Difference between revisions of "Subway ride"

Subway ride
State dimension:	1
Differential states:	2
Discrete control functions:	1
Interior point equalities:	4

Latest revision as of 17:21, 22 February 2016

The subway ride optimal control problem goes back to work of [Bock1982]Address: Taiwan
Author: Bock, H.G.; Longman, R.W.
Booktitle: Proceedings of the American Astronomical Society. Symposium on Engineering Science and Mechanics
Title: Computation of optimal controls on disjoint control sets for minimum energy subway operation
Year: 1982
for the city of New York. In an extension, also velocity limits that lead to path-constrained arcs appear. The aim is to minimize the energy used for a subway ride from one station to another, taking into account boundary conditions and a restriction on the time.

Mathematical formulation

The MIOCP reads as

$\begin{array}{llll} \displaystyle \min_{x, w} & \int_{0}^{t_f} & L(x, w) \; \mathrm{d} t \\[1.5ex] \mbox{s.t.} & \dot{x}_0 & = & x_1, \\ & \dot{x}_1 & = & f_1(x,w), \\ & x(0) &=& (0, 0)^T, \\ & x(t_f) &=& (2112, 0)^T, \\ & w(t) &\in& \{1, 2, 3, 4\}. \end{array}$

The terminal time $t_f=65$ denotes the time of arrival of a subway train in the next station. The differential states $x_0(\cdot)$ and $x_1(\cdot)$ describe position and velocity of the train, respectively. The train can be operated in one of four different modes, $w(\cdot) = 1$ series, $w(\cdot) = 2$ parallel, $w(\cdot) = 3$ coasting, or $w(\cdot) = 4$ braking that accelerate or decelerate the train and have different energy consumption.

Acceleration and energy comsumption are velocity-dependent. Hence, we will need switching functions $\sigma_i( x_1 ) = v_i - x_1$ for given velocities $v_i, i=1..3$ .

The Lagrange term reads as

\begin{array}{rcl} L(x, 1) &=& \left\{ \begin{array}{cl} e \; p_1 & \mbox{if } \sigma_1 \ge 0 \\ e \; p_2 & \mbox{else if } \sigma_2 \ge 0 \\ e \; \sum_{i=0}^{5} c_i (1) \left( \frac{1}{10} \gamma\ x_1 \right)^{-i} \quad & \mbox{else} \end{array} \right. \\ L(x, 2) &=& \left\{ \begin{array}{cl} \infty & \mbox{if } \sigma_2 \ge 0\\ e \; p_3 & \mbox{else if } \sigma_3 \ge 0 \\ e \; \sum_{i=0}^{5} c_i (2) \left( \frac{1}{10} \gamma\ x_1 - 1 \right)^{-i} & \mbox{else} \end{array} \right. \\ L(x, 3) &=& L(x, 4) = 0. \end{array}

The right hand side function $f_1(x,w)$ reads as

\begin{array}{rcl} f_1(x, 1) &=& \left\{ \begin{array}{cl} f_1^{1A} := \frac{g \; e \; a_1}{W_{\mathrm{eff}}} & \mbox{if } \sigma_1 \ge 0 \\ f_1^{1B} :=\frac{g \; e \; a_2}{W_{\mathrm{eff}}} & \mbox{else if } \sigma_2 \ge 0 \\ f_1^{1C} := \frac{g \; (e \; T(x_1, 1) - R(x_1)}{W_{\mathrm{eff}}} & \mbox{else} \end{array} \right. \\ f_1(x, 2) &=& \left\{ \begin{array}{cl} 0 & \mbox{if } \sigma_2 \ge 0\\ f_1^{2B} := \frac{g \; e \; a_3}{W_{\mathrm{eff}}} & \mbox{else if } \sigma_3 \ge 0 \\ f_1^{2C} := \frac{g \; (e \; T(x_1, 2) - R(x_1)}{W_{\mathrm{eff}}} & \mbox{else} \end{array} \right. \\ f_1(x, 3) &=& - \frac{g \; R(x_1)}{W_{\mathrm{eff}}} - C, \\ f_1(x, 4) &=& - u = - u_{\mathrm{max}}. \end{array}

The braking deceleration $u(\cdot)$ can be varied between $0$ and a given $u_{\mathrm{max}}$ . It can be shown that only maximal braking can be optimal, hence we fixed $u(\cdot)$ to $u_{\mathrm{max}}$ without loss of generality.

