Optimization of transportation routes

Abstract

This invention provides a computer implementation method, program, and information processing device for optimizing routes in a transportation network. [Solution] The method includes obtaining an analysis of geographical regions, generating probabilistic predictions, checking whether at least a subset of the generated probabilistic predictions has been previously solved by a trained machine learning model, if at least a first subset of the generated probabilistic predictions has been previously solved, retrieving the previously used trained machine learning model, running the trained machine learning model to determine a first set of proposed routes for the transport network, and if at least a second subset of the generated probabilistic predictions has not been previously solved, training a new machine learning model to solve at least a second subset of the generated probabilistic predictions to determine a second set of proposed routes for the transport network.

Description

[Technical Field]

[0001] Embodiments of the present invention described herein relate to systems and methods for optimizing transportation routes, and more particularly to computer implementation methods, computer programs, and information processing devices. [Background technology]

[0002] Today, cities around the world are aiming to expand the capacity of their public transport systems due to population growth, while considering budgetary constraints and environmental impacts in providing more sustainable transport options. According to the International Energy Agency, transport will account for approximately 22.96% of global energy-related CO2 emissions in 2022.

[0003] In recent years, well-designed bus rapid transit (BRT) systems have become a viable alternative to more expensive rail-based public transport systems (light rail transit, LRT, trains, or subways) around the world. However, once a BRT system is operational, its success often depends on the routes offered to passengers.

[0004] Therefore, the Bus Rapid Transit Route Design Problem (BRTRDP) is the problem of finding a set of routes and frequencies that minimize operating costs and passenger costs (travel time) while simultaneously satisfying the technical constraints of the system, such as requirements for travel, bus frequency, and lane capacity.

[0005] A key approach to solving BRTRDP is to analyze and understand people's mobility needs. Non-patent document 2 presents a web service that automates the generation of origin-destination (OD) matrices for mass transit systems. Another approach can be found in non-patent document 3, which proposes an algorithm that helps construct an initial set of routes based on a large set of geospatial data in relation to reducing the average length cost function for the optimization problem.

[0006] Traditionally, BRTRDP has been treated as a traditional optimization problem. Some examples of different approaches are as follows:

[0007] Non-patent document 1 presents a model for optimizing bus routes and service headways serving an area using commuter (many-to-one) travel patterns. This approach uses a pure grid network model applicable to irregular grid networks, demonstrating that the optimal bus route is sensitive to the demand distribution across the service area.

[0008] Non-patent document 2 proposes a route optimization method for long-distance commuter bus services to improve the attractiveness of public transport as a sustainable mode of transportation. This study presents origin-destination (OD) demand analysis and design optimization for express bus service (EBS) routes.

[0009] Non-patent document 3 proposes a three-dimensional macroscopic basic diagram, which is expressed as an objective function used to determine the optimal design parameters for route design.

[0010] Non-patent document 4 proposes an optimization for bus-feeder-route model design using a genetic algorithm to solve defined constraints on the optimization problem to be solved.

[0011] Alternatively, lastly, Non-Patent Document 5 proposes a modeling approach for designing bus routes with short-turn service patterns that take into account various operator and passenger objectives, such as minimizing capacity surplus, capacity deficit, and passenger time-related costs, using traditional optimization problem solvers. [Prior art documents] [Non-patent literature]

[0012] [Non-Patent Document 1] Chien S., Dimitrijevic BV, Spasovic LN Optimization of Bus Route Planning in Urban Commuter Networks 6, 53-79 (2003). https: / / doi.org / 10.5038 / 2375-0901.6.1.4. [Non-Patent Document 2] Ren, H.; Wang, Z.; Chen, Y. Optimal Express Bus Routes Design with Limited-Stop Services for Long-Distance Commuters. Sustainability 2020, 12, 1669. https: / / doi.org / 10.3390 / su12041669 [Non-Patent Document 3] Dakic I., Leclercq L., Menendez M. On the optimization of the bus network design: An analytical approach based on the three-dimensional macroscopic fundamental diagram. Transportation Research Part B: Methodology 149, 393-417 (2021). https: / / doi.org / 10.1016 / j.trb.2021.04.012. [Non-Patent Document 4] Cao Y, Jiang D, Wang S. Optimization for Feeder Bus Route Model Design with Station Transfer. Sustainability. 2022; 14(5):2780. https: / / doi.org / 10.3390 / su14052780 [Non-Patent Document 5] Yanik, S., Yilmaz, S. Optimal design of a bus route with short-turn services. Public Transportation 15, 169-197 (2023). https: / / doi.org / 10.1007 / s12469-022-00303-6 [Non-Patent Document 6] Leaning Heuristics for the TSP by Policy Gradient, Deudon et al., 2018 [Non-Patent Document 7] Attention, Learn to Solve Routing Problems!, Kool et al. 2019 [Non-Patent Document 8] An Efficient Graph Convolutional Network Technique for the Traveling Salesman Problem, Joshi, 2019 [Overview of the project] [Problems that the invention aims to solve]

[0013] Improved methods are needed to optimize the transportation system. [Means for solving the problem]

[0014] Summary of the Invention The present invention is described in the appended claims.

[0015] According to a first aspect, a computer-implemented method for optimizing routes in a transportation network for a geographical area is disclosed herein. The method comprises: obtaining an analysis of the geographical area, the analysis including a graph showing the relationship between the current transportation needs of transportation users and tourist hotspots in the geographical area, the graph including a set of nodes and a set of edges, each node indicating a location within the geographical area, and each edge indicating a route between two locations; generating a probabilistic prediction for each node and each edge of the graph, the probabilistic prediction indicating the importance of each location and each route within the transportation network; checking whether at least a subset of the generated probabilistic predictions have been previously solved by a trained machine learning model; if at least a first subset of the generated probabilistic predictions have been previously solved, retrieving the trained machine learning model previously used to solve the at least first subset of the generated probabilistic predictions and executing the trained machine learning model to determine a first plurality of proposed routes for the transportation network; and if at least a second subset of the generated probabilistic predictions have not been previously solved, training a new machine learning model to solve the at least second subset of the generated probabilistic predictions and determining a second plurality of proposed routes for the transportation network.

[0016] Other features of the disclosure are described below and recited in the appended claims.

