Research On Weak Fault Identification Of Distribution Network Based On Artificial Intelligence

The reliability of power system is one of the most concerned problems about the safety of distribution network.Because the distribution network is the end of the system to supply power to the users,it has the characteristics of wide area coverage,complex lines and many operation modes,resulting in faults in the distribution network frequently.Therefore,in the actual system,rapid fault detection is an important way to reduce times of fault.However,a series of weak fault problems with weak fault characteristics,such as single-phase grounding fault of small current grounding system,high resistance fault caused by various grounding media and intermittent arc fault caused by cable aging,are the most difficult problems to deal with.Therefore,this paper mainly studies the identification of weak characteristic faults in distribution network.The main work of this paper is as follows:Firstly,this paper defines the weak characteristic fault,analyzes the zero sequence current characteristics of weak characteristic fault in medium voltage distribution network under two grounding modes,including single-phase low resistance grounding fault and single-phase high resistance arc grounding fault.At the same time,a medium voltage distribution network model is built on PSCAD,and the mathematical models of low resistance fault and high resistance arc grounding fault are built to explain the fault mechanism.Combined with zero sequence current and three-phase voltage waveform,the characteristics of the two fault types are compared and analyzed.Then,this paper proposes a weak characteristic fault identification network based on LSTM,establishes the mapping relationship between zero sequence current information and fault types,and proposes a transfer learning method based on small samples,which makes the identification network more meaningful in engineering.In the construction of data set,the transition resistance,fault initial phase angle,fault distance and other fault parameters are fully considered,so that the network has a certain tolerance to various fault conditions.In this paper,LSTM network is trained in ungrounded neutral system,and transfer learning is carried out in arc suppression coil system.In the process of transfer,the actual needs of small samples and low frequency sampling rate are considered,and the generalization ability of the network is verified.Finally,this paper proposes a multi-task-learning method based on LSTM,which realizes the same identification network to complete the two sub-tasks of fault type and fault area discrimination at the same time.This method effectively reduces the training time,and improves the fault detection rate by using the correlation between data.At the same time,the zero sequence current data of multiple distribution systems are integrated,and the types of samples are increased,so as to improve the generalization ability of the model.To sum up,aiming at the problem of weak feature fault identification in distribution network,this paper investigates and analyzes the existing research results and shortcomings.On the basis of modeling and fault characteristic analysis,LSTM neural network is used to propose a weak fault identification network based on transfer learning and multi-task-learning.At the same time,it has a great engineering application prospect.