Occurring forces are

\begin{array}{rcl} R(x_1) &=& c a \; {\gamma}^2 {x_1}^2 + b W {\gamma} x_1 + \frac{1.3}{2000}W + 116, \\ T(x_1,1) &=& \sum_{i=0}^{5}b_i(1) \left( \frac{1}{10} {\gamma} x_1 - 0.3 \right) ^{-i}, \\ T(x_1,2) &=& \sum_{i=0}^{5}b_i(2) \left( \frac{1}{10} {\gamma} x_1 - 1 \right) ^{-i}. \end{array}

Details about the derivation of this model and the assumptions made can be found in [Bock1982]Address: Taiwan
Author: Bock, H.G.; Longman, R.W.
Booktitle: Proceedings of the American Astronomical Society. Symposium on Engineering Science and Mechanics
Title: Computation of optimal controls on disjoint control sets for minimum energy subway operation
Year: 1982
or in [Kraemer-Eis1985]Address: Bonn
Author: Kr\"amer-Eis, P.
Publisher: Universit\"at Bonn
Series: Bonner Mathematische Schriften
Title: Ein Mehrzielverfahren zur numerischen Berechnung optimaler Feedback--Steuerungen bei beschr\"ankten nichtlinearen Steuerungsproblemen
Volume: 166
Year: 1985
.

Parameters

Symbol	Value	Unit	Symbol	Value	Unit
$W$	78000	lbs	$v_1$	0.979474	mph
$W_{\mathrm{eff}}$	85200	lbs	$v_2$	6.73211	mph
$S$	2112	ft	$v_3$	14.2658	mph
$S_4$	700	ft	$v_4$	22.0	mph
$S_5$	1200	ft	$v_5$	24.0	mph
$\gamma$	$\frac{3600}{5280}$	$\frac{\mbox{sec}}{\mbox{h}} / \frac{\mbox{ft}}{\mbox{mile}}$	$a_1$	6017.611205	lbs
$a$	100	ft $^2$	$a_2$	12348.34865	lbs
$n_{\mathrm{wag}}$	10	-	$a_3$	11124.63729	lbs
$b$	0.045	-	$u_{\mathrm{max}}$	4.4	ft / sec $^2$
$C$	0.367	-	$p_1$	106.1951102	-
$g$	32.2	$\frac{\mbox{ft}}{\mbox{sec}^2}$	$p_2$	180.9758408	-
$e$	1.0	-	$p_3$	354.136479	-

Parameter	Value	Parameter	Value
b_0(1)	-0.1983670410E02	c_0(1)	0.3629738340E02
b_1(1)	0.1952738055E03	c_1(1)	-0.2115281047E03
b_2(1)	0.2061789974E04	c_2(1)	0.7488955419E03
b_3(1)	-0.7684409308E03	c_3(1)	-0.9511076467E03
b_4(1)	0.2677869201E03	c_4(1)	0.5710015123E03
b_5(1)	-0.3159629687E02	c_5(1)	-0.1221306465E03
b_0(2)	-0.1577169936E03	c_0(2)	0.4120568887E02
b_1(2)	0.3389010339E04	c_1(2)	0.3408049202E03
b_2(2)	0.6202054610E04	c_1(2)	0.3408049202E03
b_3(2)	-0.4608734450E04	c_3(2)	0.8108316584E02
b_4(2)	0.2207757061E04	c_4(2)	-0.5689703073E01
b_5(2)	-0.3673344160E03	c_5(2)	-0.2191905731E01