Brief Description of the Drawings

[0017] Embodiments of the invention are now further described, by way of example only, with reference to the accompanying drawings. [Figure 1] Shows a system architecture and modules according to some embodiments of the invention. [Figure 2]A flowchart according to some embodiments of the present invention. [Figure 3] A flowchart according to some embodiments of the present invention. [Figure 4] Shows a problem definition module according to some embodiments of the present invention. [Figure 5] Shows exemplary inputs and outputs for the problem definition module according to some embodiments of the present invention. [Figure 5-2] (Continuation of FIG. 5) [Figure 5-3] (Continuation of FIG. 5) [Figure 5-4] (Continuation of FIG. 5) [Figure 6] Shows a tourist characteristic calibration module according to some embodiments of the present invention. [Figure 7] Shows an example of a conversion from a sparse high-dimensional non-Euclidean space to a low-dimensional space for embedding setup into a tourist fusion heuristic according to some embodiments of the present invention. [Figure 8] Shows exemplary inputs and outputs for the tourist characteristic calibration module according to some embodiments of the present invention. [Figure 8-2] (Continuation of FIG. 8) [Figure 8-3] [[ID=3)]](Continuation of FIG. 8) [Figure 8-4] (Continuation of FIG. 8) [Figure 8-5] (Continuation of FIG. 8) [Figure 8-6] (Continuation of FIG. 8) [Figure 8-7] (Continuation of FIG. 8) [Figure 8-8] (Continuation of FIG. 8) [Figure 9] Shows a tourist impact assessment module according to some embodiments of the present invention. [Figure 10] Shows a flowchart according to some embodiments of the present invention. [Figure 11] Shows an increase in the efficiency of embodiments of the present invention when compared to current solutions. [Figure 12]This document describes an optimal transport network route solver module according to several embodiments of the present invention. [Figure 13] Flowcharts illustrating several embodiments of the present invention are shown. [Figure 14] The following are exemplary inputs and outputs for an optimal transport network route solver module according to several embodiments of the present invention. [Figure 14-2] [Continued from Figure 14] [Figure 14-3] [Continued from Figure 14] [Figure 14-4] [Continued from Figure 14] [Figure 14-5] [Continued from Figure 14] [Figure 14-6] [Continued from Figure 14] [Figure 15] The following are examples of apparatus according to several embodiments of the present invention. [Figure 16] This flowchart illustrates several embodiments of the present invention. [Figure 17] This flowchart illustrates several embodiments of the present invention. [Figure 18] This flowchart illustrates several embodiments of the present invention. [Modes for carrying out the invention]

[0018] Overview The use of transportation systems like Bus Rapid Transit (BRT) is gradually expanding worldwide. By providing dedicated bus lanes, BRT systems function as an alternative to urban transport systems, thereby speeding up travel times. Nevertheless, numerous inherent strategic and operational concerns remain that need to be addressed.

[0019] Embodiments of this disclosure focus on addressing the problem of designing optimal transportation (e.g., BRT) routes for attractions within a Digital Twin (DT) that take into account the impact on travelers without the know-how of local experts and improve fixed models that do not evolve with changing people dynamics.

[0020] Embodiments of this disclosure aim to address the following issues. Currently, transportation network and route design heavily relies on local and expert knowledge. Transportation network and route design involves a strong element of human intervention, where decisions are made by experts or rules and constraints are defined manually. The models defined for the transportation network and route design are handcrafted and fixed. It is difficult to assess the impact of transportation networks on tourists. It is difficult to incorporate the impact of tourists into transportation network and route design models. It is impossible to solve non-deterministic polynomial (NP) hard problems optimally on a large scale.

[0021] Embodiments of this disclosure attempt to address the above-mentioned problems by facilitating the following: • No handcrafted heuristic methods (specifications defining the problem at hand). Instead of experts manually designing heuristics and rules, embodiments of this disclosure learn by using artificial intelligence (AI). • High-speed inference. Traditional solvers often have long execution times for large-scale problems. In embodiments of this disclosure, once the model is trained, it significantly reduces execution time. • Explainability. Converting people's movement data to OD (Origin-Destination) allows for an intuitive representation of the relationships between different locations of interest. • Automating the movement of people from relevant travel for tourists, and aligning travel within the transportation network.

[0022] While several existing approaches have attempted to address BRTDP (see, for example, the background section above), none of the existing studies have considered the impact of BRT design on tourism impacts in areas where the tourism sector has a significant economic impact. Furthermore, current technology solutions rely on the know-how of transportation and local experts to define potential BRT routes, providing handcrafted, fixed solutions that do not scale for large-scale scenarios.

[0023] One application of embodiments of the present invention is a BRT system as described above. The present invention is described below primarily in relation to this application. However, embodiments of the present invention focus on optimizing transportation routes in areas where tourism has a significant economic impact, and this may have applications across all types of transportation networks beyond BRT systems. Embodiments of the present invention can be equally used for any transportation system that would benefit from optimizing routes with tourism in mind. Other examples of transportation systems that may be improved using embodiments of the present invention include rail transportation systems (e.g., subways, metros, trams, streetcars, commuter rail services, monorails), other bus transportation systems (e.g., standard bus services, shuttle buses), and water transportation systems (e.g., ferries, water taxis).

[0024] The following non-exclusive and illustrative list explains some technical terms. ·AI…Artificial Intelligence AR…Auto Regressive BRT…Bus Rapid Transit BRTRDP…Bus Rapid Transit Route Design Problem COP…Combinatorial Optimization Problems ·DT…Digital Twin EBS…Express Bus Service LRT…Light Rail Transit MLE…Maximum Likelihood Estimation • NAR…Non-Auto Regressive OD…Origin-Destination

[0025] Various aspects and details of these key concepts are explained below as mere examples, with reference to Figures 1-18.

[0026] This disclosure enables the acquisition of optimal transportation network routes for areas with a significant economic impact from tourists (i.e., tourist-driven regions). This disclosure relates to an automated system for providing optimal transportation network routes that maximize tourist comfort and the economic impact of the tourism sector by covering tourist needs from a transportation perspective. Advantageously, the automated system understands people's preferences and behavior and provides a computationally efficient process, without relying on local or expert knowledge.

[0027] This disclosure describes two main aspects: (1) a system that enables an understanding of people's dynamics and covers the assessment of the impact of tourists on key locations, and (2) a system that identifies locations with a large tourist impact and finds optimal transport network routes in a more computationally efficient way than traditional approaches.

[0028] As shown in Figure 1, the present invention may include four main components.

[0029] ●S000…Problem Definitioner: This component constructs a network representation of current transportation needs based on an analysis of people's behavior. It analyzes people's behavior and movements to determine individual movements and define locations that are important to people. Movements and locations may be filtered based on relevance; for example, rarely used movements may be removed from the transportation data. This analysis is the first step in constructing a complete problem definition with potential transportation network needs.

[0030] ●S100…Calibration of Tourist Characteristics: This component constructs a network representation of current transportation needs and tourist hotspots in the geographic area being analyzed. It constructs a representation of tourist hotspots using input data of (1) accommodation, (2) tourist place, and (3) sightseeing spot, which are combined through a logical process. The representation of tourist hotspots is combined with the representation of current transportation needs to generate a tourism fusion network / heuristic in the form of a graph.

[0031] ●S200…Tourist Impact Assessment: This component trains the model by passing a tourist fusion network graph with an initial set of embeddings and calibrating the correct impact for each defined location (i.e., a node in the graph). Finally, the tourist impact decoding module provides the updated tourist fusion network / heuristic in the form of an updated graph that defines the heuristics for the most relevant locations based on tourist features / hotspots.

[0032] ●S300…Optimal Transportation Network Route Solver: This component uses an updated tourist fusion network to understand heuristics about transportation needs and predict the probability of movement based on heuristics defined for tourists. Finally, this module determines the optimal transportation network route and provides the optimal route represented within the network.

