A network travel reliability evaluation and prediction method based on community

Abstract

The present invention discloses a method for evaluating and predicting the reliability of network travel based on a community, comprising the following steps: S1: pre-processing online car-hailing data to construct a travel trajectory data set for online car-hailing; S2: extracting the traffic status characteristics of each time period according to the trajectory data set; S3: calculating the time deviation degree of each trip and modeling the travel style of online car-hailing; S4: performing aggregate processing to evaluate the network travel reliability of each community; S5: using a clustering algorithm to divide into multiple reliability levels; S6: setting a neural network prediction model and using the data set to train the neural network model; S7: using the historical characteristics and reliability of the community to predict the reliability of each community in the entire network. The present invention can capture the spatial correlation between different communities when the community division is irregular, and can capture the temporal correlation between the research community and itself and other communities; it takes into account travel demand, is more in line with the needs of system evaluation, and has a fast prediction speed and high accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic management, and in particular to a community-based network travel reliability evaluation and prediction method. Background Art

[0002] As the travel distance increases, travelers face more uncertain events, such as bad weather and traffic accidents. The direct consequence of this uncertainty is that the actual travel time exceeds expectations, resulting in economic and time losses for travelers, which in turn reduces the travel experience. For system evaluation, identifying low reliability helps improve the transportation system. There are many definitions of travel time reliability by scholars at home and abroad. According to the research object, it can be divided into link-based, travel-based, and region-based. These definitions can provide the status of the road network and the distribution of travel time. Some scholars consider the impact of network traffic flow, but these literatures only use traffic flow as an explanatory variable for reliability changes. In fact, the purpose of system evaluation decisions is to meet more needs. It is very important to develop a new indicator that considers both traditional reliability and travel demand reliability indicators.

[0003] Generally, real-time reliability information is very helpful for improving the traffic system. If real-time reliability can be obtained, the government can take more accurate measures to optimize the performance of the traffic system. It is actually very difficult to obtain information on large-scale road networks. Many scholars obtain real-time reliability through prediction. Ding Chuyin et al. used long short-term memory network (LSTM) to predict reliability. However, this method requires high-dimensional features, which are difficult to obtain in practice; in addition, this method is not suitable for face-based traffic state prediction. In face-based traffic state prediction, CNN and its extended network are widely used. However, this network requires input to be grids of the same size when capturing spatial correlation (Haiyang Yu et al., XianZhou et al.). Obviously, this division method will lose the original structure and functional characteristics of each traffic zone. It is necessary to propose a new prediction method based on traffic zones. At the same time, when capturing the temporal correlation of features, the heterogeneity of features is often ignored, such as speed and flow rate evolve with traffic flow, while demand is not affected by traffic flow. When demand is used as a feature, a new processing method is needed. Summary of the invention

[0004] The purpose of the present invention is to provide a cell-based network travel reliability evaluation and prediction method to solve the above defects.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for evaluating and predicting network travel reliability based on a cell, comprising the following steps:

[0007] S1. Preprocess the online car-hailing data and construct an online car-hailing travel trajectory dataset;

[0008] S2, extracting traffic status characteristics of each time period based on the trajectory data set;

[0009] S3, calculate the time deviation of each trip and model the travel style of online ride-hailing;

[0010] S4, aggregate all trips departing from the same community and evaluate the network travel reliability of each community in the study area;

[0011] S5, using clustering algorithm to divide multiple reliability levels;

[0012] S6. Setting a neural network prediction model and using the data set to train the neural network model;

[0013] S7. Use the historical characteristics and reliability of the cells to predict the reliability of each cell in the entire network.

[0014] Preferably, step S1 specifically includes: map matching: matching all GPS data points to roads; data cleaning: deleting or repairing abnormal data; track restoration: separating different orders and restoring the track data of the orders;

[0015] Preferably, the criteria for judging abnormal data in the data cleaning step include: longitude and latitude, timestamp, speed, and passenger status.

