Flexible bus route planning method based on user travel behavior prediction

Abstract

The invention provides a flexible bus route planning method based on user travel behavior prediction. The invention discloses a flexible bus route planning method based on user travel behavior prediction. The method comprises the following steps: S1, collecting and integrating multi-source travel data; s2, user behavior analysis and demand prediction; s3, dynamic service area division; s4, generating and optimizing a flexible line; s5, performing real-time dynamic scheduling and adjustment; s6, ticket price dynamic generation and feedback; and S7, service effect evaluation and iterative optimization. The flexible bus route planning method based on user travel behavior prediction provided by the invention has the advantages of being capable of obtaining balance between global optimum and calculation efficiency, reducing the vehicle empty driving rate, improving the route operation cost recovery rate, quickly responding to newly added demands and shortening the average waiting time of passengers.

Description

Technical Field

[0001] The present invention relates to the technical field of bus route planning, and in particular to a flexible bus route planning method based on user travel behavior prediction. Background Art

[0002] With the acceleration of urbanization and the growth of population density, public transportation systems are facing increasingly severe operational pressures. Traditional public transportation systems mostly adopt a fixed route and fixed frequency operation mode. Their core flaw is the lack of responsiveness to dynamic travel demand. Specifically,

[0003] Temporal and spatial mismatch between demand and supply: Fixed routes rely on historical passenger flow statistics and are difficult to adapt to real-time demand fluctuations (such as unexpected events and weather changes). This results in congestion during peak hours and high idle rates (generally exceeding 40%) during off-peak hours, resulting in significant resource waste.

[0004] Limitations of static scheduling algorithms: Existing scheduling systems are mostly based on rule engines or simple heuristic algorithms (such as shortest path first). They cannot effectively handle multi-objective optimization problems (such as balancing waiting time, cost, and passenger satisfaction), and are particularly prone to falling into local optimality in complex road networks.

[0005] Inefficient data utilization: Although mobile internet technologies provide massive amounts of travel data (such as GPS and transportation card records), traditional methods lack the ability to integrate multi-source heterogeneous data and fail to fully explore user behavior patterns, resulting in low demand forecast accuracy (root mean square error typically exceeds 15 people / hour).

[0006] Rigid design of fares and services: Current fare strategies mostly adopt a single pricing model (such as charging by mileage or zone), which lacks dynamic response to passengers' willingness to pay, time sensitivity and demand elasticity, making it difficult to balance public welfare and market-oriented operation needs.

[0007] In recent years, some studies have attempted to introduce demand-responsive public transportation to improve flexibility, but existing technologies still have significant bottlenecks.

[0008] Therefore, it is necessary to provide a flexible bus route planning method based on user travel behavior prediction to solve the above technical problems. Summary of the Invention

[0009] The technical problem solved by the present invention is to provide a flexible bus route planning method based on user travel behavior prediction, which can achieve a balance between global optimization and computational efficiency, reduce vehicle idle rate, improve route operation cost recovery rate, quickly respond to new demand, and shorten the average waiting time of passengers.

[0010] To solve the above technical problems, the present invention provides a flexible bus route planning method based on user travel behavior prediction, comprising the following steps: S1: multi-source travel data collection and integration: obtaining user travel demand data through mobile terminal applications, public transportation card data, and reservation platforms, wherein the travel demand data includes scheduled travel demand, real-time travel demand, historical travel records, user preference information, and geographic positioning data;

[0011] S2: User Behavior Analysis and Demand Forecasting: Based on the travel demand data collected in S1, the machine learning model analyzes user travel behavior characteristics and predicts the spatiotemporal distribution of travel demand within the future target period. The spatiotemporal distribution includes the departure point, destination, travel time window, and number of passengers.

[0012] S3: Dynamic service area division: Based on the spatiotemporal distribution of travel demand predicted in S2, combined with the urban road network topology and traffic status data, the customized bus service area is divided and the potential station set and station coverage within the service area are determined;

[0013] S4: Flexible Route Generation and Optimization: Based on the potential station set and real-time demand distribution, a dynamic programming algorithm is used to generate an initial set of candidate routes. The optimal route solution is then selected using a multi-objective optimization model, taking into account vehicle capacity constraints, maximum passenger waiting time thresholds, and route operating costs. The objective functions of the multi-objective optimization model include minimizing average passenger waiting time, minimizing vehicle idle rate, and minimizing route operating costs.