[0033] In summary, embodiments of the present invention may include one or more of the following steps. 1. Combine tourist information with transportation information. 2. Incorporate the combined information into the graph embedding (for example, this can be done using the encoding process shown in Figure 7). 3. Train a model to recalibrate the graph weights based on connected location and proximity ("Tourist Impact Assessment"). 4. Decode the low-dimensional information from the tourist impact assessment into interpretable information for estimating migration probabilities (this can be done by reversing the encoding process in Figure 7). 5. Find the optimal routes based on the routes that have the greatest impact on tourists.

[0034] These four components are shown in Figure 1 as being collectively housed on platform 100. Platform 100 communicates with network 102. Network 102 communicates with transport specialists 104, cell tower 106, and satellite 108. Cell tower 106 communicates with IoT vehicle sensors 110, mobile phones 112, and road network sensors 114. The satellite communicates with GPS receiver 116 and CCTV 118.

[0035] Figure 2 illustrates method 200. One or more of its steps can be performed according to embodiments of the present invention.

[0036] In the first step S001, Method 200 collects GPS data (e.g., from GPS receiver 116 via network 102) and data on the movement of people within a defined geographical area / region (e.g., from IoT vehicle sensors 110, mobile phones 112, and road network sensors 114 via network 102). This data is sometimes collectively referred to as “transportation data.” The defined area is the area in which the transport network operates. In step S002, this data is processed and formatted to understand the dynamics of people and generate movement undertaken by groups (“movement information”). Then, in step S003, the movement information is disaggregated and supplemented with demographic information to filter it for a target group. In step S004, geographic information is included, which may include filtered movement (e.g., movement may be filtered so that infrequent movement is excluded from the data). Then, in step S005, heuristics for transport needs based on group movement are defined.

[0037] Steps S101-S107 are performed in parallel with defining the transportation needs for the group (step S005). Firstly, step S101 can acquire accommodation information and generate a map showing the geolocation / concentration of accommodation spots within the defined area. Secondly, step S102 can acquire information about tourist places (e.g., restaurants, shops, and tourist areas) and generate a map showing the geolocation / concentration of tourist spots within the defined area. Thirdly, step S103 can acquire information about sightseeing spots within the defined area and generate a map showing the geolocation / concentration of sightseeing spots. Accommodation spots, tourist places, and sightseeing spots are sometimes collectively referred to as "tourist hotspots." Finally, the three information layers (accommodations, tourist places, and sightseeing spots) are combined into one based on proximity logic that calculates the relevance [importance] of each location. This new heuristic may be called tourist feature fusion or "tourist data."

[0038] Currently, to calibrate the impact of transportation needs on tourists and conduct a tourist impact assessment, transportation needs for the group (transportation data) and the characteristics of tourists (tourist data) are combined. To implement this combination, transportation data is used to set up initial embeddings for origins and destinations within the geographical area. Then, tourist data is incorporated into the embeddings to generate a tourist fusion network / heuristic in the form of a graph. Here, the nodes of the graph represent locations within the geographical area, and the edges of the graph represent routes between those locations.

[0039] Finally, the tourist fusion network is used to train a model that performs tourist impact assessment by decoding the heuristics of the tourist fusion network (by converting compressed low-dimensional information into "significant" information for tourist impact assessment; the reverse of the process described later in relation to Figure 7) and recalibrating the features of the embedding based on connected locations and proximity (i.e., reweighting the nodes of the graph with respect to non-significant information, in other words, low-dimensional spatial information). As a result, the embedding is updated to obtain updated tourist fusion heuristics.

[0040] Once calibrated, embodiments of the present invention interpret the updated tourist fusion heuristic and decode the heuristic into a probabilistic prediction for the transportation needs of each defined location and between locations. As a final step, embodiments of the present invention define a discrete solution containing the optimal route based on a solver that searches for all connected locations and their probabilistic predictions.

[0041] Tourist Impact Assessment The aspect of the present invention for evaluating the impact of tourists includes a problem definition (S000), tourist characteristic calibration (S100), and tourist impact evaluation (S200). The aspect of evaluating the impact of tourists addresses the following technical challenges. 1. How to understand people's dynamics and transportation needs from transportation data, and how to define heuristics to represent the needs of transportation networks. 2. How can we automatically estimate the tourist importance of a geographical area by combining different locations that are important to tourists? 3. How to automatically define heuristics about the impact of tourists on the transportation network, based on transportation needs and the importance of tourists.

[0042] Referring to Figure 3, the three main stages of the tourist impact assessment aspect are referred to here as the problem definition S000, tourist characteristics calibration S100, and tourist impact assessment S200.

[0043] Problem Definition Module The problem definition module S000 (shown in Figure 4) provides transportation data regarding transportation needs for a geographical region / area. The module collects people's movement data 400 from different devices and GPS signals operating within the geographical region. This information is processed, transformed, and aggregated by the people data parser 406 to generate individual movements. These movements are analyzed by the OD generator 408 using statistical inference-based methods to generate OD 410 through spatial clustering methods such as trip chaining or Maximum Likelihood Estimation (MLE) techniques.

[0044] Once the OD is generated, the problem definition engine 412 collects information 402 about the geographical area to be analyzed (for example, polygons of coordinates (latitude and longitude) that define areas such as TAZ (traffic analysis zone), LSOA (Lower Layer Super Output Area), MSOA (Middle Layer Super Output Area)) and data 404 about the transportation network (for example, the current or future transportation network), and constructs heuristics (specifications for defining the problem) that include information about current transportation needs based on the analysis of people's behavior.

[0045] Relevance filtering 414 may be used to define important locations within a geographical area and to select significant transportation needs based on relevance and quantity, with the aim of focusing on relevant transportation needs. For example, infrequent travel may not be considered.

[0046] The output of the problem definitioner S000 is a heuristic definition 416 that represents the transportation needs of transportation users within a geographical area. This may be in graph form, as shown in Figure 4.

[0047] Figure 5 shows exemplary inputs and outputs for the problem definition module S000. This process is the first step in constructing a complete problem definition with potential transport network needs.

[0048] Tourist characteristics calibration To construct a complete problem definition with potential transportation network needs, the tourist feature calibration module S100 creates a heuristic that describes transportation needs aggregated into tourist features. This output becomes part of the input required by the tourist impact assessment module S200 to understand the impact of potential tourists in each area aligned with transportation needs.

[0049] As shown in Figure 6, the tourism feature calibration module S100 receives transportation needs as a heuristic definition 416 (sometimes simply called “transportation data”). The initial tourist relevance 600 then initializes a set of embeddings 602 for each location and their connections to other locations, with the aim of incorporating tourist hotspots / features (indicated by the tourist data 416) to measure the tourist impact on each location.

[0050] The tourist feature calibration 630 incorporates tourist hotspots (indicated by tourist data). The tourist data may include accommodation data 604, relevant tourist destination data 606, and / or tourist spot data 608. Area information 610 may also be used. Firstly, the accommodation spot builder 612 receives accommodation data 604 about a geographical area, calculates the concentration of accommodations, the types of accommodations, and geolocates the accommodation spots. This may be shown on the accommodation map 618. Similarly, the tourist destination builder 614 receives relevant tourist destination data 606 about a geographical area, classifies those locations, calculates the capacity of each location, and geolocates them. This may be shown on the tourist destination map 620. Finally, the tourist spot builder 616 collects information on tourist spots 608, the types of spots, and geolocates them. This may be shown on the tourist map 622.

