SlideShare a Scribd company logo
1 of 20
Deep Reinforcement Learning based
Dynamic Optimization of Bus
Timetable
Ankit Sharma
Bus Timetable Optimization
 Bus timetable optimization is a key issue to reduce operational cost of bus
companies and improve the service quality.
 Heuristic algorithms work in offline and does not account for people flow change.
https://doi.org/10.48550/arXiv.2107.07066
Bus Timetable Optimization
 The optimization of the bus timetable aims to consider the interests of both
passengers and the bus company, and set the departure time of buses to meet
the demand of passenger flow.
 The main quantitative indicators are bus congestion and the waiting time of
passengers, while the interests of bus companies are mainly affected by the
number of departures (departure intervals) in the timetable.
https://doi.org/10.48550/arXiv.2107.07066
Deep Reinforcement Learning Formulation
 Actions – No Departure(0), Departure(1)
 States – Time(hr-𝑿𝟏𝒕,),Time(min-𝑿𝟐𝒕,),Load Rate(𝑿𝟑𝒕), Waiting Time(𝑿𝟒𝒕),
Carrying Capability(𝑿𝟓𝒕), Stranded Passengers(𝑿𝟔𝒕).
 Reward(from reference paper):
1 − (𝑿𝟓𝒕) − 𝛼 ∗ 𝑿𝟒𝒕 − 𝛽 ∗ 𝑿𝟔𝒕, (𝑎𝑐𝑡𝑖𝑜𝑛 = 0)
(𝑿𝟓𝒕) − 𝛽 ∗ 𝑿𝟔𝒕, (𝑎𝑐𝑡𝑖𝑜𝑛 = 1)
Alternate reward functions were also tested.
https://doi.org/10.48550/arXiv.2107.07066
, , ,
Bus Environment
 Bus timetable considered as episodic task (star-end schedule) with 6 states changing w.r.t
time.
𝑺𝒕 = [𝑿𝟏𝒕, 𝑿𝟐𝒕, 𝑿𝟑𝒕, 𝑿𝟒𝒕, 𝑿𝟓𝒕, 𝑿𝟔𝒕]
 𝑋1𝑡 − 𝑡ℎ/24, 𝑋2𝑡 −
𝑡𝑚
60
 𝑋3𝑡 −
𝑀𝑎𝑥 𝑃𝑎𝑠𝑠𝑒𝑛𝑔𝑒𝑟𝑠
𝑀𝑎𝑥 𝐵𝑢𝑠 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦
 𝑋4𝑡 − Normalized Waiting time of all passengers
 𝑋5𝑡 −
𝑁𝑒𝑒𝑑 𝑜𝑓 𝐶𝑎𝑟𝑟𝑦𝑖𝑛𝑔 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦
𝐶𝑎𝑟𝑟𝑦𝑖𝑛𝑔 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑜𝑓 𝑣𝑒ℎ𝑖𝑐𝑙𝑒
 𝑋6𝑡 − 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑡𝑟𝑎𝑛𝑑𝑒𝑑 𝑝𝑎𝑠𝑠𝑒𝑛𝑔𝑒𝑟𝑠
Assumptions*:
• 𝑋6𝑡 was assumed to be exponential distribution after every departure.
• Based on number of stranded passengers awaiting bus or remainder post departure
are used for calculation of 𝑋4𝑡, 𝑋5𝑡
• 𝑋3𝑡 is calculated based on max bus capacity and no. of stranded passengers at the
time of departure
• Episode considered from 06:00-10:00Am with decision point at every 5 mins
* Real data was not available
http://www.muxingyun.com/en/digital-factory
DRL - Algorithms
Deep SARSA
S.No. Parameter Value
1. DNN – Hidden
Layers
2
2. DNN-Hidden
Units
188
3. Activation
Function
ReLU
4. Epsilon 0.2
5. Gamma .99
6. Experience
memory size
10000
7. Batch Size 32
8. Learning rate .001
DQN
S.No. Parameter Value
1. DNN – Hidden
Layers
2
2. DNN-Hidden
Units
188
3. Activation
Function
ReLU
4. Epsilon 0.2
5. Gamma .99
6. Experience
memory size
10000
7. Batch Size 32
8. Learning rate .001
Bus Timetable-Fixed Interval(30 mins)
TEST - 1
Testing with Deep SARSA & DQN
SARSA
Episodes 1500
Epsilon 0.2
DQN
Episodes 1500
Epsilon 0.2
Reward: Same as in paper.
Testing with Deep SARSA & DQN
Load Factor
Fixed Interval(30 mins): 8, SARSA Departures:23,DQN Departures:23
Normalized Waiting Time
Bus Timetable
DQN
Deep SARSA
Testing with Deep SARSA & DQN
Required Carrying Capacity
Fixed Interval(30 mins): 8, SARSA Departures:23,DQN Departures:23
Stranded Passenger
Bus Timetable
DQN
Deep SARSA
TEST - 2
Testing with Deep SARSA & DQN
SARSA
Episodes 1500
Epsilon 0.2
DQN
Episodes 1500
Epsilon 0.2
Reward: Modified to reduce number of departures and increase load rate to account
for bus agency. Condition added to have Load Rate > 0.7
Testing with Deep SARSA & DQN
Load Factor
Fixed Interval(30 mins): 8, SARSA Departures:15,DQN Departures:15
Normalized Waiting Time
Bus Timetable
DQN
Deep SARSA
Testing with Deep SARSA & DQN
Required Carrying Capacity
Fixed Interval(30 mins): 8, SARSA Departures:15,DQN Departures:15
Stranded Passenger
Bus Timetable
DQN
Deep SARSA
TEST - 3
Testing with Deep SARSA & DQN
SARSA
Episodes 1500
Epsilon 0.2
DQN
Episodes 1500
Epsilon 0.2
Reward: Modified to reduce number of departures and increase load rate more to
account for bus agency. Condition added to have Load Rate > 0.8
Testing with Deep SARSA & DQN
Load Factor
Fixed Interval(30 mins): 8, SARSA Departures:11,DQN Departures:11
Normalized Waiting Time
Bus Timetable
DQN
Deep SARSA
Testing with Deep SARSA & DQN
Required Carrying Capacity
Fixed Interval(30 mins): 8, SARSA Departures:11,DQN Departures:11
Stranded Passenger
Bus Timetable
DQN
Deep SARSA
Conclusion
 Timetable created with fixed interval of 30 mins had waiting time and
stranded passenger going up and couldn’t be contained. Load Rate was also
100% which doesn’t go well with travelers due to heavy congestion.
 With right reward function DQN and Deep SARSA were able to understand
stranded passenger rate and reduce waiting time and number of stranded
passengers while keeping Load Rate less than 90% with just 3 more
departures in 4 hours (6:00-10:00AM)
 With Deep SARSA & DQN real time decision based on number of stranded
passengers can be taken without re-computing the whole problem.

