Studies On Stator Single-phase Grounding Protection Of Powerformers Based On Adaptive VMD And LightGBM

With the profound adjustment of the world energy structure and the continuous strengthening of environmental resource constraints,the energy technology revolution triggered by a new round of technological revolution represented by information and digital technology focuses on promoting clean energy power generation equipment towards large capacity,high safety and high reliability.direction of development.The birth of Powerformers adapts to the development of energy transformation.Powerformer stator windings are made of XLPE cables.This special structure makes the output voltage at the machine end reach a theoretical 400kV,which can reduce or eliminate the need for booster equipment.It has significant advantages over traditional generators in terms of cost and efficiency.However,the stator single-phase grounding,as one of the most common fault types in the actual operation of Powerformers,is easy to develop into a more harmful inter-turn and inter-phase short circuit,which affects the safe and stable operation of the power system and causes huge economic losses.For a long time,how to accurately and quickly detect and isolate faulty Powerformers has been a hot and difficult research topic.For the reliability of Powerformer stator single-phase grounding protection,this paper proposes a Powerformer stator single-phase grounding protection scheme using improved variational mode decomposition and light gradient boosting machine.This paper starts from the operating characteristics of Powerformers,describes the key fault electrical parameters according to the characteristics of its stator windings,establishes a fault equivalent circuit model,and analyzes the occurrence of internal and external single-phase grounding of the generators in the power supply system of multiple Powerformers running in parallel.The zero-sequence equivalent current characteristics at the time of fault provide basic support for the protection scheme.Secondly,in view of the problem of insufficient self-adaptability of the number of traditional VMD decomposition,the paper introduces sample entropy to improve it,and uses the improved VMD to decompose the fault zero-sequence current signal detected from the Powerformer end into several intrinsic modal function.Then,this paper extracts the features of the VMD decomposition results from the perspectives of time domain,frequency domain,time-frequency domain and waveform analysis,constructs a fusion feature vector that can truly reflect the operating state of the Powerformer,and uses it as the input of the LightGBM model.A Powerformer with an internal fault was eventually identified.In addition,this paper sets up various combinations of different neutral point grounding methods,fault closing angle,fault location,and transition resistance to simulate the situation at the time of fault in actual operation,and adds different degrees of noise interference to the original signal to test all the anti-noise performance of the protection scheme is improved,and the reliability of the proposed scheme in this fault identification task is verified by comparing with the support vector machine,random forest,k-neighbor,and gradient boosting machine model in the discrimination accuracy and training discrimination time.The scheme proposed in this paper has been trained and verified on 2052 sets of simulation data sets.The results show that the scheme can effectively improve the accuracy of fault discrimination,and has better anti-noise performance than other models,which can significantly reduce the failure time effects of various fault conditions protection.