Data-driven train operation models based on data mining and driving experience for the diesel-electric locomotive

https://doi.org/10.1016/j.aei.2016.07.004Get rights and content

Abstract

Traditional control methods in automatic train operation (ATO) models have some disadvantages, such as high energy consumption and low riding comfort. To alleviate these shortcomings of the ATO models, this paper presents three data-driven train operation (DTO) models from a new perspective that combines data mining methods with expert knowledge, since the manual driving by experienced drivers can achieve better performance than ATO model in some degree. Based on the experts knowledge that are summarized from experienced train drivers, three DTO models are developed by employing K-nearest neighbor (KNN) and ensemble learning methods, i.e., Bagging-CART (B-CART) and Adaboost.M1-CART (A-CART), into experts systems for train operation. Furthermore, the DTO models are improved via a heuristic train parking algorithm (HPA) to ensure the parking accuracy. With the field data in Chinese Dalian Rapid Rail Line 3 (DRRL3), the effectiveness of the DTO models are evaluated on a simulation platform, and the performance of the proposed DTO models are compared with both ATO and manual driving strategies. The results indicate that the developed DTO models obtain all the merits of the ATO models and the manual driving. That is, they are better than the ATO models in energy consumption and riding comfort, and also outperform the manual driving in stopping accuracy and punctuality. Additionally, the robustness of the proposed model is verified by a number of experiments with some steep gradients and complex speed limits.

Introduction

In recent years, urban rail transit has been developed rapidly due to its superiorities of high-speed, punctuality and safety in public transportation systems [1]. Most large cities all over the world are expanding their subway systems to relieve the pressure of public transportation. The efficiency of a subway system is largely determined by the train driving strategies. In most new established subway systems, automatic train operation (ATO) systems, that control a train to accelerate, coast or brake automatically, have replaced the manual driving methods [2]. The ATO models has been developed for many years. For example, fuzzy control and predictive fuzzy control were developed to overcome the complexity of train control model [3]. Chang proposed a genetic algorithm (GA) to optimize train movements using appropriate coast control [4]. Hou proposed iterative learning control (ILC) theory to make the train tracking the given guidance trajectory more precisely by iterative learning process [5]. To overcome the traction/braking saturation of train dynamic model, Song [6], [7] designed a computationally inexpensive tracking control method in which a single-coordinate dynamic model that reflected in-train forces was derived.

It is known that subway train operation system requires multiple objectives, including operation safety, train parking error, energy consumption, and passenger comfort, etc [8]. The train control can be regarded as regression problem with several variables. The simple linear models like auto-regression models are very fast but produce poor performance in punctuality, safety, passage comfort, and energy efficiency. The neural network based nonlinear regression models, such as SVM, ANFIS, and current popular deep learning models, are too complex against the real-time response [9], [10]. Therefore, this paper develop a empirical model to find a good balance between efficiency and complexity. In other words, they are more practical for the train control. In addition, most current ATO models are developed to track a target speed curve as precisely as possible. Thus, they need to frequently adjust the control output, which will reduce the riding comfort and increase energy consumption [11], [12], [13].

A large amount of train operation data are recorded everyday by the ATO systems and manual driving. For research studies, the field data from Dalian Rapid Rail Line 3 (DRRL3) was collected. From statistical analysis on the practical data, one interesting fact is discovered that the manual driving by experienced drivers is better than the automatic driving in energy consumption and riding comfort [14]. Some preliminary works by using data mining techniques have been done to design some human driving rules [15], [16]. On the foundation of the preliminary works, the contributions of this paper can be summarized as follows: (1) Combining the driving experiences, the automatic train operation, data mining techniques, and a heuristic train parking algorithm (HPA) together, three new DTO models based on KNN, B-CART and A-CART, are proposed; (2) all the three DTO models are trained and evaluated with real-world data sets from DRRL3 to evaluate their effectiveness; (3) the robust analysis is developed to verify the performance of the models in different kinds of extreme situations. In brief, this work aims to establish effective data-driven train operation (DTO) models from a new point of data mining to improve the overall performances and robustness of subway train operations [17].

