Research On Intelligent Hierarchical Fault Diagnosis Method Of Power Grid Based On Multi-source Information

With the increasing scale of power grid,the structure of grid is more complex,and the connection between the regional power grids is closer.Thus Local network fault is more likely to cause Cascading Failures.If the Power Dispatching and Control center cannot accurately and effectively deal with the system fault,the fault range will further expand and threaten the reliable power supply and steady operation of the system.The accurate and efficient fault diagnosis system can provide auxiliary decision support for emergency control and fault recovery after failure,and speed up the process of accident handling and power supply recovery.Due to the transfer restriction of different data in time and space,the Power Dispatching and Control center cannot get the corresponding fault information synchronously.However,the existing fault diagnosis methods cannot be effectively applied to real-time emergency control after power grid failure.On one hand,they are based on the assumption that all information has been uploaded,and can only be used for the analysis after accident.On the other hand,they show low fault tolerance in the case of loss or error of alarm information and maloperation or rejection of protection and circuit breaker.To satisfy the demands of real-time demands for Dispatching and Control and ensure the reliable power supply and steady operation of the system,the paper focuses on intelligent hierarchical fault diagnosis method of power grid based on multi-source information.According to the on the characteristics and the arrival time priority of the fault information,the paper puts forward the framework of hierarchical fault diagnosis system,and makes futher research on the intelligent methods of fault region division,fault element diagnosis and fault type identification.For the first layer of the hierarchical fault diagnosis system,this paper introduces the information of switch tripping and the WAMS,and proposes a fast fault region recognition method based on the breadth first search.In order to overcome the defects of traditional network topology analysis method in the case of loss and distortion for switch information,the information in WAMS is introduced to identify the boundary circuit breakers of fault area.And then search along the direction of the outage side to quickly determine the fault area.The fault cases of IEEE 39 bus system proves the effectiveness of the method.For the second layer of the hierarchical fault diagnosis system,this paper introduces the time constraints of protection and circuit breaker action information,and proposes a fault element diagnosis method based on the temporal weighted fuzzy time Petri net.Firstly,the paper studies the construction method of hierarchical temporal weighted fuzzy Petri net of bus and line.In order to overcome low fault tolerance of traditional fault diagnosis method in the case of loss or error of alarm information and maloperation or rejection of protection and circuit breaker,the credibility of the protection and circuit breaker operation is modified by reverse and positive timing reasoning taking the fault occurrence time as the event-starting point.The superiority of the method is illustrated by the simulation example of the IEEE 39 node system,actual failure cases of power grid and the comparison with the existing fault diagnosis methods.For the third level of fault diagnosis system,this paper introduces fault recorder information and proposes the fault type recognition method based on deep belief network.In order to overcome the dependence of shallow intelligent methods on signal processing and artificial experience,and the lack of ability to extract and express the features of complex power systems,this paper takes the fault type recognition DBN model of transmission line as an example to learn and extract fault characteristics automatically from the original time-domain signal of fault recorder data.The fault feature extraction and recognition ability of the DBN model are studied and analyzed from three aspects: the reconstruction ability of the DBN network,the visual results of the feature extraction layer by layer,and the comparison of the fault type recognition results with the traditional artificial neural network.At the end of this paper,the main research work and the results are summarized,and the directions of further research work are prospected.