WO2023109720A1 - 城市公共交通规划方法 - Google Patents

城市公共交通规划方法 Download PDF

Info

Publication number
WO2023109720A1
WO2023109720A1 PCT/CN2022/138218 CN2022138218W WO2023109720A1 WO 2023109720 A1 WO2023109720 A1 WO 2023109720A1 CN 2022138218 W CN2022138218 W CN 2022138218W WO 2023109720 A1 WO2023109720 A1 WO 2023109720A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
urban
points
travel
public transportation
Prior art date
Application number
PCT/CN2022/138218
Other languages
English (en)
French (fr)
Inventor
叶可江
苏林煜
须成忠
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2023109720A1 publication Critical patent/WO2023109720A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/18—Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]

Definitions

  • the invention relates to a method for urban public traffic planning.
  • the existing urban public transportation planning methods mainly fall into the following categories.
  • One is to plan the urban public transportation system by using artificial intelligence based on experience and residents’ travel intentions obtained from questionnaire surveys.
  • the other is to plan the urban public transportation system using traditional statistical methods and some machine learning methods.
  • the public transportation system is planned, and the third is to use deep learning methods to plan the public transportation system.
  • the BiLSTM neural network model based on the attention mechanism, from the candidate public transport sites obtained above, calculate the distance from the current public transport site to the next different public transport site Probability; g. select several adjacent sites with higher probability as alternative sites, connect the lines from the starting point to the adjacent sites, and then connect the lines with public transport sites in all the lines obtained to obtain new public transport lines, Then, the new public transport line is compared with the existing public transport system line to check whether the expected goal is achieved.
  • the step a specifically includes:
  • Step S11 Calculate the city's carbon emissions:
  • Q the total amount of daily average resident travel, the unit is 10,000 person-times
  • Step S12 To minimize the above-mentioned urban carbon emissions under the constraints and meet the travel needs of urban residents, obtain the traffic demand constraints and land resource constraints respectively:
  • the traffic demand constraints are:
  • Q the total travel volume of motor vehicles in a single day, the unit is person-time
  • w the daily average total amount of non-motorized travel such as walking
  • Step S13 According to the objective function obtained in step S11 and the two constraint conditions obtained in step S12, use the objective programming method to obtain the optimization range that should be obtained for various transportation modes.
  • the step b specifically includes:
  • Step S21 Collect travel data of private cars and taxis
  • Step S22 Process the data into a vector format.
  • the step c specifically includes:
  • Step S31 clearing incomplete data collected in the data
  • Step S32 directly remove the data with default values
  • Step S33 Comparing the trajectory points in the travel data with the latitude and longitude of the boundary of the area to be planned, and removing the data whose start and end points are not within the urban planning range.
  • step d specifically includes:
  • Step S41 Mapping the GPS values in the collected data to the map
  • Step S42 Map the GPS points into a coarse-grained space.
  • each GPS point l i is located in a cell C j . All GPS points in a single unit are regarded as the same object, and the original GPS point expressed as embedded points
  • step e specifically includes:
  • Step S51 using the various GPS points obtained above, to obtain candidate public transport stations through a clustering algorithm
  • Step S52 Using the spherical distance to represent the similarity between two GPS points, the calculation formula is as follows:
  • R represents the average radius of the earth, which is about 6378137m
  • ⁇ lat represents the latitude difference between two GPS points
  • ⁇ lng represents the longitude difference between two GPS points
  • Dist approximately represents the real spherical distance between two GPS points
  • Step S53 Before grouping each point, some groups that are very far away from the current point are excluded, and the exclusion process does not need to calculate the distance to each group one by one;
  • Step S54 Select those points far away from the data-intensive area as reference points
  • Step S55 According to the selected reference point, calculate the distance between the current point and the reference point before each grouping, exclude groups with farther distances, merge the points of each group with all the points contained in all reachable groups, and then merge Use a clustering algorithm on the subsequent data domain;
  • Step S56 Calculate the center of the urban traffic hotspot area by the following formula, and use this as the specific location of the public transport candidate site;
  • n is the number of all data objects in a certain cluster
  • Dist(i, j) is the distance between data object i and data object j.
  • step f specifically includes:
  • the BiLSTM neural network model based on the attention mechanism is used to predict the probability of different sites to the next adjacent site.
  • the BiLSTM neural network model outputs the final probability through the fully connected layer and the Softmax classifier, and the formula involved in the fully connected layer is as follows:
  • W FC and b FC are parameter matrices that can be learned by the fully connected layer; followed by the final prediction layer, using Softmax as a multi-class logistic regression classifier to obtain the probability distribution of candidate destinations.
  • Softmax as a multi-class logistic regression classifier
  • the loss function is as follows:
  • the step g specifically includes:
  • Step S71 For each public transportation station, use it as a starting point, use the BiLSTM neural network model of step f to obtain the probability of going to different stations; then select several adjacent stations with higher probability as alternative stations, and connect the starting point routes to adjacent stations;
  • Step S72 After completing the calculation of all public transport stations, connect the lines with public transport stations in all the obtained lines, then optimize the existing public transport lines according to the optimization range obtained in step S13, and optimize Compare the new public transportation lines with the existing public transportation lines, and adjust the existing ground conventional bus lines and rail transit lines;
  • Step S73 Input the adjusted route into the digital twin city of the optimized city for modification, and check whether the expected goal is achieved.
  • this application can optimize the public transportation system, optimize and adjust the urban public transportation system through the deep learning model method, reduce carbon emissions, and promote carbon neutrality. Moreover, this application has higher computing efficiency and shorter training time. At the same time, because the map embedding method is used, it has better robustness. The attention mechanism used in the end enables the model to notice the correlation between different sites. Instead of simply focusing on where the site appears in the sequence.
  • Fig. 1 is the flow chart of urban public transportation planning method of the present invention
  • Fig. 2 is the overall flowchart of the GDBSCAN * algorithm that the embodiment of the present invention provides
  • Fig. 3 is the schematic diagram of BiLSTM neural network model provided by the embodiment of the present invention.
  • FIG. 4 is a schematic diagram of the use of the attention mechanism in the model provided by the embodiment of the present invention.
  • FIG. 5 is a schematic diagram of a BiLSTM neural network model based on an attention mechanism provided by an embodiment of the present invention.
  • FIG. 1 it is a work flow chart of a preferred embodiment of the urban public transportation planning method of the present invention.
  • Step S1 analyze the existing urban passenger transport system, and calculate the urban carbon emissions. Mainly from the three aspects of environmental protection, traffic development and residents' travel, with the minimization of traffic carbon emissions as the optimization goal, and under the constraints of meeting urban traffic demand and road resource allocation, a multi-objective hybrid optimization model is established. Get the optimization range that should be obtained by various transportation modes. in particular:
  • Step S11 Calculate the city's carbon emissions.
  • This embodiment adopts the relatively mature IPCC transportation carbon emission measurement method, according to the IPCC guidelines report, the formula is as follows:
  • Q the total amount of daily average resident travel, the unit is 10,000 person-times
  • c i carbon emission factor of transportation mode i, unit is kg/(person ⁇ km).
  • this embodiment calculates the average daily carbon dioxide emissions of passenger transport modes in major cities in Beijing. Taking 2018 as an example, the total carbon emissions of conventional public transport are 1374.0 tons, the total carbon emissions of rail transit are 850.6 tons, the total carbon emissions of taxis are 1355.1 tons, and the total carbon emissions of private cars are 14312.3 tons. Ton. Obviously, high private car ownership is an important reason for the high total carbon emissions of this travel mode.
  • Q the total travel volume of motor vehicles in a single day, the unit is person-time
  • l i is the average travel distance of transportation mode i, in km;
  • w the daily average total amount of non-motorized trips such as walking, in units of times
  • L the average distance of a single trip of urban residents, in km
  • Z i dynamic land area per capita of mode i, unit m2/person
  • Step S13 According to the objective function obtained in step S11 and the two constraint conditions obtained in step S12, use the objective programming method to obtain the optimization range that should be obtained for various transportation modes.
  • the goal of this embodiment is to make the proportion of conventional bus travel increase by about 10%, and that of rail transit increase by about 80%.
  • step S2 according to the optimization range obtained in step S1, the existing public transportation lines are added and modified to achieve the proportion of urban residents using public transportation.
  • the existing public transportation includes: ground conventional public transportation and rail transportation
  • This embodiment achieves this goal by conducting research on travel data in existing cities. Collect travel data of taxis and private cars in cities to be optimized. in particular:
  • Step S21 Collect travel data of private cars and taxis.
  • Step S22 Process the data into a vector format to facilitate subsequent processing.
  • Step S3 process the data obtained in step S2, so as to prevent these data from affecting the analysis results.
  • Step S31 Preprocessing the data.
  • Step S32 directly remove data with default values.
