Traffic congestion prediction device and traffic congestion prediction method

Abstract

The goal is to obtain accurate, real-time information on traffic congestion using a simpler configuration. [Solution] The traffic congestion prediction device 1 includes a learning unit 11 that constructs a multivariate probability model using the movement amounts of each of several vehicles and estimates the parameters of the constructed multivariate probability model to generate a matrix representing the conditional dependency between the movement amounts of each of several vehicles, with each matrix corresponding to a set of observation periods along a time series, and a determination unit 12 that determines whether or not traffic congestion occurs on a road based on a comparison of the values ​​in each of the multiple matrices generated by the learning unit 11 that indicate the conditional dependency between the movement amounts of each of several vehicles.

Description

[Technical Field]

[0001] The present invention relates to a traffic congestion prediction device and a traffic congestion prediction method. [Background technology]

[0002] Conventionally, technologies have been known that provide road congestion information based on vehicle detection information from sensors installed on roads. For example, Patent Document 1 discloses a congestion prediction device that measures the average traffic volume and average speed of a predetermined section of a road using sensors installed on the road, and predicts congestion from the correlation distribution data between the average traffic volume and average speed.

[0003] However, the technology disclosed in Patent Document 1 requires sensors to be placed along the road, and furthermore, sensor failures or malfunctions could reduce the accuracy of traffic congestion prediction.

[0004] Another conventional traffic congestion prediction technology is known to use GPS-based mobile communication terminal location information and vehicle location information to understand and predict traffic congestion. For example, Patent Document 2 discloses a technology that predicts congestion if, in addition to vehicle location information, the number of times the vehicle brakes and accelerator is pressed exceeds a predetermined number within a certain period.

[0005] However, the technology disclosed in Patent Document 2 requires that, in order to improve prediction accuracy, the vehicle needs to be equipped with a device that includes a brake pressure sensor and an accelerator pedal position sensor, etc., in order to collect vehicle operation information in addition to vehicle location information.

[0006] Furthermore, for example, Google Maps® estimates traffic congestion using location information and speed information obtained from users whose smartphone's location function is enabled. However, since user permission is required to collect user location information, the data used for traffic congestion prediction may be insufficient due to user settings and other factors, potentially leading to a decrease in prediction accuracy. [Prior art documents]

Patent Literature

[0007]

Patent Literature 1

Patent Literature 2

SUMMARY OF THE INVENTION

Problem to be Solved by the Invention

[0008] As described above, in the conventional art, it has been difficult to grasp a traffic congestion state with high accuracy and in real time with a simpler configuration.

[0009] The present invention has been made to solve the above-described problem, and an object thereof is to grasp a traffic congestion state with high accuracy and in real time with a simpler configuration.

Means for Solving the Problem

[0010] To solve the above-mentioned problems, the traffic congestion prediction device according to the present invention comprises: a first acquisition unit configured to acquire location information of each of the plurality of vehicles transmitted from each of the plurality of vehicles traveling on a road; a second acquisition unit configured to calculate the length of each of the plurality of vehicles' movement vectors based on the location information of each of the plurality of vehicles acquired by the first acquisition unit and acquire it as a time-series movement amount sequence; a learning unit configured to construct a multivariate probability model using movement amounts obtained by sampling the movement amount sequence of each of the plurality of vehicles acquired by the second acquisition unit based on observation periods consisting of multiple time periods, and to estimate the parameters of the constructed multivariate probability model, thereby generating a matrix representing the conditional dependency between the movement amounts of each of the plurality of vehicles, with each matrix corresponding to each of the multiple observation periods along a time series; and a determination unit configured to determine whether or not traffic congestion occurs on the road based on a comparison of values ​​in each of the multiple matrices generated by the learning unit that indicate the conditional dependency between the movement amounts of each of the plurality of vehicles.

[0011] Furthermore, in the traffic congestion prediction device according to the present invention, the learning unit may estimate the parameters of a multivariate normal distribution model with a sample mean of 0 for multivariate data where the amount of movement of the plurality of vehicles for each time period of the observation period is a variable, by maximum likelihood estimation, thereby generating the precision matrix of the multivariate normal distribution model as the matrix showing the conditional dependency relationship between the amount of movement of the plurality of vehicles.

[0012] Furthermore, in the traffic congestion prediction device according to the present invention, the determination unit may be configured to determine that traffic congestion is occurring on the road when there is a certain number of components in the plurality of matrices where the value of each corresponding component, which indicates the conditional dependency between the amount of movement of the plurality of vehicles, is non-zero and the change in value is within an acceptable range.

