-point(s2p)model with a self-attentive mechanism is given,and the encoder part of the traditional model is replaced with a self-attentive model and a fully connected feedforward network,which has better decomposition effect on some long sequence data.The experimental results show that the event detection accuracy,recall rate and F1 score of the pro-posed LSTM combined with the improved difference sliding window algorithm are 93.50％,90.55％ and 92.00％,re-spectively;the average errors of the three decomposition performance metrics,NDE,MAE and SAE,of the proposed s2p model based on the self-attentive mechanism are smaller than those of the traditional s2p model by 0.076,4.111,and 0.223,respectively.
NETL
NSTL
万方数据