  • Step S33 Comparing the trajectory points in the travel data with the latitude and longitude of the boundary of the area to be planned, and removing the data whose start and end points are not within the urban planning range.
  • Step S4 for the processed data, map matching is performed on the original GPS track points in the itinerary.
  • this embodiment relocates the original taxi GPS track points to the traffic road network on the electronic map through map matching technology to obtain more robust GPS track points.
  • Step S41 Map the GPS values in the collected data to a map for subsequent processing.
  • Step S42 At this time, there is a problem in the original GPS point, that is, when the sampling rate is low or there is noise, there will be uncertainty, and the original longitude and latitude values are prone to deviation.
  • the solution of this embodiment is to map GPS points into a coarse-grained space. That is, the city is divided into g*g grid units. At this time, each GPS point l i is located in a cell C j . In order to make the GPS track point more robust, all GPS points in a single unit are regarded as the same object. In this way, the original GPS point can be represented as embedded points
  • Step S5 performing a clustering operation on the start and end points in the map-matched data to obtain candidate public transport stations.
  • Step S51 Using the various GPS points obtained above, a clustering algorithm is used to obtain candidate public transport stations.
  • Step S52 There are many formulas for calculating the distance between two points, such as Euclidean distance and Manhattan distance.
  • the spherical distance is used to represent the similarity between two GPS points, and the calculation formula is as follows:
  • R represents the average radius of the earth, about 6378137m
  • ⁇ lat represents the latitude difference between two GPS points
  • ⁇ lng represents the longitude difference between two GPS points
  • Dist approximately represents the true spherical distance between two GPS points.
  • Step S53 The traditional DBSCAN algorithm takes too long. Although the improved GDBSCAN algorithm based on the DBSCAN algorithm saves some time, it still takes a lot of time to calculate the distance in the grouping process. In order to further improve the clustering speed, this embodiment uses the characteristics of GPS data to improve the grouping conditions to obtain the GDBSCAN* algorithm.
  • the overall flowchart of the GDBSCAN* algorithm is shown in FIG. 2 .
  • the exclusion process does not need to calculate the distance to each group one by one, so a lot of time and resources can be saved. This step needs to be completed with reference points, and the selection method of reference points will be described in detail in step S54.
  • Step S54 Theoretically, the reference point can be random, but considering that the original intention of selecting the reference point is to minimize the distance calculation, it is still necessary to follow certain rules when selecting the reference point. It has been verified by experiments that when selecting points in data-intensive areas as reference points, the clustering speed is slow, so when selecting reference points, you should try to choose those positions far away from data-intensive areas, so that you can better improve Clustering speed.
  • Step S55 After selecting the reference point, calculate the distance between the current point and the reference point before each grouping, and use this distance to exclude groups that are farther away.
  • the distance between the master point and the reference point of each group is fixed, which can be stored and used multiple times when grouping other points. Using the above two distances to exclude groups that are far away before grouping can greatly reduce distance calculations.
  • the master point is the first point that joins the group when the group is created.
  • Step S56 In this embodiment, the center of the urban traffic hotspot area is calculated by the following formula, and used as the specific location of the candidate public transport station.
  • n is the number of all data objects in a certain cluster
  • Dist(i, j) is the distance between data object i and data object j, and the distance is calculated through step S52.
  • the data object within the cluster with the smallest sum of distances from other data objects is taken as the center point of the cluster, which represents the position of a candidate public transport station in the city. This means that the latitude and longitude coordinates of the center points of the clusters calculated by this formula are the latitude and longitude coordinates of the candidate public transport stations calculated in this embodiment.
  • Step S6 Using the BiLSTM neural network model based on the attention mechanism, calculate the probability from the current public transportation station to the next different public transportation station from the candidate public transportation stations obtained above. Specifically include:
  • Step S61 After adopting the embedding method above, the track space described by GPS has been represented by a series of stations.
  • the BiLSTM neural network model based on the attention mechanism is used to predict the probability from different sites to the next adjacent site.
  • Step S62 The BiLSTM neural network model is shown in Figure 3.
  • the trajectory data e i at time T i is passed into the BiLSTM neural network model.
  • the traditional LSTM structure can extract the information contained in it.
  • the visited locations are in a certain direction. have different discriminative abilities, such as at intersections, so it is believed that the contribution weight of GPS should consider such abilities, rather than simply looking at their locality in the trajectory.
  • the attention mechanism can capture the weight of hi at different positions in the sequence, while the one-way LSTM lacks the trajectory information of the subsequent sequence, for example, e1 ⁇ e2 ⁇ e3 lacks the information of the subsequent location, and the reverse also However, as e5 ⁇ e4 ⁇ e3, thus, BiLSTM and attention mechanism are mutually supportive in learning the latent sequence dependencies of trajectory data. Therefore, this embodiment chooses to use the BiLSTM neural network model based on the attention mechanism to simulate the previous and future environment of each GPS point.
  • -> represents the direction
  • ⁇ represents the sigmoid activation function
  • in , f n , on , g n and h n represent the input gate, forget gate, output gate, modulation gate and hidden state respectively
  • the parameters W i , W f , W o and W c represent the weight matrix of the above, respectively
  • ⁇ represents the product of elements. Since the backward process and the forward process are the same in principle, but the sequence order is reversed, the derivation process of the backward process only needs to change -> to ⁇ -.
  • the total output H of the BiLSTM unit for the input sequence X is expressed as follows:
  • h n is the BiLSTM output at step n, denoting the forward and reverse hidden states.
  • Attention mechanism has been widely used in sequence modeling and transduction models before, allowing to model the dependence of a position on other positions, rather than its position in the input sequence. Location.
  • the purpose of using the attention mechanism in the embodiment is to capture the correlation between the passing locations and the locations to determine the next stop.
  • Step S64 The use of the attention mechanism in this model is shown in Figure 4.
  • the original GPS data without map matching is also synchronously imported into the model, and features are extracted together with the embedded data, which are recorded as h and h′ respectively.
  • ⁇ and ⁇ represent the addition and multiplication of corresponding elements, respectively. More specifically, after the input sequence is input to BiLSTM, H i is sent to the perceptron, and m i is obtained after processing by the perceptron, and then m 1 , m 2 ,...,m N are processed by Softmax to derive each GPS The correlation degree between points, and then assign weights to each point regularization, and finally get the final expression of the embedded sequence X.
  • the derivation formulas involved are as follows:
  • W h and b h are the weights in the perceptron.
  • FIG. 5 shows the BiLSTM neural network model based on the attention mechanism.
  • the final probability is output through the fully connected layer and the Softmax classifier.
  • the formula involved in the fully connected layer is as follows:
  • W FC and b FC are parameter matrices that can be learned by the fully connected layer.
  • Softmax is used as a multi-class logistic regression classifier to obtain the probability distribution of candidate destinations. For an input partial transition T P , the probability that the j-th candidate destination d j is the true destination (y) of T P Obtained by performing a Softmax classifier on the raw output. The final prediction is the candidate destination with the highest probability in the following equation.
  • This embodiment uses cross-entropy as the loss function, because this is usually used to calculate the distance between the predicted probability distribution and the true probability distribution in the Softmax classifier.
  • the loss function is as follows:
  • Step S7 For each site, use it as the starting point to obtain the probability of going to different sites. Select several adjacent sites with higher probability as alternative sites, and connect the lines from the starting point to the adjacent sites. Then connect the lines that have public transport stations in all the lines obtained to obtain new public transport lines, then compare the new public transport lines with the existing public transport system lines, consider the existing public transport Optimize and change the system, such as establishing bus stations, establishing subway stations, etc. After modification, test in the digital twin city of the optimized city to see if the expected goal is achieved. in particular:
  • Step S71 For each public transportation station, use it as a starting point, and use the BiLSTM neural network model in step S6 to obtain the probability of going to different stations. Then select several adjacent sites with higher probability as candidate sites, and connect the starting point to the line of adjacent sites.
  • Step S72 After completing the calculation of all public transport stations, link all the obtained lines with public transport stations, so as to obtain some new lines with large demand. Then optimize the existing public transport lines according to the optimization index obtained in step S13, compare the optimized public transport lines with the existing public transport lines, and adjust the existing ground conventional bus lines and rail transit lines.
  • Step S73 Input the adjusted route into the digital twin city of the optimized city for modification, and check whether the expected goal is achieved.
  • This application takes the urban passenger transport system as the object, optimizes the travel structure through the multi-objective optimization model to obtain the optimization target, and then uses the deep learning method to complete the planning and adjustment of the urban public transport system to achieve the goal of reducing carbon emissions and helping carbon neutrality .