[0013] Furthermore, in the traffic congestion prediction device according to the present invention, the first acquisition unit may acquire location information of each of the plurality of vehicles associated with the location registration request signal transmitted from each of the plurality of vehicles, which is stored in the data repository and is included in the core network for the integrated management of subscriber information.

[0014] Furthermore, the traffic congestion prediction device according to the present invention may also include a notification unit configured to notify the system of the occurrence of traffic congestion when the determination unit determines that traffic congestion has occurred on the road.

[0015] To solve the above-mentioned problems, the traffic congestion prediction method according to the present invention comprises: a first acquisition step of acquiring location information of each of the plurality of vehicles attached to a location registration request signal transmitted from each of the plurality of vehicles traveling on a road; a second acquisition step of calculating the length of each of the movement vectors of the plurality of vehicles based on the location information of each of the plurality of vehicles acquired in the first acquisition step and acquiring it as a series of movement amounts along a time series; a learning step of constructing a multivariate probability model using the movement amounts obtained by sampling the series of movement amounts of each of the plurality of vehicles acquired in the second acquisition step based on observation periods consisting of multiple time periods, and estimating the parameters of the constructed multivariate probability model to generate a matrix that represents the conditional dependency between the movement amounts of each of the plurality of vehicles, and which corresponds to each of the multiple observation periods along a time series; and a determination step of determining whether or not traffic congestion occurs on the road based on a comparison of the values ​​indicating the conditional dependency between the movement amounts of each of the plurality of vehicles in each of the multiple matrices generated in the learning step.

[0016] Furthermore, in the traffic congestion prediction method according to the present invention, the learning step may be performed by estimating the parameters of a multivariate normal distribution model with a sample mean of 0 for multivariate data where the amount of movement of the plurality of vehicles for each time period of the observation period is a variable, by maximum likelihood estimation, thereby generating the precision matrix of the multivariate normal distribution model as the matrix that shows the conditional dependency between the movement vectors of the plurality of vehicles.

[0017] Furthermore, in the congestion prediction method according to the present invention, the determination step may determine that congestion is occurring on the road if there is a certain number of components in the plurality of matrices where the value of each corresponding component, which indicates the conditional dependency between the amount of movement of the plurality of vehicles, is non-zero and the change in value is within an acceptable range.

[0018] Furthermore, in the traffic congestion prediction method according to the present invention, the first acquisition step may acquire location information of each of the plurality of vehicles associated with the location registration request signal transmitted from each of the plurality of vehicles, which is stored in the data repository and is included in the core network for the integrated management of subscriber information.

[0019] Furthermore, the traffic congestion prediction method according to the present invention may also include a notification step for notifying the occurrence of traffic congestion if it is determined in the determination step that traffic congestion has occurred on the road. [Effects of the Invention]

[0020] According to the present invention, whether or not traffic congestion has occurred on a road is determined based on a comparison of values ​​that indicate the conditional dependency between the respective travel amounts of multiple vehicles in each of the multiple precision matrices generated by the learning unit. Therefore, it is possible to grasp the traffic congestion situation with high accuracy and in real time with a simpler configuration. [Brief explanation of the drawing]

[0021] [Figure 1]Figure 1 is a block diagram showing the configuration of a traffic congestion prediction system equipped with a traffic congestion prediction device according to an embodiment of the present invention. [Figure 2] Figure 2 is a diagram illustrating the overview of the traffic congestion prediction system included in the traffic congestion prediction device according to this embodiment. [Figure 3] Figure 3 is a diagram illustrating the overview of the traffic congestion prediction system included in the traffic congestion prediction device according to this embodiment. [Figure 4] Figure 4 is a diagram illustrating the configuration of the learning unit of the traffic congestion prediction device according to this embodiment. [Figure 5] Figure 5 is a diagram illustrating the learning unit of the traffic congestion prediction device according to this embodiment. [Figure 6] Figure 6 is a diagram illustrating the learning unit of the traffic congestion prediction device according to this embodiment. [Figure 7] Figure 7 is a diagram illustrating the learning unit of the traffic congestion prediction device according to this embodiment. [Figure 8] Figure 8 is a block diagram showing the hardware configuration of the traffic congestion prediction device according to this embodiment. [Figure 9] Figure 9 is a sequence diagram showing the operation of a traffic congestion prediction system equipped with a traffic congestion prediction device according to this embodiment. [Modes for carrying out the invention]

[0022] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to Figures 1 to 9.

