Research On Fault Prediction Method Of Data Driven Traction Power Supply System

Traction power supply system is the energy guarantee for the normal operation of electrified railway.Maintaining the good running state of the traction power supply system is an important task for railway power supply department.Prediction of the fault condition of the traction power supply system is beneficial to understand the possible operation of the traction power supply system in the future,provides the basis for the maintenance plan formulation,and provides a reference for the targeted preventive measures.There are a lot of useful information contained in historical fault records of traction power supply system.Mining of historical information will help to understand the state law of traction power supply system and provide guidance for the future operation of traction power supply system.Based on the collected fault data,this thesis classifies the original data into two categories: historic historical fault data and historical short-term fault data.It not only considers the impact of recent data on the forecast results,but also considers the effect of historical law over the same period.The guiding role of the recorded data was to predict the faults of the traction power supply system using three improved time series prediction methods:The improved fuzzy time series forecasting method uses the triangular fuzzy number to separately process the two types of historical data to obtain two sets of fuzzy numbers.Then,the inequality equations for reducing the ambiguities are converted into linear programming equations to solve two sets of fuzzy coefficients.The two groups of fuzzy coefficients are respectively assigned a weight to obtain the final forecast result.Using the power supply system fault statistical data of a power supply segment for 11 years to verify,the results show that the measured values all fall within the prediction interval,and the model has a certain degree of practicality.The improved EWMA forecasting method takes into account the difference in the distance prediction data from different distances to predict future results.For two different types of historical data,the two sets of data are assigned different weights according to their time and distance.The recent data is far more important than time.Finally,use two parts of the weighted data to make the final forecast.Using the power supply system fault statistical data of a power supply segment for 11 years to verify,the results show that the actual measured value and the model predicted value have little difference,and the model has certain practicality.The improved grey forecasting method uses two kinds of historical data to establish corresponding models respectively,and combines two parts of forecasting results respectively to assign a weight together to obtain the final forecasting results;and finally from three aspects of residual error,relative error and level deviation.Test the forecast results.Using the power supply system fault statistical data of a power supply segment for 11 years to verify,the results show that,except for the total number of faults measured in July 2012 and the predicted value,there is a small difference between them and they have a certain degree of accuracy and can be used as reference.