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Tourism & Hospitality (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Computational Linguistics (AREA)
  • Strategic Management (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Marketing (AREA)
  • Medical Informatics (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Traffic Control Systems (AREA)

Abstract

本发明涉及一种城市公共交通规划方法，包括：测算城市碳排量，计算得到各种交通出行方式所应该得到的优化幅度；收集待优化城市的出租车、私家车出行数据；对收集的出行数据进行处理；对行程中原始GPS轨迹点进行地图匹配；对地图匹配后数据中的起终点进行聚类操作，得到候选公共交通站点；使用基于注意力机制的BiLSTM神经网络模型，从上述得到的候选公共交通站点中，计算当前公共交通站点到下一个不同公共交通站点的概率；将得到的新的公共交通线路与现有公共交通体系线路进行对比，查看是否达到预期目标。本申请能够优化公共交通系统，通过深度学习的模型方法优化调整城市公共交通系统，降低碳排放，促进碳中和。

Description

城市公共交通规划方法 技术领域
本发明涉及一种城市公共交通规划方法。
背景技术
减少碳排放对于国家来说非常重要，中国提出碳中和是2060年，而西方是2050年，如果我们在2060年达到碳中和，而西方社会比我们提前十年左右，那就意味着有可能出现一种局面——西方在国际贸易中要计算碳足迹，如果中国的产品往回追溯产品碳足迹，不符合国际社会要求就不能出口，这对于我们国家的经济安全、外贸安全是非常严峻的考验。因此加速达到碳达峰，实现碳中和，已经是迫在眉睫。
而交通运输行业是主要“碳源”之一。人类频繁的交通活动是造成的城市二氧化碳过量排放的一个重要因素。目前我国城市正处于快速扩张时期，交通结构呈现私人交通占据较大比重并持续增长的特点，交通管理者面临持续增长的交通二氧化碳排放量、日渐恶化的拥堵路况、空气污染以及能源短缺等诸多问题。但受制于交通需求的个性化和多样性，仅从环保视角简单的调整现行交通结构可能影响到居民的日常出行及城市经济活动。有关数据显示，我国交通运输业碳排放占我国碳排放总量的10.4％。从我国交通运输领域目前碳排放结构来看，公路是主体，占85％以上；铁路占0.68％；海运和航空大约为6％左右。而现阶段，我国许多城市的公共交通系统的发展状况不是很理想，由于公共交通线路布设不合理、服务水平不优等原因，公交出行在居民出行选择中的优先度不高，全国多数城市公交机动化出行分担率不足40％，这导致更多 的居民倾向于出租车/私家车出行，提高了碳排放。
现有的城市公共交通规划方法主要有一下几类，一是利用人工，基于经验及问卷调查得到的居民出行意愿来对城市公共交通系统进行规划，二是传统统计方法和一些机器学习方法对城市公共交通系统进行规划，三是利用深度学习方法来对公共交通系统进行规划。
现有的方法中比如人工统计规划，依赖于问卷调查来的数据，耗时费力且效率不高，规划的结果并不合理；传统统计方法对公共交通系统进行规划和利用机器学习对居民起终点出行数据及既有地面公共交通系统线网的数据分析，缺乏对城市公共交通系统发展的考虑，无法将城市中最大交通流量的出租车出行客流考虑在内，浪费了包含大量信息的出租车的轨迹数据；同时由于现有技术深度学习方法的一些自身局限性，例如粒子群优化算法和蚁群算法，容易出现收敛速度较慢，计算量较大，且容易陷入局部最优解等问题。
发明内容
有鉴于此，有必要提供一种城市公共交通规划方法。
本发明提供一种城市公共交通规划方法，该方法包括如下步骤：a.对现有城市客运系统进行分析，测算城市碳排量，从环境保护、交通发展和居民出行三个方面，以交通碳排量最小化为优化目标，且在考虑满足城市交通需求、道路资源配置两个方面的约束下，计算得到各种交通出行方式所应该得到的优化幅度；b.根据得到的得到的优化幅度，收集待优化城市的出租车、私家车出行数据；c.对收集的出行数据进行处理；d.针对处理后的数据，对行程中原始GPS轨迹点进行地图匹配；e.对地图匹配后数据中的起终点进行聚类操作，得到候选公共交通站点；f.使用基于注意力机制的BiLSTM神经网络模型，从上述得到的候选公共交通站点中，计算当前公共交通站点到下一个不同公共交通站点的概率； g.选取较高概率的几个临近站点作为备选站点，连接起点到临近站点的线路，接着对得到的所有线路中存有公共交通站点的线路进行连结，得到新的公共交通线路，然后将所述新的公共交通线路与现有公共交通体系线路进行对比，查看是否达到预期目标。
具体地，所述的步骤a具体包括：
步骤S11：对城市碳排量进行测算：
Figure PCTCN2022138218-appb-000001
其中：Q——日平均居民出行总量，单位为万人次；
l i——交通方式i的平均通行距离，单位为km；
x i——交通方式i年内的方式分担率；
c i——交通方式i的碳排放因子；
步骤S12：使得上述城市碳排放量在约束条件下最小化，且满足城市居民出行需求的情况下，分别得到交通需求约束及土地资源约束条件：
所述交通需求约束条件为：
Figure PCTCN2022138218-appb-000002
其中：Q——单日机动方式总出行量，单位为人次；
x i出行总量中交通方式i的比例，
l i交通方式i的平均出行距离，
w——步行等非机动方式日均出行总量，
P——城市常住人口，
L——城市居民单次出行平均距离，
U——城市居民单日平均出行次数，
所述土地资源约束条件为：
Figure PCTCN2022138218-appb-000003
式中x i——出行总量中交通方式i得比例，
Z i——方式i的人均动态占地面积，
Z——城市人均占用道路面积；
步骤S13：根据步骤S11得到的目标函数与步骤S12中得到的两个约束条件，使用目标规划法求解得到各种交通出行方式所应该得到的优化幅度。
具体地，所述的步骤b具体包括：
步骤S21：收集私家车、出租车出行数据；
步骤S22：将数据处理为向量格式。
具体地，所述的步骤c具体包括：
步骤S31：清除数据中采集不完整的数据；
步骤S32：将存在缺省值的数据直接去除；
步骤S33：根据出行数据中的轨迹点与待规划区域边界经纬度进行比较，去除起终点不在城市规划范围内的数据。
具体地，所述的步骤d具体包括：
步骤S41：将收集到的数据中的GPS数值映射到地图上；
步骤S42：将GPS点映射到粗粒度的空间中。
具体地，所述的步骤S42具体包括：
将城市划分为g*g个网格单元，此时每个GPS点l i位于一个单元C j中，将单个单元内的所有GPS点视为同一对象，将原来的GPS点
Figure PCTCN2022138218-appb-000004
Figure PCTCN2022138218-appb-000005
表示为嵌入后的点
Figure PCTCN2022138218-appb-000006
具体地，所述的步骤e具体包括：
步骤S51：利用上述得到的各类GPS点，通过聚类算法得到候选公共交通站点；
步骤S52：采用球面距离表示两个GPS点间的相似性，计算公式如下：
Dist＝Δσ*R
Figure PCTCN2022138218-appb-000007
其中，R代表地球的平均半径，约为6378137m，Δlat表示两GPS点的纬度差，Δlng表示两GPS点间的经度差，Dist近似表示两个GPS点间的真实球面距离；
步骤S53：对各个点进行分组前先将距离当前点非常远的一些组排除掉，并且排除过程不需要逐个计算和各个组的距离；
步骤S54：选择那些远离数据密集区域的位置的点作为参照点；
步骤S55：根据选择的参照点，每次分组前先计算当前点和参照点的距离，排除距离较远的组，将各组的点与其所有可达组内包含的所有点进行合并，在合并后的数据域上使用聚类算法；
步骤S56：通过如下公式计算城市交通热点区域的中心，并以此作为公共交通候选站点的具体位置；
Figure PCTCN2022138218-appb-000008
其中，n为某一类簇中所有数据对象的数量；Dist(i,j)为数据对象i 和数据对象j的距离。
