基于注意力机制与ConvBiLSTM的非侵入式负荷分解

针对非侵入式负荷分解准确率低的问题，提出一种新的非侵入式负荷分解方法。首先，针对在训练时难以对含有大量数据的电器设备的工作状态做标签问题，设计了引入轮廓系数和平方误差和共同作为评价指标的k-means聚类来确定负荷状态数，构建了状态码表示所有电器的运行状态。其次，利用卷积层和双向长短期记忆网络对特征进行提取，并引入注意力机制选取对分解任务重要性程度高的电器状态码，然后，通过全连接层进行分类，得到各时刻下的状态码，进而得到各用电设备实际功率。最后利用公开AMPds2数据集进行验证，结果表明所提方法，具有较高的负荷分解准确率。

Non-intrusive load decomposition based on attention mechanism and ConvBiLSTM

Aiming at the problem that the accuracy of non-intrusive load decomposition still needs to be improved, a non-intrusive load decomposition method was proposed. First, aiming at the difficulty in labelling the operation status of electric equipment with a large amount of data during training, k-means clustering introducing the silhouette coefficient and the sum of squared error as evaluation indicators was designed to determine the number of load status. A status code was constructed to represent the operation state of all electrical appliances. Second, features were extracted by means of ConvBiLSTM(convolution layer and bidirectional long short-term memory network). Attention mechanism was introduced to select the electrical status codes with high importance to the decomposition task, and then these codes were classified through the full connection layer to obtain the status codes at each moment, and then the actual power of each electrical equipment was obtained. Finally, the public AMPds2 data set was used to verify the proposed method. Results show that the proposed method has higher load decomposition accuracy.