Optimizing Traffic-Signal Timing on Big Bend Boulevard:
Data, Simulation and Validation

CSE 5403 Algorithms for Nonlinear Optimization · Washington University in St. Louis · Fall 2026

Progress report from the data and simulation workstream · November 17, 2026

Summary. 4 County signals on Big Bend Blvd · weekday 5–6 PM · real County timing + counts · SUMO simulation validated (6/6 GEH < 5) · County plan 251 veh-h/h → model plan 179 (29% less) → simulator-optimized 170 (32% less). Preliminary: optimizer is our stand-in for the group's code.

Explain this section

Question: can the four lights on Big Bend Blvd be timed so drivers wait less at evening rush hour?

Split: Michael + Noah build the math model and optimizer. Gabriel + Diego get real data, simulate the street, and test every timing.

Delay = total time all drivers lose at the lights in one hour (veh-h/h). Today's County timing: 251. Best found: 170.

Status: data, simulation and tests done. Report and the group's own optimizer still to come.

1. Corridor and scope

Explain this section

The four intersections we study, the traffic at each, and the distance between them.

Lights this close affect each other: timed well, cars from one light arrive on green at the next.

Terms: NB/SB/EB/WB = northbound/southbound/eastbound/westbound. Approach = traffic arriving from one direction. τ = drive time between lights.

4 signals, north → south · main road Big Bend (NB/SB), cross streets (EB/WB) · design hour weekday 5–6 PM · 13,738 veh/h total approach volume (summed over the 4 intersections).

2026-11-17T16:05:59.865937 image/svg+xml Matplotlib v3.11.2, https://matplotlib.org/ Wydown Boulevard Clayton Road Forest Park Parkway Forsyth Boulevard Pershing Avenue Washington University (Danforth Campus) Clayton Map data © OpenStreetMap contributors N 1 1. Big Bend & Forest Park Pkwy (#051) 2 2. Big Bend & Forsyth (#040) 3 3. Big Bend & Wydown (#063) 4 4. Big Bend & Clayton Rd (#034)
Figure 1. Study area, St. Louis County (Clayton / University City): the four signals and the Big Bend corridor (bold).
Schematic of the four signals with PM approach volumes and spacing
Figure 2. Schematic: PM approach volumes (veh/h) and spacing.
linklength (m)speed (m/s)travel time τ (s)
051 → 04046613.434.8
040 → 06345515.729.1
063 → 03497015.762.0

Table 1. Links between consecutive signals (length from County coordinates; posted speeds from MoDOT 2025).

2. Data

Explain this section

The County emailed one document per light: the light's timing settings and hand-counted traffic (direction and turn of every car).

Scanned count sheets were typed in, then checked by program against the sheet totals. Everything matched.

Every number is tagged measured (real source), derived (calculated from measured) or assumed (textbook value).

Terms: L/T/R = share turning left/straight/right. Cycle = one full round of a light (120 s here). Split = seconds a phase gets. Offset = how the lights' cycles are staggered. Lost time = seconds per cycle spent on yellow and all-red.

2.1 Sources

sourcewhat we usestatus
St. Louis County timing sheets (4 signals)PM pattern 37: cycle, splits, offsets, yellow/all-red, phase sequencemeasured
St. Louis County turning counts (4 signals)5–6 PM volumes by approach + L/T/R · Forsyth 2017, FPP 2021, Wydown 2025, Clayton 2026measured · verified vs sheet totals
MoDOT 2025lanes, speed limitsmeasured
County signal inventorysignal IDs, ownership, coordinatesmeasured
OpenStreetMapapproach lanes at the stop linemeasured

Every number is tagged measured / derived / assumed (appendix).

2.2 PM peak-hour volumes

intersectionapproachveh/hLTRlanessource
Forest Park Pkwy (#051)NB5850.160.710.132051-forest-park-pkwy-2021-06-30.csv
Forest Park Pkwy (#051)SB7300.160.690.142051-forest-park-pkwy-2021-06-30.csv
Forest Park Pkwy (#051)EB9630.250.690.073051-forest-park-pkwy-2021-06-30.csv
Forest Park Pkwy (#051)WB10810.140.660.203051-forest-park-pkwy-2021-06-30.csv
Forsyth (#040)NB8540.190.680.132040-forsyth-2017-03-03.csv
Forsyth (#040)SB9770.100.860.032040-forsyth-2017-03-03.csv
Forsyth (#040)EB7100.070.420.513040-forsyth-2017-03-03.csv
Forsyth (#040)WB7450.450.270.293040-forsyth-2017-03-03.csv
Wydown (#063)NB7430.040.810.152063-wydown-2025-09-26.csv
Wydown (#063)SB16300.040.930.032063-wydown-2025-09-26.csv
Wydown (#063)EB2510.150.490.362063-wydown-2025-09-26.csv
Wydown (#063)WB2980.270.530.202063-wydown-2025-09-26.csv
Clayton Rd (#034)NB6220.200.670.122034-clayton-rd-2026-06-24.csv
Clayton Rd (#034)SB16260.210.770.022034-clayton-rd-2026-06-24.csv
Clayton Rd (#034)EB8920.120.630.254034-clayton-rd-2026-06-24.csv
Clayton Rd (#034)WB10310.230.660.114034-clayton-rd-2026-06-24.csv

