Generating of multi-criteria alternatives for

advertisement
Dec. 2009, Volume 8, No.12 (Serial No.78)
China-USA Business Review, ISSN 1537-1514, USA
Generating of multi-criteria alternatives for decision-making in electric
light rail control
Аnatoly Levchenkov1,2, Мikhail Gorobetz1,2, Leonids Ribickis1,3, Peteris Balckars2
(1. Institute of Industrial Electronics and Electrical Engineering, Riga Technical University, Riga LV-1658, Latvia;
2. Institute of Railway Transport, Riga Technical University, Riga LV-1658, Latvia;
3. Latvian Academy of Science, Riga LV-1050, Latvia)
Abstract: The purpose of this paper is to study and evaluate genetic algorithms and scheduling theory about
the methods of controlling public electric transport. Scheduling theory can be defined as criteria of time
constraints for target functions of the optimal electric transport control task and for electrical process diagnostics
task. Mathematical models and procedures are developed for the control of electrical processes and the
optimization of railway traffic control, using scheduling theory and multi-criteria of decision making. Functional
dependencies between electrical processes and dynamics of electric transport flow states are investigated in the
research. Genetic algorithms are investigated for getting dynamic multi-criteria optimization of processes in real
time mode. Efficiency of proposed methods about procedures for optimal electric transport flow control is
analyzed and. Safety is considered as an inherent component of the total transport system, which embraces
infrastructures, goods and containers, transport users and operators, vehicles and vessels, measures at policy and
legislative levels includes decision support and validation tools. Integrated safety and security systems for surface
transport systems includes advanced modeling, simulation and engineering tools, aiming at the improvement of
safety and security performance of transport systems. Development will include methodologies and designing
environments for risk based design and approval, roll-over test development and pre-normative research towards
regulations based on virtual testing methods as well as mechatronic systems for dynamic stability, reliable and
effective braking systems for vehicles. The paper is based on authors’ previous scientific work which researches
the intelligent device systems and their applications in transport systems. Intelligent devices are controllers, which
have interface to work in global network and wireless networks, and are programmed to use methods of the
artificial intelligence. Intelligent devices have possibilities to negotiate with each other and to coordinate their
work to get better decision. The results of experiments show the possibility of the developed systems to prevent
accidents and to avoid different problems by intelligent diagnostic and coordination devices.
Аnatoly Levchenkov, Dr.sc.ing., professor, Institute of Industrial Electronics and Electrical Engineering, Institute of Railway
Transport, Riga Technical University; research fields: intelligent transport systems, evolutionary programming for networks of
embedded systems.
Мikhail Gorobetz, Dr.sc.ing., senior researcher, Institute of Industrial Electronics and Electrical Engineering, Institute of Railway
Transport, Riga Technical University; research fields: intelligent transport systems, genetic programming for networks of embedded
systems.
Leonids Ribickis, Dr.hab.sc.ing., professor, Latvian Academy of Science, Institute of Industrial Electronics and Electrical
Engineering, Riga Technical University; research fields: energy saving multifunctional AC converters, power electronic converters in
hydrogen power, performance optimisation services for electric drives.
Peteris Balckars, Dr.sc.ing., professor, Institute of Railway Transport, Riga Technical University; research fields: group traction
drive of diesel-hydraulic locomotives, diagnostics of vibration for diesel locmotives, standard development in railway transport.
49
Generating of multi-criteria alternatives for decision-making in electric light rail control
Key words: decision-making; genetic algorithms; scheduling theory; intelligent electric transport system
1. Introduction
This paper presents the usage of intelligent transport control system. Such system may be used for city
electric rail transport (Gorobetz M., Levchenkov A. & Ribickis L, 2006; Rankis I., Gorobetz M. & Levchenkov A,
2006) and for railway transport (M. Gorobetz, P. Balckars, A. Levchenkov & L. Ribickis, 2008).
Traffic jams is an actual problem for light rail transport. Therefore for the optimal control of traffic lights, it
is necessary to reduce energy consumption and idle time, and increase movement speed of public transport in
cities. It gives possibility to increase public transport flow and improve service level.