More Related Content

Similar to Deep_Reinforcement_Learning_based_Dynamic_Timetable.pptx

The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)
theijes
 
queuingtheory-091005084417-phpapp01 (2).pdf
queuingtheory-091005084417-phpapp01 (2).pdfqueuingtheory-091005084417-phpapp01 (2).pdf
queuingtheory-091005084417-phpapp01 (2).pdf
Aditya Mane
 
Pedestrian dead reckoning indoor localization based on os-elm
Pedestrian dead reckoning indoor localization based on os-elmPedestrian dead reckoning indoor localization based on os-elm
Pedestrian dead reckoning indoor localization based on os-elm
Alwin Poulose
 

Similar to Deep_Reinforcement_Learning_based_Dynamic_Timetable.pptx (20)

Service analytics: A proposal to LTA Singapore for on-demand bus trials
Service analytics: A proposal to LTA Singapore for on-demand bus trialsService analytics: A proposal to LTA Singapore for on-demand bus trials
Service analytics: A proposal to LTA Singapore for on-demand bus trials
 
Augmenting Decisions of Taxi Drivers through Reinforcement Learning for Impro...
Augmenting Decisions of Taxi Drivers through Reinforcement Learning for Impro...Augmenting Decisions of Taxi Drivers through Reinforcement Learning for Impro...
Augmenting Decisions of Taxi Drivers through Reinforcement Learning for Impro...
 