[0023] [Configuration of the traffic congestion prediction system] First, an overview of a traffic congestion prediction system equipped with a traffic congestion prediction device 1 according to an embodiment of the present invention will be described. As shown in Figure 1, the traffic congestion prediction system predicts the occurrence of traffic congestion on main roads such as expressways. The main roads for which traffic congestion prediction is performed are defined by the respective communication areas A1, A2, A3, ..., An (communication areas A1 to An are not distinguished, and are collectively referred to as "communication area A") of a plurality of base stations 3_1, 3_2, 3_3, ..., 3_n (when base stations 3_1 to 3_n are not distinguished, they are collectively referred to as "base station 3") that construct a 5G mobile communication system.

[0024] As shown in Figure 1, the communication areas A1 to An of each base station 3_1 to 3_n are arranged to cover each section that makes up the main road. Vehicles traveling on the main road are equipped with communication terminals 2. Vehicles include automobiles, motor vehicles, and motorcycles. Communication terminals 2 can be implemented as tablet computers, laptop computers, wearable devices, smartphones, and other mobile communication terminals.

[0025] The communication terminal 2 can be realized by a computer equipped with a processor, main memory, communication interface, auxiliary storage, and input / output I / O, and a program that controls these hardware resources. The communication terminal 2 also includes a location information acquisition means including a GPS module, and acquires its current GPS location information. Furthermore, the communication terminal 2 can acquire the direction of travel relative to main roads, for example, based on a car navigation system or the GPS location information of the communication terminal 2. In this embodiment, there are M units of the communication terminal 2 (where M is a positive integer of 2 or more).

[0026] The communication terminal 2 is equipped with a SIM (Subscriber Identity Module) 20 and connects to the network based on subscriber identification information (IMSI: International Mobile Subscriber Identity) stored in the SIM 20. In this case, the communication terminal 2 is uniquely identified by the IMSI. In this embodiment, as an example, one communication terminal 2 is installed in each of the M vehicles, and the vehicles are uniquely identified by the IMSI of the communication terminal 2.

[0027] The communication terminal 2 transmits a location registration request signal to the connected base station 3 when it reaches the adjacent communication area A as the vehicle moves, at regular intervals, and when powered on. In the payload of the location registration request signal transmitted by the communication terminal 2 at regular intervals (e.g., 1 ms), the communication terminal 2 adds the current GPS location information of its own terminal acquired by the GPS module. In this embodiment, in addition to the location registration request signal and GPS location information, the communication terminal 2 transmits information indicating the direction of travel of the communication terminal 2 on the main road to the AMF 40 via the base station 3.

[0028] The traffic congestion prediction system according to this embodiment comprises a traffic congestion prediction device 1, a communication terminal 2 mounted on a vehicle, a base station 3, and an AMF (Access and Mobility Management Function) 40, UDM (Unified Data Management) 41, and UDR (Unified Data Repository) 42 provided by the 5G core network 4. The traffic congestion prediction device 1 is connected to the UDR 42 via a network NW such as a WAN (Wide Area Network), LAN (Local Area Network), or the Internet.

[0029] Base station 3 connects to the core network 4 and forwards user packets from communication terminals 2 located in communication areas A1 to An. Base station 3 also forwards location registration request signals transmitted periodically by communication terminals 2 mounted on vehicles to the core network 4. The geographical location of the main road can be determined by the identification information of base station 3 or the identification information of communication area A.

[0030] The AMF40 is a mobility management control device that performs mobility management and bearer control, including setting up and releasing sessions for packet communication and controlling handovers. The AMF40 grasps the transition order between communication areas A1 to An where communication terminal 2 is located as mobility management information for the main road under its management.

[0031] UDM41 is a subscriber information management device that stores subscriber profile information and other data. UDR42 is a data repository that comprehensively stores and manages subscriber information for communication terminal 2. The location registration request signal transmitted by communication terminal 2 mounted on the vehicle is received by AMF40, AMF40 transmits the received signal to UDM41, and UDM41 performs location registration using the IMSI of communication terminal 2. Furthermore, the location registration request signal, IMSI, and information indicating the direction of travel transmitted by communication terminal 2 are stored in UDR42 along with identification information regarding the base stations 3_1~3_n and communication areas A1~An located in the area, as well as the transmission timestamp of the location registration request signal and GPS location information attached to the location registration request signal.

[0032] In this embodiment, the transition order of communication areas A1 to An, as determined by the AMF40, is used as information to identify the road traveled and the direction of travel of the vehicle. Combined with GPS location information at the timestamp of the location registration request signal stored in the UDR42, this information is acquired by the traffic congestion prediction device 1 as location information of the communication terminal 2 at regular intervals.