Railway transport has no problem of traffic jams, but railway traffic flow is limited by safety criteria.
Therefore, routing and scheduling task is actual for railway transport system. As well optimal braking control and
safety of braking process are very important, the analysis of human behavior (Hasegawa Y., Tsunashima H.,
Marumo Y. & Kojima T, 2009) and simulation of train’s braking (Luo R. & Zeng J, 2009) is investigated.
Nowadays, the human factor plays a considerable role in traffic system control both in railway and in city
traffic transports. But some conflict problems in transport systems need immediate reaction and decision, which
are impossible to be solved by human timely.
As a result, railway crashes and accidents take place, as well as traffic jams in cities are uncontrollable,
which are followed by increasing of additional costs of public electric transport, such as excessive energy
consumption, idle time, schedule infringements and so on.
Intelligent transport control system give possibility to make traffic control safer and more cost-effective. It
may find optimal solution of the conflict by a decision support system, faster than human (Greivulis J.,
Levchenkov A. & Gorobetz M, 2007). As well in case of emergency, it may prevent crashes and accidents without
human intervention.
The authors propose to use genetic algorithms (S. J. Russel & P. Norvig, 2006) for optimization, and artificial
neural networks (Haykin S., 2006) for control of electric rail transport system. Genetic algorithms may be used for
optimal routing and scheduling tasks. Neural networks may be used to prevent breakdowns and accidents, and
such kind of controllers may be integrated in working infrastructure for optimal speed control of railway traffic.
2. Mathematical model
2.1 Model of tracks and points
Rail ways may be represented as graph R = {C , S} , where rails are divided into sections s, and each section
s ∈ S is connected with each other by two connectors s =< ci , c j > .
Each section s ∈ S has constant length l s , curve a s , and speed limit v * s .
Each point p ∈ C connecting set W of three or more sections and set of possible states of point D p ,
n
where d p =< si , s j > means opened both directions from si to s j and from s j to si which is following for
different point types: single point: D p = {< si , s j >, < si , s k >} ; dual point: D p = {< si , s j >, < si , s k >, < si , s m >} ;
cross point: D p = {< s i , s j >, < s i , s k >, < s m , s j >, < s m , s k >} . Each state of point d p ∈ D p has speed limit
v *d p , and there is maximal switching time for each point d p ∈ D p - t d p .
50
Generating of multi-criteria alternatives for decision-making in electric light rail control
2.2 Model of railway signals
Railway signal G is an object with fixed coordinates x0, and y0 is connected to fixed position on the track.
Type of signal:
Each signal g ∈ G has following states of signals L g ⊆ { R , Y , YG , G ,V ,W } , where R represents red, and
rolling stock must stop before the signal; Y represents yellow, can move and be ready to stop, the next signal is red;
YG represents yellow and green, next two sections are free; G represents green, V represents violet, W represents
moonlight white.
Each signal set up speed limits for next block-section: vdef : maximal predefined speed on the section, v0: 0
kmh, stop; v1: < 50 kmh, movement on turnouts 1/9 and 1/11 types; v2: < 80 kmh for movement on turnout 1/18
type; v3: <120kmh for movement on turnout 1/22 type.
2.3 Rolling stock model
Following parameters of system’s electromechanical elements are proposed for mathematical model (M.
Gorobetz, P. Balckars, A. Levchenkov & L. Ribickis, 2008):
Following parameters of system’s electromechanical elements are proposed for mathematical model:
U—railway contact net voltage; E—counter-electromotive force of the drive of the train; Ia—armature current of
the drive of the train; Ra—armature resistance of the drive of the train; La—armature inductance of the drive of the
train; If—field current of the drive of the train; Ra—field resistance of the drive of the train; La—field inductance
of the drive of the train; Laf—mutual field-armature inductance of the drive of the train; KE—voltage constant;
KT—torque constant; Te—electrical torque of the drive; TL—torque load; Rb–braking resistance; m—mass of
rolling stock; g—gear box ratio; F—friction force; B—mechanical braking force; J—inertia of the drive;
Bm—viscous friction coefficient of the drive; Tf—Coulomb friction torque of the drive; ω1—angular velocity of
the motor; ω2—angular velocity of output shaft of transmission; T2—torque on the output shaft of transmission;
P1—power on the input shaft of transmission; P2—power on the output shaft of transmission; r—wheel radius;
F—force on the wheel periphery; v—linear velocity on the wheel periphery; t—time.
