Research On Data Mining Technology And Application Of Electrified Railway Measured Data

In the past 40 years of reform and opening-up,China’s railways,especially the high-speed railways,have achieved rapid development and become a shining "Chinese business card." From a single route to the whole country,from "catching up" to "leading",from "being able to run" to "running well",the rapid development of high-speed railway has also brought power quality problems such as harmonics and negative sequence.Furthermore,as a special load of the power system,the safety,stability and reliability of high-speed railway must be ensured,while the impact of its energy consumption,environment and economy should not be ignored.Therefore,it is necessary to research on traction load forecasting,which plays a basic role in these studies.In this paper,based on the actual measured load,the mathematical characteristic indexes and the characteristic quantities of electrified railway are calculated when analyzing its digital characteristics.And then,the histogram is used to describe its probability distribution.In order to fit the probability distribution of the measured load,many estimation algorithms of data fitting are used based on parameter estimation and non-parametric estimation theory.The coefficient of determination is taken into consideration so that the results of different algorithms can be compared and the merits and applicability can be evaluated.A large number of fitting experiments show that each fitting algorithm has its advantages and disadvantages,but the fitting accuracies of all algorithms are over 85%,which meets the accuracy requirements in engineering applications.Clustering,one of the data mining algorithms,is used to tell the difference between these measured samples and it includes hard clustering,soft clustering and neural network clustering.With the characteristic quantities of traction load calculated,100 sets of measured samples are divided into seven categories by clustering,and several indexes such as silhouette values and pseudo-F statistics are used to select the best algorithm by evaluating the clustering results.The clustering results show that the K-means algorithm has the best indexes and it is the most suitable algorithm for traction load clustering.Therefore,the traction load model database is established based on the results of K-means algorithm.In the process of load forecasting of the new traction substation,it is necessary to take some practical factors into consideration,such as the limited technical information and incomplete load characteristics.Therefore,based on the clustering results,classification algorithms are applied to classifying the new load so that it can be matched and distributed into the model database.Specifically,based on the design information provided by Railway Group,the characteristic quantities of newly-built railway are calculated as much as possible,and they are treated as boundary conditions for matching and classification.Meanwhile,there are several indexes which can be applied for evaluating the results,such as Recall,Precision and Accuracy.According to the calculation of various indexes,the Native Bayes algorithm has the best effect,and it is an algorithm with 100% accuracy.Therefore,in this paper,it is the proper algorithm for matching the new load with the model database.After the new load is reasonably matched and divided,characteristic quantities such as the probability distribution of some known loads which belong to the same cluster can be treated as a reference.According to the aforementioned estimation fitting theory,the charged traction load current can be estimated,and the load process with current change can be predicted by Monte Carlo sampling.Based on the forecasting load,the temperature rise and life loss model of the traction transformer are also established in this paper,and they are differentiated so as to facilitate the practical application.The example proves that by establishing these models,the capacity of the new traction transformer can be optimized.