8000 GitHub - MG-Choi/ABTS: Agent Based Travel Scheduler library to predict individual travel schedules
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

MG-Choi/ABTS

BranchesTags

Repository files navigation

ABTS

Agent Based Travel Scheduler library to predict individual travel schedules

License

ABTS / version 0.0.3

  • install:
!pip install ABTS
import ABTS as abts
This algorithm is a simulation model for predicting individual travel schedules, named as ABTS (Agent-Based Travel Scheduler). It utilizes four parameters for prediction, and it can decompose past travel data into trip purposes and age groups using historical OD (Origin-Destination) data. Future predictions are also possible. More detailed descriptions of the model can be found in the following two papers.

  • Choi, M., & Hohl. A. (2024). Derivation of Spatiotemporal Risk Areas and Travel Behaviors During Pandemic Through Reverse Estimation of Mobility Patterns by Agent-Based Modeling, Spatial and Spatio-temporal Epidemiology. Under review (1st round)

  • Choi, M., Seo, J., & Hohl. A. (2024). ABTS: Agent-Based Travel Scheduler. In progress

Figure 1 shows the framework of ABTS. ABTS utilizes a three-step process to generate the initial outcome of individual travel schedules: 1) Comprehensive Travel Classifier, 2) Land-Use Estimator, and 3) Individual Travel Schedule. The output of ABTS includes individual daily trip chains that consists of each trip purpose, occurrence time, destination, duration, and mode, tailored according to age group.
Figure1. Framework_of_ABTS

Table 1. Classification in ABTS

In the comprehensive travel classifier, trip purposes, trip modes, and age groups are classified based on their respective criteria as shown in Table 1. For example, trip purposes are divided into 9 major trip purposes and 13 sub trip purposes.
Table 1. Classification in ABTS

Table 1. Classification in ABTS

In the Land-Use estimator, land-use data is generated. We have constructed the land-use into 13 categories, similar to the sub trip purposes in Table 1, and organized them into shp files. The source of the data is described in the 0. Preprocessing stage below.

Run the model with example data

Here, we will run the process of the Individual Travel Schedule Generator using example data. The case area is Milwaukee County, and we will demonstrate the process of decomposing and predicting past travels by trip purpose and age for September 2020.

0. Preprocessing (using sample simulation in library)

Here, we show the preprocessing steps for 'NHTS' data and 'SafeGraph' data in order to run the ABTS model. The original data can be downloaded from the two websites described below, and the data used in this example is also downloaded from these sites. However, it seems that SafeGraph Neighborhood data is currently available from other sources, yet remains accessible.

  • NHTS data: NHTS data: The source is the Federal Highway Administration (FHWA), and it uses data from the 2017 National Household Travel Survey (https://nhts.ornl.gov/downloads).

  • SafeGraph data: The source is SafeGraph, and the data is named Neighborhood Patterns (https://docs.safegraph.com/docs/neighborhood-patterns). This data represents the movements of people by CBG (Census Block Group) as OD (Origin-Destination) for a randomly selected week of the month, recording how many people moved from one CBG to another. However, it does not contain information on trip purposes or ages, collecting only counts. We used this data to create a probabilistic model and applied it to an algorithm for selecting Destination CBGs based on people's trip purposes.

0.0. Data import

We uploaded the origin data on Google drive (https://drive.google.com/drive/folders/1YlsNPeTIC-0hjgYNnatuhmIGWQVSvg0-?usp=sharing)
  • 'trippub.csv': NHTS data (2017)
  • 'Milwaukee_parcels.shp': Milwaukee landuse data
  • 'neighbor_2020_09.xlsx': SafeGraph Neighborhood data in Milwaukee (2020/09)
path = "[Your Path where you save three datasets]"
trippub = pd.read_csv(path + 'trippub.csv') # NHTS
neighbor_2020_09 =  pd.read_excel(path + 'neighbor_2020_09.xlsx') # SafeGraph
landUse = gpd.read_file(path + 'Milwaukee_parcels.shp') # Landuse

0.1. Preprocess NHTS data

0.1.1. Organize columns of NHTS data

This function reorganizes the NHTS dataset by selecting columns for analysis, mapping certain categorical codes to more meaningful values, and introducing new columns to better represent the data. It focuses on clarifying trip purposes, transportation modes, and travel days based on NHTS coding schemes.
trippub_organized = abts.organize_columns(trippub, print_progress = True)
display(trippub_organized.head())
HOUSEID PERSONID PERSONID_new HHSTFIPS R_AGE_IMP R_AGE_new WHYFROM WHYTO TRPMILES DWELTIME STRTTIME ENDTIME TRAVDAY TRAVDAY_new TDAYDATE TRPPURP_new TRPTRANS_new
0 30000007 1 300000071 37 67 Seniors 1 19 5.244 295 1000 1015 2 Weekday 201608 S_d_r Car
1 30000007 1 300000071 37 67 Seniors 19 1 5.149 -9 1510 1530 2 Weekday 201608 Home Car
2 30000007 2 300000072 37 66 Seniors 3 1 84.004 540 700 900 2 Weekday 201608 Home Car
3 30000007 2 300000072 37 66 Seniors 1 3 81.628 -9 1800 2030 2 Weekday 201608 Work Car
4 30000007 3 300000073 37 28 Adult 1 8 2.25 330 845 900 2 Weekday 201608 S_d_r Car
  • PERSONID_new: Newly assigned unique ID
  • HHSTFIPS: Milwaukee code
  • R_AGE_IMP: Age
  • R_AGE_new: new classification of age group
  • Information on the remaining columns can be found in the Format Library provided at https://nhts.ornl.gov/downloads.