具体地，所述的步骤f具体包括：
采用基于注意力机制的BiLSTM神经网络模型预测不同站点到邻近的下一个站点的概率。
具体地，所述的BiLSTM神经网络模型通过全连接层和Softmax分类器来输出最终的概率，全连接层涉及到的公式如下：
Figure PCTCN2022138218-appb-000009
Figure PCTCN2022138218-appb-000010
其中，W FC和b FC都是全连接层可学习的参数矩阵；接着是最后的预测层，采用Softmax作为多类logistic回归分类器，得到候选目的地的概率分布，对于输入部分转移T P，第j个候选目的地d j作为T P的真实目的地的概率
Figure PCTCN2022138218-appb-000011
通过对原始输出执行Softmax分类器来获得，最终预测结果是下列等式中概率最高的候选目的地：
Figure PCTCN2022138218-appb-000012
Figure PCTCN2022138218-appb-000013
使用交叉熵作为损失函数，损失函数如下：
Figure PCTCN2022138218-appb-000014
具体地，所述的步骤g具体包括：
步骤S71：对于每一个公共交通站点，将其作为起点，利用步骤f的BiLSTM神经网络模型，得到其到不同站点的概率；接着从中选取较高概率的几个临近站点作为备选站点，连接起点到临近站点的线路；
步骤S72：完成对所有的公共交通站点的计算后，对得到的所有线路中存有公共交通站点的线路进行连结，接着根据步骤S13中得到的优化幅度对现有公共交通线路进行优化，将优化后的公共交通线路与现有公共交通线路进行对比，调整现有地面常规公交线路和轨道交通线路；
步骤S73：将调整的路线输入优化城市的数字孪生城市中进行修改，查看是否达到预期目标。
与现有技术相比，本申请能够优化公共交通系统，通过深度学习的模型方法优化调整城市公共交通系统，降低碳排放，促进碳中和。并且本申请运算效率更高，训练时间更短，同时因为使用了地图嵌入方法，使得具有更好的鲁棒性，最后使用到的注意力机制使得模型可以注意到不同站点之间的关联性，而并非单纯的关注该站点在序列中出现的位置。
附图说明
图1为本发明城市公共交通规划方法的流程图；
图2为本发明实施例提供的GDBSCAN*算法的总体流程图；
图3为本发明实施例提供的BiLSTM神经网络模型示意图；
图4为本发明实施例提供的注意力机制在模型中的使用情况示意图；
图5为本发明实施例提供的基于注意力机制的BiLSTM神经网络模型的示意图。
具体实施方式
下面结合附图及具体实施例对本发明作进一步详细的说明。
参阅图1所示，是本发明城市公共交通规划方法较佳实施例的作业流程图。
步骤S1，对现有城市客运系统进行分析，测算城市碳排量。主要从环境保护、交通发展和居民出行三个方面，以交通碳排量最小化为优化目标，且在考虑满足城市交通需求、道路资源配置两个方面的约束下，建立多目标混合优化模型，得到各种交通出行方式所应该得到的优化幅度。具体而言：
步骤S11：对城市碳排量进行测算。本实施例采用较为成熟的IPCC交通碳排放测量方法，根据IPCC指南报告，公式如下：
Figure PCTCN2022138218-appb-000015
其中：Q——日平均居民出行总量，单位为万人次；
l i——交通方式i的平均通行距离，单位为km；
x i——交通方式i年内的方式分担率；
c i——交通方式i的碳排放因子，单位为kg/(人·km)。
本实施例以北京市为例，根据《北京市交通发展年鉴》中的数据，得到主要几种出行方式的人均二氧化碳排放因子，如下表所示：
Figure PCTCN2022138218-appb-000016
基于上述公式，本实施例对北京市主要城市客运交通方式的日均二氧化碳排放量进行核算。以2018年为例，得到常规公交的碳排放总量为1374.0吨，轨道交通的碳排放总量为850.6吨，出租车的碳排放总量为1355.1吨，私人小汽车的碳排放总量为14312.3吨。显然高私人小汽车保有量是导致该出行方式碳排放总量较高的的重要原因。
步骤S12：使得上述碳排放量在约束条件下最小化，首先无论如何优化，都应当满足城市居民出行需求，分别得到交通需求约束及土地资源约束条件。具体包括：
得到交通需求约束：
Figure PCTCN2022138218-appb-000017
其中：Q——单日机动方式总出行量，单位为人次；
x i出行总量中交通方式i的比例，单位％；
l i交通方式i的平均出行距离，单位km；
w——步行等非机动方式日均出行总量，单位为次；
P——城市常住人口，单位人；
L——城市居民单次出行平均距离，单位km；
U——城市居民单日平均出行次数，单位为次
除了要满足城市居民出行需求以外，同时随着城市化进程的推进，交通资源逐渐成为限制城市交通建设的一大重要因素，突出的土地资源不足的问题成为制约城市交通发展的关键因素。土地资源约束条件要求所有交通方式的平均人均动态占地面积不得多于城市的人均道路面积，具体为：
Figure PCTCN2022138218-appb-000018
式中x i——出行总量中交通方式i得比例，单位％；
Z i——方式i的人均动态占地面积，单位㎡/人；
Z——城市人均占用道路面积，单位㎡/人
步骤S13：根据步骤S11得到的目标函数与步骤S12中得到的两个约束条件，使用目标规划法求解得到各种交通出行方式所应该得到的优化幅度。
本实施例的目标为使得选择常规公交出行占比增幅10％左右，而轨道交通增幅80％左右。
步骤S2，根据步骤S1得到的得到的优化幅度，增设修改现有公共交通线路，以达到城市居民使用公共交通占比。所述现有公共交通包括：地面常规公交和轨道交通
本实施例通过对现有城市内的出行数据进行研究以期达到这一目标。收集待优化城市的出租车、私家车出行数据。具体而言：
步骤S21：收集私家车、出租车出行数据。
步骤S22：将数据处理为向量格式方便后续处理。
步骤S3，对步骤S2中得到的数据进行处理，以免这些数据对分析结果产生影响。
首先清除数据中采集不完整的数据，然后去除一些异常点数据，例如剔除掉起终点经纬度超出边界的数据。
具体包括：
步骤S31：对数据进行预处理。
步骤S32：将存在缺省值的数据直接去除。
步骤S33：根据出行数据中的轨迹点与待规划区域边界经纬度进行比较，去除起终点不在城市规划范围内的数据。
步骤S4，针对处理后的数据，对行程中原始GPS轨迹点进行地图匹配。
为了增强数据的鲁棒性，本实施例通过地图匹配技术将原始出租车GPS轨迹点重新定位到电子地图上的交通路网中，得到更具有鲁棒性的GPS轨迹点。
步骤S41：将收集到的数据中的GPS数值映射到地图上，以便进行后续处理。
步骤S42：此时原始GPS点存在一个问题，即当由于采样率低或者有噪声时会产生不确定性，原始经纬度值容易产生偏差。
本实施例的解决方法是将GPS点映射到粗粒度的空间中。即将城市划分为g*g个网格单元，此时每个GPS点l i位于一个单元C j中,为了使GPS轨迹点更具鲁棒性，则将单个单元内的所有GPS点视为同一对象。通过这种方式，原来的GPS点
Figure PCTCN2022138218-appb-000019
可以被表示为嵌入后的点
Figure PCTCN2022138218-appb-000020
步骤S5，对地图匹配后数据中的起终点进行聚类操作，得到候选公共交通站点。
使用设计的改进后的DBSCAN聚类算法对城市热点区域进行聚类处理，将其生成结果作为候选公共交通站点(包括地面公交车站和地铁站)。具体包括：
步骤S51：利用上述得到的各类GPS点，通过聚类算法得到候选公共交通站点。
步骤S52：两点间距离计算公式有很多，如欧式距离、曼哈顿距离等。本实施例采用球面距离表示两个GPS点间的相似性，计算公式如下：
Dist＝Δσ*R
Figure PCTCN2022138218-appb-000021
其中，R代表地球的平均半径，约为6378137m，Δlat表示两GPS点的纬度差，Δlng表示两GPS点间的经度差，Dist近似表示两个GPS点间的真实球面距离。
步骤S53：传统的DBSCAN算法耗时太长。而基于DBSCAN算法改进的GDBSCAN算法虽然节省了部分时间，但在分组过程中依然需要花费大量时间进行距离计算。为进一步提高聚类速度，本实施例利用GPS数据的特点对分组条件进行改进得到GDBSCAN*算法。所述GDBSCAN*算法的总体流程图如图2所示。在GDBSCAN*算法中，对各个点进行分组前先将距离当前点非常远的一些组排除掉，并且排除过程不需要逐个计算和各个组的距离，那么就可以节省大量的时间和资源。这一步骤需要借助参照点来完成，参照点的选择方法将在步骤S54中详细讲述。