[0016] Preferably, the step S2 specifically includes:

[0017] (1) The constructed online ride-hailing travel trajectory dataset is used to extract the OD point travel time and order start time of all online ride-hailing trips;

[0018] (2) Calculate the average travel time between different time periods and different OD pairs as the driver's expected travel time, and calculate the travel volume between different time periods and different OD pairs as the travel intensity between communities;

[0019] (3) Obtain the average speed and demand of all trips from each community to other communities in the study area at different time periods;

[0020] Preferably, in step S3, the expression of the time deviation degree is: in, and They represent the deviation degree and actual travel time of the kth trip from cell i to cell j respectively; represents the expected travel time from cell i to cell j in the hth period, which is standardized to make different trips comparable;

[0021] The travel pattern is: T = {O, D, h, d, t, M}, where O and D represent the starting and ending points respectively, h and d represent the time period and date of the trip respectively, t represents the actual travel time, and M represents the degree of deviation from the actual travel time.

[0022] Preferably, the step S4 specifically includes:

[0023] First, the average timeout rate is calculated. The expression of the average timeout rate is:

[0024] Where α represents the tolerance level of travelers to timeouts. The larger the value, the stronger the tolerance of travelers in this area to unreliable travel time. α The upper α quantile corresponding to the deviation degree of all trips represents the threshold of the degree of deviation accepted by the travelers. Trips exceeding this threshold are considered unreliable.

[0025] Secondly, the network travel reliability of each cell is obtained. The network travel reliability is expressed as:

[0026] Among them I ij is the travel intensity from community i to community j.

[0027] Preferably, in step S4, the average timeout rate and network travel reliability can be aggregated over any time span.

[0028] Preferably, in step S5, the clustering algorithm is a k-means clustering algorithm.

[0029] Preferably, in step S6, the neural network is a graph convolutional neural network with heterogeneous features (GCNHF), comprising a spatiotemporal convolution block, a temporal convolution layer, and a fully connected layer; wherein the spatiotemporal convolution block is formed by sequentially connecting a temporal convolution layer, a spatial convolution layer, and a temporal convolution layer;

[0030] Preferably, in step S7, the historical characteristics of the community include travel demand, travel speed, and reliability.

[0031] The beneficial effects of the present invention are:

[0032] The invention discloses a network travel reliability evaluation and prediction method based on cells, which can capture the spatial correlation between different cells when the cells are irregularly divided, and can capture the temporal correlation of reliability distribution between different time steps; it takes travel demand into consideration, is more in line with the needs of system evaluation, and has fast prediction speed and high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 : Flow chart of the method of the present invention;

[0034] Figure 2 : A cell division diagram of embodiment 1 of the present invention;

[0035] Figure 3 : Frequency distribution histogram of the degree of deviation of Example 1 of the present invention;

[0036] Figure 4 : Reliability evaluation diagram of Example 1 of the present invention;

[0037] Figure 5 : A neural network framework diagram of Example 1 of the present invention;

[0038] Figure 6 : A diagram showing the structure of a spatiotemporal convolution block according to Embodiment 1 of the present invention;

[0039] Figure 7 : A graph showing changes in accuracy during the prediction process of Example 1 of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with the embodiments. It should be noted that these are merely examples and illustrations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they should be deemed to fall within the protection scope of the present invention.

[0041] Embodiment 1:

[0042] like Figure 1-7 As shown, a cell-based network travel reliability evaluation and prediction method specifically includes the following steps:

[0043] S1. Preprocess the online car-hailing data and construct an online car-hailing travel trajectory dataset.

[0044] The GPS data of online ride-hailing vehicles within the Fourth Ring Road of Beijing from October 31, 2016 to December 30, 2016 were used. The specific data format is shown in Table 1. A total of 364 traffic zones were divided. See the attached figure for details. Figure 2 .

[0045] First, map matching is performed, and ArcGIS is used to match GPS points to the road network according to the closest distance. Secondly, abnormal data is cleaned according to speed, direction angle, GPS coordinates, and driving status, including deleting data that deviates from normal values ​​and filling in missing values; the criteria for judging abnormal data in the above data cleaning steps include but are not limited to: longitude and latitude, timestamp, speed, passenger status, etc. Finally, orders are separated according to vehicle operation status and vehicle trajectories are restored in timestamp order. For two GPS points that are far apart, the shortest path principle is adopted to connect them; orders with travel time within 120 seconds are merged.

