Prediction Of Low Voltage Trip In Distribution Substation Area Based On Random Forest Theory

As the basic power supply unit for power grid connection users,the distribution substation area has a critical impact on the power supply reliability of the power grid.With the continuous deepening of China’s requirements for the construction of intelligent distribution networks,the intelligence level of the distribution substation area also determines the intelligence level of the entire distribution network.In the existing researches,scholars mainly forecast the operation status of distribution substation area based on the heavy overload and load conditions,and there are not studies on low-voltage trip faults in distribution substation area.With the gradual application of big data information technology and artificial intelligence technology in the power industry,it has accumulated a large amount of valuable data in the aspects of power grid operation,equipment condition monitoring,and customer electricity consumption information collection.Combining these data resources and related technologies,this paper innovatively builds a model of low-voltage trip faults prediction in distribution substation area.Based on the summary of related research,this paper proposes a pred iction model based on random forest theory based on the characteristics of low-voltage trip prediction in distribution substation area.The main work is as follows:First of all,this article extracts from multiple data sources the data required for the model of prediction,and performs preliminary visual analysis on the obtained data.From the extracted data set,it selects the characteristic variables which affecting the low-voltage trips in distribution substation area.Those variables are analyzed by using Pearson correlation coefficient method and kernel density estimation,and the correlation between the characteristic factors and the trip event is obtained initially.Secondly,due to the time-sequence characteristics of low-voltage trip faults,this paper uses a sliding time window method to divide the original data set into the training set,the test set,and the valid set.For the data quality problems in the training sample set,an isolated forest algorithm is used to process for outlier rejection.In view of the imbalance of the trip faults samples,which greatly affects the performance of the prediction model,this paper innovatively proposes a mixed resampling method based on the combination of SMOTE-NC oversampling and NCL＿Plus undersampling.The mixed resampling method performs sampling processing on the training sample set that has undergone outlier processing to obtain new training samples with a more reasonable sample.Then,the new training set is substituted into the prediction model based on random forest theory for training and the test set is predicted.In order to get better model performance,this paper uses Bayesian Optimization algorithm to quickly find the optimal parameters of the prediction model.The comparison results with other models show that the prediction model proposed in this paper has better prediction performance.Finally,the prediction model proposed in this paper is used to predict the trip probability of the valid set.By analyzing the difference of the prediction model result under different probability thresholds,a method for assessing the trip risk level of the distribution substation area is established,which can provide references for power supply managers to a certain extent.