[0014] S5: Real-time dynamic scheduling and adjustment: During vehicle operation, new travel demand and road condition changes are monitored in real time, and local adjustments to planned routes are made through dynamic matching algorithms, including the insertion of new stations, route detour optimization, and vehicle scheduling priority allocation.

[0015] S6: Dynamic fare generation and feedback: Develop differentiated fare strategies based on route operating costs, passenger willingness to pay, and public subsidy policies, and push fare information and route adjustment notifications in real time through user terminals;

[0016] S7: Service effectiveness evaluation and iterative optimization: Collect actual operation data and user satisfaction feedback, use simulation tools to verify the route planning effect, and iteratively update the machine learning model in S2 and the optimization parameters in S4 through a data closed loop.

[0017] Preferably, the machine learning model in S2 is an integrated learning model, including a fusion of random forest, gradient boosting tree and time series prediction model, for capturing the periodicity and randomness characteristics of user travel; the specific method of user behavior analysis and demand forecasting is:

[0018] Based on historical travel records and real-time demand data, a time series decomposition model is used to predict the number of passengers from departure point i to destination j in the future time period t. Its expression is:

[0019]

[0020] in, is the historical average demand for the same period. is the real-time traffic status correction factor, is the weather influence coefficient, α, β, γ are weight parameters, is a random disturbance term.

[0021] Preferably, the dynamic programming algorithm in S4 adopts a hybrid optimization method combining a greedy strategy and a genetic algorithm; the objective function of the multi-objective optimization model is defined as:

[0022]

[0023] in, is the waiting time of the kth passenger, C empty is the number of empty vehicles, C total is the total number of departures, L route is the total operating cost of the line, λ1, λ2, and λ3 are normalized weight coefficients.

[0024] Preferably, the dynamic matching algorithm in S5 dynamically adjusts the vehicle route based on the urgency of passenger demand and the remaining capacity of the vehicle, and uses a sliding time window mechanism to limit the frequency of real-time adjustment; the dynamic matching algorithm calculates the priority score Q of the vehicle route adjustment by the following formula:

[0025]

[0026] Among them, U urgency is the urgency of demand, U urgency Negatively correlated with the remaining response time, R detour is the ratio of the detour distance, and ω1 and ω1 are the dynamic adjustment weights.

[0027] Preferably, the differentiated fare strategy in S6 includes segmented pricing based on travel distance, dynamic floating pricing based on time sensitivity, and a discount mechanism for high-frequency users; the fare P of the dynamic floating pricing strategy is calculated as follows:

[0028]

[0029] Among them, P base is the base fare, D current is the demand density of the current period, D avgis the average demand density during the same period in history, and η is the price elasticity coefficient.

[0030] Preferably, the dynamic service area division in S3 adopts a Voronoi diagram algorithm to generate a coverage polygon with the potential site as the center, and dynamically adjust the polygon boundary according to the demand density.

[0031] Preferably, the vehicle capacity constraint in S4 includes a hard constraint and a flexible constraint: the hard constraint is that the single passenger volume does not exceed the rated capacity of the vehicle; the flexible constraint is that the overload ratio is allowed to be no more than 10%, and additional vehicle dispatch is triggered.

[0032] Preferably, the service effect evaluation in S7 adopts Monte Carlo simulation to simulate the line load rate, passenger waiting time and operating cost under different demand scenarios, and generate a Pareto front to guide parameter optimization.

[0033] Preferably, the ensemble learning model fuses the prediction results of each sub-model through a weighted voting mechanism, and the weights are dynamically adjusted according to the prediction errors of the models in historical data.

[0034] Preferably, the sliding time window mechanism is set to trigger a local path optimization every 5 minutes, and the number of sites changed involved in a single adjustment does not exceed 3.