Table 2. Measured approach volumes and turning shares, 5:00–6:00 PM. Lanes are approach lanes at the stop line.

2.3 Current signal timing (County, PM pattern)

intersectionsheetpatterncycle (s)offset (s)phase splits (s)Big Bend blockcross blocklost time L (s)
Forest Park Pkwy (#051)October 21, 202637120101:18 2:42 3:18 4:42 5:18 6:42 7:18 8:42606023.4
Forsyth (#040)December 10, 202437120441:15 2:41 3:34 4:30 5:19 6:37566421.4
Wydown (#063)December 10, 202437120681:19 2:47 3:20 4:34 5:15 6:51 7:20 8:34665425.2
Clayton Rd (#034)May 12, 202637120231:27 2:39 3:20 4:34 5:24 6:42 7:23 8:31665427.0

Table 3. County timing, weekday PM. Splits include yellow and all-red. Lost time is the total clearance along the phase sequence.

3. Method

Explain this section

Workflow:

  1. Read and check the County documents.
  2. Build one data file, every number tagged with its source.
  3. Give the math team its input.
  4. Build the virtual street (lights, lanes, distances, speed limits).
  5. Add traffic matching the real counts.
  6. Prepare timing plans: County, textbook methods, optimized.
  7. Simulate each plan; measure delay.
  8. Check the simulation against the counts.
  9. Produce tables, charts and this site.

One command reruns all of it (~2 min).

Model: chooses cycle length, Big Bend's green time, and offsets to minimize total delay, within limits (60–150 s cycle, minimum green, no road over 90% full). Michael + Noah's part.

Simulation: SUMO, a traffic simulator that moves every car individually. Each plan runs 10 times (random arrivals) and is averaged; 15 min warm-up, then one measured hour. Lights run the County's real sequence including left arrows; right turn on red allowed; small flows between lights stand in for side streets.

Where it runs: Gabriel's laptop (pipeline/ folder). This site only displays results.

3.1 Pipeline

stepwhatinputoutput
1Parse + verify County dataCounty PDFs (timing sheets, turning counts)count CSVs, timing JSON; every count checked against the sheet's printed totals
2Build corridor datasetconfig, raw + measured datacorridor.json: every number tagged measured / derived / assumed
3Optimizer inputcorridor.jsonoptimizer_input.json / .csv for the model
4SUMO networkcorridor.json4 signals at real spacing, real approach lanes + left bays, posted speeds
5SUMO demandcorridor.jsonentry flows, measured turn shares, mid-block balancing flows
6Timing planscorridor.json, County timingbaselines, County current plan, formula optimum
7Simulateplans10 seeded runs per plan: delay, stops, throughput
8Validateruns vs countsGEH per approach
9Tables + figureseverything abovereport tables, figures, this page

Table 4. One command (./run_all.sh) reruns all steps. Simulation + Code: live mini-simulation and every program →

3.2 Timing model

Variables: cycle C, Big Bend green gi, offsets · objective: Webster + platoon delay · constraints: 60 ≤ C ≤ 150 s, min greens, x ≤ 0.9 · solver: multistart SLSQP.

3.3 Simulation

settingvalue
softwareSUMO 1.28
network4 signals at real spacing · real approach lanes · left-turn bays · posted speeds
demandmeasured volumes + turn shares · mid-block balancing flows (Table 5)
signalsCounty phase sequence (incl. left arrows), scaled to each plan · right turn on red
run15 min warm-up + 60 min measured · 10 seeds per plan
metricdelay = time lost on corridor links + time waiting to enter (veh-h per hour)
linkdirectionpropagated veh/hcounted veh/hbalancing veh/h
051-040SB716977+261
051-040NB845585-260
040-063SB15381630+92
040-063NB698854+156
063-034SB16811626-55
063-034NB638743+105

Table 5. Mid-block balancing flows: counted volume into a signal minus the volume propagated from the previous signal.

4. Validation

Explain this section

Simulated car counts vs the County's real counts.