The rest of the paper is organized as follows. Section 2 describes the problem of train operations for the diesel-electric locomotives. In Section 3, the domain driving experiences are summarized, and three models, i.e., DTOK,DTOB and DTOA are developed based on KNN, B-CART and A-CART, respectively. In Section 4, a heuristic parking algorithm (HPA) will be proposed to improve the parking accuracy. In Section 5, a number of criteria are presented to evaluate the performance of the three models. In Section 6, the models are evaluated and analyzed in detail by using the field operation data in DRRL3. Furthermore, the robustness of the DTO models are verified with steep gradient and complex speed limits. Finally, a conclusion is arrange in Section 7.

Section snippets

Problem statement

In general, a train control model can be expressed asMu=M×dvdt-fr(v)-fg(s),dsdt=v,v<vl,where M,v and s are the mass, velocity and position of a train, u represents the output of the train controller and vl represents the speed limit, which is the speed that a train cannot exceed. In addition, fr(v)=αv2+βv+γ describes the resistances produced by friction. fg(s)=Mgsin(r) represents the resistances aroused by gradients, and r is the slope angle [18]. Then, Eqs. (1), (2) can be written asu=dvdt-fr(v

DTO framework

As shown in Fig. 2, the framework of DTO models is motivated by achieving a data-driven control system. The inputs consist of an expert knowledge database, an offline database and an online database. The driving experiences, data mining algorithms and a heuristic parking algorithm are developed in the second column. Combined with K-nearest neighbor, Bagging-CART and Adaboost.M1-CART, three DTO models, i.e. DTOk,DTOB and DTOA, are proposed, respectively. The DTO models are developed to achieve

Heuristic parking algorithm

In the stopping stage, we consider the usage of the precise location information provided by balises, which play an important role in European train control system (ETCS) and Chinese train control system (CTCS) [28]. The balises are usually installed and distributed with certain functions in front of stations in urban rail transit systems. When a train is achieving, these equipments can collect and transfer the precise location information rapidly and efficiently to the train. There are five

Simulation platform and evaluation criteria

The train simulation platform (TSP) is established with Matlab/Simulink. The platform includes five modules, input module, generator module, actuator module, train model module and display module. The simulation system structure of TSP is demonstrated in Fig. 4.

To evaluate the performance of the DTO models, a number of criteria will be discussed. As mentioned before, train operation is a multi-objective problem in terms of the punctuality, the energy consumption, the riding comfort, the service

Field data description and model training

To illustrate the proposed DTO models, we present numerical examples based on the actual field data from the DRRL3, which connects the center of Chinese Dalian City, Economic Development Zone and Jinzhou Zone. The length of the line is 49.15 km, and its run route is shown in Fig. 5.

In the driving process of a train, the vehicle-mounted equipments receive online information of train operations. Additionally, a large volume of data are recorded in the on-board computers from manual driving. In

Conclusion

In recent years, high-speed railways have been constructed or under construction in many countries due to the high efficiency. However, control algorithm design for high-speed train is much more challenging because of the complexity of high-speed train model and the high requirements for control accuracy. To alleviate these shortcomings of the ATO models, this paper presents three data-driven train operation (DTO) models from a new perspective that combines data mining methods with expert

Acknowledgements

This work has been partially funded by the Start Funding for Minjiang Chair Professor under grand 510146, and by the start funding of Fuzhou University under grand 510206. The Project of Fujian Province Key Laboratory of Network Computing and Intelligent Information Processing under Grant No. 2009J1007. Application Demo of Industry Network Control System in Rail Transportation, Information Technology Application “Doubling Plan” by Ministry of Industry, China.

References (28)

  • C. Chang, S. Sim, Optimising train movements through coast control using genetic algorithms, in: IEE...
  • H. Sun et al.

    Coordinated iterative learning control schemes for train trajectory tracking with overspeed protection

    IEEE Trans. Autom. Sci. Eng.

    (2013)
  • Q. Song et al.

    Computationally inexpensive tracking control of high-speed trains with traction/braking saturation

    IEEE Trans. Intell. Transp. Syst.

    (2011)
  • Q. Song et al.

    Data-based fault-tolerant control of high-speed trains with traction/braking notch nonlinearities and actuator failures

    IEEE Trans. Neural Networks

    (2011)
  • Cited by (0)

    View full text