Reference Solutions

The optimal trajectory for this problem has been calculated by means of an indirect approach in [Bock1982]Address: Taiwan
Author: Bock, H.G.; Longman, R.W.
Booktitle: Proceedings of the American Astronomical Society. Symposium on Engineering Science and Mechanics
Title: Computation of optimal controls on disjoint control sets for minimum energy subway operation
Year: 1982
[Kraemer-Eis1985]Address: Bonn
Author: Kr\"amer-Eis, P.
Publisher: Universit\"at Bonn
Series: Bonner Mathematische Schriften
Title: Ein Mehrzielverfahren zur numerischen Berechnung optimaler Feedback--Steuerungen bei beschr\"ankten nichtlinearen Steuerungsproblemen
Volume: 166
Year: 1985
, and based on the direct multiple shooting method in [Sager2009]Author: Sager, S.; Reinelt, G.; Bock, H.G.
Journal: Mathematical Programming
Number: 1
Pages: 109--149
Title: Direct Methods With Maximal Lower Bound for Mixed-Integer Optimal Control Problems
Url: http://mathopt.de/PUBLICATIONS/Sager2009.pdf
Volume: 118
Year: 2009
.

Time $t$	$w(\cdot)$	$f_1 =$	$x_0$ [ft]	$x_1$ [mph]	$x_1$ [ftps]	Energy
$0.00000$	1	$f_1^{1A}$	0.0	0.0	0.0	0.0
$0.63166$	1	$f_1^{1B}$	0.453711	0.979474	1.43656	0.0186331
$2.43955$	1	$f_1^{1C}$	10.6776	6.73211	9.87375	0.109518
$3.64338$	2	$f_1^{2B}$	24.4836	8.65723	12.6973	0.147387
$5.59988$	2	$f_1^{2C}$	57.3729	14.2658	20.9232	0.339851
$12.6070$	1	$f_1^{1C}$	277.711	25.6452	37.6129	0.93519
$45.7827$	3	$f_1(3)$	1556.5	26.8579	39.3915	1.14569
$46.8938$	3	$f_1(3)$	1600	26.5306	38.9115	1.14569
$57.1600$	4	$f_1(4)$	1976.78	23.5201	34.4961	1.14569
$65.0000$	-	$-$	2112	0.0	0.0	1.14569

Variants

The given parameters have to be modified to match different parts of the track, subway train types, or amount of passengers. A minimization of travel time might also be considered.

The problem becomes more challenging, when additional point or path constraints are considered.

Point constraint

We consider the point constraint

x_1 \le v_4 \mbox{ if } x_0 = S_4

for a given distance $0 < S_4 < S$ and velocity $v_4 > v_3$ . Note that the state $x_0(\cdot)$ is strictly monotonically increasing with time, as $\dot{x}_0 = x_1 > 0$ for all $t \in (0, T)$ .

The optimal order of gears for $S_4 = 1200$ and $v_4 = 22 / \gamma$ with the additional interior point constraints (\ref{FASOPOINTCON}) is $1, 2, 1, 3, 4, 2, 1, 3, 4$ . The stage lengths between switches are 2.86362, 10.722, 15.3108, 5.81821, 1.18383, 2.72451, 12.917, 5.47402, and 7.98594 with $\Phi = 1.3978$ . For different parameters $S_4 = 700$ and $v_4 = 22 / \gamma$ we obtain the gear choice 1, 2, 1, 3, 2, 1, 3, 4 and stage lengths 2.98084, 6.28428, 11.0714, 4.77575, 6.0483, 18.6081, 6.4893, and 8.74202 with $\Phi = 1.32518$ .

Path constraint

A more practical restriction are path constraints on subsets of the track. We will consider a problem with additional path constraints

x_1 \le v_5 \; \mbox{ if } \; x_0 \ge S_5.

The additional path constraint changes the qualitative behavior of the relaxed solution. While all solutions considered this far were bang-bang and the main work consisted in finding the switching points, we now have a path-constrained arc. The optimal solutions for refined grids yield a series of monotonically decreasing objective function values, where the limit is the best value that can be approximated by an integer feasible solution. In our case we obtain 1.33108, 1.31070, 1.31058, 1.31058, ...

