Machine Learning Based Outlier Detection Algorithm for Distributed Flexible Sensing Module with Non-stationary Multi-Parametric Data

As the most basic component of the power system, the power consumption information collection system plays a crucial role in the development of smart grid. The article constructs a distributed flexible sensing terminal application system based on AMI architecture, including power metering online monitoring, expansion module and metering accuracy inspection module. In order to cope with the detection of multi-parametric data outliers in the non-stationary environment, this paper utilizes the PCA algorithm to downscale the original data, and then combines it with the maximum likelihood method to screen the power metering feature data. The improved DBSCAN clustering algorithm is introduced to cluster the features, and the optimized local outlier factor algorithm is used to realize the design of multi-parametric data outlier detection algorithm. It is found that the mean value of the profile coefficient of the improved DBSCAN clustering algorithm is 0.705, which improves the performance by 81.23% compared with the original DBSCAN clustering algorithm, and the accuracy of the outlier detection algorithm can reach up to 0.845. Combining machine learning techniques with AMI architecture can realize multi-parametric data outlier detection in non-stationary environment of distributed flexible sensing module, which lays the foundation for optimizing grid configuration.