[0046] Table 1: GPS data format for online ride-hailing

[0047] VID LON LAT speed bearing state year month day hour minute second 0 21954 116.4836 39.89572 75 180 1 2016 10 31 0 50 0 1 431328 116.3731 39.95534 72 358 1 2016 10 31 0 50 0 2 403201 116.3833 39.90581 73 86 1 2016 10 31 0 50 0 3 431013 116.4023 39.93203 40 270 0 2016 10 31 0 50 0 4 155475 116.3739 39.96538 60 184 0 2016 10 31 0 50 0 5 156476 116.3505 39.90158 60 176 1 2016 10 31 0 50 0 6 174902 116.385 39.94758 67 88 1 2016 10 31 0 50 0

[0048] Among them, VID corresponds to the vehicle number, LON and LAT correspond to longitude and latitude, speed corresponds to the instantaneous speed of the vehicle, bearing corresponds to the direction angle of the vehicle, state is the vehicle operation status, 0 corresponds to no load, 1 corresponds to carrying a passenger, year, month, day, hour, minute, second correspond to the year, month, day, hour, minute, second when the GPS point was recorded, respectively.

[0049] S2. Extract traffic status characteristics of each time period based on the trajectory data set, including:

[0050] (1) Through the constructed online ride-hailing travel trajectory dataset, the OD point travel time and order start time of all online ride-hailing trips are extracted, with a time interval of 20 minutes and each hour is divided into three intervals. The extraction results are shown in the following table (Table 2):

[0051] Table 2: OD point travel time and order start time of online ride-hailing trips

[0052] O D month day hour Interval TravelTime 246 3 10 31 0 0 272 305 353 10 31 1 0 1044 97 260 10 31 6 2 218 238 305 10 31 7 0 384 175 305 10 31 7 1 142 202 50 10 31 7 2 542 43 317 10 31 10 0 55 305 257 10 31 11 1 985

[0053] Among them, in the above Table 2, "O" is the starting cell number; "D" is the end cell number; "month, day, hour, Interval" correspond to "month, day, hour, time period" respectively, among which the time period information indicates the starting time of the current trip in which 20-minute interval in an hour, such as hour = 0, Interval = 0 means that the travel time is in the interval of 0:00-0:20; "TravelTime" is the travel time, in seconds.

[0054] (2) The average travel time between different time periods and different OD pairs is calculated as the driver's expected travel time. The travel volume between different time periods (in this example, the time period is divided into 20 minutes, and a day is divided into 72 time periods) and different OD pairs is calculated as the travel intensity between the sub-districts. The matrix size is (72, 364, 364), which corresponds to the time period when the trip is generated, the starting sub-district number, and the ending sub-district number;

[0055] (3) Obtain the average speed and total demand of all trips from each community to other communities in the study area at different time periods, corresponding to the two characteristics of each community.

[0056] S3. Calculate the time deviation of each trip and model the travel style of online ride-hailing vehicles.

[0057] The travel time deviation degree expression is defined as in, and They represent the deviation degree and actual travel time of the kth trip from the i-th cell to the j-th cell, represents the expected travel time from cell i to cell j in the hth period (in this case, the average travel time is used instead). Due to the different distances of different trips, the travel time may vary greatly, so the travel time needs to be standardized to make different trips comparable;

[0058] Based on the information of each trip, the online car-hailing travel style is modeled as T = {O, D, h, d, t, M}, where O and D represent the start and end cell numbers, h and d represent the time period and date of the trip, t represents the actual travel time, and M represents the degree of deviation of the actual travel time from the expected travel time. For example, T = {246, 3, 30, 20, 1200, 0.2} means that the trip starts from cell 246 to cell 3, the departure time is the 30th time period (9:40-10:00) on the 20th day, and the actual travel time is 1200 seconds, which exceeds the expected travel time by 0.2 (20%).

[0059] S4. All trips departing from the same community are aggregated and the network travel reliability of each community in the study area is evaluated.

[0060] First calculate the average excess ratio: Secondly, the Network Travel Reliability of each community is calculated:

[0061] Among them, α represents the tolerance level of travelers for timeouts. The larger the value, the stronger the tolerance of travelers in this area for unreliable travel time. In this example, 0.2 is taken, which means that 80% of the trips are reliable (see Figure 2). Figure 3 ). α The upper α quantile corresponding to the deviation degree of all trips represents the threshold of the degree of deviation accepted by the travelers. Trips exceeding this threshold are considered unreliable. In this example, M α It is about 0.144, which means that travelers can accept a travel time that exceeds their expectations by up to 14.4%. ij is the travel intensity from community i to community j, which is represented by the average travel volume in this example.

