An individual MAAS travel path decision generation method based on multi-source data

Abstract

The application provides a kind of individual MAAS travel path decision generation method based on multi-source data, comprising the following steps: collecting traffic state data, user behavior data, environmental factor data, user preference data, real-time event data;Build a multi-objective optimization model including objective function and constraint condition;Introduce dynamic crossover rate and dynamic mutation rate to improve the NSGA-III framework, use the improved NSGA-III framework to solve the multi-objective optimization model, and get the non-dominated solution set;Based on the non-dominated solution set, combined with Top-K preference screening to generate personalized candidate paths, and the selected path from the personalized candidate paths is real-time adapted to traffic changes through the event-driven dynamic adjustment mechanism.The application solves the problems of insufficient data utilization, insufficient personalization and low multi-mode coordination efficiency in the prior art.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and travel service technology, specifically to a method for generating individual MAAS travel route decisions based on multi-source data. Background Technology

[0002] With the increasing complexity of urban transportation systems and the diversification of user travel needs, traditional single-mode route planning methods (such as driving and public transportation) are no longer sufficient to meet the personalized needs of MAAS (Mobility as a Service) scenarios. MAAS emphasizes seamless connection and collaborative services among multiple modes of transportation (such as subways, buses, shared bicycles, ride-hailing services, and walking), with its core being to provide users with a "one-stop" travel solution. However, existing technologies have the following problems:

[0003] (1) Single data source and insufficient dynamism: Traditional route planning relies on static road network data or single real-time traffic data (such as road congestion), ignoring the collaborative influence of multi-source heterogeneous data such as weather, user historical behavior, events (such as concerts and sports events), and real-time public transportation schedules, resulting in a disconnect between route planning and actual scenarios.

[0004] (2) Insufficient consideration of individual differences: Existing methods are mostly based on single-objective model solutions and do not fully explore traffic features, user features, etc., resulting in a mismatch between route recommendations and actual user needs.

[0005] (3) Weak multi-mode collaborative optimization capability: Existing technologies cannot effectively coordinate the various conditions and constraints of different modes of transportation. They often adopt the method of independently planning conditions and constraints and then simply splicing them together, failing to achieve global optimization among multiple modes in the MAAS scenario. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method for generating individual MAAS travel route decisions based on multi-source data.

[0007] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0008] A method for generating individual MAAS travel route decisions based on multi-source data includes the following steps:

[0009] Step 1: Collect traffic status data, user behavior data, environmental factor data, user preference data, and real-time event data;

[0010] Step 2: Based on traffic status data, user behavior data, and environmental factor data, construct a multi-objective optimization model that includes objective functions and constraints. The objective functions include minimizing total travel time, minimizing total cost, minimizing comfort penalty, and minimizing carbon emissions. The constraints include time window constraints, transfer constraints, and resource constraints.

[0011] Step 3: Introduce dynamic crossover rate and dynamic mutation rate to improve the NSGA-III framework, and use the improved NSGA-III framework to solve the multi-objective optimization model to obtain the non-dominated solution set;

[0012] Step 4: Based on the non-dominated solution set, personalized candidate paths are generated by combining Top-K preference screening. The selected path determined from the personalized candidate paths is adapted to traffic changes in real time through an event-driven dynamic adjustment mechanism.

[0013] Furthermore, the minimization of total travel time The calculation formula is:

[0014]

[0015] in, This represents the total travel time across all traffic segments. This represents the total time spent transferring between different modes of transportation. This is the total time spent on all walking segments during the trip.

[0016] Minimize total cost The calculation formula is:

[0017]

[0018] in, This is the sum of the base fares for all modes of transportation; Additional costs during peak periods and when supply and demand are imbalanced;

[0019] Minimize comfort penalty The calculation formula is:

[0020]

[0021] Among them, the crowding level is a quantitative indicator ranging from 0 to 100; This indicates that when the distance walked in a single trip is less than 500 meters, the penalty is... When the walking distance is greater than or equal to 500 meters, the penalty is 0; A quantitative classification of environmental factors;

[0022] Minimize carbon emissions The calculation formula is:

[0023]

[0024] in, The driving length of each traffic segment, The carbon emission factor per unit length.

[0025] Furthermore, the time window constraint expression is as follows:

[0026]

[0027] in, Plan departure time for users. To pass the objective function Calculated total travel time The latest time the user needs to arrive; This is for buffer time;

[0028] The transfer constraint expression is as follows:

[0029]

[0030] in, This represents the total walking distance across all transfer points.