[0051] All geolocated tourist features / hotspots are used by tourist feature fusion 624 to perform a tourist relevance analysis based on weights and proximity 626 for each level, simultaneously at three levels. Firstly, the weights of each tourist feature for each location are normalized for accommodation, tourist destination, and tourist feature. Once their values ​​are normalized, all geolocated hotspots are classified using an unsupervised clustering technique with geolocation features and a predetermined proximity threshold setup 626. Here, for every cluster group, the spots minimize the number of clusters where all clusters are at a distance shorter than a predetermined proximity distance. Then, relevance weights for each hotspot are set based on the weights of neighboring spots belonging to the same cluster. Finally, tourist feature fusion 624 estimates the impact of each area by using the clustering technique. The results are provided as a tourist feature fusion map 628. In effect, tourist hotspots may also be grouped based on their proximity; for example, all hotspots within 2 miles may be grouped into a single hotspot.

[0052] Finally, tourist feature calibration 630 incorporates tourist data from tourist feature fusion map 628 into the initial embedding 602 to include transportation needs and tourist influences in a new heuristic called tourist fusion heuristic 632. This heuristic may be in the form of a graph where locations within a geographic area are represented by nodes in the graph and connections between locations are represented by edges in the graph.

[0053] Figure 7 shows the sparse, high-dimensional non-Euclidean space (where v) derived from the original tourist feature fusion map 628 G(V,E). i How graph embeddings use a transformation from {f1,f2,f3,f4}) to a lower-dimensional space for embedding setup, and the tourist fusion heuristic 632 z i This indicates whether it is encoded into {e1,e2,e3} (encoding process). Here, Φ(vi )=z i ∈R L i = 1, 2, ... |V|. This transformation may be reversed and used to decode the tourist fusion heuristic 632 for tourist impact assessment.

[0054] Figure 8 shows exemplary inputs and outputs for the tourist feature calibration module S100.

[0055] Tourist Impact Assessment The tourist fusion heuristic 632 provides a graph showing the current transportation needs of transportation users and tourist hotspots within a geographical area. However, transportation needs and tourist hotspots are not correlated. The objective of the tourist impact assessment module S200 shown in Figure 9 is to provide a tourist fusion heuristic 910 that combines transportation needs and tourist hotspots to define the relationship between tourist hotspots and transportation needs, thereby showing the tourist relevance for a given transportation need.

[0056] Figure 10 shows a flowchart illustrating the steps performed by the tourist impact assessment module S200, where the tourist fusion heuristic checker 900 reads the tourist fusion heuristic 632 (s1000) and checks it using the Tourist Fusion Heuristic (TFH) repository 902 if a trained model exists for the same geographic area (s1002). If one exists, it checks whether it contains the same location and the same features (i.e., the same graph nodes, the same tourist hotspots, and the same embedding values) (s1004). If the location and features are all the same (s1006), the updated tourist fusion heuristic is retrieved from the TFH repository 902 (s1008) and provided as the updated tourist fusion heuristic 910.

[0057] If the geographical areas are different, TFH decomposition 904 analyzes the tourist fusion heuristic 632 and divides it into n independent heuristics (also called “subgraphs”) (s1014). Then, TFH parallel evaluation 906 runs an anisotropic aggregation machine learning model that calibrates the location embeddings by sharing neighborhood embeddings for other connected locations and, in that sense, aggregating the tourist impact per region and per other connected location (s1016). In other words, the machine learning model reads the embeddings between connected nodes and nearby nodes and rebalances the embedding information. This is done by receiving subgraphs with embeddings and using this information along with edges to calibrate the embeddings based on node proximity, connections (edges) between nodes, and their proximity. This method effectively reduces the complexity and computational cost of large-scale, high-density tourist fusion heuristics. Finally, TFH merge 908 aggregates the n updated heuristics (subgraphs) into one as an updated tourist fusion heuristic 910 (s1018). The trained tourist fusion heuristics are then saved for future use (s1020).

[0058] For the tourist fusion heuristic, if locations share the same geographic area but have different locations and / or characteristics, the tourist fusion heuristic checker iterates through all locations (s1010), annotates those different locations (or the same location but with different characteristics), and forms a new heuristic with dependent (connected) and independent (disconnected) terms. Then, for each annotated location, a one-hop connected location is added to calibrate the tourist influence (s1012), the heuristic is divided into n independent heuristics (subgraphs) (s1014), and TFH parallel evaluation 906 (s1016) and TFH merge 908 (s1018) are performed to provide an updated tourist fusion heuristic 910.

[0059] Examples of suitable anisotropic and attention-based graph neural networks (GNNs) for use with embodiments of the present invention include the machine learning algorithm approach described in Non-Patent Document 6, the model described in Non-Patent Document 7, and the approach outlined in Non-Patent Document 8.

[0060] Figure 11 compares the efficiency of evaluating tourist fusion heuristics updated using different methods. The first approach 1100 could be a sequential execution of anisotropic aggregation to calibrate the embedding of locations, and as the number of locations and connections between locations increases, the time t0 and the number of resources for execution increase exponentially. In the second approach 1102, the anisotropic aggregation for calibrating the embedding of locations is divided into independent heuristics (subgraphs), and n executions are run in parallel, reducing complexity and the required time, where the total time t t = max(t1, t2, t3). Finally, in the third approach 1104, the anisotropic aggregation for calibrating the embedding of locations is only executed for updated locations and their neighborhoods, thus removing redundant executions for non-updated locations and embeddings (in this example, the evaluation shown by t2 has been previously executed, so time can be saved by retrieving its solution from the repository). This keeps the total time at least (in the worst-case scenario) parallel execution t t = max(t1, t3) and reduces the number of executions by omitting redundant executions. The total execution time is (t t = t1 + t3). The total execution time for the parallel approach is (t t = t1 + t2 + t3).

[0061] Optimal Route Solver Returning to FIG. 3 for reference, this section focuses on the optimal route solver S300. The optimal route solver S300 is shown in more detail in FIG. 12. FIG. 13 shows a flowchart for the optimal transport network route solver module S300.

[0062] The optimal route solver S300 interprets the updated tourist fusion heuristic 910 to generate probabilistic predictions for each location and connection between locations defined in the heuristic. The probabilistic predictions indicate how important each location / connection is. The optimal route solver also determines the proposed / optimal routes among all specified locations and connections, given a set of probabilistic predictions.

[0063] For this purpose, the tourist movement predictor 1200 receives an updated tourist fusion heuristic 910 (s1300). Here, geographical regions are analyzed to show relevant locations for transport and tourism, along with the relationships between them. The relationships of each element are also provided in the embeddings. The embeddings for each element are encoded. The tourist movement predictor 1200 decodes this encoded information (s1302) and generates predicted probabilities for each element using techniques such as non-autoregressive decoding (NAR) or autoregressive decoding (AR) (s1304).

number

[0064] The learned context variables are learned by either a NAR or AR model, depending on the approach (NAR is used for independent nodes or edges belonging to the solution, while AR is used conditionally to traverse the graph). The training process for the model is performed by using a loss function and comparing the predicted embeddings with the actual embeddings to calibrate the vertex context.