2.4 Neural network model for accident prevention
Control part of rolling stock is represented as neural network mathematical model:
Input data set: X = {x1, x2, …, xn}; Weights: W = {w1, w2, …, wn, wn+1}; Fitness function: F = x1*w1 +
x2*w2 + … + xn*wn + wn+1; Result classes: C = {c1, c2, …, cm}; Set of states: I = {i1, i2, ... }; Set of properties: Ei
= {x1, x2, …, xn}; Set of clusters: C = {c1, c2, ... }; Members of cluster: Mc = {m1, m2, ... }; Prototype-vectors: Pc
= {p1, p2, ..., pn}; Sum vector: Sm = Pc ∩ Em = {s1, s2, ..., sn}; Vigilance: ρ (0 < ρ ≤ 1).
This model is proposed to detect dangerous part of the track, where accident is possible.
2.5 Target function for optimal braking control task
Multi-criteria target function for braking:
⎧ F br ( DL, CL, EL) → min
⎪
⎪⎪ DL = Δs → 0
⎨CL = a (t ) → a *
⎪
⎪ EL = da = const
⎪⎩
dt
where Δs —distance between closed section and rolling stock–danger level criteria (DL); a (t ) —deceleration of
rolling stock; a * —optimal deceleration for passengers–comfort level criteria (CL); da / dt —changes of
br
deceleration and braking torque–optimal energy consumption criteria (EL); F —function for braking process
optimization.
51
Generating of multi-criteria alternatives for decision-making in electric light rail control
2.6 Target function for routing and scheduling task of the rolling stock
Routing task for accident prevention consists of generating of new route and schedule for rolling stocks V
moving on points P.
Target function for scheduling and routing is to arrange points for each train to reach a destination and
assigning of time moments t to each train and each point.
• Train’s schedule: σ v : P → {t v1 , t v 2 ,..., t vs } ⊂ ℜ
• Point’s schedule: σ p1 : V → {t p1 , t p 2 ,..., t pm } ⊂ ℜ
3. Computer experiment
Computer experiment of this work extends experiment of proposed safety device working with one rolling
stock (Gorobetz M., Ribickis L., Balckars P., Greivulis J. & Levchenkov A, 2009).
Current experiment is proposed for modelling about crash prevention of two trains moving towards each
other. The purpose of this experiment is to detect dangerous section, and not allow trains to enter this section.
The model consists of 3 series block-sections; 2 rolling stocks of ER-2 type; 4 railway signals.
Each rolling stock and signal is equipped with receiving and transmitting devices that gives for possibility in
multi-agent system.
Electrical part of the model consists of DC drive, with characteristics of 8 DC motors, 1 switch to connect or
disconnect electric drive to electric contact network. 2 pairs of switches are for acceleration and for braking that
changes direction of field current if flow. Braking branch contains braking resistance. Output of DC drive is
electrical torque which handles mechanical part of rolling stock.
Mechanical part of rolling stock model gets electrical torque of electric drive as input transmitted to gears of
rolling stock and wheels (see Fig. 1).
Fig. 1
Model of railway system
Control part of rolling stock (see Fig. 2) contains two artificial neural network ANN controllers.
52
Generating of multi-criteria alternatives for decision-making in electric light rail control
Fig. 2
Fragment of control part of computer model
ANN controller (J. Greivulis, A. Levchenkov & M. Gorobetz, 2008) for braking way calculation is trained to
predict estimated distance of rolling stock which will be passed if braking process will begin at current speed.
What input this controller is wireless signal from railroad signal, current relative position of rolling stock and
linear speed of rolling stock. And what output this controller is a distance to traffic light and braking way
transmitted to motion controller of rolling stock. Cascade-forward back-propagation neural network (see Fig. 3) is
used for braking way calculation. It is trained using Levenberg-Marquardt algorithm.