[0035] Compared with related technologies, the flexible bus route planning method based on user travel behavior prediction provided by the present invention has the following beneficial effects:

[0036] The present invention provides a flexible bus route planning method based on user travel behavior prediction. Through multi-source data collection and integrated learning models, it can efficiently integrate historical travel records, real-time demand and user preferences, and accurately predict the spatiotemporal distribution of travel demand in future periods. The machine learning model's ability to capture the periodic and random characteristics of user behavior significantly reduces prediction deviations, providing a reliable basis for subsequent route planning. Dynamic service area division can quickly identify high-demand hot spots, covering more than 90% of predicted demand points, and reducing service blind spots. By combining genetic algorithms, greedy strategies with passenger waiting time, empty driving rate, and operating cost minimization, it can achieve a balance between global optimization and computational efficiency. By dynamically generating a set of candidate routes and screening the optimal The solution can reduce the vehicle idle rate and improve the line operation cost recovery rate; through vehicle capacity constraints, it can further ensure flexible resource allocation, which not only guarantees operational safety but also reduces vehicle idleness due to demand fluctuations; through the dynamic matching algorithm, the route can be adjusted in real time based on the urgency of passenger demand and the remaining capacity of the vehicle; combined with the sliding time window mechanism, it can quickly respond to new demand and shorten the average waiting time of passengers while ensuring line stability; through differentiated fare strategies, it comprehensively considers operating costs and willingness to pay, which not only increases corporate profits but also improves user stickiness through the high-frequency user discount mechanism; through the Monte Carlo simulation tool, 20 scenarios such as extreme weather and sudden congestion can be simulated to verify the stability and disaster recovery capabilities of route planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of the flexible bus route planning method based on user travel behavior prediction provided by the present invention. DETAILED DESCRIPTION

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0039] Please refer to Figure 1 ,in, Figure 1 This is a flow chart of the flexible bus route planning method based on user travel behavior prediction provided by the present invention. The flexible bus route planning method based on user travel behavior prediction includes the following steps: S1: Multi-source travel data collection and integration: Acquiring user travel demand data through mobile terminal applications, public transportation card data, and reservation platforms. The travel demand data includes scheduled travel demand, real-time travel demand, historical travel records, user preference information, and geolocation data. The user preference information is extracted by analyzing the user's historical travel frequency, frequently visited locations, travel time preferences, and transportation selection tendencies. The geolocation data is calibrated through the integration of GPS, base station positioning, and Wi-Fi fingerprint technology, with a positioning accuracy error of no more than 50 meters.

[0040] S2: User Behavior Analysis and Demand Forecasting: Based on the travel demand data collected in S1, a machine learning model is used to analyze user travel behavior characteristics and predict the spatiotemporal distribution of travel demand within a future target period. The spatiotemporal distribution includes the departure point, destination, travel time window, and number of passengers. The training dataset of the machine learning model covers weekdays, holidays, extreme weather, and emergency scenarios, and the model input features are dynamically updated using a sliding window method.

[0041] The machine learning model is an integrated learning model, including a fusion of random forest, gradient boosting tree and time series prediction model, which is used to capture the periodic and random characteristics of user travel. The specific method of user behavior analysis and demand forecasting is as follows:

[0042] Based on historical travel records and real-time demand data, a time series decomposition model is used to predict the number of passengers from departure point i to destination j in the future time period t. Its expression is:

[0043]

[0044] in, is the historical average demand for the same period. is the real-time traffic status correction factor, is the weather influence coefficient, α, β, γ are weight parameters, is a random disturbance term; the weight parameter is dynamically adjusted during the training phase through the Bayesian optimization algorithm, and the weather influence coefficient Quantitative calculation based on the precipitation probability, temperature and wind level released by the Meteorological Bureau;

[0045] The ensemble learning model fuses the prediction results of each sub-model through a weighted voting mechanism, and the weights are dynamically adjusted according to the prediction error of the model in historical data; the prediction error is measured using the root mean square error (RMSE), and the weight update formula is:

[0046]

[0047] Among them, δ = 0.1 is used to prevent the denominator from being zero;

[0048] S3: Dynamic service area division: Based on the spatiotemporal distribution of travel demand predicted in S2, combined with the city's road network topology and traffic status data, a customized bus service area is divided and the potential station set and station coverage within the service area are determined. The traffic status data includes real-time congestion index, road construction information, and traffic accident warnings, which are synchronized in real time with the city transportation management center through the API interface;

[0049] The dynamic service area division uses a Voronoi diagram algorithm to generate coverage polygons centered on potential sites, and dynamically adjust the polygon boundaries based on demand density. The Voronoi diagram generation process introduces constraints to ensure that the overlap rate of adjacent site coverage areas does not exceed 5%, and the maximum number of edges per polygon is 6 to simplify computational complexity.