A Dynamic Logistic Dispatching System With Set-Based Particle Swarm Optimization
A Dynamic Logistic Dispatching System With Set-Based Particle Swarm OptimizationA Dynamic Logistic Dispatching System With Set-Based Particle Swarm Optimization
A Dynamic Logistic Dispatching System With Set-Based Particle Swarm Optimization
 
The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)
 
Cost minimization model
Cost minimization modelCost minimization model
Cost minimization model
 
full report
full reportfull report
full report
 
queuingtheory-091005084417-phpapp01 (2).pdf
queuingtheory-091005084417-phpapp01 (2).pdfqueuingtheory-091005084417-phpapp01 (2).pdf
queuingtheory-091005084417-phpapp01 (2).pdf
 
A Longitudinal Control Algorithm for Smart Cruise Control with Virtual Parame...
A Longitudinal Control Algorithm for Smart Cruise Control with Virtual Parame...A Longitudinal Control Algorithm for Smart Cruise Control with Virtual Parame...
A Longitudinal Control Algorithm for Smart Cruise Control with Virtual Parame...
 
Collision Warning....SAMRIDDHA SHIL & MY TEAM
Collision Warning....SAMRIDDHA SHIL & MY TEAMCollision Warning....SAMRIDDHA SHIL & MY TEAM
Collision Warning....SAMRIDDHA SHIL & MY TEAM
 
Traffic state estimation with multi-sensor data for large networks with macro...
Traffic state estimation with multi-sensor data for large networks with macro...Traffic state estimation with multi-sensor data for large networks with macro...
Traffic state estimation with multi-sensor data for large networks with macro...
 
Automated Parameterization of Performance Models from Measurements
Automated Parameterization of Performance Models from MeasurementsAutomated Parameterization of Performance Models from Measurements
Automated Parameterization of Performance Models from Measurements
 
TRAFFIC SIMULATION AT TOLL ROAD SECTION USING VISSIM SOFTWARE
TRAFFIC SIMULATION AT TOLL ROAD SECTION USING VISSIM SOFTWARETRAFFIC SIMULATION AT TOLL ROAD SECTION USING VISSIM SOFTWARE
TRAFFIC SIMULATION AT TOLL ROAD SECTION USING VISSIM SOFTWARE
 
PRESENTATION
PRESENTATIONPRESENTATION
PRESENTATION
 
Contoh Desain Slide Presentasi Ilmiah Kreatif dan Menarik #1
Contoh Desain Slide Presentasi Ilmiah Kreatif dan Menarik #1Contoh Desain Slide Presentasi Ilmiah Kreatif dan Menarik #1
Contoh Desain Slide Presentasi Ilmiah Kreatif dan Menarik #1
 
Traffic signal
Traffic signalTraffic signal
Traffic signal
 
How Analytic Reporting Can Identify and Solve Paratransit Service Shortcomings
How Analytic Reporting Can Identify and Solve Paratransit Service ShortcomingsHow Analytic Reporting Can Identify and Solve Paratransit Service Shortcomings
How Analytic Reporting Can Identify and Solve Paratransit Service Shortcomings
 
Deep reinforcement learning for traffic light cycle control
Deep reinforcement learning for traffic light cycle controlDeep reinforcement learning for traffic light cycle control
Deep reinforcement learning for traffic light cycle control
 
PROPOSED INTELLIGENT TRANSPORT SYSTEM DEPLOYMENTS IN KAJANG CITY
PROPOSED INTELLIGENT TRANSPORT SYSTEM DEPLOYMENTS IN KAJANG CITYPROPOSED INTELLIGENT TRANSPORT SYSTEM DEPLOYMENTS IN KAJANG CITY
PROPOSED INTELLIGENT TRANSPORT SYSTEM DEPLOYMENTS IN KAJANG CITY
 
Battery Powered and Hybrid Drive Opportunities in Heavy Duty, Large Capacity,...
Battery Powered and Hybrid Drive Opportunities in Heavy Duty, Large Capacity,...Battery Powered and Hybrid Drive Opportunities in Heavy Duty, Large Capacity,...
Battery Powered and Hybrid Drive Opportunities in Heavy Duty, Large Capacity,...
 
