基于隐马尔可夫模型的非侵入式负荷监测泛化性能改进

隐马尔可夫模型(HMM)是非侵入式负荷监测常用的算法.由于电压波动与负荷自身电气特性变化等原因,负荷的测量状态如功率可能持续变化,运行过程中出现新的状态转移,但当前基于HMM的非侵入式负荷监测方法并未考虑如何处理该情况,缺乏状态辨识与功率分解的泛化能力.针对这一问题,本文提出并构建二元参数隐马尔科夫模型(BPHMM),结合DBSCAN聚类算法,基于有功功率和稳态电流对负荷状态进行聚类,降低了因电压波动和噪声数据对负荷状态聚类结果造成干扰的可能性;改进维特比算法使其考虑到HMM模型参数更新以实现对负荷状态预测泛化性能的改进;考虑到功率的随机波动性,基于极大似然估计原理构建功率计算优化模型并实现负荷的功率分解.本文采用公共数据集AMPds2对所述方法进行验证,测试算例验证了所述方法的有效性.

Generalization performance improvement of non-intrusive load monitoring based on hidden Markov model

Hidden Markov model (HMM) is a common algorithm for non-intrusive load monitoring. Due to voltage fluctuation and changing load electrical characteristics, the measured state of load, such as power, may continue to change, and new state transition occurs during operation. However, the current non-intrusive load monitoring method based on HMM does not consider how to deal with this situation and lacks the generalization ability of state identification and power decomposition. To solve this problem, this paper proposes and constructs a binary-parameter hidden Markov model (BPHMM). Combined with DBSCAN clustering algorithm, the load state is clustered based on active power and steadystate current to reduce the possibility of interference caused by voltage fluctuation and noise data. Viterbi algorithm is improved to take into account the updating of HMM parameters to improve the generalization performance of load state prediction. Considering the random fluctuation of power, based on the principle of maximum likelihood estimation, the optimal model of active power calculation is constructed to realized the load power decomposition. In this paper, the public data set AMPds2 is used to verify the proposed method, and the test example shows that the effectiveness of the proposed method is verified.