[0065] These probabilities are sent to the optimal route solver 1202, which communicates with the solver repository 1204 to check if there is already a trained machine learning model for that scenario (s1306). If so, the trained model is retrieved and run (s1316) to determine the proposed / optimal routes for the transport network. This is more advantageous than current solvers for combinatorial optimization problems, which require long execution times each time an optimal solution is obtained.

[0066] If no trained model exists for the scenario, the optimal route solver 1202 trains a model based on predicted probabilities; that is, the model is trained to predict the optimal routes based on probabilities by converting the probabilities into discrete decisions using search techniques. For efficiency, the training phase is decomposed (s1308) and executed in parallel (s1310). Once the training phase is complete, all trained models are merged into one (s1312) to produce a unique solver, which is stored in the solver repository 1204 for future use (s1314). Thus, redundant training that consumes time and resources is reduced. Finally, the solver is run (s1316) to obtain the optimal discrete solution 1206. Search techniques such as greedy search or beam search and sampling can be applied to obtain the solution.

[0067] The optimal discrete solution 1206, along with information related to the geographical area 1210 (e.g., polygons defining a map that may be a TAZ, LSOA, or MSOA) and the transport network 1212, is processed by the optimal BRT route module 1208 (s1318) to link the optimal discrete solution 1206 to the transport network infrastructure (e.g., roads, rails, etc.) that should be used by the transport network. The output of this processing is the proposed / optimal routes for the transport network 1214. The output may be in the form of a map with the roads and related routes marked.

[0068] Figure 14 shows exemplary inputs and outputs for the optimal transport network route solver S300.

[0069] Figure 16 is a flowchart of a computer implementation method 1600 for optimizing routes in a transport network for a geographical area. Method 1600 includes steps 1602, 1604, 1606, 1608, and optionally step 1610.

[0070] In step 1602, transport data related to a geographical area is acquired. The transport data represents the current transport needs of transport users within the geographical area. Preferably, the transport data may be in the format of a graph / heuristic definition (e.g., heuristic definition 416). The transport data may be filtered to remove transport data related to movements that occur less frequently than a predetermined threshold (e.g., relevance filtering 414).

[0071] In step 1604, tourist data relating to a geographical area is obtained. The tourist data indicates tourist hotspots within the geographical area. Preferably, the tourist data may be in the format of a graph / heuristic definition (e.g., tourist fusion map 628).

[0072] In step 1606, a graph (e.g., tourist fusion heuristic 632) is generated based on transport data and tourist data. The graph includes a set of nodes and a set of edges, where each node represents a location within a geographical area and each edge represents a route between two locations. The graph shows the current transport needs of transport users and tourist hotspots within the geographical area.

[0073] In step 1608, the relationship between tourist hotspots and current transportation needs is determined (for example, by the tourist impact assessment module S200). This determination includes at least the following steps: a) Divide the graph into multiple subgraphs (for example, using TFH decomposition 904); b) Evaluate each subgraph in parallel to determine the relationship between tourist hotspots and current transportation needs (for example, using TFH parallel evaluation 906); c) Combine the evaluated subgraphs (e.g., using TFH Merge 908) to generate an updated graph (e.g., updated tourist fusion heuristic 910) that represents the relationship between tourist hotspots and current transportation needs for geographical areas.

[0074] Referring to Figure 17 for more details, the decision step s1608 may include the following steps: In step 1702, the method checks whether there is a trained machine learning model associated with the geographical region.

[0075] Otherwise, in step 1704, the graph is divided into multiple subgraphs. In step 1706, each subgraph is then evaluated in parallel to determine the relationship between tourist hotspots and current transportation needs (Figure 17 shows three subgraphs evaluated in parallel, but this is purely illustrative; the graph may be divided into any number of subgraphs, each evaluated in parallel). In step 1708, the evaluated subgraphs are combined. The combination of evaluated subgraphs generates an updated graph representing the relationship between tourist hotspots and current transportation needs for geographical areas.

[0076] If a trained machine learning model associated with a geographical area exists, in step 1710, the method checks whether the graph has been previously evaluated by the trained machine learning model to determine the relationship between tourist hotspots and current transportation needs.

[0077] If so, in step 1712, the corresponding evaluated graph is retrieved and provided as an updated graph showing the relationship between tourist hotspots and current transportation needs for geographical areas.

[0078] Otherwise, in step 1714, the graph is divided into multiple subgraphs. Then, for each parallel subgraph (Figure 17 shows two subgraphs being evaluated in parallel at this point, but this is purely illustrative; the graph may be divided into any number of subgraphs, each being evaluated in parallel), in step 1716, the method checks whether the subgraph has been previously evaluated by a trained machine learning model to determine the relationship between tourist hotspots and current transportation needs. If the subgraph has been previously evaluated, in step 1718, the corresponding evaluated subgraph is retrieved. If the subgraph has not been previously evaluated, in step 1720, the subgraph is evaluated to determine the relationship between tourist hotspots and current transportation needs. In step 1722, the evaluated subgraphs are combined to generate an updated graph representing the relationship between tourist hotspots and current transportation needs for geographical areas.

[0079] Therefore, at each "end" in the flowchart, the output is the evaluated graph that forms the updated graph mentioned above.

[0080] In any of the above evaluations of a graph / subgraph, the evaluation may involve using a graph convolutional network (GCN) to weight the nodes of the subgraph based on neighboring nodes and / or connected nodes. The GCN may be trained by passing a graph specification (e.g., tourist fusion heuristic 632) to the GCN model during TFH parallel evaluation 906. Training takes place at the time of evaluation. In situations where no suitable trained model is available (e.g., s1706, s1720), a new GCN model may be trained on the details of that situation. Once the model is trained, it is run to weight the nodes of the subgraph (i.e., "evaluate" the subgraph). The GCN may use anisotropic aggregation.

[0081] The updated graph may be output to the user; for example, the updated graph may be displayed to the user via display 995.

[0082] Referring back to Figure 16, in step 1610, the updated graph is used to determine several proposed routes for the transport network, and optionally, the proposed routes can be output to the user (for example, via display 995). Step 1610 may include estimating the probability of each of several possible movements within a geographical area based on the updated graph, and determining the proposed routes based on those probabilities.

[0083] Step 1610 may include steps 1804–1812, as described immediately below in relation to Figure 18.

[0084] Figure 18 is a flowchart of a computer implementation method 1800 for optimizing routes in a transport network for a geographical area. Method 1800 includes steps 1802, 1804, 1806, 1808, 1810, 1812, and optionally step 1814.

[0085] In step 1802, an analysis of the geographical area is obtained (for example, the updated tourist fusion heuristic 910). The analysis includes a graph showing the relationship between the current transport needs of transport users and tourist hotspots within the geographical area. The graph includes a set of nodes and a set of edges, where each node represents a location within the geographical area and each edge represents a route between two locations. Step 1802 may include steps 1602-1608, described in relation to Figure 16, and / or steps 1702-1722, described in relation to Figure 17.

[0086] Depending on the format of the analysis, it may need to be decoded before step 1804. For example, if the analysis is in the form of heuristics, the heuristics may need to be decoded (for example, by reversing the process described in relation to Figure 7) in order to obtain meaningful information that underlies the probabilistic predictions. See, for example, step 1302 described in relation to Figure 13.