Fig. 3 Cascade-forward back-propagation neural network for braking way calculation
Fig. 4
Acceleration and linear velocity dynamics of rolling stocks
53
Generating of multi-criteria alternatives for decision-making in electric light rail control
ANN controller for motion control uses input signals from sensors of mechanical and electrical part of rolling
stock—linear speed, field current. It outputs command signals for acceleration and braking.
Fig. 4 presents dynamics acceleration and velocity of rolling stocks. As the restriction signal has
worked—braking for both rolling stocks was enabled simultaneously.
The specific environment is developed by the authors for the dynamic modelling of intelligent agents for
safety improving algorithms.
As an object of experiment, the crash on Riga central station is taken (see Fig. 5).
The algorithms for safety presented in the paper were assessed by the simulation of the situation.
The results of using algorithms of artificial intelligence in public electric transport systems show the
possibility to avoid crashes and detect dangerous points on the way. They advance safety improvement,
optimization of energy usage, profit increasing, idle time minimization and coordination.
Fig. 5 Layout of railway elements to area of the crash
4. Experimental device
The result of this work is train emergency braking device.1 Invented device is proposed to increase safety on
railway transport. It gives possibility to stop rolling stock automatically and timely before closed signal.
Different from known devices that actuate brake only after the passing of closed signal, invented device
provides a train emergency braking and stopping before a closed section, even if it is not equipped with automatic
locomotive signaling. The device also provides distance control and the emergency braking way calculation.
Train emergency braking device contains antenna installed at the beginning of each block-section, the
generator, a train-mounted antenna, data reception unit, a turning angle pickup meter, connected to a train wheel
axle box, distance meter, wheel set tyre diameter meter, speed meter, acceleration meter, brake pressure meter,
brake efficiency meter, data storage unit with the additional input for data base programming and train braking
system. It also contains data reception unit, which input is connected to the train-mounted satellite communication
receiver, but output is connected to the object distance detector. It is also connected to the program braking unit
and a train braking system through the amplifier, but program braking unit has parallel inputs connected to the
braking way detector and to data storage unit, while inputs of braking way detector are parallel connected to the
outputs of brake efficiency meter, speed meter, distance meter and wheel set tyre diameter meter. The generator
connected to each block-section satellite communication receiver, whose output connects to the antenna installed
at the beginning of each block-section which is in connection with the train-mounted antenna.
1
Patent Nr. (2009). P-09-93. Train emergency braking device. M. Gorobetz, J. Greivulis, A. Levchenkov, P. Balckars, L. Ribickis.
54
Generating of multi-criteria alternatives for decision-making in electric light rail control
Fig. 6
Fragment of train emergency braking device
Train emergency braking device determines the starting location of block-section by satellite communication
receiver installed at the beginning of each block-section and determines train location by the train-mounted
satellite communication receiver. Object distance detector identifies a distance between the train and the
beginning of block-section, but braking way detector determines braking way. Program braking unit evaluates
braking way with the distance between the train and the beginning of block-section and actuates train braking
system using amplifier.
Fig. 6 presents the demonstrator of this device, which can be installed on the train. There are two traffic
lights, the electric motor, sensors, and wireless communication equipment are installed on the demonstrator.
According to the traffic light signal, the controller selects the appropriate engine speed. When burning a red
light, the control system automatically stops the engine. In response to the light sensor, the control unit in addition
to the fan is turned on and switches to another mode of operation.
Remote monitoring and control of the processes is possible to use wireless communication. In real system, it
could be dispatching control centers, from which it is possible to switch both signals and also take over control of
the train speed.
5. Conclusions
Results of experiment show the possibility to use neural network controller for speed control of electric drive
depending on the distance to stop. The results of the modeling show the possibility of the developed system to
prevent accidents and to avoid different problems by control devices. Neural network controllers allow to control
speed of rolling stock and to stop train in time preventing accidents and collisions.
Main advantages of using wireless communication between devices of rolling stock and railway
infrastructure provide additional safety level to railway system.
(to be continued on Page 64)
55
Download