Train Energy-Saving Scheme Optimized On Case Intelligence with Synthesis-Reasoning Technology in Urban Rail Transit

Abstract

Train energy consumption in URT has been attracted much greater concerns for it becomes more serious with the large scale operation and expansion of operation network. One of the important ways for energy-saving propulsion is to find the energy-efficient train speed curve, which is a complicated CSP (constraint satisfaction problem) with uncertainty, and cannot be solved effectively with such inconsistent constrains. The case intelligent based on CBR (case-based reasoning) is proposed in this paper for its problem-solving ability, for which the domain expertise is rich while rule knowledge deficient, to construct a flexible system integrated with efficient machine learning components and acquire the train operation preferences from the former stored cases. The experiments testing on the spot indicates that the system performs well in synthesis-reasoning, which can conquer the complexity and uncertainty of real problem from both RBR (Rule-based reasoning) and CBR, to minimize the energy consumption for train traction with punctuality and safety demands.

Full text

IOP Conference Series: Materials Science and Engineering
PAPER • OPEN ACCESS
Train Energy-Saving Scheme Optimized On Case Intelligence with
Synthesis-Reasoning Technology in Urban Rail Transit
To cite this article: Jianyang Li et al 2018 IOP Conf. Ser.: Mater. Sci. Eng. 452 032055
View the article online for updates and enhancements.
This content was downloaded from IP address 139.81.116.208 on 13/12/2018 at 17:18

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd
IMMAEE 2018
IOP Conf. Series: Materials Science and Engineering452 (2018) 032055 IOP Publishing
doi:10.1088/1757-899X/452/3/032055
1
Train Energy-Saving Scheme Optimized On Case Intelligence
with Synthesis-Reasoning Technology in Urban Rail Transit
Jianyang Li1, a, Hongseng Wu2, b, Benkun Zhu2
1School of Electrical Engineering and Automation Zhenjiang Institute of Technology
Zhenjiang, China
2School of Computer and Information Engineering Hefei University of Technology
Hefei, China
alijianyang@sina.com, bHongseng1 @sina.com
Abstract. Train energy consumption in URT has been attracted much greater concerns
for it becomes more serious with the large scale operation and expansion of operation
network. One of the important ways for energy-saving propulsion is to find the
energy-efficient train speed curve, which is a complicated CSP (constraint satisfaction
problem) with uncertainty, and cannot be solved effectively with such inconsistent
constrains. The case intelligent based on CBR (case-based reasoning) is proposed in
this paper for its problem-solving ability, for which the domain expertise is rich while
rule knowledge deficient, to construct a flexible system integrated with efficient
machine learning components and acquire the train operation preferences from the
former stored cases. The experiments testing on the spot indicates that the system
performs well in synthesis-reasoning, which can conquer the complexity and
uncertainty of real problem from both RBR (Rule-based reasoning) and CBR, to
minimize the energy consumption for train traction with punctuality and safety
demands.
1. Introduction
The problem of energy consumption in URT becomes more serious along with the large scale operation
and expansion of operation network, and takes terrible proportion of energy consumption. Through a
decade development URT has become the mainstream of public transport in most of China large cities
for green transportation, and more than forty URT lines have been built in Beijing, Shanghai,
Guangzhou and other cities with a mileage of 5,000 kilometers. Many more electric energies are
demanding for URT operation, causing most of their tickets income paid for the electricity bills.
Resulting in high budget deficit from URT operating losses, it was absolutely essential way to reduce
the energy consumption in each URT, which has caused the most prominent problem from daily
operations.
Energy-saving train operation can be formulated as a problem of optimal control, which aims to
calculate the optimal reference speed profile compromising on all kinds of control parameters, and a
large number of studies have been carried out from both analytical and numerical methods since1960s.
[1] Considers the problem of determining an optimal driving strategy in control with a generalized
equation of motion. [2] Uses the Pontryagin principle to find necessary conditions and shows these

