Midterm: Toward a Self-Sizing Mobility-on-Demand System Simulator

For our midterm project involving Citi Bike, my group (Aaron Arntz, T.K. Broderick) was interested in exploring how to simulate a self-sizing network. However, before we could do that, we needed to create a simulation that somewhat accurately depicted the behavior of the real system.

T.K. sculpted our overall message, tone and delivery – what was important about the system we were we aiming to simulate, and how would we connect our simulator to real world data to make it relevant? Aaron mined the Citi Bike System Data for clues about real-world behavior and refined them into data points describing stations and bikes. And I incorporated T.K.'s and Aaron's findings into a system simulator which I wrote from scratch in NetLogo.

In order to have as accurate a simulator as possible, we decided to populate NetLogo with the actual number of stations in their actual geography. The latitude and longitude of stations in the simulator are mapped to Cartesian coordinates so that the shapes of Manhattan and Brooklyn are easily distinguished.



To populate the stations' data, we used two types of information: (1) a snapshot of station capacities at a particular moment in time and (2) aggregate data from all trips in August 2014. Each station has the following fields:

  • Station ID (1)
  • # of available bikes (1)
  • # of available docks (1)
  • Master launch rate (2)
  • Launch table (2)

The last two fields are the most important for achieving realistic behavior. The master launch rate is the percentage of trips from a given station out of all trips. The launch table is a collection of destination station IDs whose frequencies match the percentage of trips from a given station to a given station. Note that the master launch rates and the launch table rates always add up to 1, respectively, as they account for all trips taken in August 2014.

Below is the launch table for station 339 (Avenue D & E 14 St), constructed from the percentage of trips from station 339 to any of the other station IDs in the list. Note that selecting a random item from the list will result in the selection of a given station ID with the same probability as actually occurred from this particular station in August 2014.

[339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 339 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 295 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 361 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 259 279 279 279 279 279 279 279 279 279 279 279 279 279 279 279 279 279 279 279 307 307 307 307 307 307 307 307 307 307 307 307 307 307 307 307 332 332 332 332 332 332 332 332 332 332 332 332 332 475 475 475 475 475 475 475 475 475 475 475 475 475 501 501 501 501 501 501 501 501 501 501 501 501 507 507 507 507 507 507 507 507 507 507 306 306 306 306 306 306 306 306 306 306 351 351 351 351 351 351 351 351 351 351 385 385 385 385 385 385 385 385 385 385 412 412 412 412 412 412 412 412 412 412 337 337 337 337 337 337 337 337 337 504 504 504 504 504 504 504 504 504 375 375 375 375 375 375 375 375 375 519 519 519 519 519 519 519 519 519 497 497 497 497 497 497 497 497 326 326 326 326 326 326 326 326 502 502 502 502 502 502 502 502 3002 3002 3002 3002 3002 3002 3002 3002 445 445 445 445 445 445 445 284 284 284 284 284 284 284 415 415 415 415 415 415 415 280 280 280 280 280 280 280 315 315 315 315 315 315 315 285 285 285 285 285 285 401 401 401 401 401 401 293 293 293 293 293 293 511 511 511 511 511 511 2009 2009 2009 2009 2009 455 455 455 455 455 518 518 518 518 518 393 393 393 393 393 341 341 341 341 341 350 350 350 350 350 263 263 263 263 454 454 454 454 428 428 428 428 335 335 335 335 312 312 312 312 363 363 363 363 296 296 296 296 264 264 264 264 347 347 347 347 224 224 224 224 297 297 297 297 433 433 433 161 161 161 266 266 266 250 250 250 342 342 342 2003 2003 2003 265 265 265 394 394 394 150 150 150 531 531 531 505 505 505 331 331 331 438 438 438 311 311 311 490 490 490 382 382 382 260 260 260 2012 2012 2012 410 410 410 317 317 376 376 379 379 302 302 380 380 168 168 528 528 127 127 316 316 2023 2023 304 304 540 540 236 236 174 174 400 400 486 486 404 404 360 360 473 473 432 432 305 305 496 496 2006 2006 237 237 532 532 512 512 153 153 301 301 483 435 2022 494 358 345 521 310 308 491 320 392 487 349 232 294 411 482 444 212 291 459 173 128 303 252 340 79 457 300 461 318 405 116 2017 229 427]

To better represent the stations' states, they are color-coded to match their current usage. Green stations are balanced (i.e. less than 95% full and greater than 95% empty), red stations are too full (i.e. greater than 95% full), and blue stations are too empty (i.e. less than 5% full). Here is the state of the system before any simulation has occurred:



Our focusing on stock rebalancing issues meant defining the concept of station balance, or the potential use of the station compared to its actual use. The goal of a self-sizing system would be the spread the usage evenly (or better yet, optimally) over all stations so as to minimize congestion due to either a shortage of bikes or a shortage of docks.

The final piece of the simulator was to model the basic behavior of a user (bike) arriving at a station that is completely full. We guessed that a realistic radius for seeking out an alternate station is 0.4 miles, or about 8 minutes walking. The bikes in the simulator wait a short amount of time, then proceed to another station within this radius if no docks are available.

After running the simulator many times it was clear that our model highlights one of the basic truths of the system: on average more bikes are launched from Manhattan stations than Brooklyn stations, and this causes an imbalance if no restocking is performed. This can be seen in the progression below (the first image is the same initial system state shown above).


We were pleased to identify some of the same stations that our previous visualization picked out. This gives further proof that their are certain stations that may be candidates for self-sizing and/or elimination.



Having this simulator as a base for our future exploration of the Citi Bike system is an important step, and we are pleased that we were able to make such progress for our midterm project.

No comments:

Post a Comment

Speak now...