Protection For Powerformers Based On S-transform And Random Forest Algorithm

Powerformer can be directly connected with the high-voltage transmission network without the step-up transformer compared with the conventional generator.It has the advantages of high operating efficiency,small floor space and good environmental benefits.However,the existing protection methods and devices based on single fault characteristic when multiple powerformers are operated in parallel are susceptible to various external factors,and the selectivity and sensitivity are poor,and the combination of various fault characteristics can be to complement each other.Therefore,based on the S-transformation and random forest algorithm theory,the fusion of various fault features is expected to improve the performance of powerformers stator single-phase-to-ground protection.In this paper,the following main contents are used to study the principle of selective protection of powerformers.Firstly,the research status of generator single-phase-to-ground protection based on different principles at home and abroad is expounded,and the difficulties in this field are pointed out.Secondly,the paper analyzes the internal fault model of powerformers and the zero-sequence current distribution characteristics under different fault points around the difference between high-voltage generator and traditional generator stator winding structure.Thirdly,this paper introduces the S-transform time-frequency analysis technology and the basic principle of random forest algorithm.Then,based on the existing protection principle,this paper applies refining.In the fault characteristics of powerformers protection,and under a single fault condition,the single fault feature quantity is simulated and analyzed one by one.It is pointed out that a single protection criterion only uses part of the fault information for fault identification,and is not applicable to complex fault conditions.With the help of the multi-decision tree model of random forest algorithm,the protection discriminant model is established by using the training samples constructed by the fusion of multiple feature quantities,and proposes a new high-voltage generator protection method under neural network.Finally,with ATP-EMTP and MATLAB software,the proposed protection scheme is used for simulation verification and applicability analysis.The discriminant situation of three typical neural networks under different noise influences and the applicability of three different principle methods are compared.The simulation results show that the proposed method has higher accuracy of fault identification under different fault conditions,which makes up for the deficiency of single fault feature quantity discriminant method,and does not need to set the setting value,which effectively reduces the impact of system operation mode on protection.The further development and operation of powerformers have created feasible conditions.