IMMAEE 2018
IOP Conf. Series: Materials Science and Engineering452 (2018) 032055 IOP Publishing
doi:10.1088/1757-899X/452/3/032055
2
conditions yield key equations. [3] Uses the Kuhn–Tucker equations to find key equations that
determine the optimal switching times. [4] Obtains the optimal solution for the operation of a train on
a variable grade profile subject to speed restrictions.
Due to the complexity of the above analytical methods, many researches have constructed large of
simulating methods on energy-saving, such as dynamic programming, genetic algorithm, fuzzy control,
artificial neural networks, quadratic sequence planning, ant colony optimization, etc. [5-8].
Singaporean scholar uses genetic algorithms to generate inertial control tables to optimize the
operational control of their MRT system [9, 10], Australia SCG developed their online operation
guidance system METROMISER, which can calculate the train operation process to optimize driving
in real-time [11, 12].
Researches also shows that the factors affecting train energy consumption mainly include traction
and braking performance, train weights and speed, metro line parameters, signaling blocking mode
and train operation mode, which are difficult to obtain the optimal solution due to the complexity of
the train operating environment along with real-time passenger flow changing. So far, the precise
calculation of traction and energy consumption is still a difficult task, causing researches either
simplify the calculating model, or assume running in specific conditions omitting certain constrains.
From the view of recognition, problem-solving can be used by the traditional Rule-based reasoning
(RBR), just like what we have described the both researches have done, that is a kind abstract thinking
of human merely. Experts in the decision-making really uses more imaginative thinking - Case-based
reasoning (CBR)- to perform analogy model for creative reasoning [13], which is a kind of inferential
study strategy allowing people to process their reasoning course for new problem-solving wherever
they have similar characters. Case-intelligent system based on CBR performs well in weak area full of
domain expertise but lack knowledge like fault diagnosis, help-desk support, online e-commerce and
online decision guides, etc. [14]. As many researches have bothered by the real complex constrains on
the URT train optimized operation, which are more complicated and with inconstant impact-factors,
here we present the case-intelligent learning to solve such problems.
2. Synthesis- Reasoning modeling
2.1. Cases Collection
As we know one URT train is running from the same route station A to B many times a day, about
which its VOBC (Vehicle on-board Computer) records these propulsion parameters, large amounts of
propulsion data are stored for long time running; in the meantime this kind of URT trains in the same
line are also stored huge propulsion data. The big data of the URT train operation implicitly gives many
empirical knowledge, which are effective to optimized operation but to be difficultly obtained with
brief knowledge – the running ‘rule’ for the optimal solution accounting for the complexity of the train
operating environment.
People often use the analogy reasoning models and assumptions to study new concepts and find
new knowledge like this:
Object X has attributes a, b, c, d, e.
Object B has attributes a, b, c, e.
It suggests us that B may have similar attributes d. The so-called mapping analog is to compare two
similar things, searching for their similar relations at a certain level and as a reason to map the
problem space, through which solves the new issues by appropriate knowledge transformation, and the
matching methods can be composed by partial similar attributes, partial matching feature, or even by
interpretable matching. Naturally we can define URT train propulsion cases learning as follows:
1) Suppose a URT train serials run from station A to station B, each VOBC control the train
routing and record its real speed curve, S1,S2,…,Sn, from which we can find the best speed curve Si
with the least energy consumption.

IMMAEE 2018
IOP Conf. Series: Materials Science and Engineering452 (2018) 032055 IOP Publishing
doi:10.1088/1757-899X/452/3/032055
3
2) Another URT train X running from station A to station B, can reuse the best speed curve Si by
analogy mapping and case adaptation to perform CBR process, for they have the most similar running
constrains.
3) Case library collects both kind of running big-data from the same route AB, the one propulsion
cases is drilling from itself, the others come from the same kind of train, for they have the most similar
dynamic presentation under the same signal system.
4) case library can also collects propulsion cases from the similar kinds of train and similar speed
curve under different route by transfer learning, either from the different kinds of train ideally but not
mentioned in this paper just now, for the learning is really so time-consuming that cannot meet the
real-time ATO (Automatic Train Operation) demands.
2.2. System Modeling
In general, CBR system implements four processes such as Retrieve, Reuse, Revise, and Retain, well
known as the 4R. According to the problem (target case) space, it obtains the similar former cases (base
case) from the source case library, to deal with the similar circumstances, appropriately adapting to new
situation for the new problem-solving. The former cases can also be used to evaluate the new issues,
new statues and programs of problem-solving, and prevent the potential errors in the future.
Vehicle data base stores primary train parameters, such as Maximum Train Acceleration/Maximum
Train Service Brake Rate/ Train Length /Max wheel diameter /Min wheel diameter Normal brake
average decelerate, etc. Line data base stores such Metro Line parameters involving train operations
like types of switches, state and position/ Permanent Speed Limit / Proximity plates / Axle detection
point / Minimum radius of plane curves of the guideway /Maximum gradient at vertical sections of
different guideways/ Wayside Radio Units and Access Points, etc.
As fig 1 described, analogy in the minds of human beings plays a very important role, for which
people’s knowledge are gradually built up. The new problem are mapping and compared with the
original knowledge which have been carried out of case library, and can reason from similar
knowledge transfer (synthesis reasoning), just like what we have been searching for the running rule of
the optimal solution. In order to improve the CBR system retrieval efficiency with our synthesis
reasoning process, the best way is to integrate three organizational strategies, which performs well in
our system described in paper [15].
Figure 1. The improved system model