The test is the six points between lights, where the simulation must produce the right traffic on its own.

Result: 6 of 6 match (GEH under 5, the standard threshold). The simulation reproduces the real street.

Simulated vs counted volumes, GEH < 5 = pass · interior approaches: 6/6 pass.

intersectionapproachkindcounted veh/hsimulated veh/hGEH
Forest Park Pkwy (#051)NBinterior5856111.08
Forest Park Pkwy (#051)SBentry7307300.00
Forest Park Pkwy (#051)EBentry9639630.00
Forest Park Pkwy (#051)WBentry108110810.00
Forsyth (#040)NBinterior8548720.62
Forsyth (#040)SBinterior9779910.45
Forsyth (#040)EBentry7107100.00
Forsyth (#040)WBentry7457450.00
Wydown (#063)NBinterior7437741.14
Wydown (#063)SBinterior163015083.09
Wydown (#063)EBentry2512510.00
Wydown (#063)WBentry2982980.00
Clayton Rd (#034)NBentry6226220.00
Clayton Rd (#034)SBinterior162615092.96
Clayton Rd (#034)EBentry8928920.00
Clayton Rd (#034)WBentry103110310.00

Table 6. Simulated vs counted volumes under the County's current timing (mean of 10 seeds).

5. Results

Explain this section

Every timing plan, ranked by delay.

  • study1-sumo-loop (170): best timing found by searching inside the simulation.
  • formula-optimum-standin (179): the math model's best (our stand-in for Michael + Noah's code).
  • sumo-coordinated: SUMO's built-in timing tool.
  • webster: 1958 textbook formula.
  • county-current (251): what runs on Big Bend today.
  • equal-split: equal green for both streets (naive baseline).

Columns: vs County = % change from today (negative is better). Stops/veh = average stops per car. Throughput = cars served (same for all plans, so the comparison is fair).

Result: the model's timing cuts delay 29% vs today. Provisional until Michael + Noah's code is plugged in.

plancycle (s)SUMO delay, veh-h/h (mean ± sd)vs Countystops / vehthroughput veh/hformula delaysource
study1-sumo-loop124.7169.9 ± 5.6-32%1.697861115.1Study 1: Nelder-Mead on SUMO delay (120 evaluations x 2 seeds), warm start = formula optimum
formula-optimum-standin124.8178.9 ± 3.5-29%1.747854114.2stand-in for Michael+Noah: formula model, multistart SLSQP (formula.py)
sumo-coordinated150.0203.3 ± 3.7-19%1.627842128.6baseline: SUMO tlsCoordinator.py offsets on the common Webster plan
webster-per-light150.0225.7 ± 4.4-10%2.017825—baseline: Webster (1958) cycle + splits per light, uncoordinated
webster-common150.0230.2 ± 3.8-8%1.777858145.8baseline: common Webster cycle, Webster splits, offsets 0
county-current120.0251.3 ± 3.0+0%2.267852963.2MEASURED: St. Louis County timing sheets, pattern 37 (PM weekday)
equal-split150.0310.4 ± 8.4+24%2.2877691830.3baseline: equal green to main and side, common Webster cycle, offsets 0

Table 7. All plans, 10 seeds each. formula-optimum-standin = our stand-in for the group's optimizer.

2026-11-17T16:05:59.944658 image/svg+xml Matplotlib v3.11.2, https://matplotlib.org/ 0 50 100 150 200 250 300 350 corridor delay in SUMO (vehicle-hours per hour) study1-sumo-loop formula-optimum-standin sumo-coordinated webster-per-light webster-common county-current equal-split 170 179 203 226 230 251 310
Figure 3. Simulated corridor delay by plan; County's current timing in black (error bars: one standard deviation over seeds).

6. Study 1: simulator-in-the-loop vs formula model

Explain this section

Two ways to find a timing: the formula (~1 s, simplified) vs searching inside the simulation (~7 min, realistic).

Left chart: 30 random timings. If the formula judged like the simulation, the dots would form a rising line. They don't (Spearman 0.19; 1 = perfect agreement).

Right chart: the simulation search improving, then levelling off.

