适于低功率状态的多特征融合负荷分解方法OA

Non-intrusive load disaggregation is a way to analyze energy consumption characteristics and load profiles of individual appliances in power systems,which has been widely applied due to its advantages of low cost and easy implementation.Existing load disaggregation methods often attempt to obtain reliable semantic information by stacking more and more pooling and downsampling processing.However,the tiny target features with less noticeable power fluctuations may gradually being lost in the process of the forward propagation,and thus reducing the method's performance in decomposing low-power states of multi-state appliances.Therefore,a load disaggregation method combining multi-feature fusion is proposed in this paper,which designs skip-connection feature fusion between encoder and decoder,as well as between decoder interiors,to fully obtain power fluctuation features of different scales.By combining contextual information and multi-scale spatial pattern characteristics,the proposed method improves the model's perfermance for pattern recognition of target appliances.Experimental results demonstrate that the proposed method reduces the mean absolute error(MAE)of load disaggregation for all target appliances and multi-state appliances by approximately 36%and 47%respectively,while evaluating unknown household data.Additionally,the load was successfully decomposed to low-power states.The experimental results verify the effectiveness of the proposed approach.