IMMAEE 2018
IOP Conf. Series: Materials Science and Engineering452 (2018) 032055 IOP Publishing
doi:10.1088/1757-899X/452/3/032055
4
2.3. Synthesis Reasoning Technology
The problem of optimized URT energy-saving operation is a extremely complex CSP, which consists
of a set of variables V = {V1, ..., Vn}, corresponding to the domain of values r={ r1, r2, ... rm } of each
variable, and a set of constraints C = {C1, ..., Ck}. As many researches have done, only when we
should search an assignment for each variable to fit for all constraints, can it get the optimized energy-
saving operation. But each constraint describes a legal combination of a subset of variables with a
particular fickle value, which is changeable with URT running environment. For example, monitoring
the train real speed is the most important to train safety protection, when the train spin/ slide, or wheel
diameter attrition causing by the steel surface humidity, the changing process can’t be a steady with a
definite value for they must vary from a range. That’s why many researches must simplify the CSP, and
intelligent simulation performs well for such problem-solving.
The traditional view of reasoning is a process by its causality (expressed as a ‘rule’- the reasoning
chain, RBR) to derive the conclusion. But for the real URT train operation, the optimized problem is
about uncertain and incomplete, for which traditional knowledge processing RBR can only work well
on the basis of sufficient complete and clear understanding, once the information is missing or blurred,
its reasoning ability will be drastically reduced. Therefore, the system integrated with RBR and CBR
is expected to construct our case-intelligent system, to give full play to their advantages, for the naive
CBR method does not guarantee the good performance of the system efficiency in the real URT
operation, which needs to be cooperated with RBR where the rule is acquired by machine learning
(ML) technology. There are many methods to combine CBR and RBR for problem-solving in our
system, for they have excellent flexibility.
3. Experiments and Explanation
3.1. Experimental Planning
To meet the train safety demands is the chief task to process any test on the spot, where the URT train
operating modes are divided into two general classes: ATC modes - the train is controlled by the ATC
system, and manual modes - the train is under the control of a driver. Considering many kinds of trains
running all over the world taken from different signaling systems, their inconsistent Abbreviations and
Acronyms may confuse us, so we briefly take a look.
When Mode Selection Switch is in the ATO position, VOBC enters the ATO mode and controls
the vehicle without driver intervention; CM position (Cab Manual mode) or other ATPM (ATP
Manual mode) has ATP and IATP protection mode, the train functions of acceleration, coasting,
deceleration, stopping, and door opening are under the direct manual control of the Train Operator and
are supervised by the ATP system with the driving information shown on the TOD, so our
experiments are taken on the spot with CM or ATPM mode.
our experimental CBR software system is designed in components for easily integrated with many
ML tools, such as RS and RBF in our experiments. Due to the train RAMS requirements, our software
cannot be directly inserted into VOBC system software, so the train energy consume and traction
force calculating can be only reflected with the position of driver operation handle instead. A digital
electricity meter is recording the real-time energy consume with a period of 3 seconds, so it can be
added up for the interval real energy consume to demonstrate our testing results. Thus, each case has
such attributes: Case={start point, end point, start point speed, end point speed, line conditions,
passenger flows, Traction Force, etc. }. Summarizing the testing principles as follows:
Testing on the spot under the real train operation in CM /ATPM mode;
The position of driver operation handle reacting the traction force;
A digital electricity meter recording each consume in 3 seconds periodically.
3.2. AW0-3 Learning
Acquiring Characteristic Performance Curve of Traction Force (aw0, aw1, aw2, aw3) is the basic
requirement for traction computing, for they have a different loads in real train operation involving in