Result: the formula gets close fast but can't reliably rank timings. Simulation search finds ~5% better timings at ~300× the compute.

testresult
Does the formula rank plans like SUMO? ({s1['prediction']['n_plans']} random plans)weakly · Spearman ρ = {s1['prediction']['spearman']:.2f}
Formula optimizer{fo:.0f} veh-h/h · {s1['formula_optimizer']['seconds']} s
Simulator-in-the-loop (Nelder–Mead, {s1['sumo_loop']['evaluations']} evals){sl:.0f} veh-h/h · ~{s1['sumo_loop']['seconds']//60} min
2026-11-17T16:05:59.972270 image/svg+xml Matplotlib v3.11.2, https://matplotlib.org/ 125 130 135 140 145 150 155 formula model delay (veh-h/h) 180 200 220 240 260 280 SUMO delay (veh-h/h)
Figure 4. Formula vs simulated delay, 30 random feasible plans.
2026-11-17T16:05:59.997802 image/svg+xml Matplotlib v3.11.2, https://matplotlib.org/ 0 20 40 60 80 100 120 SUMO evaluations 166 168 170 172 174 176 178 180 best delay so far (veh-h/h)
Figure 5. Simulator-in-the-loop optimization: best delay found vs evaluations.

7. Interfaces for the optimizer

Explain this section

The files passed between our part and Michael + Noah's: we send traffic data; they send back timings; we simulate them.

If their format differs, one small file changes.

In: optimizer_input.json (q, s per approach; τ per link; parameters) · Out: plan file:

{ "name": "...", "C": 124.7,
  "lights": { "051": {"g_main": 37.0, "offset_s": 0.0},
              "040": {"g_main": 49.8, "offset_s": 52.7}, ... } }
g_main   = Big Bend green (s); cross-street green = C - L_i - g_main
offset_s = start of the Big Bend block relative to Forest Park Pkwy (s)

8. Modelling choices for the group to confirm

Explain this section

Choices that affect the report's claims. The group should agree.

  1. Real lost time and capacity (21–27 s per cycle, from the County's sequence) fed to the model. Needed for its timings to work. Michael + Noah to approve.
  2. Left-turn arrows: all four real lights have them. Revisit "future work".
  3. Simplified virtual street: the full map jammed the simulation; we built a clean one from real distances and lanes.
  4. Assumptions where nothing was measured (e.g. left-turn lane length), all labelled.
  1. Lost time + capacity computed along the real signal sequence (L ≈ 21–27 s) · needs Michael + Noah's OK · switchable back.
  2. Real signals use left-turn arrows · simulation includes them · revisit "future work".
  3. Simulated network generated from measured geometry (full OpenStreetMap version jammed).
  4. Assumed: 1,800 veh/h/lane · p = 0.8 · 90 m left bays · right turn on red.

9. Status

Explain this section

Done: data, simulation, testing, Study 1.

Open: Michael + Noah's optimizer, group sign-off on section 8, our report sections, final report and submission.

itemstatus
Real corridor data: County timing + turning counts for all 4 signals, verifiedDone
SUMO simulation of the corridor, validated against countsDone
All timing plans simulated (10 seeds each); results tables + figuresDone
Study 1 (simulator-in-the-loop vs formula model)Done (first full run)
Reproducible code package + README (one command)Done (data/simulation part)
Michael + Noah's optimizer (signal_opt.py) plugged inOpen: stand-in used
Group agreement on the modelling choices in §8Open
Written report sections (Data, Setup, Study 1)Open
Full report (8–12 pp), Contributions, submission + code to TAOpen

10. Reproducing the results

Explain this section

The commands the grader uses to rerun everything from the original County documents.

Appendix: the source of every number (measured / derived / assumed, with evidence).

cd pipeline
./run_all.sh                                   # data, network, plans, 10-seed runs, validation, tables
.venv/bin/python src/study1.py --n-random 30 --budget 120   # Study 1

Inputs: data/, config/ · outputs: build/ (regenerated) · Simulation + Code: all code, settings and data, plus a zip →