[0087] In step 1804, probabilistic predictions are generated for each node and edge of the graph (for example, by the tourist movement predictor 1200). The probabilistic predictions indicate the importance of each location and each route within the transport network.

[0088] In step 1806, the method checks whether at least a subset of the generated probabilistic predictions have been previously solved by the trained machine learning model. For example, this can be checked by checking the solver repository 1204 for the saved trained machine learning model.

[0089] If at least a first subset of the generated probabilistic predictions has been previously solved, in step 1808, the trained machine learning model previously used to solve at least a first subset of the generated probabilistic predictions is retrieved.

[0090] If at least a second subset of the generated probabilistic predictions has not been previously solved, in step 1812, a new machine learning model is trained to solve at least a second subset of the generated probabilistic predictions to determine a second set of proposed routes for the transport network. The training phase may be decomposed and run in parallel for efficiency. See the explanation of steps 1308-1314 related to Figure 13.

[0091] In step 1812, the trained machine learning model (either the extracted model or the newly trained model) is used to determine the proposed route for the transport network (e.g., the optimal discrete solution 1206).

[0092] In an optional step 1814, the proposed route may be output to the user, for example, displayed to the user via the display 995.

[0093] The proposed route (e.g., solution 1206) is processed together with information related to geographical areas (e.g., area information 1210) and transportation networks (e.g., transportation network information 1212) to generate an optimal transportation network solution (e.g., optimal BRT route 1214), which is output to the user, for example, via display 995.

[0094] The transport network from which the optimal route is generated may be any suitable transport network, for example, the transport network may include a Bus Rapid Transit (BRT) system, a rail transport system, a bus transport system, and / or a water transport system.

[0095] advantage The advantages of the embodiments of the invention described herein include the following: • Determine the optimal routes for sustainable tourism without relying on local expertise. • Provides a dynamic learning process for generating the optimal route. • Discover characteristics that are difficult for experts in the field to discover due to their scalability and complexity. This method does not rely on handcrafted heuristic methods designed by experts. This method achieves fast inference by using parallel execution to match transportation needs with the impact of tourists, and by significantly reducing the complexity and time required for the initial problem. The method used is explainable. This method provides a transformation from people's movement and tourist data into an intuitive representation of the relationships between different locations of interest. This method provides an automated conversion from people's movements to relevant transportation networks.

[0096] hardware Figure 15 is a block diagram of an information processing device 10 or computing device 10 embodying the present invention, such as a data storage server, which may be used to implement some or all of the operation of a method embodying the present invention and to perform some or all of the tasks of the apparatus of the embodiment. The computing device 10 may be used to carry out any of the method steps described above, for example, any of methods 1600, 1700, or 1800.

[0097] The computing device 10 comprises a processor 993 and memory 994. Advantageously, the processor 993 includes a graphics processing unit (GPU) processor. Optionally, the computing device also includes a network interface 997 for communicating with other such computing devices, for example, other computing devices in embodiments of the invention. Optionally, the computing device also includes one or more input mechanisms, such as a keyboard and a mouse 996, and one or more display units, such as a monitor 995. These elements can facilitate user interaction. These components are connectable to each other via a bus 992.

[0098] Memory 994 may include computer-readable media, and this term may refer to one or more media configured to carry computer-executable instructions (e.g., a centralized or distributed database and / or associated caches and servers). Computer-executable instructions may include instructions and data accessible by a general-purpose computer, a dedicated computer, or a dedicated processing device (e.g., one or more processors) for performing one or more functions or operations. For example, computer-executable instructions may include instructions for implementing the methods disclosed herein or any method steps disclosed herein, e.g., the methods illustrated in Figures 16-18 or any method steps. Thus, the term “computer-readable storage medium” may also include any medium capable of storing, encoding, or carrying a set of instructions for machine execution, and causing a machine to execute any one or more of the method steps of this disclosure. Thus, the term “computer-readable storage medium” may be interpreted to include, but not limited to, solid-state memory, optical media, and magnetic media. Such computer-readable media may include, but are not limited to, non-temporary computer-readable storage media, including random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), compact disk read-only memory (CD-ROM) or other optical disk storage, magnetic disk storage or other magnetic storage devices, and flash memory devices (e.g., solid-state memory devices).

[0099] The processor 993 is configured to control the computing device and perform processing operations, for example, by executing computer program code stored in memory 994 to perform any of the steps of the method described herein. Memory 994 stores data to be read and written by the processor 993 and may store the transport data, tourist data, and / or trained models described above, and / or programs for performing any of methods 1600, 1700, and 1800. As referred to herein, the processor may include one or more general-purpose processing devices such as a microprocessor, a central processing unit, etc. The processor may include a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, or a processor that implements a processor or combination of instruction sets. The processor may also include one or more dedicated processing devices such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a network processor, etc. In one or more embodiments, the processor is configured to execute the operations and instructions for performing the operations described herein. The processor 993 may be thought to include any of the modules described above. Any operation described as being performed by a module may be performed by the computer, for example, by the processor 993, as a method.

[0100] The memory 994 and processor 993 may be collectively configured to provide a problem definition module S000, a tourist characteristics calibration module S100, a tourism impact assessment module S200, and an optimal transport network route solver module S300.

[0101] The display unit 995 may display a representation of data stored by the computing device, such as updated graphs, proposed routes, and / or optimal transport network solutions, as described above.

[0102] The network interface (network I / F) 997 may be connected to a network such as the Internet and can connect to other such computing devices via the network. The network I / F 997 may control data input from and output to other devices via the network.

[0103] Other peripheral devices such as microphones, speakers, printers, power supply units, fans, cases, scanners, and trackballs may also be included in the computing device.

[0104] A method of embodying the present invention may be performed on a computing device / apparatus 10 as shown in Figure 15. Such a computing device does not need to have all the components shown in Figure 15, but may consist of a subset of those components. For example, apparatus 10 may comprise a processor 993 and memory 994 connected to the processor 993. Alternatively, apparatus 10 may comprise a processor 993, memory 994 connected to the processor 993, and a display 995. A method of embodying the present invention may be performed by a single computing device that communicates with one or more data storage servers over a network. The computing device may be data storage itself that stores at least a portion of the data.

[0105] The method for embodying the present invention may be carried out by a plurality of computing devices working together. One or more of these computing devices may be data storage servers that store at least a portion of the data.

[0106] The present invention may be implemented in digital electronic circuits, or in computer hardware, firmware, software, or a combination thereof. The present invention may be implemented as a computer program or computer program product, i.e., a computer program tangibly embodied in a non-temporary information carrier, such as a machine-readable storage device, or in a propagating signal, for execution by or control of the operation of one or more hardware modules.

[0107] A computer program may be in the form of a standalone program, a computer program portion, or multiple computer programs, and may be written in any form of programming language, including compiled or interpreted languages, and may be deployed in any form, such as a standalone program or as modules, components, subroutines, or other units suitable for use in a data processing environment. A computer program may be deployed to run on a single module, or on multiple modules distributed across a single site or multiple sites and interconnected by a communication network.

