非侵入式负荷监测是智能用电的未来发展趋势，其中负荷的分解与辨识是实现该技术的重要环节。鉴于变分模态分解（variational mode decomposition,VMD）在信号处理方面的优势，提出一种基于VMD-FastICA（variational mode decomposition and fast independent component analysis)和VMD-Entropy-PSOSVM(variational mode decamposition-entropy-particle swanm optimization fo optimizing support vector machines)的负荷识别算法。该方法利用VMD对总负荷功率信号进行分解得到多个模态分量（intrinsic mode functions,IMF），再依据峭度准则和奇异值分解对分解后的模态分量重构，将单通道盲源分离虚拟成多通道盲源分离，输入快速独立分量分析（fast independent component analysis,FastICA）进行负荷信号分离，求取分解负荷波形模态分量的能量与能量熵。构建多维特征矩阵输入建立粒子群算法优化支持向量机（particle swarm optimization for optimizing support vector machines,PSO-SVM），进行负荷的分类辨识。采用开源数据集(reduced electricity dataset, REDD)对实验算法进行仿真，与其他算法相比，验证算法在分解和识别上都具有较好的效果。
英文摘要:
Non-intrusive load monitoring is one of the important technologies of intelligent power consumption, among which load decomposition and identification is an important link to realize the technology. In view of the advantage of variational mode decomposition (VMD) in signal processing, a load identification algorithm based on variational mode decomposition and fast independent component analysis (VMD-FastICA) and variational mode decamposition-entropy-particle swanm optimization fo optimizing support vector machines (VMD-Entropy-PSOSVM) is proposed. The total load power signal is decomposed using VMD to obtain multiple intrinsic mode functions (IMF), and then, the IMF is reconstructed based on the cliff criterion and singular value decomposition to virtualize single-channel blind source separation into multi-channel blind source separation into fast independent component analysis （FastICA） for load signal separation. Then, the energy and energy entropy of the modal components of the decomposed load waveform are obtained, and the multi-dimensional feature matrix input is constructed to establish a particle swarm optimization-support vector machine particle swarm optimization for optimizing support vector machines (PSO-SVM) for classification and identification of the load. The experimental algorithm is simulated using the reduced electricity dataset (REDD), and it is verified that the algorithm has better results in both decomposition and recognition compared to other algorithms.