基于多特征序列融合的负荷辨识方法

针对当前利用低频采样实现非侵入式负荷辨识存在的准确率低的问题,提出了基于多特征序列融合的负荷辨识方法。该方法首先建立负荷存在可能性的整数规划模型,进行初辨识以降低负荷辨识的维度。然后,根据滑动窗口算法获得组合功率序列和原始功率序列,提取其统计特征和奇异值特征,进而利用概率神经网络获得隐马尔可夫模型的观测值序列。同时,利用隐马尔可夫模型对负荷序列信息进行融合,计算观测序列和组合功率序列之间的相似度,从而完成在低频采样下的负荷辨识,并获得各个家用负荷的耗电量。最后,通过单负荷辨识、多负荷辨识、不同采样率辨识和各居民用户负荷辨识的仿真实验,得到负荷准确率和辨识精度的平均值均在85%以上,证明了所提算法的合理性和即时性能够达到在低频采样下负荷的辨识要求。

Load Identification Method Based on Multi-feature Sequence Fusion

Aiming at the problem of low recognition accuracy for non-intrusive load identification method at the low sampling rate, a load identification method based on multi-feature sequence fusion is proposed. First, an integer programming model is developed to solve the possibility of load existence, thus reducing the dimensions of load identification process. According to the sliding window method, the combined power sequence and the original power sequence are obtained, from which the statistical characteristic value and the contour singular value are extracted. Then the probabilistic neural network(PNN)is used to obtain the observed sequence of the hidden Markov model(HMM). Meanwhile, the information of load sequence is fused by HMM, and the similarity between the observed sequence and the combined power sequence is calculated. Thus the load identification at a low sampling rate is completed and the power consumption of each household load is obtained. Finally, through the simulation experiments of single load identification, multi-load identification, load identification with different sampling rates and load identification of each user, the average results of the load accuracy and the identification accuracy are more than 85%, which has verified that the rationality and immediacy of the proposed method can meet the requirement of load identification at the low sampling rate.