[0108] The method steps of the present invention may be performed by one or more programmable processors that execute a computer program to perform the functions of the present invention by acting on input data and generating an output. The apparatus of the present invention may be implemented as programmed hardware or as a dedicated logic circuit, including, for example, an FPGA (Field Programmable Gate Array) or an ASIC (Application-Specific Integrated Circuit).

[0109] Processors suitable for executing computer programs include, for example, both general-purpose and dedicated microprocessors, and any one or more processors in any type of digital computer. Generally, a processor receives instructions and data from read-only memory or random-access memory or both. An essential element of a computer is a processor for executing instructions, coupled with one or more memory devices for storing instructions and data.

[0110] The above-described embodiments of the present invention may be used, advantageously, independently of other embodiments or in any feasible combination with one or more other embodiments.

[0111] Various modifications, whether by adding, deleting, or replacing features, may be made to the embodiments described above to provide further embodiments, any and all of which are intended to be covered by the appended claims.

[0112] With regard to embodiments including the above examples, the following additional information is disclosed. (Note 1) A computer-implemented method for optimizing routes in a transport network for a geographical area, the method being: A step of obtaining an analysis of the said geographical area, wherein the analysis includes a graph showing the relationship between the current transportation needs of transportation users and tourist hotspots in the said geographical area, the graph including a set of nodes and a set of edges, where each node indicates a location in the said geographical area, and each edge indicates a route between two locations; A step of generating a probabilistic prediction for each node and edge of the graph, wherein the probabilistic prediction indicates the importance of each location and each route within the transport network; The steps include checking whether at least a subset of the generated probabilistic predictions has been previously solved by a trained machine learning model; If at least a first subset of the generated probabilistic predictions has been previously solved, the trained machine learning model previously used to solve the at least first subset of the generated probabilistic predictions is retrieved, and the trained machine learning model is run to determine a first set of proposed routes for the transport network; If at least a second subset of the generated probabilistic predictions has not been previously solved, the process includes the step of training a new machine learning model to solve the at least second subset of the generated probabilistic predictions to determine a second set of proposed routes for the transport network. method. (Note 2) The method according to Appendix 1, further comprising the step of outputting the first and / or second multiple suggested routes to the user. (Note 3) The method according to Appendix 1 or 2, further comprising processing the first and / or second proposed routes using information relating to the geographical area and the transportation network to generate an optimal transportation network solution, and optionally outputting the optimal transportation network solution to a user. (Note 4) To obtain the aforementioned analysis of the aforementioned geographical area: A step of obtaining transportation data related to the geographical area, wherein the transportation data indicates the current transportation needs of transportation users within the geographical area; A step of obtaining tourist data related to the geographical area, wherein the tourist data indicates tourist hotspots within the geographical area; A step of generating a preliminary graph based on the transportation data and the tourist data, wherein the graph includes a set of nodes and a set of edges, each node indicating a location within the geographical area, each edge indicating a route between two locations, and the graph shows the current transportation needs of transportation users and tourist hotspots within the geographical area; A step of determining the relationship between the aforementioned tourist hotspot and the aforementioned current transportation needs, wherein the determination involves: The aforementioned preliminary graph is divided into multiple subgraphs; To determine the relationship between the aforementioned tourist hotspots and the current transportation needs, each subgraph was evaluated in parallel; To generate the graph showing the relationship between the current transportation needs of transportation users and tourist hotspots in the aforementioned geographical area, the process includes combining evaluated subgraphs. Steps and The method described in any one of the appendices 1 to 3, including the method described in the appendices 1 to 3. (Note 5) The above decision is: To check whether there are any trained machine learning models associated with the aforementioned geographical region; If you have a trained machine learning model associated with the aforementioned geographical region: In order to determine the relationship between the tourist hotspot and the current transportation needs, check whether the preliminary graph has been previously evaluated by the trained machine learning model; If the preliminary graph has been evaluated previously, retrieve the corresponding evaluated graph and provide it as the graph; If the aforementioned preliminary graph has never been evaluated before: The aforementioned preliminary graph is divided into multiple subgraphs; For each subgraph, in parallel: In order to determine the relationship between the tourist hotspot and the current transportation needs, check whether the subgraph has been previously evaluated by the trained machine learning model; If the subgraph has been evaluated before, retrieve the corresponding evaluated subgraph; If the subgraph has not been previously evaluated, evaluate the subgraph to determine the relationship between the tourist hotspot and the current transportation needs; The evaluated subgraphs are combined to generate the aforementioned graph; If there is no machine learning model trained for the aforementioned geographical region: The aforementioned preliminary graph is divided into multiple subgraphs; To determine the relationship between the aforementioned tourist hotspots and the current transportation needs, each subgraph was evaluated in parallel; The evaluated subgraphs are combined to generate the aforementioned graph. The method described in Appendix 4, including the method described in Appendix 4. (Note 6) The method according to Appendix 4 or 5, wherein the evaluation includes weighting the nodes of the subgraph based on neighboring nodes and / or connected nodes using a graph convolutional network (GCN). (Note 7) The method described in Appendix 6, wherein the evaluation includes using anisotropic aggregation. (Note 8) The method according to any one of the appendices 4 to 7, further comprising the step of outputting the aforementioned graph to the user. (Note 9) The method according to any one of the appendices 4 to 8, wherein the transport data is filtered to remove transport data related to movements that occur at a frequency lower than a predetermined threshold. (Note 10) The transportation network is a method according to any one of the items in Appendix 1 to 9, including a bus rapid transit (BRT) system, a rail transport system, a bus transport system, and / or a water transport system. (Note 11) A computer program that, when executed on a computer, causes the computer to perform the method described in any one of the appendices 1 to 10. (Note 12) An information processing apparatus comprising a memory and a processor connected to the memory, wherein the processor is configured to perform the method described in any one of the appendices 1 to 10. [Explanation of symbols]