The plot shows two possible integer realizations, with a trade-off between energy consumption and number of switches. Note that the solutions approximate the optimal driving behavior (a convex combination of two operation modes) by switching between the two and causing a touching of the velocity constraint from below as many times as we switch.

Reference solution plots
Optimal solution with 1 touch point.
Optimal solution with 3 touch points.

The differential state velocity of a subway train over time. The dotted vertical line indicates the beginning of the path constraint, the horizontal line the maximum velocity. Left: one switch leading to one touch point. Right: optimal solution for three switches. The energy-optimal solution needs to stay as close as possible to the maximum velocity on this time interval to avoid even higher energy-intensive accelerations in the start-up phase to match the terminal time constraint $t_f \le 65$ to reach the next station.

Source Code

Model descriptions are available in

Muscod code at Subway ride (Muscod)

References

There were no citations found in the article.

@@ Line 6: / Line 6: @@
 }}
-The ''subway ride'' optimal control problem goes back to work of <bibref>Bock1982</bibref> for the city of New York. In an extension, also velocity limits that lead to path-constrained arcs appear.
+The ''subway ride'' optimal control problem goes back to work of <bib id="Bock1982" /> for the city of New York. In an extension, also velocity limits that lead to path-constrained arcs appear.
 The aim is to minimize the energy used for a subway ride from one station to another, taking into account boundary conditions and a restriction on the time.
@@ Line 15: / Line 15: @@
 <center>
 <math>
-\begin{array}{llcl}
+\begin{array}{llll}
-  \displaystyle \min_{x, w} & & & \int_{0}^{t_f} L(x, w) \; \mathrm{d} t   \\[1.5ex]
+  \displaystyle \min_{x, w} &   \int_{0}^{t_f} & L(x, w) \; \mathrm{d} t   \\[1.5ex]
-  \mbox{s.t.} & \dot{x}_0(t) & = & x_1(t), \\
+  \mbox{s.t.} & \dot{x}_0 & = & x_1, \\
-  & \dot{x}_1(t) & = & f_1(x,w),  \\
+  & \dot{x}_1 & = & f_1(x,w),  \\
   & x(0) &=& (0, 0)^T, \\
   & x(t_f) &=& (2112, 0)^T, \\
-  & w(t) &\in&  \{1, 2, 3, 4\}, \quad t \in [0, t_f].
+  & w(t) &\in&  \{1, 2, 3, 4\}.
 \end{array}
 </math>
 </center>
-The terminal time <math>t_f=65</math> denotes the time of arrival of a subway train in the next station. The differential states <math>x_0(\cdot)</math> and <math>x_1(\cdot)</math> describe position \review{and velocity of the train, respectively}. The train can be operated in one of four different modes, <math>w(\cdot) = 1</math> series, <math>w(\cdot) = 2</math> parallel, <math>w(\cdot) = 3</math> coasting, or <math>w(\cdot) = 4</math> braking
+The terminal time <math>t_f=65</math> denotes the time of arrival of a subway train in the next station. The differential states <math>x_0(\cdot)</math> and <math>x_1(\cdot)</math> describe position and velocity of the train, respectively. The train can be operated in one of four different modes, <math>w(\cdot) = 1</math> series, <math>w(\cdot) = 2</math> parallel, <math>w(\cdot) = 3</math> coasting, or <math>w(\cdot) = 4</math> braking that accelerate or decelerate the train and have different energy consumption.
-that accelerate or decelerate the train and have different energy consumption.
 Acceleration and energy comsumption are velocity-dependent. Hence, we will need switching functions <math>\sigma_i( x_1 ) = v_i - x_1</math> for given velocities <math>v_i, i=1..3</math>.
@@ Line 63: / Line 62: @@
 </math></center>
-Details about the derivation of this model and the assumptions made can be found in <bibref>Bock1982</bibref> or in <bibref>Kraemer-Eis1985</bibref>.
+Details about the derivation of this model and the assumptions made can be found in <bib id="Bock1982" /> or in <bib id="Kraemer-Eis1985" />.
+== Parameters ==
+{| border="1" align="center" cellpadding="5" cellspacing="0"
+|- bgcolor=#c7c7c7
+! Symbol !! Value !! Unit !! Symbol !! Value !! Unit
+|-
+!<math>W</math>                !! 78000 !! lbs !! <math>v_1</math> !! 0.979474 !! mph
+|-
+!<math>W_{\mathrm{eff}}</math> !! 85200 !! lbs !! <math>v_2</math> !! 6.73211 !! mph
+|-
+!<math>S</math>                !! 2112  !! ft  !! <math>v_3</math> !! 14.2658 !! mph
+|-
+!<math>S_4</math>              !!  700  !! ft  !! <math>v_4</math> !! 22.0 !! mph
+|-
+!<math>S_5</math>              !! 1200  !! ft  !! <math>v_5</math> !! 24.0 !! mph
+|-
+!<math>\gamma</math>           !! <math>\frac{3600}{5280}</math> !! <math>\frac{\mbox{sec}}{\mbox{h}} / \frac{\mbox{ft}}{\mbox{mile}}</math> !! <math>a_1</math> !! 6017.611205 !! lbs
+|-
+!<math>a</math>                !! 100   !! ft<math>^2</math> !! <math>a_2</math> !! 12348.34865 !! lbs
+|-
+!<math>n_{\mathrm{wag}}</math> !! 10    !! -   !! <math>a_3</math> !! 11124.63729 !! lbs
+|-
+!<math>b</math>                !! 0.045 !! - !! <math>u_{\mathrm{max}}</math> !! 4.4 !! ft / sec<math>^2</math>
+|-
+!<math>C</math>                !! 0.367 !! - !! <math>p_1</math> !! 106.1951102 !! -
+|-
+!<math>g</math>                !! 32.2  !! <math>\frac{\mbox{ft}}{\mbox{sec}^2}</math> !! <math>p_2</math> !! 180.9758408 !! -
+|-
+!<math>e</math>                !! 1.0   !! - !! <math>p_3</math> !! 354.136479 !! -
+|}
+{| border="1" align="center" cellpadding="5" cellspacing="0"
+|- bgcolor=#c7c7c7
+! Parameter !! Value !! Parameter !! Value
+|-
+! b_0(1) !! -0.1983670410E02 !!  c_0(1) !! 0.3629738340E02
+|-
+! b_1(1) !! 0.1952738055E03 !! c_1(1) !! -0.2115281047E03
+|-
+! b_2(1) !! 0.2061789974E04 !! c_2(1) !! 0.7488955419E03
+|-
+! b_3(1) !! -0.7684409308E03 !! c_3(1) !! -0.9511076467E03
+|-
+! b_4(1) !! 0.2677869201E03 !! c_4(1) !!  0.5710015123E03
+|-
+! b_5(1) !! -0.3159629687E02 !! c_5(1) !! -0.1221306465E03
+|-
+! b_0(2) !! -0.1577169936E03 !! c_0(2) !! 0.4120568887E02
+|-
+! b_1(2) !! 0.3389010339E04 !! c_1(2) !! 0.3408049202E03
+|-
+! b_2(2) !! 0.6202054610E04 !! c_1(2) !! 0.3408049202E03
+|-
+! b_3(2) !! -0.4608734450E04 !! c_3(2) !! 0.8108316584E02
+|-
+! b_4(2) !! 0.2207757061E04 !! c_4(2) !! -0.5689703073E01
+|-
+! b_5(2) !! -0.3673344160E03 !! c_5(2) !! -0.2191905731E01
+|}
 == Reference Solutions ==
-The optimal trajectory for this problem has been calculated by means of an indirect approach in <bibref>Bock1982</bibref><bibref>Kraemer-Eis1985</bibref>, and based on the direct multiple shooting method in <bibref>Sager2009</bibref>.
+The optimal trajectory for this problem has been calculated by means of an indirect approach in <bib id="Bock1982" /><bib id="Kraemer-Eis1985" />, and based on the direct multiple shooting method in <bib id="Sager2009" />.
+{| border="1" align="center" cellpadding="5" cellspacing="0"
+|- bgcolor=#c7c7c7
+! Time <math>t</math> !! <math>w(\cdot)</math> !! <math>f_1 = </math> !! <math>x_0</math> [ft] !! <math>x_1</math> [mph] !! <math>x_1</math> [ftps] !! Energy