IMMAEE 2018
IOP Conf. Series: Materials Science and Engineering452 (2018) 032055 IOP Publishing
doi:10.1088/1757-899X/452/3/032055
5
traction force directly. So, approximately estimating their weights by four kinds of load conditions as
table 1 shown ,and get the real curve that each train must perform the four groups testing firstly by
simulation and then on the spot running.
Table 1. Four exteme weights evaluation of the train
Statue Number of Passengers weight(t)
AW0(no load) 0 220
AW1(full seat) 336 240
AW2(overload,6men/m2) 1860 331
AW3(overload,9men/m2) 2592 375
(Assume each passenger weights 60kg)
According to Newton’s theory, the train acceleration curve must be a straight line under a certain
load, which is a consistent rule in RBR. But in fact, the real curve isn’t a direct line, even not with a
simple fitting function, for the train traction force varies depending on different conditions such as
train speed and load as fig 2 shown. The more troubled problem is there isn’t consistent cooperation
with load varying, so the train acceleration curve must be plotted under different loads, especially with
extreme loads- from empty load AW0 to overload AW3. When they are used for traction calculating,
interpolating the approximate value is a feasible way from the similar loads.
Figure 2. The four groups of traction force in exteme weights
That is a normal and common way for evaluation the URT train speed due to the inconsistent rule,
from which such RBR system has to just find a approximate value instead. Let’s think over CBR
system, for each case can be drilled for the rule, the four curves of extreme load can be easily used in
our CBR system for they stored with cases. Furthermore, each case acquired from the real implicates
the whole running conditions for the train operating, and later they can be reasoning in CBR system
cycles in spite of those trouble constraints.
3.3. Results with Outlook
In our experiments, the URT train runs from Station A to Station B about 1300m, whose basic slope
data and speed limits are shown in fig 3. Considering of the control for energy-saving within a effective
passenger’s travel speed, the basic rules and application conditions on energy-efficient control of train
operation are concluded as follows:
The coasting mode is the key for the train energy-saving control, which is fully used if it is possible.

IMMAEE 2018
IOP Conf. Series: Materials Science and Engineering452 (2018) 032055 IOP Publishing
doi:10.1088/1757-899X/452/3/032055
6
The full power mode is applied when the train needs to run at the maximum accelerated speed
(from the starting stage and also from a low speed to a high speed) considering of time limitation.
The full braking mode is mainly applied to the braking deceleration stage before the train stops for
the time limitation.
Figure 3. different running speed curves in three methods
The proposed running curve can be divided into such stages: from start up full power, coasting1,
(routing full power, coasting2), to end up braking. Depending on train speed and load, the maximum
train acceleration usually varies; real cases are merged in each running statement for the same routing
stage, which is classified inML- RS module; RBF is used to find the most similar case for the
optimized operation in each stage, which at last is changed into a feasible operation method for
population.
As table 2 indicates, CBR routing can be used for real URT train operation in 26.92% energy
reduction ratio, and can meet the time limitation demands. Although GA Simulation has a better
performance for they can touching the speed limit while real running must have a few speed surplus
caring about train braking caused by safety problem, in other words that is an idealist value which can
encourage us promote our methods furthermore.
Table 2. Performance in different operation
Operation mode Energy consume(kWh) Time(s)
Punctuality real operation 57.239 84
GA optimized simulation 32.327 95
CBR Real testing 41.831 95
4. Conclusion
Many researches have found different ways for URT optimal running, to minimize the energy for
propulsion from both analytical and numerical methods, and drawn a lots of useful design for industrial
implementation. The train speed is affected by many inconsistent factors with uncertain constrains,
which are so complex that researches have to omit some constrains for simplify their model. In order to
achieve such multiplex tasks under complex environments with complicated operation, our case
intelligent system has a good flexibility to integrate many components, which can use synthesis
reasoning technology from both RBR and CBR for problem-solving. The system performs well on the
spot and indicates that our future work to tune the speed better from system control.

IMMAEE 2018
IOP Conf. Series: Materials Science and Engineering452 (2018) 032055 IOP Publishing
doi:10.1088/1757-899X/452/3/032055
7
Acknowledgments
This research is financially supported by the China Postdoctoral Fund of HFUT, and the Natural
Science Project of Jiangsu Province YJ2017012. All the persons involved in the research projects are
thanked for their help.
References
[1] I. A. Asnis and A. V. Dmitruk, Solution of the problem of the energetically optimal control of
the motion of a train by the maximum principle, U.S.S.R. Comput. Math. Math. Phys., 25 (6):
37–44, 1985.
[2] R. R. Liu and I. M. Golovitcher, Energy-efficient operation of rail vehicles, Transportation
Research Part A: Policy and Practice, 37 (10): 917–932, 2003.
[3] E. Khmelnitsky, On an Optimal Control Problem of Train Operation, IEEE Transactions on
Automatic Control, 45 (7): 1257–1266, 2000.
[4] P. G. Howlett, P. J. Pudney and X. Vu, Local energy minimization in optimal train control,
Automatica, 45 (11): 2692–2698, 2009.
[5] R. Franke, P. Terwiesch and M. Meyer, An algorithm for the optimal control of the driving of
trains, in Proceedings of the 39th IEEE Conference on Decision and Control, 2000: 2123–
2127.