Study On Protection Scheme Of DG Distribution Network Based On Region Partition

With the development of distributed generation(DG)technology matures, the flexibility of the network power supply is improved, which in a certain extent alleviated the pressure of the network power supply, but the trend of network is no longer one-way flow also followed. Taking into account the future network structure more flexibility, in the face of failure when the DG "one size fits all" is not a perfect solution.In this paper, we consider using layers of simplification, step by step decomposition to form a new protection scheme, the regional division, the intelligent equipment unit and neural network blend together in application to the protection scheme. The main contents are as follows:The characteristics of typical distributed power supply are analyzed, and the influence of DG on the power flow distribution and short circuit current of the system is analyzed according to the location of the distributed power distribution network. Consider the distribution feeder type structure characterized by tree model, the distribution network tree analogy to change, then the similarity of the feeder structure transformation for multi tree similarity problem, select the appropriate constraints, given general suitable of islanding process. After the hierarchical network decomposition tree model of distribution network adding weights to search, in order to get a reasonable islanding scheme, paving the way for the subsequent processing of the typical structure of feeder automation.Expounds the existing single point information protection configuration can’t solve the problem of DG trend reverse flow, adopting advanced feeder in the division unit and intelligent switch more information and application in the region, form the coordination between the various intelligent unit action strategy, communication technology is validated by computer simulation interactive solutions whether meet the requirements.According to the regional division to verify the derived results can be locked in a fault section fault, but the fault section often contains a plurality of feeder bus, by using BP(back propagation) neural network second judgment to the locked section, to determine the fault location specific feed line and fault type. In sample selection, rely on the amountof current for different faults have different characterization and sequence component is not affected by the oscillation system characteristics will meet network adjustment decisions and a large number of samples is substituted into the training, in order to get a reasonable weight value and threshold value matrix in this paper deals with the problem.Finally, the training results are verified by the prediction of the sample data.