Pedestrian dead reckoning indoor localization based on os-elm
Pedestrian dead reckoning indoor localization based on os-elmPedestrian dead reckoning indoor localization based on os-elm
Pedestrian dead reckoning indoor localization based on os-elm
 

Recently uploaded

如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
zifhagzkk
 
原件一样伦敦国王学院毕业证成绩单留信学历认证
原件一样伦敦国王学院毕业证成绩单留信学历认证原件一样伦敦国王学院毕业证成绩单留信学历认证
原件一样伦敦国王学院毕业证成绩单留信学历认证
pwgnohujw
 
Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...
Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...
Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...
Valters Lauzums
 
Audience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptxAudience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptx
Stephen266013
 
edited gordis ebook sixth edition david d.pdf
edited gordis ebook sixth edition david d.pdfedited gordis ebook sixth edition david d.pdf
edited gordis ebook sixth edition david d.pdf
great91
 
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
Amil baba
 
Displacement, Velocity, Acceleration, and Second Derivatives
Displacement, Velocity, Acceleration, and Second DerivativesDisplacement, Velocity, Acceleration, and Second Derivatives
Displacement, Velocity, Acceleration, and Second Derivatives
23050636
 
如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证
如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证
如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证
ju0dztxtn
 
如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证
如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证
如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证
a8om7o51
 

Recently uploaded (20)

如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
如何办理(Dalhousie毕业证书)达尔豪斯大学毕业证成绩单留信学历认证
 
Seven tools of quality control.slideshare
Seven tools of quality control.slideshareSeven tools of quality control.slideshare
Seven tools of quality control.slideshare
 
社内勉強会資料_Object Recognition as Next Token Prediction
社内勉強会資料_Object Recognition as Next Token Prediction社内勉強会資料_Object Recognition as Next Token Prediction
社内勉強会資料_Object Recognition as Next Token Prediction
 
NOAM AAUG Adobe Summit 2024: Summit Slam Dunks
NOAM AAUG Adobe Summit 2024: Summit Slam DunksNOAM AAUG Adobe Summit 2024: Summit Slam Dunks
NOAM AAUG Adobe Summit 2024: Summit Slam Dunks
 
Jual Obat Aborsi Bandung (Asli No.1) Wa 082134680322 Klinik Obat Penggugur Ka...
Jual Obat Aborsi Bandung (Asli No.1) Wa 082134680322 Klinik Obat Penggugur Ka...Jual Obat Aborsi Bandung (Asli No.1) Wa 082134680322 Klinik Obat Penggugur Ka...
Jual Obat Aborsi Bandung (Asli No.1) Wa 082134680322 Klinik Obat Penggugur Ka...
 
原件一样伦敦国王学院毕业证成绩单留信学历认证
原件一样伦敦国王学院毕业证成绩单留信学历认证原件一样伦敦国王学院毕业证成绩单留信学历认证
原件一样伦敦国王学院毕业证成绩单留信学历认证
 
SCI8-Q4-MOD11.pdfwrwujrrjfaajerjrajrrarj
SCI8-Q4-MOD11.pdfwrwujrrjfaajerjrajrrarjSCI8-Q4-MOD11.pdfwrwujrrjfaajerjrajrrarj
SCI8-Q4-MOD11.pdfwrwujrrjfaajerjrajrrarj
 
Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...
Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...
Data Analytics for Digital Marketing Lecture for Advanced Digital & Social Me...
 
Audience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptxAudience Researchndfhcvnfgvgbhujhgfv.pptx
Audience Researchndfhcvnfgvgbhujhgfv.pptx
 
Northern New England Tableau User Group (TUG) May 2024
Northern New England Tableau User Group (TUG) May 2024Northern New England Tableau User Group (TUG) May 2024
Northern New England Tableau User Group (TUG) May 2024
 
How to Transform Clinical Trial Management with Advanced Data Analytics
How to Transform Clinical Trial Management with Advanced Data AnalyticsHow to Transform Clinical Trial Management with Advanced Data Analytics
How to Transform Clinical Trial Management with Advanced Data Analytics
 
Predictive Precipitation: Advanced Rain Forecasting Techniques
Predictive Precipitation: Advanced Rain Forecasting TechniquesPredictive Precipitation: Advanced Rain Forecasting Techniques
Predictive Precipitation: Advanced Rain Forecasting Techniques
 
edited gordis ebook sixth edition david d.pdf
edited gordis ebook sixth edition david d.pdfedited gordis ebook sixth edition david d.pdf
edited gordis ebook sixth edition david d.pdf
 
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
NO1 Best Kala Jadu Expert Specialist In Germany Kala Jadu Expert Specialist I...
 