[0062] The above average timeout rate and network trip reliability can be aggregated by any time span (every 20 minutes is used in this example).

[0063] S5. Use clustering algorithm to divide into multiple reliability levels.

[0064] The k-means clustering algorithm is used to divide the data into four categories. The specific steps are as follows: first, randomly assign cluster centers; classify the data according to the given cluster centers; calculate new cluster centers according to the degree of dispersion of each cluster; repeat the above steps until convergence or the maximum number of iterations is reached. The division results are shown in Table 3. The time period of 9:00-9:20 on October 31, 2016 was selected. The evaluation results are shown in the attached figure. Figure 4 .

[0065] Table 3: Reliability level classification details

[0066] grade reliable Generally reliable Unreliable Very unreliable NTR 0.0-1.57 1.57-4.93 4.93-11.63 11.63-∞

[0067] Depend on Figure 4 It can be seen that the most unreliable areas in the current period are located on the West Fourth Ring Road and the North Fourth Ring Road, indicating that a large number of travel demands in these areas are not met during the current period.

[0068] S6. Setting a neural network prediction model and using the data set to train the neural network model.

[0069] The above neural network is a graph convolutional neural network with heterogeneous features (GCNHF), which includes a spatiotemporal convolution block, a temporal convolution layer, and a fully connected layer (see Figure 2). Figure 5 ). The above-mentioned spatiotemporal convolution blocks are composed of a temporal convolution layer, a spatial convolution layer, and a temporal convolution layer connected in sequence (see Figure 2). Figure 6 ). The feature dimensions of each layer’s input and output are given by (C i →C o )express.

[0070] The input module receives three characteristics of the cell: demand, level, and speed. It inputs the demand of the current period in the previous k days (3 in this example) and the level and speed information of the previous q periods (5 in this example), and inputs the two types of data into the network. In this example, the cells are divided into four levels according to the size of the NTR index, which are coded as 0, 1, 2, and 3 respectively.

[0071] All cell features were linearly normalized.

[0072] The shortest path distance matrix between cells (a symmetric matrix of size (364,364)) is used to extract the adjacency matrix and normalize it. The specific extraction method is: select all cell pairs with a distance greater than 5km and set the distance to positive infinity, use the average circumscribed circle radius of the cell (about 500 meters) to represent the distance between the cell and itself, take the inverse of all elements in the obtained matrix A, and obtain the adjacency matrix after Laplace transformation (L) and normalization.

[0073] The temporal convolution layer is a time-gated convolution layer. It captures the dependencies of two types of features (long-term demand features and short-term speed and level features) in the time dimension through three convolution kernels of size 1*3. No padding is performed during the convolution process, that is, a new time slice is generated for every three consecutive time slices. The time dimension decreases by 2 each time a temporal convolution layer is passed. After a total of 3 temporal convolution layers, the final time dimension is k+q-2*3=2.

[0074] The spatial convolution layer is a traditional graph convolution layer, which is expressed by the formula Forward propagation, where C i and C o are the number of input and output channels respectively, 64 and 16 in this example.

[0075] After a spatiotemporal convolution module and a spatial module, the time dimension of the feature is 2, the number of channels is 64, and the four-dimensional feature is output through the fully connected layer, corresponding to the four levels of scores of the community.

[0076] Using mean square error As the loss function of back propagation. During back propagation, the level label is encoded using one-hot encoding, that is, a four-dimensional vector is used to represent the level. For example, the corresponding encoding of a level 0 cell is 1, 0, 0, 0; the corresponding encoding of a level 1 cell is 0, 1, 0, 0, and so on.

[0077] S7. Use the historical characteristics and reliability of the cells to predict the reliability of each cell in the entire network.

[0078] The historical characteristics of the community include, but are not limited to, travel demand, travel speed, reliability, etc.

[0079] Using 80% of the data set as the training set and 20% as the prediction set, the final prediction level of each cell is where y i is the score obtained for the ith level. Where n is the number of custom categories, which is 4 in this example.