步骤S54：理论上，参照点可以是随机的，但考虑到选择参照点的初衷是要尽量减少距离计算，因此在选择时还是需要遵循一定的规律。经过实验验证，当选择数据密集区域位置中的点作为参照点时，聚类速度较慢，故在进行参照点选择时，应尽量选择那些远离数据密集区域的位置，这样就能更好的提升聚类速度。
步骤S55：选定参照点后，每次分组前先计算当前点和参照点的距离，利用这个距离将较远的组排除。而每组的master点与参照点的距离是固定的，可以存储起来在后续为其它点进行分组时多次使用，在分组 前利用上述两个距离排除较远的组可以大量减少距离计算。其中，master点是创建组时加入该组的第一个点。
分组后，所有数据点都被分到了不同的以eps为半径(eps值通过DBSCAN算法得到)的圆形区域中，而每个组的可达组也是知道的。之后将各组的点与其所有可达组内包含的所有点进行合并，然后在合并后的数据域上使用聚类算法。
步骤S56：本实施例通过如下公式计算城市交通热点区域的中心，并以此作为公共交通候选站点的具体位置。
Figure PCTCN2022138218-appb-000022
在此公式中，n为某一类簇中所有数据对象的数量；Dist(i,j)为数据对象i和数据对象j的距离，所述距离通过步骤S52计算得到。对于任何一个类簇，本实施例把类簇内与其他数据对象距离之和最少的数据对象作为该类簇的中心点，即代表城市候选公共交通站点的位置。这也就意味着，通过此公式计算得到的类簇中心点的经纬度坐标就是本实施例所计算的候选公共交通站点的经纬度坐标。
步骤S6：使用基于注意力机制的BiLSTM神经网络模型，从上述得到的候选公共交通站点中，计算当前公共交通站点到下一个不同公共交通站点的概率。具体包括：
步骤S61：在采用上述嵌入的方法后，GPS描述的轨迹空间已由一系列的站点代为表示。本实施例采用基于注意力机制的BiLSTM神经网络模型预测不同站点到邻近的下一个站点的概率。
步骤S62：BiLSTM神经网络模型如图3所示，将T i时刻的轨迹数据e i传入到BiLSTM神经网络模型当中，传统的LSTM结构可以提取其中蕴含的信息，然而已访问的地点在确定方向时具有不同的辨别能力，例如在十字路口处时，因此认为GPS的贡献权重应当考虑这样的能力，而不是单纯的看他们在轨迹中的局部性。而注意到注意力机制可以捕捉不同位置上的hi在序列中的权重情况，而单向的LSTM缺乏后序的轨迹信息，例如e1→e2→e3缺乏之后的地点的信息，而反向的亦然，如e5→e4→e3，因此，在学习轨迹数据的潜在序列依赖性时，BiLSTM和注意力机制是相互支持的。故本实施例选择使用基于注意力机制的BiLSTM神经网络模型来模拟每个GPS点之前和未来的环境。
步骤S63：不同于标准LSTM，BiLSTM通过同时执行向前和向后过程来集成先前和未来的顺序其中特征。其中，单向的LSTM每个输入序列表示为x＝{e 1,e 2,…,e N}(e∈R 3×V)，其中N代表输入序列长度大小，3代表三种日子的类型，V为输入向量的维度。进一步的，前向推导公式如下所示：
Figure PCTCN2022138218-appb-000023
Figure PCTCN2022138218-appb-000024
Figure PCTCN2022138218-appb-000025
Figure PCTCN2022138218-appb-000026
Figure PCTCN2022138218-appb-000027
Figure PCTCN2022138218-appb-000028
其中，->代表方向，σ代表sigmoid激活函数，i n、f n、o n、g n和h n分别代表输入门、遗忘门、输出门、调制门和隐藏状态，而参数W i、W f、W o和W c分别表示上述们的权重矩阵，⊙表示元素的乘积。由于后向过程和前向过程在原理上是相同的，但序列顺序上是相反的，因此后向过程的推导过程只需要将->改变为<-即可。最后，输入序列X的BiLSTM单元的总输出H表示如下：
Figure PCTCN2022138218-appb-000029
其中，h n是在n步的BiLSTM输出，表示正向和反向隐藏状态。
注意力机制在BiLSTM模型中的作用：此前注意力机制已广泛运用在序列建模和转导模型当中，允许对某个位置与其它位置的依赖性进行建模，而不是其在输入序列中的位置。实施例使用注意力机制目的是为了捕获已经经过的地点对确定下一站的地点之间的关联性。
步骤S64：注意力机制在本模型中的使用情况如图4所示。为了使实验模型更具有鲁棒性，本实施例将未经过地图匹配的原始GPS数据也同步传入模型，同经过嵌入过后的数据一起提取特征，分别记为h和h′。其中⊕和⊙分别代表对应元素相加和相乘。更具体地说，将输入序列输入到BiLSTM之后，将H i发送到感知机，经过感知机处理后得到m i，然后对m 1,m 2,…,m N经过Softmax处理，推导出各个GPS点之间的关联度，然后对各个点正则化赋予权值，最终得到嵌入序列X地最终表达式。而涉及到的各项推导公式如下：
m i＝tanh(W hh i+b h)+tanh(W hh′ i+b h)
Figure PCTCN2022138218-appb-000030
Figure PCTCN2022138218-appb-000031
其中，W h和b h是感知机中的权重。
步骤S65：图5示出了基于注意力机制的BiLSTM神经网络模型，本实施例通过全连接层和Softmax分类器来输出最终的概率。全连接层涉及到的公式如下：
Figure PCTCN2022138218-appb-000032
Figure PCTCN2022138218-appb-000033
其中，W FC和b FC都是全连接层可学习的参数矩阵。接着是最后的预测层，本实施例采用Softmax作为多类logistic回归分类器，得到候选目的地的概率分布。对于输入部分转移T P，第j个候选目的地d j作为T P的真实目的地(y)的概率
Figure PCTCN2022138218-appb-000034
通过对原始输出执行Softmax分类器来获得。最终预测结果是下列等式中概率最高的候选目的地。
Figure PCTCN2022138218-appb-000035
Figure PCTCN2022138218-appb-000036
本实施例使用交叉熵作为损失函数，因为这通常用于计算Softmax分类器中预测的概率分布与真实概率分布之间的距离。损失函数如下：
Figure PCTCN2022138218-appb-000037
步骤S7：对于每一个站点，将其作为起点，得到其到不同站点的概 率。选取较高概率的几个临近站点作为备选站点，连接起点到临近站点的线路。接着对得到的所有线路中存有公共交通站点的线路进行连结，得到新的公共交通线路，然后将所述新的公共交通线路与现有公共交通体系线路进行对比，考虑对现有的公共交通体系进行优化改变，例如建立公交车站点、建立地铁站点等。修改后在优化城市的数字孪生城市进行检测，查看是否达到预期目标。具体而言：
步骤S71：对于每一个公共交通站点，将其作为起点，利用步骤S6的BiLSTM神经网络模型，得到其到不同站点的概率。接着从中选取较高概率的几个临近站点作为备选站点，连接起点到临近站点的线路。
步骤S72：完成对所有的公共交通站点的计算后，对得到的所有线路中存有公共交通站点的线路进行连结，这样可以得到一些新的需求量较大的线路。接着根据步骤S13中得到的优化指标对现有公共交通线路进行优化，将优化后的公共交通线路与现有公共交通线路进行对比，调整现有地面常规公交线路和轨道交通线路。
步骤S73：将调整的路线输入优化城市的数字孪生城市中进行修改，查看是否达到预期目标。
本申请以城市客运交通系统为对象，通过多目标优化模型进行出行结构优化得到优化目标，然后使用深度学习方法完成对城市公共交通系统进行规划调整，以达到降低碳排放，助力碳中和的目标。
虽然本发明参照当前的较佳实施方式进行了描述，但本领域的技术人员应能理解，上述较佳实施方式仅用来说明本发明，并非用来限定本发明的保护范围，任何在本发明的精神和原则范围之内，所做的任何修饰、等效替换、改进等，均应包含在本发明的权利保护范围之内。