What is Insertion Sort. Its basic information
What is Insertion Sort. Its basic informationWhat is Insertion Sort. Its basic information
What is Insertion Sort. Its basic information
 
Displacement, Velocity, Acceleration, and Second Derivatives
Displacement, Velocity, Acceleration, and Second DerivativesDisplacement, Velocity, Acceleration, and Second Derivatives
Displacement, Velocity, Acceleration, and Second Derivatives
 
如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证
如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证
如何办理英国卡迪夫大学毕业证(Cardiff毕业证书)成绩单留信学历认证
 
Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...
Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...
Data Visualization Exploring and Explaining with Data 1st Edition by Camm sol...
 
Credit Card Fraud Detection: Safeguarding Transactions in the Digital Age
Credit Card Fraud Detection: Safeguarding Transactions in the Digital AgeCredit Card Fraud Detection: Safeguarding Transactions in the Digital Age
Credit Card Fraud Detection: Safeguarding Transactions in the Digital Age
 
如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证
如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证
如何办理加州大学伯克利分校毕业证(UCB毕业证)成绩单留信学历认证
 

Deep_Reinforcement_Learning_based_Dynamic_Timetable.pptx

  • 1. Deep Reinforcement Learning based Dynamic Optimization of Bus Timetable Ankit Sharma
  • 2. Bus Timetable Optimization  Bus timetable optimization is a key issue to reduce operational cost of bus companies and improve the service quality.  Heuristic algorithms work in offline and does not account for people flow change. https://doi.org/10.48550/arXiv.2107.07066
  • 3. Bus Timetable Optimization  The optimization of the bus timetable aims to consider the interests of both passengers and the bus company, and set the departure time of buses to meet the demand of passenger flow.  The main quantitative indicators are bus congestion and the waiting time of passengers, while the interests of bus companies are mainly affected by the number of departures (departure intervals) in the timetable. https://doi.org/10.48550/arXiv.2107.07066
  • 4. Deep Reinforcement Learning Formulation  Actions – No Departure(0), Departure(1)  States – Time(hr-𝑿𝟏𝒕,),Time(min-𝑿𝟐𝒕,),Load Rate(𝑿𝟑𝒕), Waiting Time(𝑿𝟒𝒕), Carrying Capability(𝑿𝟓𝒕), Stranded Passengers(𝑿𝟔𝒕).  Reward(from reference paper): 1 − (𝑿𝟓𝒕) − 𝛼 ∗ 𝑿𝟒𝒕 − 𝛽 ∗ 𝑿𝟔𝒕, (𝑎𝑐𝑡𝑖𝑜𝑛 = 0) (𝑿𝟓𝒕) − 𝛽 ∗ 𝑿𝟔𝒕, (𝑎𝑐𝑡𝑖𝑜𝑛 = 1) Alternate reward functions were also tested. https://doi.org/10.48550/arXiv.2107.07066 , , ,
  • 5. Bus Environment  Bus timetable considered as episodic task (star-end schedule) with 6 states changing w.r.t time. 𝑺𝒕 = [𝑿𝟏𝒕, 𝑿𝟐𝒕, 𝑿𝟑𝒕, 𝑿𝟒𝒕, 𝑿𝟓𝒕, 𝑿𝟔𝒕]  𝑋1𝑡 − 𝑡ℎ/24, 𝑋2𝑡 − 𝑡𝑚 60  𝑋3𝑡 − 𝑀𝑎𝑥 𝑃𝑎𝑠𝑠𝑒𝑛𝑔𝑒𝑟𝑠 𝑀𝑎𝑥 𝐵𝑢𝑠 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦  𝑋4𝑡 − Normalized Waiting time of all passengers  𝑋5𝑡 − 𝑁𝑒𝑒𝑑 𝑜𝑓 𝐶𝑎𝑟𝑟𝑦𝑖𝑛𝑔 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝐶𝑎𝑟𝑟𝑦𝑖𝑛𝑔 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑜𝑓 𝑣𝑒ℎ𝑖𝑐𝑙𝑒  𝑋6𝑡 − 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑡𝑟𝑎𝑛𝑑𝑒𝑑 𝑝𝑎𝑠𝑠𝑒𝑛𝑔𝑒𝑟𝑠 Assumptions*: • 𝑋6𝑡 was assumed to be exponential distribution after every departure. • Based on number of stranded passengers awaiting bus or remainder post departure are used for calculation of 𝑋4𝑡, 𝑋5𝑡 • 𝑋3𝑡 is calculated based on max bus capacity and no. of stranded passengers at the time of departure • Episode considered from 06:00-10:00Am with decision point at every 5 mins * Real data was not available http://www.muxingyun.com/en/digital-factory
