Island detection method for distributed generation with online self-learning ability

Abstract

The invention relates to a distributed generation island detection method with an on-line self-learning ability. The distributed generation island detection method includes the following steps that (1) original samples are obtained on line by using a microgrid SCADA system; (2) the original samples are sampled again in a self-adaptive mode by using an on-line clustering method containing micro clusters; (3) weights of all sample subsets are calculated according to a multiple-classifier model, the superior is selected, the inferior is eliminated to weed out sample sets with the classification accuracy rate lower than a set threshold value, and preferred sample sets are obtained; (4) a classifier model is generated through on-line training according to the preferred sample sets; (5) by using the classifier model obtained through on-line self-learning, the classifier model used for real-time island detection is updated in an asynchronous mode. Compared with the prior art, the distributed generation island detection method has the advantages of being high in accuracy and robustness, good in stability and adaptability and the like.

Description

technical field [0001] The invention relates to a distributed power generation technology, in particular to a distributed power generation island detection method with online self-learning capability. Background technique [0002] Distributed generation (distributed generation, DG) is an important part of smart grid. Considering the safety of personnel and equipment, stable operation of the system, and power quality, distributed generation is generally required to have an island detection function. Since data mining technology can effectively solve the problem of threshold setting in island detection, the island detection method based on data mining has been paid more attention in recent years. In existing studies, the classifier models for island detection are all obtained through offline training. Due to the fluctuation of distributed power sources and local loads, as well as the changes in the topology and operation mode of the distribution network containing distribute...

AI Technical Summary

Problems solved by technology

Due to the fluctuation of distributed power sources and local loads, as well as the changes in the topology and operation mode of the distribution network containing distributed power sources, the statistical characteristics of the samples will change unpredictablely with the passage of time or changes in the environment. It will cause the classification accuracy of the classifier model obtained offline to gradually decrease

Images