Islanding Detection For Dc Microgrid Based On Random Forest Classification

With the rapid economic growth and the steady development of society,the demand for energy continues to increase,and the resulting environmental pollution problems are becoming more and more serious.In today's world still supported by non-renewable energy as the main economic development,the energy crisis and the environmental crisis have become important issues that all countries need to face.In this context,research on the microgrid is of great significance to improve the current energy structure and the efficiency of distributed power sources.At present,micro-grids mainly exist in the form of AC.With distributed power sources and a variety of DC power-using devices being widely used in micro-grids,DC micro-grids are developing rapidly,and DC micro-grids have ushered in a broad development space.The operation mode of the DC microgrid is divided into grid-connected and isolated islands.After the circuit breaker between the microgrid and the large power grid is tripped,an island is formed.The operation of the island can be divided into planned islands and unplanned islands.Unplanned islands will cause some harm to users or systems.When the distributed power generation system is connected to the grid,it is in an island state,which affects the safe and normal operation of the power system.Anti-island equipment must detect islands within an acceptable time limit.Therefore,accurate detection of islands is a necessary condition for safe and stable operation of DC microgrids.This paper has carried out related research around the operation of DC microgrid islands.The traditional island detection methods mainly include local active method,local passive method and remote method.These methods have the problems of large detection dead zone,affecting power quality of inverter output,high cost and complicated design.Data mining technology mainly has two functions: one is to query the historical operation information of the power system,and the other is toestablish potential links between the query data to solve the prediction and decision-making problems of isolated islands.This paper proposes a DC microgrid island detection method based on random forest classification.First,a simulation model of the DC microgrid was built to obtain the status information of the DC bus side voltage,current,output active power,etc.in the system grid-connected and island operating modes,and the data was cleaned;then,the key features reflecting the island micro-grid operation of the DC microgrid were extracted To generate feature vector sets;finally,a DC microgrid island detection method based on random forest classification is proposed.The results show that this method can improve the DC microgrid island detection accuracy.First,this article introduces the data preprocessing work,including data cleaning,feature extraction and feature index analysis.Wash a large amount of original data and extract features,use the method of calculating the Euclidean distance to eliminate redundant data,repeat the combination of data,select the voltage,current,output active power of the DC bus side,and the first-order backward of the three Six island characteristic indexes,such as difference,are used as detection features.At the same time,feature index analysis is performed on the six detection features,and each feature is given reasonable weight to generate a feature vector set.Secondly,the basic principles of the random forest classification method are studied,including the random selection of training sample subsets,the construction of CART decision trees and the voting of island detection results,in terms of the random selection of training sample subsets,the ratio of training and test sets Adjustments were made to find the optimal ratio to make the detection results more accurate and reasonable.The influence of random forest trees on the detection results was studied.The best tree was selected and an island detection based on weighted random forest classification to establish a DC microgrid was constructed.model.Then,the software was used to build a DC microgrid simulation including photovoltaic power generation system,wind power generation system,lithium battery energy storage system,AC load,DC load,and AC main grid,and the simulation was used to generate a large number of DC microgrid operating voltage,current,Basic state information such as output active power,AC side voltage,current,output active power,output reactive power,etc.,pre-processes the original data,and builds a random forest classification model on this basis to achieve accurate detection of islands.Finally,in the analysis of experimental results,the following experiments were carried out.The random forest classification model without considering the weight of the characteristic index was compared with the decision tree classification method.The random forest classification model with considering the weight was compared with the random forest classification model without considering the weight of the characteristic index.Considering the weight of the random forest classification model,compared with traditional detection methods,can detect islands more accurately,which has certain scalability and practical significance.