Early Warning Of Grid Tower Failure Under Typhoon Disaster Based On Improved Random Forest

The typhoon is likely to cause failures such as barrows,broken bars and high-voltage trips in the distribution network towers,which may endanger the normal operation of the distribution network and cause a large-scale blackout accident.In the process of typhoon disasters,predicting the fault before it occurs,can provide reference for repairing materials,personnel arrangements and accidents after faults,reducing fault losses.The existing typhoon-induced grid fault prediction methods mainly include fault prediction based on dynamic monitoring of typhoon wind circle range and line fault probability assessment based on typhoon disaster mechanism model.However,due to the low precision of the typhoon wind circle monitoring,it is difficult to give the fault prediction of the distribution network line;the disaster mechanism model also needs more accurate local data collection,and the disaster causing mechanism is complicated or even unknown,and it is difficult Applicable to fault prediction of distribution towers.Considering that a large amount of historical monitoring data accumulated in the typhoon monitoring system during the operation of the distribution network is not fully utilized,and the typhoon warning of the power grid equipment belongs to the second classification problem,it is suitable to use the data mining method to model the failure rate.Therefore,this paper combines historical typhoon information with grid fault information,and establishes a grid typhoon fault warning model based on data mining method.The main research contents of this paper are as follows:Firstly,the fault mechanism model of the tower is analyzed,and the main factors affecting the fault of the tower are determined.Then,a large amount of typhoon information stored in the typhoon monitoring system and the latitude and longitude information of the weather station and the tower are extracted for preprocessing,and the model is generated according to the disaster data of the grid over the years.The original samples needed are based on the partial mutual information method to screen the original samples for feature quantity,and select the optimal influencing factors to form the sample set required by the model.Secondly,according to the category data imbalance problem of grid typhoon disaster samples,the existing unbalanced data processing methods based on data plane and algorithm level are analyzed,and the balanced data set is generated based on One-sided selection algorithm and Borderline-SMOTE hybrid sampling method.Thirdly,a typhoon warning model based on mixed sampling samples and artificial neural networks is established.Firstly,the back propagation algorithm is used to train BP neural network to adjust the weight of each layer to establish the typhoon warning model of BP neural network.Then the typhoon fault warning model based on generalized regression neural network is established,and the average absolute error is used to find the optimal smoothing of generalized regression neural network.Factor parameters,and finally use the mixed sample samples to compare the prediction effects of the two on the same data set.Fourth,a typhoon warning model based on cost-sensitive learning and random forest algorithm is established.Firstly,the cost function is determined by the sample distribution of the instance.Secondly,the decision tree is generated by using the error classification cost reduction value instead of the traditional decision tree splitting algorithm.Then,the classification performance of the decision tree is judged by using the out-of-bag data of each decision tree.Finally,the weight voting method is adopted.Determine if the model issues a warning of failure.The simulation results show that the model has higher prediction accuracy than the artificial neural network and can provide a reference for the typhoon in the coastal area.