Transcript Slide 1

SATURN and DIADEM
Practical Experience
Toni Dichev
31st Oct 2008
Overview
 What is DIADEM?
 DIADEM Approach
 Model Requirements
 Realism Tests / Convergence Criteria
 Issues & Solutions
 Practical Example
 Summary: Lessons Learned & Conclusions
DIADEM and Variable Demand
 Variable Demand Modelling Advice (VADMA) – WebTAG
Guidance
 Any change to transport conditions will, in principle, cause a
change in demand. The purpose of variable demand
modelling is to predict and quantify these changes
 Trip Frequency
 Trip Distribution
 Mode Choice
 Time of Day Choice
 DIADEM (Dynamic Integrated Assignment, and Demand
Modelling) software - allows you to implement variable
demand modelling as recommended by WebTAG
Demand Modelling Using DIADEM
Specify Demand Model Structure
Select Model Parameters and Realism Test
Prepare Forecast Networks and Reference
Trip Matrices for Demand Model
Assignments
Run DIADEM for different Time Periods
Identify Best Iteration
&
Reassign networks with best matrices
DIADEM Demand Model
Structure
Model needs to have appropriate split by journey purpose
Which Responses to Include?
Circumstances and Policy interests of assessment;
Availability of Data;
Dft Guidance what responses to consider
Realism Tests - The Essentials
Base Year Validated Network
Base Year Revised Network
Base Validated Matrix
Base Year Assigned Model
Demand Segments
Select/Tweak Sensitivity
(Lambda) Parameters
DIADEM RUN
Calculate Elasticities
Fuel Price
Journey Time
No
Realistic? –Ask
Is itQuestions
in accordance with
general experience?
Realistic?
Reasonable?- a subjective judgement
Reasonable?
with which you can “convince” others
Yes
Select Parameters
Experience from Past Studies
Judgement (subjective)
Take advise from the expert
Contact the supplier
Problems
Running times – can be extensive
High number of computers (one for each time
period/year/scenario)
Difficult to reach convergence specifically for future years –
unacceptable Gap values (0.2% recommended - new
Guidance 0.15%)
Mode Choice modelling (can’t cope with the nested Mode
Choice, doesn’t pass the changes in speeds etc. to PT)
Help with SATURN Convergence
Importance of well converged SATURN assignment (Delta
<0.1%)
MASL,NITA,NITA_M – convergence parameters
MONACO = T – reduces problems with blocking back at
single lane junctions (right turners)
NUC => 50 – improves accuracy (hence instability) of
signals
Latest versions of SATURN help greatly in achieving
convergence
Help with DIADEM Convergence
Change the assignment method
Increase the maximum number of iterations
Decrease the stopping values for the gap values
Decrease the value of the ‘maximum flow change’
parameter
Improve your assignment convergence
Help with DIADEM run time
Skimming minimum costs within DIADEM (only if SATURN
assignment convergence < 0.1%) SAVEIT = F
Assignment Method Algorithm 1 tends to converge quicker
Use of the best (high spec) machines
NITA = 10 MASL = 100 AUTOK = T KOMBI = 0
NITA_M = 5 DIDDLE = T
DIADEM – Future Year Assignments
The same convergence criteria is required – can be difficult
to achieve
The same structure of the files as for the Realism Test
Can significantly increased the running times due to
assignment and DIADEM convergence
Best matrices are reassigned to produce the final
assignment
Example – Strategic Motorway Link
SATURN assignment
Large model with assignment times > 1.5 hours
Six journey purposes
AM, IP and PM models
Three modelled years and 4 scenarios
DIADEM – distribution and frequency responses only
considered (no mode choice), WebTAG guidance used for
the Lamda Values, Method of Successive Averages
Practical Experience
Issue
Solution
Positive Effect
Negative Effect
SATURN assignment
Increased MASL/NITA
 Delta < 0.04%
 Potential to use
 Improved GAP in
Significantly increased
time of the SATURN
DIADEM GAP > 0.2
As above plus increased
iterations and change of
algorithm
Reduce DIADEM GAP <
Significantly increased
time of the DIADEM
Lamda Values WebTAG
not providing sensible
Decrease the Lamda
Improve the DIADEM
and the elasticises
Not within WebTAG
however excepted by the
Issues with elasticity
Indentify local area
up of the local journey
Able to demonstrate
for the model area
Not within WebTAG
however excepted by the
Run times too high
Higher spec computers
the model
Reduce run times
Costs
Summary
Lessons Learned
• Aware of the guidance
• How to select sensitivity parameters
• Convergence Criteria
• Running Times
SATURN and DIADEM
Practical Experience
Toni Dichev
31st Oct 2008