Research On Distribution Network Outage Prediction Based On Data Mining Technology

In recent years,with the greatly improvement of living standards of urban and rural residents,people's demand of power supply reliability is increasing.The electric companies can only passive response to the distribution network outage,which makes the distribution network reliability bottlenecks on ascension.Predicting the distribution network outage can provide the operation and maintenance decision support to the distribution network which leads the great saving of economic and social assets.Domestic related research has not yet in-depth,the research done outside the China is relatively deep but has limitations to implement.For this reason,this paper studies the method of prediction of distribution network outage based on the data mining technology.According to the actual demand of electric companies,this paper also predicted the feeder outage level and apply the feeder outage prediction results to optimize the position of rush-repair stagnation.Firstly,this paper analysis the feeder outage influencial factors comprehensively,make deep investment in distribution network information system of a city and to extraction of the necessary data for feeder outage prediction.It will direct the foundation for feeder outage prediction.Aiming at solve the quality problems of raw data,a package is developed to clean,integrate and transform the raw data.Propose a new method of outlier detection based on clustering.In this method,particle swarm optimization algorithm is used to optimize the clustering center and optimal number of clusters is determined by outline value method.This method can effectively improve the clustering effect,accurately remove the outlier samples and avoid the negative impact on the prediction model caused by the outlier samples.Secondly,the redundant and non-strong outage relevant feature variables are eliminated and a set of outage relevant feature variables is preliminarily screened based on the method of data exploration and analysis.And feature selection method is used to screen the best feature subset from the outage relevant feature variable set.Then the best feature subset is used as the input of feeder prediction model and avoid the decrease of model prediction accuracy caused by improper input variables selection.Thirdly,this paper divides the number of monthly outage of the feeder according to the actual demand of electric companies and model optimization requirements.And puts forward the method of predicting the monthly outage level of the feeder.The method chooses the random forest algorithm to construct the feeder outage prediction model because of the random forest algorithm has many advantages,such as small number of adjustment parameters,upper limit of generalization error and avoid overfitting.In order to reduce the generalization error of the model,the classification tree algorithm of the random forest algorithm is preferred and the key parameters are optimized.The model realizes the prediction of feeder monthly outage level and the prediction accuracy rate is up to 92.92%.In addition,through comparing with the C4.5 decision tree,SVM and ANN algorithm,it shows that the feeder outage prediction model built by random forest algorithm is obviously better than the model built by the three classification prediction algorithms.It is proved that the random forest algorithm is effective and accurate in the construction of feeder outage prediction model.In the end,considering the arrival time to repair and other constraints,regarding the minimum of total distance between rush-repair stagnation point and its each feeder as the optimization objective function,construct the rush-repair stagnation point position optimization model based on the results of the feeder outage prediction.Considering that most of modern optimization algorithms' parameters are difficult to select,a plant growth simulation algorithm with low dependence on initial parameters is introduced to solve the optimization problem of rush-repair stagnation point position.Experiment indicates that the new method of applying the feeder outage prediction results to the optimization of rush-repair stagnation point position compared with the traditional methods can effectively improve the repair efficiency.