[0050] S4: Flexible Route Generation and Optimization: Based on the potential station set and real-time demand distribution, a dynamic programming algorithm is used to generate an initial set of candidate routes. Combined with vehicle capacity constraints, maximum passenger waiting time thresholds, and route operating costs, the optimal route solution is selected through a multi-objective optimization model. The objective functions of the multi-objective optimization model include minimizing average passenger waiting time, minimizing vehicle idle rate, and minimizing route operating costs. The maximum passenger waiting time threshold is dynamically set according to the service level agreement (SLA), with the threshold shortened to 10 minutes during peak hours and extended to 20 minutes during non-peak hours.

[0051] The dynamic programming algorithm adopts a hybrid optimization method combining greedy strategy and genetic algorithm; the objective function of the multi-objective optimization model is defined as:

[0052]

[0053] in, is the waiting time of the kth passenger, C empty is the number of empty vehicles, C total is the total number of departures, L route is the total operating cost of the line, λ1, λ2, and λ3 are normalized weight coefficients; the crossover probability of the genetic algorithm is set to 0.8, the mutation probability is 0.02, the population size is 200, the upper limit of the number of iterations is 500, and the greedy strategy is used to quickly screen the local optimal solution to accelerate convergence;

[0054] The vehicle capacity constraints include hard constraints and flexible constraints: the hard constraint is that the number of passengers carried per trip does not exceed the rated capacity of the vehicle; the flexible constraint is that the overload ratio is allowed to be no more than 10%, and it triggers the dispatch of additional vehicles; when the flexible constraint is triggered, the system automatically calls on idle vehicles in the spare vehicle pool, and gives priority to electric buses to reduce carbon emissions;

[0055] S5: Real-time dynamic scheduling and adjustment: During vehicle operation, new travel demand and road condition changes are monitored in real time, and planned routes are adjusted locally using a dynamic matching algorithm. This includes inserting new stops, optimizing route detours, and assigning vehicle scheduling priorities. This road condition change information is integrated through the vehicle's onboard optical boresight (OBD) device, roadside unit (RSU), and third-party navigation platforms, and is updated once a minute.

[0056] The dynamic matching algorithm in S5 dynamically adjusts the vehicle route based on the urgency of passenger demand and the remaining capacity of the vehicle, and uses a sliding time window mechanism to limit the frequency of real-time adjustments. The dynamic matching algorithm calculates the priority score Q of the vehicle route adjustment using the following formula:

[0057]

[0058] Among them, U urgency is the urgency of demand, U urgency Negatively correlated with the remaining response time, R detour is the detour distance ratio, ω1 and ω1 are dynamic adjustment weights; the demand urgency U urgency The calculation formula is:

[0059]

[0060] where t remaining The latest acceptable response time for passenger demands;

[0061] The sliding time window mechanism is set to trigger local route optimization every 5 minutes, and the number of site changes involved in a single adjustment does not exceed 3. If the new demand leads to more than 3 site changes, the full route replanning process will be initiated, and a "major route adjustment" prompt message will be pushed to the user terminal.

[0062] S6: Dynamic fare generation and feedback: Develop differentiated fare strategies based on route operating costs, passenger willingness to pay, and public subsidy policies, and push fare information and route adjustment notifications in real time through user terminals. WTP is assessed through a comprehensive assessment of user historical consumption data, income level tags, and real-time demand urgency. Subsidy policies are adjusted in conjunction with local government public transportation budgets.

