Research On Transmission Line Fault Cause Identification Based On Multi-source Information

High-voltage transmission lines are responsible for long-distance power transmission in the power system.Compared with other equipment,transmission lines have the characteristics of large span,wide distribution,as well as complex changeable environment.Long-term outdoor work leads to the multiplicity and complexity of line faults.Fast and accurate fault cause identification after fault happening will help the dispatchers and inspectors with decision-making and fault-clearing,which reduces the restoration time and brings economic and social benefits.With digitalization in the power grid,increasing monitoring sources have provided fault data with more angles and more details for fault cause identification,but also brought the challenges of information redundancy and data fusion.Therefore,this thesis studies transmission line fault cause identification from the perspective of multi-source information fusion.Firstly,in order to tackle the information fusion challenge of multi-source fault data,this thesis clarifies the data application direction of feature layer fusion based on information fusion theory,and then summarizes typical fault feature researches in detailed division.According to data sources,data structure and processing differences,fault features are divided into transient waveform features and scene contextual features.These two types of features are analyzed and extracted to form an original feature pool based on mechanism analysis and experience summary.The combined used of waveform features and contextual features is regarded as a breakthrough for the information fusion research.In order to tackle the information fusion challenge of waveform features and contextual features,this thesis firstly introduces the concept of multi-view learning for line fault cause identification,and proposes a hierarchical multi-view feature selection method based on sparse representation.This method establishes an objective function model of sparse regression.Considering multi-view features’nature of hierarchical structure,Frobenius norm and l_2,1 norm are respectively introduced into the regularization terms to achieve high-level and low-level feature selection.Meanwhile,in order to improve the accuracy of the regression model with multi-classification one-hot labels,theε-dragging technique is introduced to the loss function to enlarge the boundary between different classes.Case studies based on real-life fault data demonstrates that it is not appropriate to directly apply single-view methods to information fusion.HMVFS can give full play to the advantages of multi view learning and effectively improve the fault classification performance,which verifies the enhancement effect of information fusion.Secondly,in order to tackle the challenges of intra-class variation and low accuracy in transmission line fault cause classification,this thesis introduces dynamic ensemble selection technology to solve the problem of insufficient label definition,and proposes a data-driven cause identification method for transmission line fault based on HMVFS and META-DES.The method starts with extracting fault data into two-view features as mentioned above,and then obtains the optimal feature combination;an ensemble classification model of META-DES framework is built based on the meta-learning theory.The meta-classifier in the model is designed to learn the performance indicators of base classifiers so the model can combine different base classifiers for every query sample to achieve ensemble learning.Case studies based on real-life fault data show that HMVFS and META-DES can effectively improve the model’s capacity in distinguishing multi-category fault causes and improve classification performance significantly and stably,which demonstrates the feasibility and effectiveness of two-view fusion and dynamic ensemble.To sum up,this thesis proposes a method for transmission line fault cause identification based on multi-view learning and dynamic ensemble learning,which provides a solution for making full use of multi-source fault data to achieve effective information fusion.The method is calculated and analyzed base on real-life fault data with better classification performance than other comparison methods,so it is able to provide reliable auxiliary decision-making information for rapid fault response and shortening restoration time.