[0080] The accuracy rate changes during the prediction process of the present invention, such as Figure 7 As shown in the figure, training_accuracy represents the accuracy of the model in the training set, and validation_accuracy represents the accuracy of the model in the prediction set. Figure 7 It can be seen that the model needs 500 rounds of training to converge.

[0081] The invention discloses a network travel reliability evaluation and prediction method based on cells, which can capture the spatial correlation between different cells when the cells are irregularly divided, and can capture the temporal correlation between the research cell and itself and other cells; it takes travel demand into consideration, is more in line with the needs of system evaluation, and has fast prediction speed and high accuracy.

[0082] The above is an exemplary description of the invention. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as such non-substantial improvements are made by adopting the method concept and technical solution of the present invention, or the concept and technical solution of the invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A method for evaluating and predicting network travel reliability based on a cell, characterized in that: The following steps are involved: S1. Preprocess the online car-hailing data and construct an online car-hailing travel trajectory dataset; S2. Extract traffic status characteristics of each time period based on the trajectory data set, including: (1) The OD point travel time and order start time of all online ride-hailing trips are extracted through the constructed online ride-hailing travel trajectory dataset; (2) Calculate the average travel time between different time periods and different OD pairs as the driver's expected travel time, and calculate the travel volume between different time periods and different OD pairs as the travel intensity between communities; (3) Obtain the average speed and demand of all trips from each community to other communities in the study area at different time periods; S3, calculate the time deviation of each trip and model the travel style of online ride-hailing; The expression of the time deviation degree is: in, and They represent the deviation degree and actual travel time of the kth trip from cell i to cell j respectively; represents the expected travel time from cell i to cell j in the hth period, which is standardized to make different trips comparable; The travel pattern is: T = {O, D, h, d, t, M}, where O and D represent the starting and ending points, h and d represent the time period and date of the trip, t represents the actual travel time, and M represents the degree of deviation from the actual travel time; S4: All trips departing from the same community are aggregated to evaluate the network travel reliability of each community in the study area, including: First, the average timeout rate is calculated. The expression of the average timeout rate is: Where α represents the tolerance of travelers to timeouts. The larger the α, the stronger the tolerance of travelers in the area to unreliable travel time. α The upper α quantile corresponding to the deviation degree of all trips represents the threshold of the degree of deviation accepted by the travelers. Trips exceeding this threshold are considered unreliable. Secondly, the network travel reliability of each cell is obtained. The network travel reliability is expressed as: Among them I ij is the travel intensity from community i to community j; S5, using clustering algorithm to divide multiple reliability levels; S6. Setting a neural network prediction model and using the data set to train the neural network model; S7. Use the historical characteristics and reliability of the cells to predict the reliability of each cell in the entire network.

2. A cell-based network travel reliability evaluation and prediction method according to claim 1, characterized in that: The S1 step specifically includes: map matching: matching all GPS data points to the road; data cleaning: deleting or repairing abnormal data; trajectory restoration: separating different orders and restoring the trajectory data of the orders.

3. A cell-based network travel reliability evaluation and prediction method according to claim 2, characterized in that: The data cleaning step uses criteria for judging abnormal data including: longitude and latitude, timestamp, speed, and passenger status.

4. The method for evaluating and predicting network travel reliability based on a cell according to claim 1, characterized in that: The average timeout rate and network travel reliability described in step S4 can be aggregated over any time span.

5. The method for evaluating and predicting network travel reliability based on a cell according to claim 1, characterized in that: In step S5, the clustering algorithm is a k-means clustering algorithm.

6. The method for evaluating and predicting network travel reliability based on a cell according to claim 1, characterized in that: In step S6, the neural network is a graph convolutional neural network (GCNHF) with heterogeneous features, including a spatiotemporal convolution block, a temporal convolution layer, and a fully connected layer; wherein the spatiotemporal convolution block is composed of a temporal convolution layer, a spatial convolution layer, and a temporal convolution layer connected in sequence.

7. The method for evaluating and predicting network travel reliability based on a cell according to claim 1, characterized in that: In step S7, the historical characteristics of the community include travel demand, travel speed, and reliability.

Images