[0063] The differentiated fare strategy includes segmented pricing based on travel distance, dynamic floating pricing based on time sensitivity, and a discount mechanism for high-frequency users. The fare P of the dynamic floating pricing strategy is calculated as follows:

[0064]

[0065] Among them, P base is the base fare, D current is the demand density of the current period, D avg is the average demand density over the same period in history, and η is the price elasticity coefficient. Frequent users are defined as passengers who take the bus more than 15 times a month. They receive a discount of 10%-20% of the base fare, and consumption data is recorded using blockchain technology to prevent tampering.

[0066] S7: Service Effectiveness Evaluation and Iterative Optimization: Collect actual operational data and user satisfaction feedback, use simulation tools to verify route planning results, and iteratively update the machine learning model in S2 and the optimization parameters in S4 through a closed-loop data loop. User satisfaction feedback includes trip punctuality, ride comfort, and fare reasonableness scores, collected in real time through the app's embedded questionnaire and voice interaction system.

[0067] The service effectiveness evaluation uses Monte Carlo simulation to simulate the line load rate, passenger waiting time and operating costs under different demand scenarios, and generate a Pareto front to guide parameter optimization; the random variables of the Monte Carlo simulation include the demand fluctuation range (±30%), the probability of sudden road congestion (5%-20%) and the vehicle failure rate (1%-5%), and the number of simulations is no less than 10,000 to ensure statistical significance.

[0068] Compared with related technologies, the flexible bus route planning method based on user travel behavior prediction provided by the present invention has the following beneficial effects:

[0069] The present invention provides a flexible bus route planning method based on user travel behavior prediction. Through multi-source data collection and integrated learning models, it can efficiently integrate historical travel records, real-time demand and user preferences, and accurately predict the spatiotemporal distribution of travel demand in future periods. The machine learning model's ability to capture the periodic and random characteristics of user behavior significantly reduces prediction deviations, providing a reliable basis for subsequent route planning. Dynamic service area division can quickly identify high-demand hot spots, covering more than 90% of predicted demand points, and reducing service blind spots. By combining genetic algorithms, greedy strategies with passenger waiting time, empty driving rate, and operating cost minimization, it can achieve a balance between global optimization and computational efficiency. By dynamically generating a set of candidate routes and screening the optimal The solution can reduce the vehicle idle rate and improve the line operation cost recovery rate; through vehicle capacity constraints, it can further ensure flexible resource allocation, which not only guarantees operational safety but also reduces vehicle idleness due to demand fluctuations; through the dynamic matching algorithm, the route can be adjusted in real time based on the urgency of passenger demand and the remaining capacity of the vehicle; combined with the sliding time window mechanism, it can quickly respond to new demand and shorten the average waiting time of passengers while ensuring line stability; through differentiated fare strategies, it comprehensively considers operating costs and willingness to pay, which not only increases corporate profits but also improves user stickiness through the high-frequency user discount mechanism; through the Monte Carlo simulation tool, 20 scenarios such as extreme weather and sudden congestion can be simulated to verify the stability and disaster recovery capabilities of route planning.

[0070] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A flexible bus route planning method based on user travel behavior prediction, characterized in that: include: S1: Multi-source travel data collection and integration: Obtain user travel demand data through mobile terminal applications, public transportation card data, and reservation platforms. The travel demand data includes scheduled travel demand, real-time travel demand, historical travel records, user preference information, and geographic location data; S2: User Behavior Analysis and Demand Forecasting: Based on the travel demand data collected in S1, the machine learning model analyzes user travel behavior characteristics and predicts the spatiotemporal distribution of travel demand within the future target period. The spatiotemporal distribution includes the departure point, destination, travel time window, and number of passengers. S3: Dynamic service area division: Based on the spatiotemporal distribution of travel demand predicted in S2, combined with the urban road network topology and traffic status data, the customized bus service area is divided and the potential station set and station coverage within the service area are determined; S4: Flexible Route Generation and Optimization: Based on the potential station set and real-time demand distribution, a dynamic programming algorithm is used to generate an initial set of candidate routes. The optimal route solution is then selected using a multi-objective optimization model, taking into account vehicle capacity constraints, maximum passenger waiting time thresholds, and route operating costs. The objective functions of the multi-objective optimization model include minimizing average passenger waiting time, minimizing vehicle idle rate, and minimizing route operating costs. S5: Real-time dynamic scheduling and adjustment: During vehicle operation, new travel demand and road condition changes are monitored in real time, and local adjustments to planned routes are made through dynamic matching algorithms, including the insertion of new stations, route detour optimization, and vehicle scheduling priority allocation. S6: Dynamic fare generation and feedback: Develop differentiated fare strategies based on route operating costs, passenger willingness to pay, and public subsidy policies, and push fare information and route adjustment notifications in real time through user terminals; S7: Service effectiveness evaluation and iterative optimization: Collect actual operation data and user satisfaction feedback, use simulation tools to verify the route planning effect, and iteratively update the machine learning model in S2 and the optimization parameters in S4 through a data closed loop.

2. The flexible bus route planning method based on user travel behavior prediction according to claim 1 is characterized in that: The machine learning model described in S2 is an integrated learning model, including a fusion of random forest, gradient boosting tree and time series prediction model, which is used to capture the periodic and random characteristics of user travel; the specific method of user behavior analysis and demand forecasting is: Based on historical travel records and real-time demand data, a time series decomposition model is used to predict the number of passengers from departure point i to destination j in the future time period t. Its expression is: in, is the historical average demand for the same period. is the real-time traffic status correction factor, is the weather influence coefficient, α, β, γ are weight parameters, is a random disturbance term.

3. The flexible bus route planning method based on user travel behavior prediction according to claim 1 is characterized in that: The dynamic programming algorithm in S4 adopts a hybrid optimization method combining a greedy strategy and a genetic algorithm; the objective function of the multi-objective optimization model is defined as: in, is the waiting time of the kth passenger, C empty is the number of empty vehicles, C total is the total number of departures, L route is the total operating cost of the line, λ1, λ2, and λ3 are normalized weight coefficients.

4. The flexible bus route planning method based on user travel behavior prediction according to claim 1 is characterized in that: The dynamic matching algorithm in S5 dynamically adjusts the vehicle route based on the urgency of passenger demand and the remaining capacity of the vehicle, and uses a sliding time window mechanism to limit the frequency of real-time adjustments. The dynamic matching algorithm calculates the priority score Q of the vehicle route adjustment using the following formula: Among them, U urgency is the urgency of demand, U urgency Negatively correlated with the remaining response time, R detour is the ratio of the detour distance, and ω1 and ω1 are the dynamic adjustment weights.

5. The flexible bus route planning method based on user travel behavior prediction according to claim 1 is characterized in that: The differentiated fare strategy described in S6 includes segmented pricing based on travel distance, dynamic floating pricing based on time sensitivity, and a discount mechanism for high-frequency users. The fare P of the dynamic floating pricing strategy is calculated as follows: Among them, P base is the base fare, D current is the demand density of the current period, D avg is the average demand density during the same period in history, and η is the price elasticity coefficient.

6. The flexible bus route planning method based on user travel behavior prediction according to claim 1 is characterized in that: The dynamic service area division in S3 adopts the Voronoi diagram algorithm to generate coverage polygons with potential sites as the center, and dynamically adjust the polygon boundaries according to the demand density.

7. The flexible bus route planning method based on user travel behavior prediction according to claim 1 is characterized in that: The vehicle capacity constraint in S4 includes a hard constraint and a flexible constraint: the hard constraint is that the single passenger volume does not exceed the rated capacity of the vehicle; the flexible constraint is that the overload ratio is allowed to be no more than 10%, and additional vehicle dispatch is triggered.

8. The flexible bus route planning method based on user travel behavior prediction according to claim 1 is characterized in that: The service effect evaluation described in S7 adopts Monte Carlo simulation to simulate the line load rate, passenger waiting time and operating cost under different demand scenarios, and generate a Pareto front to guide parameter optimization.

9. The flexible bus route planning method based on user travel behavior prediction according to claim 2 is characterized in that: The ensemble learning model fuses the prediction results of each sub-model through a weighted voting mechanism, and the weights are dynamically adjusted according to the prediction errors of the models in historical data.

10. The flexible bus route planning method based on user travel behavior prediction according to claim 4 is characterized in that: The sliding time window mechanism is set to trigger local path optimization every 5 minutes, and the number of sites changed in a single adjustment does not exceed 3.

Images