Claims (10)

  1. 一种城市公共交通规划方法，其特征在于，该方法包括如下步骤：
    a.对现有城市客运系统进行分析，测算城市碳排量，从环境保护、交通发展和居民出行三个方面，以交通碳排量最小化为优化目标，且在考虑满足城市交通需求、道路资源配置两个方面的约束下，计算得到各种交通出行方式所应该得到的优化幅度；
    b.根据得到的得到的优化幅度，收集待优化城市的出租车、私家车出行数据；
    c.对收集的出行数据进行处理；
    d.针对处理后的数据，对行程中原始GPS轨迹点进行地图匹配；
    e.对地图匹配后数据中的起终点进行聚类操作，得到候选公共交通站点；
    f.使用基于注意力机制的BiLSTM神经网络模型，从上述得到的候选公共交通站点中，计算当前公共交通站点到下一个不同公共交通站点的概率；
    g.选取较高概率的几个临近站点作为备选站点，连接起点到临近站点的线路，接着对得到的所有线路中存有公共交通站点的线路进行连结，得到新的公共交通线路，然后将所述新的公共交通线路与现有公共交通体系线路进行对比，查看是否达到预期目标。
  2. 如权利要求1所述的方法，其特征在于，所述的步骤a具体包括：
    步骤S11：对城市碳排量进行测算：
    Figure PCTCN2022138218-appb-100001
    其中：Q——日平均居民出行总量，单位为万人次；
    l i——交通方式i的平均通行距离，单位为km；
    x i——交通方式i年内的方式分担率；
    c i——交通方式i的碳排放因子；
    步骤S12：使得上述城市碳排放量在约束条件下最小化，且满足城市居民出行需求的情况下，分别得到交通需求约束及土地资源约束条件：
    所述交通需求约束条件为：
    Figure PCTCN2022138218-appb-100002
    其中：Q——单日机动方式总出行量，单位为人次；
    x i出行总量中交通方式i的比例，
    l i交通方式i的平均出行距离，
    w——步行等非机动方式日均出行总量，
    P——城市常住人口，
    L——城市居民单次出行平均距离，
    U——城市居民单日平均出行次数，
    所述土地资源约束条件为：
    Figure PCTCN2022138218-appb-100003
    式中x i——出行总量中交通方式i得比例，
    Z i——方式i的人均动态占地面积，
    Z——城市人均占用道路面积；
    步骤S13：根据步骤S11得到的目标函数与步骤S12中得到的两个约束条件，使用目标规划法求解得到各种交通出行方式所应该得到的优 化幅度。
  3. 如权利要求2所述的方法，其特征在于，所述的步骤b具体包括：
    步骤S21：收集私家车、出租车出行数据；
    步骤S22：将数据处理为向量格式。
  4. 如权利要求3所述的方法，其特征在于，所述的步骤c具体包括：
    步骤S31：清除数据中采集不完整的数据；
    步骤S32：将存在缺省值的数据直接去除；
    步骤S33：根据出行数据中的轨迹点与待规划区域边界经纬度进行比较，去除起终点不在城市规划范围内的数据。
  5. 如权利要求4所述的方法，其特征在于，所述的步骤d具体包括：
    步骤S41：将收集到的数据中的GPS数值映射到地图上；
    步骤S42：将GPS点映射到粗粒度的空间中。
  6. 如权利要求5所述的方法，其特征在于，所述的步骤S42具体包括：
    将城市划分为g*g个网格单元，此时每个GPS点l i位于一个单元C j中，将单个单元内的所有GPS点视为同一对象，将原来的GPS点
    Figure PCTCN2022138218-appb-100004
    Figure PCTCN2022138218-appb-100005
    表示为嵌入后的点
    Figure PCTCN2022138218-appb-100006
  7. 如权利要求6所述的方法，其特征在于，所述的步骤e具体包括：
    步骤S51：利用上述得到的各类GPS点，通过聚类算法得到候选公共交通站点；
    步骤S52：采用球面距离表示两个GPS点间的相似性，计算公式如下：
    Dist＝Δσ*R
    Figure PCTCN2022138218-appb-100007
    其中，R代表地球的平均半径，约为6378137m，Δlat表示两GPS 点的纬度差，Δlng表示两GPS点间的经度差，Dist近似表示两个GPS点间的真实球面距离；
    步骤S53：对各个点进行分组前先将距离当前点非常远的一些组排除掉，并且排除过程不需要逐个计算和各个组的距离；
    步骤S54：选择那些远离数据密集区域的位置的点作为参照点；
    步骤S55：根据选择的参照点，每次分组前先计算当前点和参照点的距离，排除距离较远的组，将各组的点与其所有可达组内包含的所有点进行合并，在合并后的数据域上使用聚类算法；
    步骤S56：通过如下公式计算城市交通热点区域的中心，并以此作为公共交通候选站点的具体位置；
    Figure PCTCN2022138218-appb-100008
    其中，n为某一类簇中所有数据对象的数量；Dist(i,j)为数据对象i和数据对象j的距离。
  8. 如权利要求7所述的方法，其特征在于，所述的步骤f具体包括：
    采用基于注意力机制的BiLSTM神经网络模型预测不同站点到邻近的下一个站点的概率。
  9. 如权利要求8所述的方法，其特征在于，所述的BiLSTM神经网络模型通过全连接层和Softmax分类器来输出最终的概率，全连接层涉及到的公式如下：
    Figure PCTCN2022138218-appb-100009
    Figure PCTCN2022138218-appb-100010
    其中，W FC和b FC都是全连接层可学习的参数矩阵；接着是最后的预测层，采用Softmax作为多类logistic回归分类器，得到候选目的地的概率 分布，对于输入部分转移T P，第j个候选目的地d j作为T P的真实目的地的概率