[0113] S000 Problem Definer S100 Tourist Characteristics Calibration S200 Tourist Impact Assessment S300 Optimal BRT Route Solver S001 Reading GPS and people's movement data S002 Generate movement from geolocation S003 Supplementing the definition of movement using demographic data S004 Includes geographical information, including filtered travel. S005 Define transportation needs for mobility Read the accommodation information for S101. S102 Reading tourist information S103 Read tourist spot information S104 Building Accommodation Heuristics S105 Constructing a tourist destination heuristic S106 Constructing a tourist spot heuristic S107 Constructing Tourist Characteristics Heuristics S108 Calibrate tourist characteristics heuristic S201 Calculate heuristics for assessing the impact of tourists. Decode S301 Tourist Influence Heuristic S302 Estimate the probability of movement S303 Found the optimal BRT route 400 GPS / people data 402 Area Information 404 Transportation Network 406 People Data Parser 408 OD Generator 410 OD 412 Problem Definition Engine 414 Relevance Filtering 416 Heuristic Definitions 600 The Importance of Early Tourists 604 Accommodation Data 606 Important tourist destination data 608 tourist spots 610 Area Information 612 Accommodation Spot Builders 614 Tourist Destination Builders 616 Tourist Attractions / Builders 618 Tourist Facilities Map 620 Tourist Spot Map 622 Tourist Map 624 Tourist Features Blended 626 Proximity, Proximity Weight 630 Tourist Characteristics Calibration 628 Accommodation Map 632 Tourist Integration Heuristic 900 Tourist Fusion Heuristic Checker 902 TFH Repository 904 TFH breakdown 906 TFH parallel evaluation 908 TFH merge 910 Updated tourist fusion heuristic Reading the Tourist Fusion Heuristic for s1000 s1002 Same area trained model s1004 Check location and features s1006 They're all the same Load the s1008 trained tourist fusion heuristic. s1010 Save different locations and features s1012 Consolidate 1 hop positions Deconstructing the s1014 Tourist Fusion Heuristic s1016 Parallel evaluation training s1018 Merge trained heuristics Save the trained Tourist fusion heuristic s1020 1100 Sequential 1102 parallel 1104 TFH parallel evaluation 1200 Tourist Movement Predictor 1202 Optimal Route Solver 1204 Solver Repository 1208 Optimal BRT Route 1214 Optimal BRT Route Read the updated tourist fusion heuristic for s1300. s1302 Decode the updated tourist fusion heuristic s1304 Generate a probabilistic prediction Is there a solver for s1306? s1308 Deconstructing solver training s1310 Parallel solver training s1312 Merge trained optimal route solvers s1314 Remember the optimal route solver. Run s1316 Optimal Route Solver s1318 Map road network to optimal route solution 1602 Obtain transportation data related to geographical area. 1604 Obtain tourist data related to geographical area 1606 Generate graphs based on transportation and tourist data. 1608 Determine the relationship between tourist hotspots and current transportation needs, and generate an updated graph representing the relationship between tourist hotspots and current transportation needs for a given geographical area. Based on the updated graphs from 1610, we determine several proposed routes for the transportation network. s1702 A pre-trained machine learning model associated with geographical regions? s1704 Split the graph into multiple subgraphs. s1706 Evaluate each subgraph. s1708 Combine subgraphs s1710 Graphs previously evaluated by a trained machine learning model? s1712 Retrieve the corresponding evaluated graph s1714 Split the graph into multiple subgraphs. Has the s1716 subgraph been evaluated previously? s1718 Retrieve the corresponding evaluated subgraph. Access subgraph s1720 s1722 Combine evaluated subgraphs s1802 Obtain analysis of geographical area s1804 Generate probabilistic predictions s1806 Has the generated probabilistic prediction been solved previously? s1808 Retrieve previously trained machine learning models used to solve stochastic predictions. s1810 Training a new machine learning model to solve stochastic predictions Run the trained machine learning model (s1812) to determine proposed routes for the transport network. s1814 Output the proposed route to the user.

Claims

1. A computer-implemented method for optimizing routes in a transport network for a geographical area, the method being: Steps include obtaining an analysis of the geographical area, wherein the analysis includes a graph showing the relationship between the current transportation needs of transportation users and tourist hotspots in the geographical area, the graph including a set of nodes and a set of edges, where each node indicates a location in the geographical area and each edge indicates a route between two locations; A step of generating a probabilistic prediction for each node and edge of the graph, wherein the probabilistic prediction indicates the importance of each location and each route within the transport network; The steps include: checking whether at least a subset of the generated probabilistic predictions has been previously solved by a trained machine learning model; If at least a first subset of the generated probabilistic predictions has been previously solved, the steps include: taking the previously trained machine learning model used to solve the at least first subset of the generated probabilistic predictions, running the trained machine learning model to determine a first set of proposed routes for the transport network; If at least a second subset of the generated probabilistic predictions has not been previously solved, the process includes the step of training a new machine learning model to solve the at least second subset of the generated probabilistic predictions to determine a second set of proposed routes for the transport network. method.

2. The method according to claim 1, further comprising the step of outputting the first and / or second multiple suggested routes to the user.

3. The method according to claim 1 or 2, further comprising processing the first and / or second proposed routes using information relating to the geographical area and the transportation network to generate an optimal transportation network solution, and optionally outputting the optimal transportation network solution to a user.

4. To obtain the aforementioned analysis of the aforementioned geographical area: A step of obtaining transportation data related to the geographical area, wherein the transportation data indicates the current transportation needs of transportation users within the geographical area; A step of obtaining tourist data related to the geographical area, wherein the tourist data indicates tourist hotspots within the geographical area; A step of generating a preliminary graph based on the transportation data and the tourist data, wherein the graph includes a set of nodes and a set of edges, each node indicating a location within the geographical area, each edge indicating a route between two locations, and the graph shows the current transportation needs of transportation users and tourist hotspots within the geographical area; A step of determining the relationship between the aforementioned tourist hotspot and the aforementioned current transportation needs, wherein the determination is: The aforementioned preliminary graph is divided into multiple subgraphs; To determine the relationship between the aforementioned tourist hotspots and the current transportation needs, each subgraph was evaluated in parallel; To generate the graph showing the relationship between the current transportation needs of transportation users and tourist hotspots in the aforementioned geographical area, the process includes combining evaluated subgraphs. Steps and The method according to claim 1, including the method described in claim 1.

5. The above decision is: To check whether there are any trained machine learning models associated with the aforementioned geographical region; If you have a trained machine learning model associated with the aforementioned geographical region: In order to determine the relationship between the tourist hotspot and the current transportation needs, check whether the preliminary graph has been previously evaluated by the trained machine learning model; If the preliminary graph has been evaluated previously, retrieve the corresponding evaluated graph and provide it as the graph; If the aforementioned preliminary graph has never been evaluated before: The aforementioned preliminary graph is divided into multiple subgraphs; For each subgraph, in parallel: In order to determine the relationship between the tourist hotspot and the current transportation needs, check whether the subgraph has been previously evaluated by the trained machine learning model; If the aforementioned subgraph has been evaluated before, retrieve the corresponding evaluated subgraph; If the subgraph has not been previously evaluated, evaluate the subgraph to determine the relationship between the tourist hotspot and the current transportation needs; The evaluated subgraphs are combined to generate the aforementioned graph; If there is no machine learning model trained for the aforementioned geographical region: The aforementioned preliminary graph is divided into multiple subgraphs; To determine the relationship between the aforementioned tourist hotspots and the current transportation needs, each subgraph was evaluated in parallel; The evaluated subgraphs are combined to generate the aforementioned graph. The method according to claim 4, including the method described in claim 4.

6. The method according to claim 4 or 5, wherein the evaluation includes weighting the nodes of the subgraph based on neighboring nodes and / or connected nodes using a graph convolutional network (GCN).

7. The method according to claim 6, wherein the evaluation includes using anisotropic aggregation.

8. The method according to claim 4, further comprising the step of outputting the graph to the user.

9. The method according to claim 4, wherein the transport data is filtered to remove transport data related to movements that occur at a frequency lower than a predetermined threshold.

10. The method according to claim 1, wherein the transport network includes a bus rapid transit (BRT) system, a rail transport system, a bus transport system, and / or a water transport system.

11. A computer program that, when executed on a computer, causes the computer to perform the method according to any one of claims 1 to 2, 4 to 5, or 7 to 10.

12. An information processing apparatus comprising a memory and a processor connected to the memory, wherein the processor is configured to perform the method described in any one of claims 1 to 2, 4 to 5, or 7 to 10.

Images