Macroscopic Modeling of Urban Networks Multi-region MFD-based Model Dynamic Route Guidance Case Study Concluding Remarks and Future Research
Optimal Dynamic Route Guidance: A Model Predictive Approach
Using Macroscopic Fundamental Diagram
Mohammad Hajiahmadi
1, Victor L. Knoop
2, Bart De Schutter
1, Hans Hellendoorn
11Delft Center for Systems and Control, Delft University of Technology, The Netherlands 2Transport and Planning Department, Delft University of Technology, The Netherlands
2
Multi-region MFD-based Model
3
Dynamic Route Guidance
High-level Optimal Routing Scheme
Objective Function
Model Predictive Control Framework
4
Case Study
Set-up
Results
Macroscopic Modeling of Urban Networks
Multi-region MFD-based Model Dynamic Route Guidance Case Study Concluding Remarks and Future Research
Modeling large-scale urban networks would be a complex task if
one wants to study and model dynamics of every single element
Control using such detailed modeling approach would be a
tedious task
Urban network partitioned into multiple regions, each represented by
an MFD
Macroscopic Modeling of Urban Networks
Multi-region MFD-based Model
Dynamic Route Guidance Case Study Concluding Remarks and Future Research
J
i: set of neighboring regions of region i
Flow from region i to region j
∈ J
iis min. of 3 elements:
1
Demand from region i to region j , D
i,j2
Supply in region j , S
jAccumulation (veh/km/lane)
Production (veh/h/lane)
Macroscopic Fundamental Diagram
0 50 100 150 0 500 1000 1500 2000 2500 (a) Accumulation (veh/km/lane) Demand (veh/h/lane) Demand 0 50 100 150 0 500 1000 1500 2000 2500 (b) Accumulation (veh/km/lane) Supply (veh/h/lane) Supply 0 50 100 150 0 500 1000 1500 2000 2500 (c)
Fig. 2. The factors determining the flow, (a) Macroscopic Fundamental Diagram, (b) demand, and (c) supply.
all the demands from regioni to every neighbouring region j, the same fraction (8) will be applied. Hence, the outflow from regioni to region j∈ Jiis formulated as:
qi,j(k) = φi(k)· ˜Di,j(k) (9)
The flow can be separated per destination. So, similar to reducing the overall flow (9), we can modulate the flow per destination (5) as:
qi,j,d(k) = φi(k)· ˜Di,j,d(k) (10)
Therefore, the accumulation in any regioni towards destina-tiond can now be updated as follows:
ni,d(k+1) = ni,d(k)+ Ts P λ∈Λi κλLλ X j∈Ji qj,i,d(k)− X j∈Ji qi,j,d(k), (11)
withTsthe sample time. Hence the total accumulation in
regioni will be: ni(k + 1) =
X
d∈D
ni,d(k + 1) (12)
In the next section, we use the presented model for predic-tion of accumulapredic-tions in the network in order to determine optimal routes.
III. HIGH-LEVEL OPTIMAL ROUTE GUIDANCE
In this section, we develop a route guidance scheme based on the high-level MFD-based model derived in the previous section. In the proposed framework, we solve the dynamic routing problem on a macroscopic level. This means that instead of taking into account individual roads and intersections, we deal with regional destinations and the way that traffic flow should be splitted towards the neighboring regions in order to avoid congestion in the intermediate regions, to decrease the overall travel time and consequently, to improve the arrival rates at the destinations. We assume a two-level structure as depicted in Fig. 3. At the top level, the optimal route guidance problem is solved based on the aggregate model presented in the previous section. At the lower level, the optimal variables (the splitting rates) that are obtained from the high-level optimization problem are taken as references, i.e. local controllers in the lower level aim at realizing the optimal splitting rates for (destination
dependent) flows of vehicles that want to travel across the regions. In the following, we elaborate on the type of optimization problem that has to be solved in the highest level in order to achieve the aforementioned goals.
A. Objective function
In order to formulate the routing problem, an objective needs to be defined. The major aim in an urban network could be maximizing the arrival rate, i.e. the number of vehicles that complete their trips and reach their destinations, or similarly minimizing the total travel delays. Over the (discrete) simulation interval[0,· · · , K − 1], the total delay criterionJTD(veh·s) is formulated as:
JTD= Ts· X i∈R KX−1 k=0 X λ∈Λi κλLλ · ni(k) . (13)
Moreover, one can introduce a penalty term on the differ-ences between average speeds of all regions as follows:
Jv(K− 1) = X i,j∈R ¯ Vi(K− 1) − ¯Vj(K− 1) 2 , (14)
with ¯Vi(K−1) the average speed in region i determined from
the MFD of that region at the end of simulation period (note that one can calculate the differences between speeds of all regions for all time steps, but it may increase the computation time of the corresponding optimization problem. Therefore, we try to normalize the speeds only at the end of the time horizon). Basically, with the values of the accumulations in each region, one can estimate an average speed for that region. Assuming an exponential function for the MFD, the average speed can be determined as follows:
¯ Vi(k) = Vfree,i· exp −12 nni(k) crit,i 2 , (15)
withVfree,ithe free-flow speed andncrit,ithe critical
accu-mulation corresponding to the maximum production. Essen-tially when there is no congestion, the average speeds in the regions are high. But in case of congestion in a region, the average speed will decrease and consequently, the travel time for vehicles inside that region will increase. By minimizing (13), the overall travel delay in the network will decrease, but it might be possible that the traffic is not distributed evenly and in some regions the average speed will be high while in others we observe low speeds. The objective function (14) Optimal Dynamic Route Guidance: A Model Predictive Approach Using MFD 6/ 18
Macroscopic Modeling of Urban Networks
Multi-region MFD-based Model
Dynamic Route Guidance Case Study Concluding Remarks and Future Research
Supply function
S
j(k) =
(
P
j,critif n
j(k)
≤ n
j,critP
j(n
j(k))
if n
j(k)
> n
j,critP
j(n
j(k)): production determined from MFD
Demand function
D
i,j(k) =
X
d∈Dα
i,j,d(k)
·
n
i,d(k)
n
i(k)
· P
in
i(k)
Update equations
n
i,d(k + 1) = n
i,d(k) +
T
sP
λ∈Λiκ
λL
λX
j∈Jiq
j,i,d(k)
−
X
j∈Jiq
i,j,d(k)
Total accumulation in region i :
n
i(k + 1) =
X
d∈D
n
i,d(k + 1)
Macroscopic Modeling of Urban Networks Multi-region MFD-based Model
Dynamic Route Guidance
Case Study Concluding Remarks and Future Research
High-level Optimal Routing Scheme Objective Function
Model Predictive Control Framework
Regional destinations
Optimal splitting traffic towards neighboring regions
Aims:
avoid congestion in intermediate regions
decrease the overall travel time
Local controller Local controller Local controller LC LC LC LC Local controller Local controller Local controller High−level MPC controller Multi−region MFD−based Model Optimization Objective, constraints ni,d(k) αi,ji ,d(k) 1
Macroscopic Modeling of Urban Networks Multi-region MFD-based Model
Dynamic Route Guidance
Case Study Concluding Remarks and Future Research
High-level Optimal Routing Scheme
Objective Function
Model Predictive Control Framework
Minimizing total travel delay:
J
TD= T
s·
X
i∈R KX
−1 k=0X
λ∈Λiκ
λL
λ· n
i(k)
Local controllers/
Optimization
Prediction
J
α
∗ i,j,d(k
c)
OD Table
n
i,d(k)
α
∗ i,j,d(k
c)
Urban regions
(multi-region) model