    Figure PCTCN2022138218-appb-100011
    通过对原始输出执行Softmax分类器来获得，最终预测结果是下列等式中概率最高的候选目的地：
    Figure PCTCN2022138218-appb-100012
    Figure PCTCN2022138218-appb-100013
    使用交叉熵作为损失函数，损失函数如下：
    Figure PCTCN2022138218-appb-100014
  10. 如权利要求9所述的方法，其特征在于，所述的步骤g具体包括：
    步骤S71：对于每一个公共交通站点，将其作为起点，利用步骤f的BiLSTM神经网络模型，得到其到不同站点的概率；接着从中选取较高概率的几个临近站点作为备选站点，连接起点到临近站点的线路；
    步骤S72：完成对所有的公共交通站点的计算后，对得到的所有线路中存有公共交通站点的线路进行连结，接着根据步骤S13中得到的优化幅度对现有公共交通线路进行优化，将优化后的公共交通线路与现有公共交通线路进行对比，调整现有地面常规公交线路和轨道交通线路；
    步骤S73：将调整的路线输入优化城市的数字孪生城市中进行修改，查看是否达到预期目标。
PCT/CN2022/138218 2021-12-13 2022-12-09 城市公共交通规划方法 WO2023109720A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111517873.6A CN114330106A (zh) 2021-12-13 2021-12-13 城市公共交通规划方法
CN202111517873.6 2021-12-13

Publications (1)

Publication Number Publication Date
WO2023109720A1 true WO2023109720A1 (zh) 2023-06-22

Family

ID=81050942

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/138218 WO2023109720A1 (zh) 2021-12-13 2022-12-09 城市公共交通规划方法

Country Status (2)

Country Link
CN (1) CN114330106A (zh)
WO (1) WO2023109720A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117195484A (zh) * 2023-08-08 2023-12-08 广东贝能达交通设备有限公司 一种轨道交通管理方法及系统
CN117272848A (zh) * 2023-11-22 2023-12-22 上海随申行智慧交通科技有限公司 基于时空影响的地铁客流预测方法及模型训练方法
CN117371629A (zh) * 2023-09-28 2024-01-09 广东艾百智能科技有限公司 基于机器学习的公交线路优化方法
CN117422602A (zh) * 2023-11-07 2024-01-19 北京城建设计发展集团股份有限公司 一种轨道交通规划多维融合方法及系统
CN117474735A (zh) * 2023-11-07 2024-01-30 北京城建设计发展集团股份有限公司 一种轨道交通网络与城市规划融合评估方法及系统
CN117593167A (zh) * 2024-01-18 2024-02-23 山东国建土地房地产评估测绘有限公司 基于大数据的智能城市规划管理方法及系统
CN118568526A (zh) * 2024-07-31 2024-08-30 浙江省交通运输科学研究院 一种基于聚类分析的综合交通运输通道识别方法
CN118643578A (zh) * 2024-08-14 2024-09-13 兰州交通大学 基于深度卷积生成对抗网络的高原地区铁路线路设计方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114330106A (zh) * 2021-12-13 2022-04-12 深圳先进技术研究院 城市公共交通规划方法
CN118211767B (zh) * 2024-05-22 2024-08-06 山东慧宇航空遥感技术有限公司 应用于城乡规划的场景建模方法及系统

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105489006A (zh) * 2015-12-15 2016-04-13 浙江工业大学 基于出租车gps数据的多尺度道路流量可视分析方法
JP2017204174A (ja) * 2016-05-12 2017-11-16 日本電信電話株式会社 目的地予測装置、目的地予測方法、及び目的地予測プログラム
CN107481511A (zh) * 2017-08-16 2017-12-15 深圳先进技术研究院 一种计算候选公交站点的方法及系统
CN107657330A (zh) * 2017-08-16 2018-02-02 深圳先进技术研究院 一种候选公交线路计算方法、系统及电子设备
CN110909434A (zh) * 2019-10-11 2020-03-24 东南大学 一种低碳导向下的城市公共交通干线网络设计方法
US20210209939A1 (en) * 2020-12-08 2021-07-08 Harbin Engineering University Large-scale real-time traffic flow prediction method based on fuzzy logic and deep LSTM
CN113486719A (zh) * 2021-06-08 2021-10-08 南京邮电大学 车辆目的地预测方法、装置、计算机设备和存储介质
CN114330106A (zh) * 2021-12-13 2022-04-12 深圳先进技术研究院 城市公共交通规划方法