Appendix: where every number comes from
whereinputvaluesourceevidence
#051 NBvolume585 veh/hmeasuredSt. Louis County turning-movement count 051-forest-park-pkwy-2021-06-30.csv, 17:00-18:00 (L 95, T 415, R 75)
#051 NBsaturation flow (model)2632 veh/hderivedeffective: q / Y_block, Y_block = 0.222 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#051 SBvolume730 veh/hmeasuredSt. Louis County turning-movement count 051-forest-park-pkwy-2021-06-30.csv, 17:00-18:00 (L 120, T 505, R 105)
#051 SBsaturation flow (model)3285 veh/hderivedeffective: q / Y_block, Y_block = 0.222 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#051 EBvolume963 veh/hmeasuredSt. Louis County turning-movement count 051-forest-park-pkwy-2021-06-30.csv, 17:00-18:00 (L 239, T 661, R 63)
#051 EBsaturation flow (model)2457 veh/hderivedeffective: q / Y_block, Y_block = 0.392 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#051 WBvolume1081 veh/hmeasuredSt. Louis County turning-movement count 051-forest-park-pkwy-2021-06-30.csv, 17:00-18:00 (L 148, T 713, R 220)
#051 WBsaturation flow (model)2758 veh/hderivedeffective: q / Y_block, Y_block = 0.392 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#051lost time L23.4 smeasuredCounty sheet, signal_sequence
#040 NBvolume854 veh/hmeasuredSt. Louis County turning-movement count 040-forsyth-2017-03-03.csv, 17:00-18:00 (L 164, T 580, R 110)
#040 NBsaturation flow (model)2556 veh/hderivedeffective: q / Y_block, Y_block = 0.334 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#040 SBvolume977 veh/hmeasuredSt. Louis County turning-movement count 040-forsyth-2017-03-03.csv, 17:00-18:00 (L 102, T 842, R 33)
#040 SBsaturation flow (model)2924 veh/hderivedeffective: q / Y_block, Y_block = 0.334 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#040 EBvolume710 veh/hmeasuredSt. Louis County turning-movement count 040-forsyth-2017-03-03.csv, 17:00-18:00 (L 51, T 296, R 363)
#040 EBsaturation flow (model)2635 veh/hderivedeffective: q / Y_block, Y_block = 0.269 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#040 WBvolume745 veh/hmeasuredSt. Louis County turning-movement count 040-forsyth-2017-03-03.csv, 17:00-18:00 (L 333, T 198, R 214)
#040 WBsaturation flow (model)2765 veh/hderivedeffective: q / Y_block, Y_block = 0.269 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#040lost time L21.4 smeasuredCounty sheet, signal_sequence
#063 NBvolume743 veh/hmeasuredSt. Louis County turning-movement count 063-wydown-2025-09-26.csv, 17:00-18:00 (L 33, T 602, R 108)
#063 NBsaturation flow (model)1639 veh/hderivedeffective: q / Y_block, Y_block = 0.453 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#063 SBvolume1630 veh/hmeasuredSt. Louis County turning-movement count 063-wydown-2025-09-26.csv, 17:00-18:00 (L 64, T 1511, R 55)
#063 SBsaturation flow (model)3596 veh/hderivedeffective: q / Y_block, Y_block = 0.453 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#063 EBvolume251 veh/hmeasuredSt. Louis County turning-movement count 063-wydown-2025-09-26.csv, 17:00-18:00 (L 37, T 124, R 90)
#063 EBsaturation flow (model)1537 veh/hderivedeffective: q / Y_block, Y_block = 0.163 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#063 WBvolume298 veh/hmeasuredSt. Louis County turning-movement count 063-wydown-2025-09-26.csv, 17:00-18:00 (L 80, T 159, R 59)
#063 WBsaturation flow (model)1824 veh/hderivedeffective: q / Y_block, Y_block = 0.163 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#063lost time L25.2 smeasuredCounty sheet, signal_sequence
#034 NBvolume622 veh/hmeasuredSt. Louis County turning-movement count 034-clayton-rd-2026-06-24.csv, 17:00-18:00 (L 126, T 419, R 77)
#034 NBsaturation flow (model)1460 veh/hderivedeffective: q / Y_block, Y_block = 0.426 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#034 SBvolume1626 veh/hmeasuredSt. Louis County turning-movement count 034-clayton-rd-2026-06-24.csv, 17:00-18:00 (L 344, T 1244, R 38)
#034 SBsaturation flow (model)3816 veh/hderivedeffective: q / Y_block, Y_block = 0.426 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#034 EBvolume892 veh/hmeasuredSt. Louis County turning-movement count 034-clayton-rd-2026-06-24.csv, 17:00-18:00 (L 110, T 558, R 224)
#034 EBsaturation flow (model)3226 veh/hderivedeffective: q / Y_block, Y_block = 0.276 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#034 WBvolume1031 veh/hmeasuredSt. Louis County turning-movement count 034-clayton-rd-2026-06-24.csv, 17:00-18:00 (L 237, T 685, R 109)
#034 WBsaturation flow (model)3729 veh/hderivedeffective: q / Y_block, Y_block = 0.276 = critical flow ratio of this block along the County phase sequence (measured volumes, lanes; 1800 veh/h/lane)
#034lost time L27.0 smeasuredCounty sheet, signal_sequence
modelp0.8assumedmodel defaults from the Nov 10 formulation
modelg_min_s10.0assumedmodel defaults from the Nov 10 formulation
modelC_min_s60.0assumedmodel defaults from the Nov 10 formulation
modelC_max_s150.0assumedmodel defaults from the Nov 10 formulation
modelx_max0.9assumedmodel defaults from the Nov 10 formulation
networkleft-turn bay length90 massumedtypical urban 75–100 m; not measured