[0031] The resource constraint expression is:

[0032]

[0033] in, , These are the real-time inventory of shared bicycles and shared electric vehicles around the starting point and transfer point of the route, respectively.

[0034] Furthermore, the improved NSGA-III framework includes initialization, genetic operations, population merging and stratification, environmental selection, iteration, and termination.

[0035] Furthermore, the formulas for the dynamic crossover rate and the dynamic mutation rate are as follows:

[0036]

[0037]

[0038] in, For dynamic crossover rate, The dynamic variation rate, The initial crossover rate, The initial mutation rate is n; n is the current iteration number. The maximum number of iterations, , All are constants greater than 0 and less than 1.

[0039] Furthermore, the personalized candidate path generation process is as follows: First, based on user preference data, a lightweight neural network model is used to obtain a user preference vector. ,and Secondly, path schemes in the non-dominated solution set. A personalized comprehensive score is obtained based on user preference vectors. Finally, based on the selection rules, K=3 personalized candidate paths are selected from the non-dominated solution set.

[0040] Furthermore, the personalized comprehensive score The calculation formula is:

[0041]

[0042] in, , , , These are the normalized total travel time, total cost, comfort penalty, and carbon emissions, respectively.

[0043] Furthermore, the filtering rules satisfy:

[0044] (1) The path schemes are sorted in ascending order of S, and the first two path schemes are selected as the comprehensive optimal solution; (2) Combine the user preference vector to select one extreme objective optimization solution. The specific selection method is: select the path scheme with the smallest objective function value corresponding to the largest user preference vector from the non-dominated solution set, and the other objective function values ​​of the path scheme are all less than or equal to the upper limit of the corresponding objective function value in the non-dominated solution set; (3) The three selected paths are checked in combination with the constraint conditions. If a path does not meet the constraint, the path scheme is replaced from the non-dominated solution set in order.

[0045] Furthermore, the event-driven dynamic adjustment mechanism is specifically as follows: subscribing to real-time event data streams via the MQTT protocol, parsing the received real-time event streams, obtaining the geographical location and severity of the road segment where the real-time event is located, and if there is an intersection between the road segment that the user has not yet passed in the selected path and the road segment, and the severity of the event exceeds a preset threshold, then an incremental fast path replanning algorithm is immediately used for path replanning.

[0046] Furthermore, the incremental fast path replanning algorithm employs the D* Lite algorithm.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] (1) Collecting multi-dimensional data such as traffic status data, user behavior data, environmental factor data, and real-time event data solves the problem of incomplete information from a single data source and improves the accuracy of route decision-making.

[0049] (2) Establish a multi-objective optimization model that integrates minimizing total travel time, minimizing total cost, minimizing comfort penalty and minimizing carbon emissions, and introduce dynamic crossover rate and dynamic mutation rate to improve the NSGA-III framework. Use the improved NSGA-III framework to solve the multi-objective optimization model to ensure that the global optimal solution is fully searched and confirmed.

[0050] (3) By quantifying preferences through user behavior data and combining them with Top-K screening, personalized candidate path recommendations are made, so that the generated personalized path solutions match the actual needs of users more closely.

[0051] (4) When facing sudden scenarios, the D* Lite algorithm is used. This algorithm uses the information from the previous path search to update the affected road segment and achieves millisecond-level response. It can adapt to pedestrian preferences and ensure travel reliability. Attached Figure Description

[0052] Figure 1 This is a flowchart of the individual MAAS travel route decision generation method based on multi-source data fusion as described in this invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0054] like Figure 1 As shown, an individual MAAS travel route decision generation method based on multi-source data fusion includes the following steps:

[0055] Step 1, Multi-source data acquisition

[0056] The collected data includes traffic status data, user behavior data, environmental factor data, user preference data, and real-time event data. Traffic status data includes road condition data (such as road congestion index and average vehicle speed), public transportation data (such as real-time location, frequency, and arrival time of subways and buses), and shared resource data (such as the location and availability of shared bicycles and shared electric vehicles, and the real-time distribution of ride-hailing services). User behavior data includes origin and destination, common modes of transportation, and travel time. Environmental factor data includes weather conditions (such as temperature, precipitation, and wind speed) and environmental quality (such as PM2.5 concentration and air quality index). User preference data includes explicit preference settings (such as cost sensitivity and time sensitivity) and implicit behavioral characteristics (such as click feedback for comfort requirements). Real-time event data includes information on traffic accidents, road closures due to construction, temporary traffic control, and large-scale events.

[0057] Step 2, Construction of multi-objective optimization model

[0058] Based on traffic status data, user behavior data, and environmental factor data, a multi-dimensional, multi-constraint, multi-objective optimization model is constructed to achieve comprehensive optimization of travel plans, taking into account travel efficiency, economic cost, experience comfort, and environmental friendliness. The core of the multi-objective optimization model is to quantify each optimization dimension through an objective function and set constraints in conjunction with actual travel scenarios.

[0059] 2.1 Objective Function Design

[0060] The multi-objective optimization model includes four objective functions, the specific quantization forms of which are as follows:

[0061] (1) Minimize total travel time Travel time is the most critical efficiency indicator for users, encompassing the time spent throughout the entire travel process. The calculation formula is as follows:

[0062]

[0063] in, This represents the total travel time across all traffic segments. This represents the total time spent transferring between different modes of transportation. This represents the total time spent on all walking segments during the trip.

[0064] (2) Minimize total cost Economic cost is a key factor in user decision-making, including fixed costs and dynamic floating costs, and the calculation formula is as follows:

[0065]

[0066] in, This is the sum of the base fares for all modes of transportation; Additional costs during peak periods and when supply and demand are imbalanced.

[0067] (3) Minimize comfort penalty Comfort is assessed by quantifying "negative experiences" to create a penalty system that reflects the degree of congestion and the rationality of walking. The calculation formula is as follows:

[0068]

[0069] The crowding level is a quantitative index ranging from 0 to 100 (0 being completely empty and 100 being extremely crowded). When the crowding level exceeds 60 (the moderate crowding threshold), The penalty increases quadratically with increasing congestion, thus intensifying the punishment in highly congested scenarios. This indicates that when a single walking distance is less than 500 meters (too short a walk may reduce the experience due to frequent transfers), the penalty is... When the walking distance is greater than or equal to 500 meters (meeting the standard walking comfort level), the penalty is 0. This represents the quantitative level of environmental factors. The higher the level, the worse the environment. The value range is [0, 10].

[0070] (4) Minimize carbon emissions Environmental targets are quantified through carbon emissions across the entire transportation chain, calculated using the following formula:

[0071]

[0072] in, This represents the travel length of each traffic segment; The carbon emission factor per unit length.

[0073] 2.2, Constraint Handling

[0074] To ensure the feasibility and practicality of the travel plan, three types of core constraints need to be set to filter out invalid solutions that do not conform to the actual scenario:

[0075] (1) Time window constraint

[0076] The time window constraint ensures that the trip is completed within an acceptable timeframe for the user; the expression is:

[0077]

[0078] in, Plan departure time for users. To pass the objective function Calculated total travel time The latest time the user needs to arrive; For buffer time, It is used to cope with uncertainties such as traffic delays.

[0079] (2) Transfer constraints

[0080] Transfer constraints limit the walking distance during the transfer process to avoid excessive walking that would degrade the experience. The expression is:

[0081]

[0082] in, This represents the total walking distance for all transfers.

[0083] (3) Resource constraints

[0084] Resource constraints, for modes of transportation such as shared bicycles that require real-time availability verification, are expressed as follows:

[0085]

[0086] in, , These are the real-time inventory of shared bicycles and shared electric vehicles around the starting point and transfer point of the route, respectively.

[0087] Step 3, Solving the multi-objective optimization model

[0088] A modified NSGA-III (Non-dominated Sorting Genetic Algorithm-III) framework is used to solve the multi-objective optimization model. The modified NSGA-III introduces dynamic crossover and mutation rates to dynamically balance global exploration and local exploitation capabilities, ensuring that the final path set possesses diversity, safety, and efficiency. The specific steps are as follows:

[0089] (1) Initialization: Randomly generate an initial parent population of size N. At the same time, based on the number of objective functions, generate a set of uniformly distributed reference points on the unit hyperplane using a structured method (such as the Das-Dennis method) to guide the distribution of the population throughout the target space.

[0090] (2) Genetic operations: Through selection, crossover, and mutation, a progeny population of size N is generated. Generally, a fixed crossover and mutation rate are used to generate the progeny population. However, in path planning, the decision variables have high dimensionality, and a fixed crossover and mutation rate is ineffective. Dynamically adjusted crossover and mutation rates can more effectively find the optimal solution. Therefore, a linear function is introduced to dynamically adjust the crossover and mutation rates during the iteration process. The formula is:

[0091]

[0092]

[0093] in, For dynamic crossover rate, The dynamic variation rate, The initial crossover rate, The initial mutation rate is n; n is the current iteration number. The maximum number of iterations, , All are constants greater than 0 and less than 1. In the early stages of optimization, a higher crossover and mutation rate is needed to break the limitations of the initial search and expand the exploration boundary to create the possibility for the algorithm to escape local optima. In the later stages of optimization, a lower crossover and mutation rate is needed to allow the algorithm to ensure that the global optimal solution is fully searched and confirmed by using more concentrated search resources.

[0094] (3) Population merging and stratification: The parent and offspring populations are merged to form a mixed population of size 2N. Then, non-dominated sorting is performed on the mixed population to divide individuals into multiple non-dominated strata. .

[0095] (4) Environmental selection: Individuals are selected according to the non-dominant layer order to enter a new parent population of size N, until The critical layer cannot accommodate the entire new parent population. By calculating the association between individuals and reference points, individuals located in sparse regions are selected to maintain population diversity.

[0096] (5) Iteration and Termination: Repeat steps (2) to (4) until the preset termination condition is met (such as reaching the maximum number of iterations). (or Pareto front convergence), the final output is a set of non-dominated solutions that approximate the true Pareto front and are uniformly distributed.

[0097] Step 4: Personalized Path Generation and Dynamic Adjustment

[0098] To ensure that the generated routes not only match users' travel habits but also adapt to dynamic traffic changes in real time, a combined mechanism of "candidate route generation + dynamic adjustment" is needed to achieve the dual goals of personalized and accurate recommendations and real-time intelligent adaptation.

[0099] 4.1 Candidate Path Generation

[0100] The non-dominated solution set output by multi-objective optimization includes various trade-offs between time, cost, comfort, and environmental protection, but users usually need only a few solutions that best suit their habits. Therefore, this embodiment combines Top-K preference filtering to generate personalized candidate paths from a massive non-dominated solution set.

[0101] First, based on user preference data, a lightweight neural network model is used to obtain user preference vectors. ,and The preference vector reflects pedestrians' emphasis on time, cost, comfort, and carbon emissions; secondly, for the path schemes in the non-dominated solution set obtained in step 3... A personalized comprehensive score is obtained based on user preference vectors. , , , , These are the normalized total travel time, total cost, comfort penalty, and carbon emissions, respectively. Finally, K=3 personalized candidate paths are selected from the non-dominated solution set according to the selection rules. The selection rules must meet the following: (1) Sort by S in ascending order and select the first 2 path schemes as the comprehensive optimal solution to ensure that they fit user preferences; (2) Combine the user preference vector to select one extreme objective optimization solution to deal with the user's temporary urgent needs. The specific selection method is: select the path scheme with the smallest objective function value corresponding to the largest user preference vector from the non-dominated solution set (e.g., ...). The maximum is selected by choosing the path with the shortest total travel time, and the other objective function values ​​of the path are all less than or equal to the upper limit of the corresponding objective function values ​​in the non-dominated solution set (e.g., the 80th percentile of each objective function value in the non-dominated solution set is used as the upper limit). This condition ensures that the extreme objective optimization solution is within an acceptable range and avoids recommending impractical paths; (3) The three selected paths are checked in conjunction with the constraints in step 2 to ensure that they meet the time window constraints, transfer constraints, and resource constraints. If a path does not meet the constraints, the path scheme is replaced from the non-dominated solution set in order.

[0102] 4.2 Dynamic Adjustment Mechanism

[0103] Urban traffic conditions change dynamically, and the selected route determined from personalized candidate routes may become invalid due to unforeseen circumstances during the journey. Therefore, this embodiment establishes an event-driven dynamic adjustment mechanism to ensure that the selected route can adapt to the latest traffic conditions in real time.

[0104] The event-driven dynamic adjustment mechanism works as follows: It subscribes to real-time event data streams via the MQTT protocol. After parsing the received real-time events, it obtains the geographical location and severity of the road segment where the event occurs. If a road segment not yet visited by the user's selected path intersects with this road segment, and the event severity exceeds a preset threshold, an incremental fast path replanning algorithm is immediately used for path replanning. This incremental fast path replanning algorithm employs the D* Lite algorithm, which utilizes information from the previous path search to update only the affected road segments, achieving millisecond-level response.

[0105] In summary, personalized route generation ensures that the plan matches user preferences, while the dynamic adjustment mechanism allows the plan to adapt to traffic changes in real time. The combination of the two achieves a full-process travel plan service that combines static optimization and dynamic adaptation.

[0106] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for generating individual MAAS travel route decisions based on multi-source data, characterized in that, Includes the following steps: Step 1: Collect traffic status data, user behavior data, environmental factor data, user preference data, and real-time event data; Step 2: Based on traffic status data, user behavior data, and environmental factor data, construct a multi-objective optimization model that includes objective functions and constraints. The objective functions include minimizing total travel time, minimizing total cost, minimizing comfort penalty, and minimizing carbon emissions. The constraints include time window constraints, transfer constraints, and resource constraints. Step 3: Introduce dynamic crossover rate and dynamic mutation rate to improve the NSGA-III framework, and use the improved NSGA-III framework to solve the multi-objective optimization model to obtain the non-dominated solution set; Step 4: Based on the non-dominated solution set, personalized candidate paths are generated by combining Top-K preference screening. The selected path determined from the personalized candidate paths is adapted to traffic changes in real time through an event-driven dynamic adjustment mechanism.

2. The individual MAAS travel route decision generation method based on multi-source data according to claim 1, characterized in that, Minimize total travel time The calculation formula is: ， in, This represents the total travel time across all traffic segments. This represents the total time spent transferring between different modes of transportation. This is the total time spent on all walking segments during the trip. Minimize total cost The calculation formula is: ， in, This is the sum of the base fares for all modes of transportation; Additional costs during peak periods and when supply and demand are imbalanced; Minimize comfort penalty The calculation formula is: ， Among them, the crowding level is a quantitative indicator ranging from 0 to 100; This indicates that when the distance walked in a single trip is less than 500 meters, the penalty is... When the walking distance is greater than or equal to 500 meters, the penalty is 0; A quantitative classification of environmental factors; Minimize carbon emissions The calculation formula is: ， in, This represents the travel length of each traffic segment; The carbon emission factor per unit length.

3. The individual MAAS travel route decision generation method based on multi-source data according to claim 2, characterized in that, The time window constraint expression is as follows: ， in, Plan departure time for users. To pass the objective function Calculated total travel time The latest time the user needs to arrive. This is for buffer time; The transfer constraint expression is as follows: ， in, This represents the total walking distance across all transfer points. The resource constraint expression is: ， in, , These are the real-time inventory of shared bicycles and shared electric vehicles around the starting point and transfer point of the route, respectively.

4. The individual MAAS travel route decision generation method based on multi-source data according to claim 1, characterized in that, The improved NSGA-III framework includes initialization, genetic operations, population merging and stratification, environmental selection, iteration, and termination.

5. The individual MAAS travel route decision generation method based on multi-source data according to claim 1, characterized in that, The formulas for the dynamic crossover rate and the dynamic mutation rate are as follows: ， ， in, For dynamic crossover rate, For dynamic variation rate, The initial crossover rate, The initial mutation rate is n; n is the current iteration number. The maximum number of iterations, , All are constants greater than 0 and less than 1.

6. The individual MAAS travel route decision generation method based on multi-source data according to claim 1, characterized in that, The personalized candidate path generation process is as follows: First, based on user preference data, a lightweight neural network model is used to obtain the user preference vector. ,and Secondly, path schemes in the non-dominated solution set. A personalized comprehensive score is obtained based on user preference vectors. Finally, based on the selection rules, K=3 personalized candidate paths are selected from the non-dominated solution set.

7. The individual MAAS travel route decision generation method based on multi-source data according to claim 6, characterized in that, The personalized comprehensive score The calculation formula is: ， in, , , , These are the normalized total travel time, total cost, comfort penalty, and carbon emissions, respectively.

8. The individual MAAS travel route decision generation method based on multi-source data according to claim 6, characterized in that, The filtering rules satisfy: (1) The path schemes are sorted in ascending order of S, and the first two path schemes are selected as the comprehensive optimal solution; (2) Combine the user preference vector to select one extreme objective optimization solution. The specific selection method is: select the path scheme with the smallest objective function value corresponding to the largest user preference vector from the non-dominated solution set, and the other objective function values ​​of the path scheme are all less than or equal to the upper limit of the corresponding objective function value in the non-dominated solution set; (3) For the three selected paths, check them in conjunction with the constraints. If a path does not meet the constraints, replace the path scheme in the non-dominated solution set according to the order.

9. The method for generating individual MAAS travel routes based on multi-source data according to claim 1, characterized in that, The event-driven dynamic adjustment mechanism is as follows: subscribe to real-time event data streams via the MQTT protocol, parse the received real-time event streams, obtain the geographical location and severity of the road segment where the real-time event is located, and if there is an intersection between the road segment that the user has not yet passed in the selected path and the road segment, and the severity of the event exceeds a preset threshold, then the incremental fast path replanning algorithm is immediately used for path replanning.

10. The method for generating individual MAAS travel routes based on multi-source data according to claim 9, characterized in that, The incremental fast path replanning algorithm uses the D* Lite algorithm.

Images