0.1.2. Preprocess NHTS

Cleans and preprocesses the NHTS dataset (After running organize_columns function) to correct data anomalies and enhance data quality for analysis. The preprocessing steps include adjusting dwelling times, calculating travel times, and ensuring data consistency across the travel records.
repaired_NHTS = abts.preprocess_NHTS(trippub_organized, print_progress = True) # will take long time
repaired_NHTS.to_csv('[yourPath]' + 'repaired_NHTS.csv') # save
display(repaired_NHTS.head())
Unnamed: 0 uniqID age Day_Type Trip_pur sta_T_hms arr_T_hms end_T_hms Dwell_T_min Trip_T_min sta_T_min arr_T_min end_T_min age_class
0 0 300000071 67 Weekday Home 0 0 1000 600 0 0 0 600 Seniors
1 1 300000071 67 Weekday S_d_r 1000 1015 1510 295 15 600 615 910 Seniors
2 2 300000071 67 Weekday Home 1510 1530 2400 510 20 910 930 1440 Seniors
3 3 300000073 28 Weekday Home 0 0 845 525 0 0 0 525 Adult
4 4 300000073 28 Weekday S_d_r 845 900 1430 330 15 525 540 870 Adult
  • age: Origin age
  • Day_Type: Weekday or Weekend
  • Trip_pur: trip purpose
  • sta_T_hms: Trip start time. (1000 -> 10:00, 1510 -> 15:10)
  • arr_T_hms: Trip arrival time
  • end_T_hms: End time of activity
  • Dwell_T_min: Dwell time (duration - minutes)
  • Trip_T_min: Travel time (duration - minutes)
  • sta_T_min: Trip start time corresponding to 'sta_T_hms' (600 -> 10:00, 910 -> 15:10)
  • arr_T_min: Trip arrival time corresponding to 'arr_T_hms'
  • end_T_min: End time of activity corresponding to 'end_T_hms'
  • age_class: age groups

0.1.3. Preprocessing for tripmode

Processes the NHTS dataset to create a new trip purpose category and calculates the probability of trip mode choices by age and newly defined trip purpose. This function aims to simplify the analysis of trip behaviors across different demographics and trip purposes by mapping detailed purposes into broader categories.
trip_mode_prop_all = abts.preprocess_NHTS_tripMode(trippub_organized) 
trip_mode_prop_all.to_csv('[yourPath]' + 'trip_mode_prop_all.csv')  # save
display(trip_mode_prop_all.head())
age_class Trip_pur Trip_mode nominator denominator Trip_modeP
0 Adult D_shop Bicy 126 15158 0.00831244
1 Adult D_shop Car 13528 15158 0.892466
2 Adult D_shop PTrans 210 15158 0.0138541
3 Adult D_shop Walk 1294 15158 0.0853675
4 Adult Home Bicy 727 46155 0.0157513
  • Trip_modeP: Probability of choosing trip mode for each age class and trip purpose

0.2. Preprocess SafeGraph data

0.2.1. DO to OD, Covert Destination area to Origin area

Converts Destination-Origin (DO) data into Origin-Destination (OD) data using the input SafeGraph Neighborhood data. The function processes columns related to work behavior and device home areas during weekdays and weekends, transforming them into a format that specifies how many devices move from one area to another.
neighbor_2020_09_OD = abts.DOtoOD(neighbor_2020_09, print_progress = True)
display(neighbor_2020_09_OD.heaD())
area work_behavior_from_area weekday_device_from_area_home weekend_device_from_area_home
3 550791863001 {'550791863001': 9, '550790144002': 4,...} {'550791863001': 43, '550791402011': 4,...} {'550791863001': 11, '550790148001': 4,...}
4 550790194001 {'550791503013': 4, '550790146002': 4,...} {'550790194001': 28, '550791009002': 6,... } {'550790194001': 16, '550791009002': 5,...}
  • area: Census Block Groups (CBG), origin home area
  • work_behavior_from_area: # people trip to workplace from 'area' (Workplace: # people)
  • weekday_device_from_area_home: # people trip from 'area' to certain CBG in weekday (certain CBG: # people)
  • weekend_device_from_area_home: # people trip from 'area' to certain CBG in weekend

0.2.2. Computing probability of trips from origin cbg to dest cbg

Computes the probability of moving from an origin CBG to a destination CBG, factoring in the spatial attractiveness weight Ws and the distance sensitivity index Wd.
Please refer to the Eqution 10 that describe how to compute this probability.
prob_trips_2020_09_ws05_wd025 = abts.compute_probabilityByk_Ws_Wd(neighbor_2020_09_OD, landUse, W_s = 0.5, W_d = 0.25)
display(prob_trips_2020_09_ws05_wd025.head())

0.2.3. Combine all probability tables by their month

The process involves pre-creating many combinations of Ws and Wd, and then concatenating them into one. This varies depending on how many combinations of Ws and Wd the user creates, hence no code has been provided for it. In this example, the combinations are set as Ws = 0.5,1.5,1 and Wd = 0.25,0.5,0.75, and they are combined into one under the name 'prob_2020_09_combined'.

0.2.4. fill empty probability

Fills empty probability distributions for trip purposes in a row of a input DataFrame with alternative values based on a predefined hierarchy of preferences. This ensures that each trip purpose category has a valid probability distribution, either specific or borrowed from a related category.
prob_2020_09_combined = abts.prob_2020_09_combined.apply(fill_values, axis = 1)
prob_2020_09_combined.to_csv('[yourPath]' + 'prob_2020_09_combined.xlsx') # save
display(prob_2020_09_combined.head())
area weekday_Work weekend_Work weekday_School weekday_University weekday_Dailycare weekday_Religion weekday_Large_shop weekday_Etc_shop weekday_Meals weekday_V_fr_rel weekday_Rec_lei weekday_Serv_trip weekday_Others weekend_School weekend_University weekend_Dailycare weekend_Religion weekend_Large_shop weekend_Etc_shop weekend_Meals weekend_V_fr_rel weekend_Rec_lei weekend_Serv_trip weekend_Others Ws Wd
1153 550790012001 {'550790012002': 0.23171, '550790019004': 0.27989,...} {'550790012002': 0.23171, '550790019004': 0.27989,...} {'550790040001': 0.05985, '550790008001': 0.086...} {'550791863001': 0.07454, '550790601012': 0.35423,...} {'550791863001': 0.00127, '550790601021': 0.04264,...} {'550790076001': 0.00092, '550790039003': 0.00814,...} {'550790001021': 0.00877, '550790602001': 0.02857,...} {'550791863001': 0.00024, '550791009002': 0.00016,...} {'550791863001': 0.00059, '550791009002': 0.00018,...} {'550791863001': 0.00013, '550790194001': 3e-05,...} {'550791863001': 0.00051, '550790194001': 3e-05,...} {'550791863001': 0.00089, '550790194001': 5e-05,...} {'550791863001': 0.00039, '550791009002': 6e-05,...} {'550790008001': 0.20682, '550790034001': 0.08553,...} {'550791863002': 0.24308, '550791853001': 0.75692} {'550790011001': 0.37633, '550790001012': 0.0137,...} {'550790008001': 0.09629, '550790034001': 0.03982,...} {'550790602001': 0.04491, '550790501011': 0.00957,...} {'550790014002': 0.02371, '550790034001': 0.01567,...} {'550790034001': 0.02218, '550790069002': 0.003,...} {'550790014002': 0.01961, '550790008001': 0.05287,...} {'550790014002': 0.02669, '550790008001': 0.04266,...} {'550790014002': 0.04832, '550790008001': 0.0546,...} {'550790014002': 0.00924, '550790008001': 0.02088,...} 0.5 0.5
5417 550790067002 {'550790067001': 1.0} {'550790067001': 1.0} {'550791854002': 0.0472, '550791857001': 0.17183,...} {'550791863002': 0.58456, '550791853001': 0.41544} {'550791854002': 0.31535, '550790011001': 0.14428,...} {'550791854002': 0.04848, '550790065003': 0.06424,...} {'550791402011': 0.00147, '550791401001': 0.00067,...} {'550791854002': 0.04594, '550791402011': 0.00069,...} {'550791854002': 0.01903, '550791402011': 0.00057,...} {'550791854002': 0.01833, '550791402011': 0.0024, ...} {'550791854002': 0.01252, '550791402011': 0.00038,...} {'550791854002': 0.05368, '550791402011': 0.00162,...} {'550791854002': 0.05892, '550791402011': 0.00068,...} {'550790602001': 0.07441, '550790070004': 0.29906,...} {'550790141001': 0.73842, '550791853001': 0.26158,...} {'550790352001': 1.0} {'550790024001': 0.01743, '550790070004': 0.21597...} {'550790602001': 0.9111, '550790044001': 0.0889} {'550791009002': 0.00877, '550790602001': 0.57703,...} {'550791009002': 0.00354, '550790108002': 0.02074,...} {'550791009002': 0.00591, '550790108002': 0.07082,...} {'550791009002': 0.00499, '550790024001': 0.10415,...} {'550791009002': 0.0049, '550790602001': 0.31572,...} {'550791009002': 0.00464, '550790024001': 0.02351,...} 1 0.25
4588 550791007002 {'550791501002': 9e-05, '550790164002': 0.00022...} {'550791501002': 9e-05, '550790164002': 0.00022,...} {'550791401001': 0.00034, '550790125002': 0.02147,...} {'550791501002': 0.0, '550791008001': 0.01605,...} {'550791201022': 0.00167, '550791002001': 0.01508,...} {'550790301002': 0.0, '550791201022': 0.00145,...} {'550791201022': 0.00345, '550791401001': 7e-05,...} {'550791009002': 0.39036, '550791201022': 0.00109,...} {'550791009002': 0.1375, '550790076002': 0.0,...} {'550791009002': 0.05497, '550790301002': 0.0,...} {'550791009002': 0.05592, '550790301002': 0.0,...} {'550791009002': 0.05516, '550790076002': 0.0,...} {'550791009002': 0.06552, '550791201022': 0.00164,...} {'550791301002': 0.03874, '550791401001': 0.00045,...} {'550791008001': 1.0} {'550791201022': 0.00619, '550791503013': 1e-05,...} {'550791301002': 0.00017, '550791201022': 0.00331,...} {'550791201022': 0.00902, '550791401001': 0.00018,...} {'550791009002': 0.42921, '550791301002': 0.01761,...} {'550791009002': 0.24759, '550791201022': 0.00241,...} {'550791009002': 0.04555, '550791301002': 0.00112,...} {'550791009002': 0.14939, '550791301002': 0.00938,...} {'550791009002': 0.13345, '550791301002': 0.00096,...} {'550791009002': 0.10349, '550791301002': 2e-05,...} 1.5 0.75
  • The output dataframe exhibits the probability of going to each destination by trip purpose for both weekdays and weekends. Each cell in the dataframe has a key representing the destination CBG and a value representing the probability. These defined probability values are used later to determine the destination.

1. Execution for ABTS

In this section, we intend to run the ABTS using the preprocessed data mentioned above. Since the preprocessed data is provided within this library, you can just import those example datasets to run the model.

1.0. Data import

import ABTS as abts

prob_2020_09_combined = abts.prob_2020_09_combined # Combined Probability of travels from O to D in Sep, 2020
repaired_NHTS = abts.repaired_NHTS # preprocessed NHTS
trip_mode = abts.trip_mode # trip mode
cbg = abts.cbg # CBG shp file
network_road = abts.network_road # road network data in Milwaukee
abts.cbg is a downloaded dataset from the Census. abts.network_road dataset is imported using the code stated below.
import osmnx as ox
network_road = ox.graph_from_place('Milwaukee, Wisconsin, USA', network_type='drive')

1.1. Trip Occurence Builder

In the trip occurrence builder, the probability and total number of trips for each individual are computed using the NHTS data. The daily trip occurrence probability, P(θ), is determined for each day type d, age group a, and trip purpose t for an individual, employing a Naïve Bayes probability model. Day types are differentiated into weekdays and weekends. The parameter Wt(t,d) modifies the occurrence probability for each trip purpose, where t and d are represented as superscripts, indicating that different values might be allocated based on the trip purpose t and the day type d.
Equation 1. The probability of a person within age group ‘a’ having ‘t’ trip occurs in a single day.

Equation 1. The probability of a person within age group ‘a’ having ‘t’ trip occurs in a single day.

Following this, the total number of trips k for each agent i is determined as shown in equation 2. Here, Wk serves as a parameter that fine-tunes the daily trip count for each individual. k(t,i) is modeled to follow a multinomial distribution, where this distribution is sampled for each trip purpose t based on the probability distribution Wt∙p(1,a,d), which is conditioned on the age group a, day type d, and trip purpose t. Furthermore, to ensure alignment with the actual data, the aggregate of all daily trip counts k(t,i) is set to match θi.
Equation 2. Counting ‘t’ trips occurring ‘k’ times for a single individual ‘i’ in a day.

Equation 2. Counting ‘t’ trips occurring ‘k’ times for a single individual ‘i’ in a day.

1.1.0. Data staging

You can run this data staging only once for generating multiple travel schedules in the same area. We set the study area to Milwaukee, so you can just save the results of here to create multiple schedules in Milwaukee.
1.1.0.1. The probability of a person with age ‘a’ having ‘t’ trips occurs in a single day
Calculate the probability of trip purpose by age group and day type (Weekday or Weekend) using Naive Bayes.
naive_result = abts.naive_bayes_prob_with_day(repaired_NHTS, age_col = 'age_class', tripPurpose_col= 'Trip_pur', travday_col= 'Day_Type')

naive_result.head()
Age Trip_pur Day_Type Prob
67 Child Others Weekend 0.0123082
18 Adult Home Weekday 0.46988
84 Teen Others Weekday 0.00842771
19 Adult Home Weekend 0.483808
21 Adult S_d_r Weekend 0.0666646
  • Now, the results exhibit probability of travels by trip purpose for each age group and day type.
1.1.0.2. The number of trips based on unique IDs, age, day type, and trip purpose derived from NHTS data
Generate a distribution of trips based on unique IDs, age, day type, and trip purpose (from origin NHTS data).
trip_count = abts.generate_trip_distribution(repaired_NHTS, id_age = 'age_class', id_col = 'uniqID', day_type_col = 'Day_Type', trip_purpose_col = 'Trip_pur')

trip_count.head()
uniqID age_class Day_Type Trip_pur count
72503 301276494 Child Weekday Home 2
240927 304228581 Seniors Weekend Home 2
362687 401351691 Adult Weekday D_shop 1
561851 407165011 Adult Weekday Home 3
76059 301334554 Teen Weekday Home 1
  • This table displayes the actual counts of travels collected by each uniq person (each recepient of survey). We will use the results from this 'data staging' to generate simulated travels for individual agent.

1.1.1. The number of trips occurring ‘k’ times for a single individual ‘i’ in a day

Generate a combined trip count distribution based on age groups, day type, and trip purpose.
# set up the number of people by age
age_pop = {'Seniors': 3, 'Adult': 3, 'MidAdult': 3, 'Child': 3, 'Teen': 3}

# the number of trips occurring ‘k’ times for a single individual ‘i’ in a day
eachTrip_simul_total_sample = abts.generate_combined_trip_count(
    naive_prob=naive_result,
    trip_count=trip_count,
    age_n_dict=age_pop,
    method='multinomial',
    Home_cbg=550790005021, #home CBG (Census Block Group)
    W_k_weekday = 0.7,
    W_k_weekend = 1.0,
    W_t_weekday={'Work': 1.2, 'S_d_r': 1, 'Rec_lei': 1.1},
    W_t_weekend={'Work': 0.7, 'S_d_r': 1, 'Rec_lei': 1.1},
    print_progress=True
)

# Parameters:
# - naive_prob: DataFrame containing Naive Bayes probabilities.
# - trip_count: DataFrame with trip count data.
# - age_n_dict: Dictionary mapping age groups to the number of individuals.
# - method: String specifying the method to distribute trips ('multinomial' or 'cdf').
# - Home_cbg: The home census block group id.
# - W_k_weekday: Weight multiplier for trip counts on weekdays.
# - W_k_weekend: Weight multiplier for trip counts on weekends.
# - W_t_weekday: Dictionary with trip purpose as keys and weights as values for weekdays.
# - W_t_weekend: Dictionary with trip purpose as keys and weights as values for weekends.
# - print_progress: Boolean flag to print progress.


eachTrip_simul_total_sample.head()
uniqID ageGroup Day_Type Week_Type TRPPURP count Home_cbg Wk_wD Wk_wK Wt_wD Wt_wK
487 6 MidAdult Friday Weekday S_d_r 0 550790005021 0.7 1 1 1
717 10 Child Saturday Weekend Others 0 550790005021 0.7 1 1 1
687 10 Child Friday Weekday Rec_lei 0 550790005021 0.7 1 1.1 1.1
878 13 Teen Friday Weekday Serv_trip 0 550790005021 0.7 1 1 1
639 11 Child Wednesday Weekday Home 2 550790005021 0.7 1 1 1
  • Results: simulated counts of each trip for each individual (agent)

1.2. Trip Chains Builder

The objective of this equation is to find, within the original NHTS data, a trip sequence O composed of trips most similar to those in the simulated trip list Si (e.g., Home, Home, Work, Rec_lei, etc.) for each individual agent in the current simulation. This process is divided into three steps (equations 3-5).
The first step is to find the optimal origin sequence Oi most similar to Si. Here, Si is a list of the types and counts of an individual's daily trips by day type, where the sequence is not yet specified. We search for the Si that is most similar to the set of all possible trip sequences from NHTS Q, and designate the trip sequence from Q with the highest similarity score as Oi. To determine similarity, the Kronecker delta function is used, scoring 1 for each matching element between the two lists and 0 for non-matching ones, with the similarity score being the sum of these values.
The second step is to randomly assign the trip sequence for Si. The first and last trips are designated as Home. Then, excluding the initial and final Home trips, the sequence of the remaining trips t in the simulated trip list S for each individual agent is shuffled randomly. As a result, the sequence of the trips, apart from the first and last Home, is determined randomly.
The third step is to reassign the sequence of trips in Si based on Oi. Here, l and j are indices traversing each trip purpose in O and S, respectively. First, it checks whether the trip following O(i,a,l) in sequence exists in S(i,a). If not, it finds the index j in S(i,a) that matches the trip purpose of O(i,a,l+1). Then, the corresponding trip purpose at S(i,a,j) is assigned to O(i,a,l)+1, rearranging S(i,a) accordingly.
Equation 3-5. Simulated sequence 'S' reassignment based on the most similar origin sequence (O) for individual (i) within age group (a)

Equation 3-5. Simulated sequence 'S' reassignment based on the most similar origin sequence 'O' for individual 'i' within age group 'a'

1.2.0. Data staging

Here, again, you run this once and use the result for simulating multiple travel schedules.
1.2.0.1. Create trip seqeunce of individuals using origin NHTS data
Creates a trip sequence column by concatenating Trip_pur values for each group defined by age_class, Day_Type, and uniqID.
trip_sequence_origin = abts.create_trip_sequence(repaired_NHTS, print_progress = True)

# Parameters:
# - df: DataFrame containing trip data (NHTS).
# - print_progress: Boolean indicating whether to print progress information.

trip_sequence_origin.head()
age_class Day_Type uniqID Trip_sequence count
125583 Seniors Weekday 303362241 Home-D_shop-Home 1
61063 MidAdult Weekday 302647372 Home-Work-Home 1
51450 MidAdult Weekday 300479321 Home-S_d_r-Home-Serv_trip-D_shop-Home-Serv_trip-Home-S_d_r-Home 1
195589 Teen Weekend 303065214 Home-S_d_r-Others-Home 1
152311 Seniors Weekday 404375522 Home-Serv_trip-D_shop-D_shop-Home 1
  • This table exhibits the trip sequence extracted from the origin NHTS data

1.2.1. Finding optimal origin sequence Oi most similar to Si

1.2.2. Randomly assign the trip sequence for Si

1.2.3. Reassign the sequence of trip in Si based on Oi

Constructs and refines trip sequences for simulated data to ensure realistic patterns by aligning them with sequences from original NHTS data. The process includes adjusting 'Home' trip counts to reflect real-world patterns, duplicating rows based on trip counts, finding and assigning the most similar trip sequence from NHTS data, and sequentially refining trip sequences to eliminate inconsistencies such as duplicated or consecutive 'Home' trips and ensuring all trip sequences start and end with 'Home'. The function applies several passes of adjustments to achieve coherent and logically ordered trip sequences that realistically simulate daily travel behavior.
simul_trip_sequence_sample = abts.makeTripSequence(eachTrip_simul_total_sample, trip_sequence_origin, print_progress = True)

# Parameters:
# - print_progress: Boolean flag indicating whether to print progress messages and bars during the execution.
    
display(simul_trip_sequence_sample[(simul_trip_sequence_sample['uniqID'] == 7) & (simul_trip_sequence_sample['Day_Type'] == 'Monday')]) # display the filtered table to show Monday trip for 1 agent with uniqID 7
uniqID ageGroup Home_cbg Day_Type Week_Type TRPPURP sequence seq_NHTS Wk_wD Wk_wK Wt_wD Wt_wK
206 7 MidAdult 550790005021 Monday Weekday Home 1 Home-Work-S_d_r-Work-S_d_r-Meals-Work-D_shop-Home 0.7 1 1 1
207 7 MidAdult 550790005021 Monday Weekday Work 2 Home-Work-S_d_r-Work-S_d_r-Meals-Work-D_shop-Home 0.7 1 1.2 0.7
208 7 MidAdult 550790005021 Monday Weekday S_d_r 3 Home-Work-S_d_r-Work-S_d_r-Meals-Work-D_shop-Home 0.7 1 1 1
209 7 MidAdult 550790005021 Monday Weekday Home 4 Home-Work-S_d_r-Work-S_d_r-Meals-Work-D_shop-Home 0.7 1 1 1
  • This agent goes to work -> S_d_r (School/Daycare/Religious activity) -> home. Let's assume he goes to daycare center to pick his child up.

