基于欠定盲源分离模型的负荷分解方法研究

Research on load decomposition method based on under determined blind source separation model

大规模分布式能源并入电网,其波动性与随机性对网侧的安全稳定带来挑战。而柔性负荷的合理调控可促进需求侧响应,柔性负荷的分解与识别是实现源荷积极互动的前提和关键。传统非侵入式负荷监测需要先验信息,在实际中较难获取。为此,提出一种基于欠定式盲源分离模型的负荷分解方法,利用开源数据集REDD进行实例验证分析。首先,对功率序列进行邻点做差处理,构建负荷功率与时间分布特征,通过聚类识别出独立负荷个数;随后,采用概率归一思想求解欠定混合矩阵,将独立负荷有功功率矩阵的输出转化为几种概率事件;最后,采用盲源分离模型将总功率信号分解为各独立负荷功率的叠加。实例分析结果表明:所提方法具有普适性,分解精度高,可满足实际需求。

To achieve non-invasive monitoring and identification of household loads and lay the foundation for source load interaction of flexible loads,this paper proposes a load decomposition method not relying on prior information.The load of household users is regarded as an unknown source signal while the usage of the load is treated as an unknown superposition method.Multiple monitoring data of household electricity meters are employed as observation signals to build a blind source separation model for load decomposition.Blind source separation of household loads is achieved based on multi period power data of smart meters.Load switching is identified by subtracting adjacent points in the time series of electricity consumption.Two-dimensional distribution characteristics of load power and usage time are built.Load features are clustered and intra cluster and inter cluster dissimilarity of load features is calculated.Contour coefficient method is employed to determine the number of independent loads.Based on the built potential independent load power matrix,the usage probability of independent loads is calculated under different combinations and the true active power of independent loads is selected according to the probability normalization idea.By combining the blind source separation model,the disconnection situation of each load in different time periods is solved.Thereby,the total power signal is decomposed into the superposition of independent load powers.Non-invasive load decomposition is achieved based on meter data.Tests is conducted by using the open-source dataset REDD.Our results show the proposed method accurately identifies the number of loads and decomposes different loads for different load combination scenarios without utilizing prior knowledge.The average absolute error of load decomposition is no more than 6.8%,and the accuracy of disconnection recognition is above 0.77.These data demonstrate that our method delivers fairly good load decomposition performances and achieves blind source separation of household user loads,laying a solid foundation for flexible load recognition and regulation.