Research on non-intrusive electrical equipment detection method based on MMoE-BiLSTM
Non-intrusive electrical equipment device detection can obtain detailed customer electricity data in a low-cost way,which helps to raise the awareness of electricity consumption among residential customers and reduce the wastefulness of residential electricity consumption in order to achieve the goal of energy saving and emission reduction.Aims at the problems of low accuracy and feature flooding of existing single-task non-intrusive electrical equipment detection methods based on low-frequency data,a multi-task non-invasive electrical equipment detection model based on the combination of multi-gate mixture-of-experts network(MMoE)and bidirectional long short-term memory network(BiLSTM)is proposed.The MMoE network is used to achieve soft sharing of the underlying parameters,learn the coupling characteristics between different electrical appliances,and fully exploit the load characteristics of the electrical appliances.The BiLSTM is used as a subtask layer to output the power sequences of all electrical appliances simultaneously in one model.The experimental results on UK-dale public dataset show that the proposed method outperforms the existing single-task methods in the detection indexes of various electrical appliance,which verifies that the proposed method has good electrical equipment detection performance.
non-intrusivedeep learningmulti-gate mixture-of-experts networkbidirectional long short-term memory networks
NETL
NSTL
万方数据