  • 6. DRL - Algorithms Deep SARSA S.No. Parameter Value 1. DNN – Hidden Layers 2 2. DNN-Hidden Units 188 3. Activation Function ReLU 4. Epsilon 0.2 5. Gamma .99 6. Experience memory size 10000 7. Batch Size 32 8. Learning rate .001 DQN S.No. Parameter Value 1. DNN – Hidden Layers 2 2. DNN-Hidden Units 188 3. Activation Function ReLU 4. Epsilon 0.2 5. Gamma .99 6. Experience memory size 10000 7. Batch Size 32 8. Learning rate .001
  • 9. Testing with Deep SARSA & DQN SARSA Episodes 1500 Epsilon 0.2 DQN Episodes 1500 Epsilon 0.2 Reward: Same as in paper.
  • 10. Testing with Deep SARSA & DQN Load Factor Fixed Interval(30 mins): 8, SARSA Departures:23,DQN Departures:23 Normalized Waiting Time Bus Timetable DQN Deep SARSA
  • 11. Testing with Deep SARSA & DQN Required Carrying Capacity Fixed Interval(30 mins): 8, SARSA Departures:23,DQN Departures:23 Stranded Passenger Bus Timetable DQN Deep SARSA
  • 13. Testing with Deep SARSA & DQN SARSA Episodes 1500 Epsilon 0.2 DQN Episodes 1500 Epsilon 0.2 Reward: Modified to reduce number of departures and increase load rate to account for bus agency. Condition added to have Load Rate > 0.7
  • 14. Testing with Deep SARSA & DQN Load Factor Fixed Interval(30 mins): 8, SARSA Departures:15,DQN Departures:15 Normalized Waiting Time Bus Timetable DQN Deep SARSA
  • 15. Testing with Deep SARSA & DQN Required Carrying Capacity Fixed Interval(30 mins): 8, SARSA Departures:15,DQN Departures:15 Stranded Passenger Bus Timetable DQN Deep SARSA
  • 17. Testing with Deep SARSA & DQN SARSA Episodes 1500 Epsilon 0.2 DQN Episodes 1500 Epsilon 0.2 Reward: Modified to reduce number of departures and increase load rate more to account for bus agency. Condition added to have Load Rate > 0.8
  • 18. Testing with Deep SARSA & DQN Load Factor Fixed Interval(30 mins): 8, SARSA Departures:11,DQN Departures:11 Normalized Waiting Time Bus Timetable DQN Deep SARSA
  • 19. Testing with Deep SARSA & DQN Required Carrying Capacity Fixed Interval(30 mins): 8, SARSA Departures:11,DQN Departures:11 Stranded Passenger Bus Timetable DQN Deep SARSA
  • 20. Conclusion  Timetable created with fixed interval of 30 mins had waiting time and stranded passenger going up and couldn’t be contained. Load Rate was also 100% which doesn’t go well with travelers due to heavy congestion.  With right reward function DQN and Deep SARSA were able to understand stranded passenger rate and reduce waiting time and number of stranded passengers while keeping Load Rate less than 90% with just 3 more departures in 4 hours (6:00-10:00AM)  With Deep SARSA & DQN real time decision based on number of stranded passengers can be taken without re-computing the whole problem.