Weak Fault Line Selection Method For Distribution Network Based On Machine Learning

Distribution network is closely related to the national economy and people’s livelihood,and its safe and reliable operation is very important to the national economy,production and life.Due to the complicated wiring of distribution network,various equipment and loads,faults occur from time to time.In order to improve the stability of power distribution,small current grounding system is widely used in distribution network.In practical application,the single-phase ground fault is the most frequent and difficult one.When this kind of weak fault occurs,the electrical signal is weak and the amplitude of characteristic quantity is small.Therefore,it is a difficult problem to correctly detect the fault and select the fault line in time.In this paper,in the case of weak fault,the weak fault detection of each line is done to determine whether the weak fault occurs,and then select the fault line.Aiming at the problem of weak fault detection and line selection,the work is as follows(1)Using the method of line selection for weak fault detection,this paper summarizes the current research status and methods of weak fault detection in distribution network,illustrates their application scope and limitations,and then points out that the method based on single feature is not enough to deal with complex distribution network conditions.(2)When the fault occurs in the small current grounding system,there will be steady-state component and transient component,the former is easy to obtain,but relatively weak,while the latter is significant,but the existence time is shorter.Aiming at the situation that the characteristics of transient steady-state have great limitations,this paper analyzes the characteristics of transient steady-state of small current grounding fault in distribution network,and selects a number of characteristics that can detect weak fault,such as zero sequence current amplitude,zero sequence current phase,transient energy,initial traveling wave,transient input zero sequence impedance,etc,After that,the data are preprocessed to remove the redundant items,which can be used as the sample data set of weak fault detection.(3)A line selection method based on machine learning is proposed.Firstly,the sample data set is formed,and the support vector machine(SVM),random forest(RF)and Bayesian(NB)classifier are trained to establish the classifier model;After the classifier model is established,the soft voting method is adopted in this paper.According to the importance of each base classifier in the integrated classifier and the credibility of the output label,the weight of the output classification result of each classifier is given,which represents the degree of influence on the final classification result.Then the output results of all classifiers are weighted and voted,and the highest value is the output result.Then,the dynamic weighting method is used to assign appropriate weights to each classifier,and the weights assigned to each classifier in the multi classifier system are not fixed,but will change according to the changes of the input vector.The trained ensemble classifier can effectively identify the faulty lines and detect the weak fault lines when the features are not obvious.(4)Simulation results verify the feasibility of the proposed route selection method based on machine learning.Firstly,the PSCAD software is used to generate the original data,and the original data is preprocessed to form the training sample data set.Then,the python software is used to write the algorithm to train the weak fault detection model integrated by the three base classifiers,and the identification sample data set is used to determine whether the weak fault in the line can be detected,and then the fault line is selected for verification.In this paper,a machine learning based on SVM,RF and Nb classifier ensemble modelis proposed,which has the ability of cascading and parallel processing samples.Finally,themodel is implemented by python programming,and the fault features are obtained bysimulation to verify the classifier.The results show that the selected feature quantity canrepresent the fault condition of single-phase to ground fault with weak fault characteristics.The integrated classifier trained by this method has a wide range of applications,and canaccurately detect faults in a variety of working conditions with high accuracy.