Macroscopic Modeling of Urban Networks Multi-region MFD-based Model
Dynamic Route Guidance
Case Study Concluding Remarks and Future Research
High-level Optimal Routing Scheme Objective Function
Model Predictive Control Framework
J
TDMPC= T
s·
RX
i =1 M·(kcX
+Np)−1 k=M·kcX
λ∈Λiκ
λL
λ· n
i(k)
– Overall optimization problem:
min
˜ αi,j,d(kc)J
TDMPCsubject to:
model equations,
0
≤ α
i,j,d(k)
≤ 1,
α
i,j,d(k) =
α
ci,j,d(k
c),
if k
∈ {M · k
c, . . . , M
· (k
c+ 1)
− 1}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Blue squares: origins
Red circles: destinations
Macroscopic Modeling of Urban Networks Multi-region MFD-based Model Dynamic Route Guidance
Case Study
Concluding Remarks and Future Research
Set-up
Results
For each region, the MFD is approximated by:
P
i= n
i· V
free· exp
−
1
2
n
in
crit 2Table :
Origin-destination demands
∗(veh/h)
Region 2
Region 8
Region 9
Region 14
Region 1
1000
1800
1750
3000
Region 4
1900
1400
1000
1400
Region 11
1700
1200
1300
1300
Region 16
2000
1000
1000
1800
Determining splitting rates
Static shortest-path (in time), Floyd-Warshall algorithm
based on average speed of regions
Shortest-path algorithm, updated every 60 seconds
Dynamic, MPC algorithm using multi-region MFD model
16 15 14 16 13 15 12 14 13 12 11 11 10 9 7 8 10 9 8 6 7 5 5 4 6 3 4 2 3 2 1 1 200 s 16 15 14 16 13 15 12 14 13 12 11 11 10 9 9 8 8 10 7 6 7 5 4 6 5 3 4 2 3 1 2 1 200 s 16 15 14 16 13 13 15 12 14 12 11 11 10 9 7 8 8 10 9 6 7 5 4 6 5 3 4 2 3 2 1 1 200 s 16 15 14 16 13 13 15 12 14 12 11 11 10 9 7 8 8 10 9 6 7 5 5 4 4 6 3 2 3 2 1 1 600 s 16 15 14 16 13 15 12 14 13 12 11 11 10 9 7 8 10 9 8 6 7 5 4 4 6 5 3 2 3 1 2 1 600 s 16 15 14 16 13 15 12 14 13 11 12 11 10 9 8 10 9 7 8 6 7 5 4 6 5 3 4 2 2 3 1 1 600s 16 16 15 15 14 13 12 14 13 7 9 8 8 12 11 11 10 10 9 6 7 5 4 6 5 3 4 2 3 2 1 1 1000 s 16 16 15 15 14 13 12 14 13 12 11 11 10 9 7 8 10 9 8 6 7 5 4 6 5 3 4 2 3 1 2 1 1000 s 16 15 15 14 16 13 12 14 13 11 12 10 11 9 7 8 10 9 8 6 7 5 4 6 5 3 4 2 3 1 2 1 1000 s 16 15 14 14 16 13 15 12 13 9 8 12 11 7 11 10 10 9 8 6 7 5 4 6 5 3 4 2 2 3 1 1 1400 s 16 16 15 12 15 14 13 13 14 12 11 11 10 9 8 8 10 9 7 6 7 5 4 4 6 5 3 2 3 1 1 2 1400 s 16 15 14 14 16 13 15 12 13 12 11 11 10 9 7 8 8 10 9 6 7 5 4 6 5 3 4 2 3 1 2 1 1400 s 16 15 15 14 16 13 14 12 13 9 8 12 11 11 10 10 9 8 7 7 6 5 4 6 5 3 4 2 3 2 1 1 1800 s 16 15 15 12 16 14 13 14 13 12 11 7 9 9 8 8 11 10 10 6 7 5 4 6 5 3 4 2 3 1 2 1 1800 s 16 15 14 16 13 15 12 12 14 13 11 11 10 9 8 10 9 7 8 6 7 5 4 6 5 3 4 2 3 1 1 2 1800 s 16 16 15 14 13 15 14 12 13 8 8 9 12 11 11 10 10 9 7 7 6 5 4 6 5 3 4 2 3 2 1 1 2400 s 16 16 15 12 15 14 13 14 13 12 11 8 9 11 10 10 9 8 7 6 7 5 5 4 6 3 4 2 3 1 2 1 2400 s 16 15 14 16 13 15 12 14 13 11 12 11 10 9 7 8 8 10 9 6 7 5 4 4 6 5 3 2 3 1 2 1 2400 s 16 16 15 14 13 15 14 12 13 8 8 9 12 11 11 10 10 9 7 3 7 6 5 5 4 6 4 2 3 2 1 1 2800 s (a) 16 15 12 1516 14 13 14 13 12 11 8 9 11 10 10 9 8 7 6 7 5 4 6 5 3 4 2 3 1 1 2 2800 s (b) 16 16 15 15 14 13 12 14 13 11 12 11 10 9 7 8 8 10 9 6 7 5 5 4 4 6 3 2 3 1 2 1 2800 s (c) Fig. 5. Results for4× 4 network, (a) Uncontrolled (fixed routes), (b) Shortest-path algorithm, (c) Optimal dynamic routing using MPC [13] B. N. Janson, “Dynamic traffic assignment for urban road networks,”
Transportation Research Part B, vol. 25B, no. 2/3, pp. 143–161, 1991.
[17] J. Maciejowski, Predictive Control with Constraints.Harlow, Eng-land: Prentice Hall, 2002.
Results for 4x4 network: (a) Uncontrolled (fixed routes), (b) Shortest-path algorithm, (c) Optimal dynamic routing using MPC