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105489006A (zh) * 2015-12-15 2016-04-13 浙江工业大学 基于出租车gps数据的多尺度道路流量可视分析方法
JP2017204174A (ja) * 2016-05-12 2017-11-16 日本電信電話株式会社 目的地予測装置、目的地予測方法、及び目的地予測プログラム
CN107481511A (zh) * 2017-08-16 2017-12-15 深圳先进技术研究院 一种计算候选公交站点的方法及系统
CN107657330A (zh) * 2017-08-16 2018-02-02 深圳先进技术研究院 一种候选公交线路计算方法、系统及电子设备
CN110909434A (zh) * 2019-10-11 2020-03-24 东南大学 一种低碳导向下的城市公共交通干线网络设计方法
US20210209939A1 (en) * 2020-12-08 2021-07-08 Harbin Engineering University Large-scale real-time traffic flow prediction method based on fuzzy logic and deep LSTM
CN113486719A (zh) * 2021-06-08 2021-10-08 南京邮电大学 车辆目的地预测方法、装置、计算机设备和存储介质
CN114330106A (zh) * 2021-12-13 2022-04-12 深圳先进技术研究院 城市公共交通规划方法

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
JIEXIA YE; JUANJUAN ZHAO; KEJIANG YE; CHENGZHONG XU: "How to Build a Graph-Based Deep Learning Architecture in Traffic Domain: A Survey", ARXIV.ORG, 11 October 2020 (2020-10-11), XP081782982 *
TIAN KELANG, GUO JINGJIE, YE KEJIANG, XU CHENG-ZHONG: "ST-MGAT: Spatial-Temporal Multi-Head Graph Attention Networks for Traffic Forecasting", 2020 IEEE 32ND INTERNATIONAL CONFERENCE ON TOOLS WITH ARTIFICIAL INTELLIGENCE (ICTAI), IEEE, 1 November 2020 (2020-11-01) - 11 November 2020 (2020-11-11), pages 714 - 721, XP093072853, ISBN: 978-1-7281-9228-4, DOI: 10.1109/ICTAI50040.2020.00114 *
WANG HAO; ZHAO JUANJUAN; YE KEJIANG; XU CHENG-ZHONG; ZHANG LIUTAO; ZHANG JUN; SUN JING: "A Destination Prediction Model for Individual Passengers in Urban Rail Transit", 2020 INTERNATIONAL CONFERENCE ON HIGH PERFORMANCE BIG DATA AND INTELLIGENT SYSTEMS (HPBD&IS), IEEE, 23 May 2020 (2020-05-23), pages 1 - 6, XP033787313, DOI: 10.1109/HPBDIS49115.2020.9130592 *

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117195484B (zh) * 2023-08-08 2024-05-03 广东贝能达交通设备有限公司 一种轨道交通管理方法及系统
CN117195484A (zh) * 2023-08-08 2023-12-08 广东贝能达交通设备有限公司 一种轨道交通管理方法及系统
CN117371629B (zh) * 2023-09-28 2024-04-26 广东艾百智能科技有限公司 基于机器学习的公交线路优化方法
CN117371629A (zh) * 2023-09-28 2024-01-09 广东艾百智能科技有限公司 基于机器学习的公交线路优化方法
CN117422602A (zh) * 2023-11-07 2024-01-19 北京城建设计发展集团股份有限公司 一种轨道交通规划多维融合方法及系统
CN117474735A (zh) * 2023-11-07 2024-01-30 北京城建设计发展集团股份有限公司 一种轨道交通网络与城市规划融合评估方法及系统
CN117474735B (zh) * 2023-11-07 2024-06-07 北京城建设计发展集团股份有限公司 一种轨道交通网络与城市规划融合评估方法及系统
CN117422602B (zh) * 2023-11-07 2024-06-07 北京城建设计发展集团股份有限公司 一种轨道交通规划多维融合方法及系统
CN117272848B (zh) * 2023-11-22 2024-02-02 上海随申行智慧交通科技有限公司 基于时空影响的地铁客流预测方法及模型训练方法
CN117272848A (zh) * 2023-11-22 2023-12-22 上海随申行智慧交通科技有限公司 基于时空影响的地铁客流预测方法及模型训练方法
CN117593167A (zh) * 2024-01-18 2024-02-23 山东国建土地房地产评估测绘有限公司 基于大数据的智能城市规划管理方法及系统
CN117593167B (zh) * 2024-01-18 2024-04-12 山东国建土地房地产评估测绘有限公司 基于大数据的智能城市规划管理方法及系统
CN118568526A (zh) * 2024-07-31 2024-08-30 浙江省交通运输科学研究院 一种基于聚类分析的综合交通运输通道识别方法
CN118643578A (zh) * 2024-08-14 2024-09-13 兰州交通大学 基于深度卷积生成对抗网络的高原地区铁路线路设计方法

Also Published As

Publication number Publication date
CN114330106A (zh) 2022-04-12

Similar Documents

Publication Publication Date Title
WO2023109720A1 (zh) 城市公共交通规划方法
Liu et al. Contextualized spatial–temporal network for taxi origin-destination demand prediction
Cai et al. Applying a deep learning approach for transportation safety planning by using high-resolution transportation and land use data
WO2023029234A1 (zh) 一种面向有缺失数据的公交车到站时间预测方法
Lai et al. Concurrent optimization of rail transit alignments and station locations
CN108388970B (zh) 一种基于gis的公交站点选址方法
Zhou et al. Research on traffic situation analysis for urban road network through spatiotemporal data mining: a case study of Xi’an, China
Zhang et al. Using street view images to identify road noise barriers with ensemble classification model and geospatial analysis
Zhang et al. How road network transformation may be associated with reduced carbon emissions: An exploratory analysis of 19 major Chinese cities
Jiang et al. Pedestrian volume prediction with high spatiotemporal granularity in urban areas by the enhanced learning model
CN112559909B (zh) 一种基于gcn嵌入空间聚类模型的商业区发现方法
Menguc et al. A data driven approach to forecasting traffic speed classes using extreme gradient boosting algorithm and graph theory
Dorosan et al. Use of machine learning in understanding transport dynamics of land use and public transportation in a developing city
Lu et al. Optimization of vehicle automatic navigation path based on remote sensing and GIS
Zhou et al. Refined taxi demand prediction with ST-Vec
Tran et al. Tleta: Deep transfer learning and integrated cellular knowledge for estimated time of arrival prediction
CN113850295A (zh) 一种基于Weighted-LeaderRank和GMM聚类的站点选址预测方法
Li et al. Using random undersampling boosting classifier to estimate mode shift response to bus local network expansion and bus rapid transit services
Lee et al. Ensemble-based methodology to identify optimal personal mobility service areas using public data
Xu et al. Mining Urban Congestion Evolution Characteristics Based on Taxi GPS Trajectories
CN116894609B (zh) 一种基于半监督图神经网络的公交轨道接驳评价方法
Yin et al. Prediction Intervals for Bus Travel Time Based on Road Segment Sharing, Multiple Routes’ Driving Style Similarity, and Bootstrap Method
Han et al. Research on Travel Time Prediction of Multiple Bus Trips Based on MDARNN
Han et al. A Deep Spatiotemporal Model for Travel Time Prediction
CN117235556B (zh) 一种面向交通管理者的多模式交通复合网络构建方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22906467

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE