Research On The Precise Parking Problem Of Subway

China is the most populous country in the world and the third largest country in the world by territory.Therefore,urban transportation is very important for China.According to the 2018 Statistical Report of the China Urban Rail Transit Association,as of the end of 2018,there were 185 urban rail transit operating lines in 35 cities in China(excluding Hong Kong,Macao and Taiwan)with a total length of 5761.4 kilometers.There are 16 cities with 4 or more operating lines,and 3 or more interchange stations.Networked operations have been implemented,and subway operating lines are 4354.3 kilometers.However,subway operations in various cities are trending topic on Weibo due to the inability to accurately parking.Relevant departments and subway operators have also been seeking breakthroughs in subway precise parking technology in recent years.Therefore,the issue of accurate subway parking has naturally become a hot research topic in the field of rail transit.The idea of studying this problem in the engineering field is to consider complex modeling and calculations,and use a large number of additional equipment,through continuous debugging of equipment parameters,sensors,etc.to achieve precise parking.However,these methods all rely on the establishment of physical models and adjustment of equipment parameters,and ignore the learning and modeling of train operating data itself.Therefore,this paper uses the actual data of subway operation in a city to consider the problem of precise parking from a statistical perspective.By analyzing and processing the data,the factors that affect the precise parking of the subway are obtained,and a prediction model is established to know the parking situation of the train in advance.For trains predicted to be inaccurate,this paper proposes a heuristic precise parking algorithm to regulate the speed of the parking braking process.The main work of this article is as follows: 1.Since the train operation data is collected by sensors in time series,in the datapreprocessing stage,this paper first performs data segmentation,and divides thdata into the driving process of the train from the outbound station to the targetstation.Second,due to the instability of the signal data,this article removesoutliers in the original data through rule-based and algorithm-based methods.2.Before carrying out the model research,this article first studies the factors thataffect precise parking.By consulting the literature,this paper has reached theconclusion that the speed and position-related variables when entering the parkingbrake are the key factors affecting accurate parking.By analyzing the originaldata and using the survival analysis theory,it is concluded that there is acorrelation between signal disturbance and precise parking.The alarm datacorresponding to the instance of signal disturbance was further carried out,and itwas found that the alarms were mostly faults of the speed measurement systemand the position sensor.3.In order to predict train arrival and stop conditions before the train enters theparking brake stage,this article first performs feature engineering.Through theprevious stage of research,30 features are constructed,and the remaining 15relevant features are left after feature selection;Since the inaccurate train arrivaland departure time is a small probability time,the positive and negative samplesare extremely imbalanced,and the imbalance ratio is about 500.Therefore,in thispaper,the idea of bagging integration is used to build a balanced data set ofmultiple positive and negative samples.Different sub-models,and then use thevoting or averaging method to get the final prediction result.With regard to theselection of the base model,this paper attempts a logistic regression model,arandom forest model,a GBDT model,and an XGBoost model.Among them,therandom forest has the best effect.Its AUC is 0.8001.The feature importanceranking of the model output is also the same as that of the previous stage.Theconclusion is the same.In addition,in order to better intervene the train,the paperalso predicts the parking position of the train.4.Before the train enters the parking brake phase,it is predicted whether the trainwill stop accurately at the station.According to the research in the previous stage,this paper constructs 30 features,and after the feature selection,there are 15related features remaining.Because train arrival and stop inaccuracy is a smallprobability event,the positive and negative samples are extremely imbalanced,and the imbalance ratio is about 500.Therefore,this paper uses the idea ofbagging integration to construct multiple data sets with balanced positive andnegative samples,and establishes a sub-model under each data set.Then use thevoting / averaging method to get the final prediction result.With regard to theselection of the base model,this paper attempts logistic regression model,randomforest model,GBDT model,and XGBoost model.The AUC of the random forestis 0.8001,and the feature importance ranking of the model output is alsoconsistent with the conclusions of the previous stage.To sum up,this paper has realized the research of train precise parking based on the data itself.The prediction module and precise parking algorithm module can complement the existing ATO system.