1.3. Trip Timing Estimator

The trip timing estimator is responsible for calculating the dwell time for each trip and the initial start time from home. Dwell time is determined by sampling from NHTS data, taking into account variables such as age group a, day type d, trip purpose t, and the count of daily trips k. This procedure recognizes patterns, for example, teenagers typically have shorter dwell times for work due to part-time employment, and a tendency for dwell times to decrease as the number of daily trips for the same purpose increases. The data categorizes trip counts into groups: 1 trip per day (62,984 individuals), 2-3 trips per day (66,126 individuals), and 4+ trips per day (68,810 individuals).
In Equation 6, Dt(a,d,k',t) signifies the list formed by sampling dwell times from the NHTS data based on age group (a), day type (d), number of trips (k), and trip purpose (t). From this list, a dwell time DTit that matches the individual's specifics is uniformly selected and imported.
Equation 7 involves sampling from the NHTS data to create a probability distribution of trip start times based on age group a, day type d, and trip purpose t. From this distribution, the trip start time for each individual i is then determined.
Equation 6. Dwell time of all trips for each individual

Equation 6. Dwell time of all trips for each individual

Equation 7. Trip start time for each individual

Equation 7. Trip start time for each individual

1.3.0. Data staging

Here, again, you run this code once and use the result for simulating multiple travel schedules.
1.3.0.1. Extracting dwell time by trip purpose using NHTS
Extracts and organizes dwell time distributions by trip count, trip purpose, and other classifications from NHTS data.
dwellTime_dict = abts.dwellTime_listFromNHTS(repaired_NHTS)
1.3.0.2. Extracting trip start time by trip purpose using NHTS
Generates a dictionary mapping the start times of trips based on age class, day type, and the second trip purpose from the NHTS data. This function groups the data and extracts start times to understand typical trip start patterns. It's designed to be run once and reused for analysis or simulation purposes.
startTime_dict = abts.startTime_listFromNHTS(repaired_NHTS)

1.3.1. Estimating dwell time

1.3.2. Estimating trip start time

Estimates and assigns dwell times and start times for simulated trip data using distributions derived from NHTS data. It first classifies each trip within the simulated data by trip count and then assigns dwell times based on age class, day type, trip count class, and trip purpose. Finally, it assigns start times to the trips using similar criteria.
# assign dwell time and start time to simul data
simul_trip_start_time_sample = abts.assignDwellStartT(simul_trip_sequence_sample, dwellTime_dict, startTime_dict, print_progress = True)
display(simul_trip_start_time_sample[(simul_trip_start_time_sample['uniqID'] == 7) & (simul_trip_start_time_sample['Day_Type'] == 'Monday')])
uniqID ageGroup Home_cbg Day_Type TRPPURP sequence Wk_wD Wk_wK Wt_wD Wt_wK trip_count_class Dwell_Time sta_T_min Trip_mode
206 7 MidAdult 550790005021 Monday Home 1 0.7 1 1 1 2 42 0 Car
207 7 MidAdult 550790005021 Monday Work 2 0.7 1 1.2 0.7 2 510 510 Car
208 7 MidAdult 550790005021 Monday S_d_r 3 0.7 1 1 1 2 3 nan Car
209 7 MidAdult 550790005021 Monday Home 4 0.7 1 1 1 2 390 nan Car
  • Now the dwell time for each trip and trip start time (sta_T_min) are assigned.
  • Here, sta_T_min is minute based time, so 0 -> 12:00 AM, 360 -> 06:00 AM, 510: 8:30 AM

1.4. Trip Mode Assigner

In the process of assigning individual trip modes Mj(i), we employ sampling from NHTS data and apply heuristic methods to define rules (Equation 8). The approach is as follows: First, if an individual i does not have any trips planned outside, then no trip mode is assigned. However, for each individual i, the trip modes for their kth trips are initially determined by sampling from a probability distribution P(m│a,t). This distribution is specifically tailored to the individual's age and the purpose of each trip.
Next, certain conditions are implemented for individuals whose first trip mode is 'Car' (C in equation 8). The initial condition dictates that the last trip mode must also be 'Car', based on the logical assumption that if one leaves in a personal vehicle, they are expected to return in it. This excludes outliers such as taxi use, focusing solely on personal vehicle use. The subsequent condition applies to non-round trips, indicating that a change in trip mode suggests the inability to return to a car left at a previous location. This requires all subsequent trip modes to be 'Car', or the individual to maintain their previous trip mode M(j-1)(i).
Equation 8. Assign Trip mode applying round trip

Equation 8. Assign Trip mode applying round trip

simul_trip_mode_sample = abts.put_tripMode(simul_trip_start_time_sample, trip_mode, print_progress = True)

display(simul_trip_mode_sample[(simul_trip_mode_sample['uniqID'] == 7) & (simul_trip_mode_sample['Day_Type'] == 'Monday')])
uniqID ageGroup Home_cbg Day_Type TRPPURP sequence Wk_wD Wk_wK Wt_wD Wt_wK trip_count_class Dwell_Time sta_T_min Trip_mode Origin Dest Ws Wd TRPPURP_det
206 7 MidAdult 550790005021 Monday Home 1 0.7 1 1 1 2 42 0 nan nan 550790005021 nan nan Home
207 7 MidAdult 550790005021 Monday Work 2 0.7 1 1.2 0.7 2 510 510 Car nan 550790026001 1.5 0.5 Work
208 7 MidAdult 550790005021 Monday S_d_r 3 0.7 1 1 1 2 3 nan Car nan 550790001022 1.5 0.25 Religion
209 7 MidAdult 550790005021 Monday Home 4 0.7 1 1 1 2 390 nan Car nan 550790005021 nan nan Home
  • Now, trip modes for each trip are assigned.

