基于特征融合与深度学习的非侵入式负荷辨识算法

针对使用单一设备特征进行负荷辨识存在的局限性,提出了一种基于特征融合与深度学习的非侵入式负荷辨识算法。通过分析设备的高频采样数据提取了V-I轨迹图像特征与功率数值特征。利用人工神经网络的高级特征提取能力,实现了V-I轨迹图像特征与功率数值特征的融合。最后以复合特征作为设备新的特征训练反向传播(BP)神经网络进行非侵入式负荷辨识。使用PLAID数据集对算法辨识效果进行了验证,并对比了不同分类算法对特征融合的有效性与负荷辨识能力。结果表明,该算法利用不同特征之间的互补性,克服了使用V-I轨迹特征无法反映设备功率大小的缺点,从而提高了V-I轨迹特征的负荷辨识能力,并且在嵌入式设备中的运算速度为毫秒级。

Non-intrusive Load Identification Algorithm Based on Feature Fusion and Deep Learning

Aiming at the limitation of using single equipment features for load identification, a non-intrusive load identification algorithm based on feature fusion and deep learning is proposed. Firstly, V-I trajectory image features and power numerical features are extracted by analyzing the high-frequency sampling data of the equipment. Then the fusion of V-I trajectory image features and power numerical features is realized by using the advanced feature extraction ability of artificial neural network (ANN). Finally, the back propagation (BP) neural network is trained to identify equipment by using fusion feature as the new feature of the equipment. The PLAID data set is used to verify the identification performance of the algorithm, and the performances of different classification algorithms are compared for feature fusion and load identification ability. The results show that the proposed algorithm makes use of the complementarity of different features, overcomes the disadvantage that V-I trajectory features cannot reflect the power of the equipment, and improves the load identification ability of V-I trajectory features. In embedded devices, the computing speed of the proposed algorithm can reach the millisecond level.