+|-
+! <math>0.00000</math> !! 1 !! <math>f_1^{1A}</math> !! 0.0 !! 0.0 !! 0.0 !! 0.0
+|-
+! <math>0.63166</math> !! 1 !! <math>f_1^{1B}</math> !! 0.453711 !! 0.979474 !! 1.43656 !! 0.0186331
+|-
+! <math>2.43955</math> !! 1 !! <math>f_1^{1C}</math> !! 10.6776 !! 6.73211 !! 9.87375 !! 0.109518
+|-
+! <math>3.64338</math> !! 2 !! <math>f_1^{2B}</math> !! 24.4836 !! 8.65723 !! 12.6973 !! 0.147387
+|-
+! <math>5.59988</math> !! 2 !! <math>f_1^{2C}</math> !! 57.3729 !! 14.2658 !! 20.9232 !! 0.339851
+|-
+! <math>12.6070</math> !! 1 !! <math>f_1^{1C}</math> !! 277.711 !! 25.6452 !! 37.6129 !! 0.93519
+|-
+! <math>45.7827</math> !! 3 !! <math>f_1(3)</math> !! 1556.5 !! 26.8579 !! 39.3915 !! 1.14569
+|-
+! <math>46.8938</math> !! 3 !! <math>f_1(3)</math> !! 1600 !! 26.5306 !! 38.9115 !! 1.14569
+|-
+! <math>57.1600</math> !! 4 !! <math>f_1(4)</math> !! 1976.78 !! 23.5201 !! 34.4961 !! 1.14569
+|-
+! <math>65.0000</math> !! - !! <math>-</math> !! 2112 !! 0.0 !! 0.0 !! 1.14569
+|}
 == Variants ==
@@ Line 96: / Line 181: @@
 The additional path constraint changes the qualitative behavior of the relaxed solution. While all solutions considered this far were bang-bang and the main work consisted in finding the switching points, we now have a path-constrained arc. The optimal solutions for refined grids yield a series of monotonically decreasing objective function values, where the limit is the best value that can be approximated by an integer feasible solution. In our case we obtain 1.33108, 1.31070, 1.31058, 1.31058, ...
+The plot shows two possible integer realizations, with a trade-off between energy consumption and number of switches. Note that the solutions approximate the optimal driving behavior (a convex combination of two operation modes) by switching between the two and causing a touching of the velocity constraint from below as many times as we switch.
+<gallery caption="Reference solution plots" widths="271px" heights="379px" perrow="2">
+ Image:subway78000PathConstraint1600-24Opt1.png| Optimal solution with 1 touch point.
+ Image:subway78000PathConstraint1600-24Opt2.png| Optimal solution with 3 touch points.
+</gallery>
+The differential state ''velocity'' of a subway train over time. The dotted vertical line indicates the beginning of the path constraint, the horizontal line the maximum velocity. Left: one switch leading to one touch point. Right: optimal solution for three switches. The energy-optimal solution needs to stay as close as possible to the maximum velocity on this time interval to avoid even higher energy-intensive accelerations in the start-up phase to match the terminal time constraint <math>t_f \le 65</math> to reach the next station.
+== Source Code ==
+Model descriptions are available in
+* [[:Category:Muscod | Muscod code]] at [[Subway ride (Muscod)]]
 == References ==
-<bibreferences/>
+<biblist />
 <!--List of all categories this page is part of. List characterization of solution behavior, model properties, ore presence of implementation details (e.g., AMPL for AMPL model) here -->

Difference between revisions of "Subway ride"

Latest revision as of 17:21, 22 February 2016

Contents

Mathematical formulation

Parameters

Reference Solutions

Variants

Point constraint

Path constraint

Source Code

References

Navigation menu

Personal tools

Namespaces

Variants

Views

More

Search

Navigation

Support

Tools