1.5. Spatial Trip Route Estimator

In the Spatial Trip Route Estimator, destinations for each trip purpose at the CBG level are identified through the interplay of distance, land use, and agent preferences. Following this, an optimal final schedule for each individual is crafted, adhering to logical and spatiotemporal constraints.
Equation 9 is deployed to probabilistically ascertain the destination D for each trip purpose. The initial probabilities are ascertained from observed monthly records of destination choices within the origin area A, based on SafeGraph data. Given that this research back-calculates past travels, observed travel data serves to establish baseline destination choices. Nevertheless, in scenarios lacking prior travel data for forecasts, this probability may be adjusted to 1, thus becoming solely influenced by alternative parameters.
Because of the aggregation of SafeGraph data by counts without trip purpose delineation, two equations are combined to compute the probability for each destination specific to the trip purpose. The inaugural equation calculates the probability from the count of locations pertinent to trip purpose t, symbolized as C(t), sourced from land use data. This probability escalates with an increase in locations tied to the trip purpose within the land use data, assigned a weight Ws, termed as 'spatial attractiveness weight'. The subsequent equation is grounded in a distance sensitivity formula deriving from the distance decay function. An elevated weight Wd signifies that the distance d(A,D) between origin and destination markedly sways the destination selection, while a diminished weight indicates lesser influence of distance.
Equation 9. Probability of trips 't' from origin area 'A' to destination 'D'

Equation 9. Probability of trips 't' from origin area 'A' to destination 'D'

1.5.0. Data staging

Here, again, you run this code once and use the result for simulating multiple travel schedules.
1.5.0.1. Ratio between straight path and network path
Calculates the average ratio between network (road) distance and straight-line distance for randomly selected pairs of points within CBGs.
Trip distances and durations are estimated using OpenStreetMap (OSM) data via the OSMnx Python library. Given the high computational demands of calculating network distances for all trips, a sampling approach was adopted. Random points across CBGs were selected to calculate the ratio of network road distance to straight-line distance, resulting in a factor of 1.223 (example). Thus, a straight-line distance of 10 km translates to an estimated trip distance of 12.23 km.
ratio_table, average_distance_ratio = abts.calculate_sampled_network_distance(cbg_gdf = cbg, network_road = network_road, num_samples = 200)

print('average_distance_ratio: ', average_distance_ratio)

# average_distance_ratio:  1.161508600010609

1.5.1. Estimate probabilistic destinations for trip purpose t

Estimates spatial trip routes by assigning origins and destinations for simulated trips. This process uses probabilities derived from real-world data to make the simulated trips more realistic in terms of spatial distribution.
option 1. Randomly pick W_s and W_d from prob_2020_09_combined.
simul_trip_routes_sample, assigned_trip_table = abts.assignDest(prob_2020_09_combined, simul_trip_mode_sample, print_progress = True) # randomly select 'W_d' and 'W_s' in prob_2020_09_combined dataframe.
option 2. Manually put W_s and W_d for each trip purpose
# set W_d and W_s manually
W_d_W_s = {'Rec_lei': {'Ws': 1.5, 'Wd': 0.5}, 
           'Meals': {'Ws': 0.5, 'Wd': 0.75}, 
           'Work': {'Ws': 1.0, 'Wd': 0.5}, 
           'Serv_trip': {'Ws': 0.5, 'Wd': 0.75}, 
           'D_shop': {'Ws': 0.5, 'Wd': 0.25}, 
           'S_d_r': {'Ws': 1.0, 'Wd': 0.5}, 
           'Others': {'Ws': 0.5, 'Wd': 0.25}, 
           'V_fr_rel': {'Ws': 1.0, 'Wd': 0.5}}

simul_trip_routes_sample, assigned_trip_table = abts.assignDest(prob_2020_09_combined, simul_trip_mode_sample,
                                                                 W_d_W_s = W_d_W_s,
                                                                 print_progress = True)

display(simul_trip_routes_sample[(simul_trip_routes_sample['uniqID'] == 7) & (simul_trip_routes_sample['Day_Type'] == 'Monday')])
uniqID ageGroup Home_cbg Day_Type TRPPURP sequence Wk_wD Wk_wK Wt_wD Wt_wK trip_count_class Dwell_Time sta_T_min Trip_mode Origin Dest Ws Wd TRPPURP_det TripDist TripTime
206 7 MidAdult 550790005021 Monday Home 1 0.7 1 1 1 2 42 0 nan 550790005021 550790005021 nan nan Home 0.552907 nan
207 7 MidAdult 550790005021 Monday Work 2 0.7 1 1.2 0.7 2 510 510 Car 550790005021 550790026001 1.5 0.5 Work 7.75753 12
208 7 MidAdult 550790005021 Monday S_d_r 3 0.7 1 1 1 2 3 nan Car 550790026001 550790001022 1.5 0.25 Religion 9.48366 14
209 7 MidAdult 550790005021 Monday Home 4 0.7 1 1 1 2 390 nan Car 550790001022 550790005021 nan nan Home 5.54624 8
  • Now destinations (CBGs) for each trip purpose are assigned.

1.5.2. Compute trip distance and duration

Estimates the distances and travel times for each trip in the simulated dataset. This function uses the geographic information from census block groups (CBG) and a predefined distance ratio to calculate network distances. It then applies average speeds based on the mode of transport and the age group of the traveler to estimate travel times.
simul_trip_Dist_Time_sample = abts.estimate_tripDist_Time(simul_trip_routes_sample, cbg, distance_ratio = average_distance_ratio, print_progress = True)

# Parameters:
# - cbg_gdf: GeoDataFrame containing CBG polygons and their FIPS codes for locating origins and destinations.

display(simul_trip_Dist_Time_sample[(simul_trip_Dist_Time_sample['uniqID'] == 7) & (simul_trip_Dist_Time_sample['Day_Type'] == 'Monday')])
uniqID ageGroup Home_cbg Day_Type sequence Wk_wD Wk_wK Wt_wD Wt_wK TRPPURP TRPPURP_det Ws Wd Origin Dest TripDist Trip_mode TripSTARTT TripTime TripENDT start_min Dwell_Time end_min
206 7 MidAdult 550790005021 Monday 1 0.7 1 1 1 Home Home nan nan 550790005021 550790005021 0.552907 nan nan nan nan 0 510 510
207 7 MidAdult 550790005021 Monday 2 0.7 1 1.2 0.7 Work Work 1.5 0.5 550790005021 550790026001 7.75753 Car 510 12 522 522 510 1032
208 7 MidAdult 550790005021 Monday 3 0.7 1 1 1 S_d_r Religion 1.5 0.25 550790026001 550790001022 9.48366 Car 1032 14 1046 1046 3 1049
209 7 MidAdult 550790005021 Monday 4 0.7 1 1 1 Home Home nan nan 550790001022 550790005021 5.54624 Car 1049 8 1057 1057 390 1447
  • Now, trip distances (km) for each trip are assigned.

1.5.3. Optimize Trips with Logical and space-time constraints

  • Applies constraints to optimize individual trips by adjusting start and end times and updating trip modes based on certain logical conditions, such as changing modes from walking to driving or public transport for longer trips.
simul_trip_result_sample = abts.optimizeTrips_constraint(simul_trip_Dist_Time_sample, dwellTime_dict, print_progress = True)

display(simul_trip_result_sample[(simul_trip_result_sample['uniqID'] == 7) & (simul_trip_result_sample['Day_Type'] == 'Monday')])
uniqID ageGroup Home_cbg Day_Type sequence Wk_wD Wk_wK Wt_wD Wt_wK TRPPURP TRPPURP_det Ws Wd Origin Dest TripDist Trip_mode TripSTARTT TripTime TripENDT start_min Dwell_Time end_min
205 7 MidAdult 550790005021 Monday 1 0.7 1 1 1 Home Home nan nan 550790005021 550790005021 0.552907 nan nan nan nan 0 435 435
206 7 MidAdult 550790005021 Monday 2 0.7 1 1.2 0.7 Work Work 1.5 0.5 550790005021 550790026001 7.75753 Car 435 12 447 447 156 603
207 7 MidAdult 550790005021 Monday 3 0.7 1 1 1 S_d_r Religion 1.5 0.25 550790026001 550790001022 9.48366 Car 603 14 617 617 5 622
208 7 MidAdult 550790005021 Monday 4 0.7 1 1 1 Home Home nan nan 550790001022 550790005021 5.54624 Car 622 8 630 630 810 1440
  • Now, the results pertinent to the time, such as TripSTARTT, TripENDT, start_min, Dwell_Time, end_min are all are re-assigned and become more reasonable.

2. Mapping the results (example)

Figure 2 depicts a choropleth map based on the number of trips originating from approximately 50 CBGs (Study area) during the month of September. It shows all trip destinations (CBG) mapped according to the volume of trips.
Following this, Figure 3 maps the workplaces of only the Adult and Mid-adult age groups for the months of September to November 2020. This was achieved by filtering for Work trips and mapping the count values at each destination (CBG). In this manner, we can decompose and examine the travel within our area of interest by trip purpose and age group through the ABTS.
Figure 2. Spatial patterns of travel counts in destination CBG in Milwaukee (2020/09~11)

Figure 2. Spatial patterns of travel counts in destination CBG in Milwaukee (2020/09)

Figure 3. Spatial patterns of travel counts for work trips in destination CBG by Adult and Mid-adult agegroup (2020/09~11)

Figure 3. Spatial patterns of travel counts for work trips in destination CBG by Adult and Mid-adult agegroup (2020/09~11)

Related Document:

  • Choi, M., & Hohl. A. (2024). Derivation of Spatiotemporal Risk Areas and Travel Behaviors During Pandemic Through Reverse Estimation of Mobility Patterns by Agent-Based Modeling, Spatial and Spatio-temporal Epidemiology. Under review (1st round)

  • Choi, M., Seo, J., & Hohl. A. (2024). ABTS: Agent-Based Travel Scheduler. In process

Author

About

Agent Based Travel Scheduler library to predict individual travel schedules

Resources

Readme
Activity

Stars

1 star

Watchers

1 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

No packages published

Languages
0