[0033] Figure 2 is a diagram illustrating the overview of the traffic congestion prediction system according to this embodiment. As shown in Figure 2, as a vehicle travels along a main road in the direction of travel, it moves sequentially from the communication area A1 of base station 3_1 covering the first section of the main road, to the communication area A2 of base station 3_2 covering the second section, to the communication area A3 of base station 3_3 covering the third section, and then to the communication area An of base station 3_n covering the nth section.

[0034] The arrows in Figure 2 indicate the location registration request signals transmitted at regular intervals (e.g., every 1 ms) by the communication terminal 2 installed in each vehicle. In sections of main roads where congestion does not occur, each vehicle travels relatively freely, so the amount of movement represented by the length of the movement vector of each vehicle, calculated based on GPS location information at regular intervals, varies from vehicle to vehicle. On the other hand, in sections where congestion occurs, many vehicles are stopped or traveling in close proximity at low speeds, so the observed values ​​of the amount of movement of each vehicle, calculated based on GPS location information at regular intervals, are concentrated within a narrow range.

[0035] [Functional block of the traffic congestion prediction system] As shown in Figure 1, the traffic congestion prediction device 1 comprises an acquisition unit (first acquisition unit, second acquisition unit) 10, a learning unit 11, a determination unit 12, a notification unit 13, and a storage unit 14.

[0036] The acquisition unit (first acquisition unit) 10 acquires the location information of each of the multiple vehicles that are attached to the location registration request signals transmitted at regular intervals from each of the multiple vehicles traveling on the main road. The acquisition unit 10 acquires GPS location information from the UDR 42 via the network NW at the transmission timestamp of each location registration request signal for each IMSI equipped with the communication terminal 2 of the M vehicles moving in the transition order from base station 3_1 to base station 3_n. Furthermore, the acquisition unit (second acquisition unit) 10 calculates the length of each movement vector of the multiple vehicles at regular intervals based on the acquired location information of each of the multiple vehicles and acquires it as a movement amount series along the time series.

[0037] Figure 3 is a schematic diagram showing an example of the movement vectors of multiple vehicles. The upper part of Figure 3 shows the vehicles traveling on a main road, and the white circles placed along the main road indicate GPS location information (G) at each point on the main road. The lower part of Figure 3 shows the movement vectors of each communication terminal 2 (2_1~2_M) equipped on M vehicles at regular intervals (1 [ms]) in section S1 where there is no congestion and section S2 where congestion is occurring.

[0038] Each circle in Figure 3 represents the GPS position information of the communication terminal 2 of each vehicle at each time point, and the arrows represent the movement vector. Each circle can be represented by the GPS position information of the center point, for example, as a value that represents the GPS position information of an arbitrary range of area. Furthermore, as can be seen from the fact that the GPS positions shown in the first circle of section S1 in Figure 3 are aligned, a common position is used as the starting point for the movement vectors of each vehicle acquired at regular intervals for vehicles traveling in the same direction on a main road. For example, from among M vehicles located within a radius of 30 [m] from a GPS position set as the starting point on a main road, it is possible to extract the vehicles traveling in a predetermined direction on the main road and obtain the amount of movement of each vehicle at regular intervals.

[0039] As shown in Figure 3, in section S1 where there is no congestion, each vehicle travels at a different speed above a certain threshold, so the length of each vehicle's movement vector differs from vehicle to vehicle, and some are relatively long. On the other hand, in section S2 where congestion occurs, many vehicles are stopped or traveling at low speeds, and the distance between vehicles is reduced, so the length of each vehicle's movement vector is shorter compared to the movement vector in section S1, and its length is almost constant.

[0040] Here, the movement vector of the communication terminal 2 (hereinafter simply referred to as "vehicle m" or "vehicle") equipped in each vehicle m is (a 1m ,a 2m ,a 3m ) is expressed as, and component a 1m ,a 2m ,a 3m These represent the x, y, and z coordinate components, respectively. At regular intervals for each vehicle m (hereinafter referred to as "time t"), k」, hereinafter referred to as). That is, at time t k This is the movement amount x, which is the length of the movement vector for each (k=1,...,K) m is at time t k and the immediately preceding time t k-1 is calculated from the difference in GPS position information at , and is expressed by the following equation (1).

Equation

[0041] In the present embodiment, since the targets are vehicles traveling on a main road in the same traveling direction, the direction of the movement vector substantially coincides with the traveling direction. Therefore, as the observation value used for constructing the multivariate probability model by the learning unit 11, the movement amount x, which is the length of the movement vector of each vehicle m m is used. That is, the acquisition unit 10 uses the above equation (1) to calculate the observation value group D of the movement amounts of the M vehicles at each time t k (k=1,...,K) k ={x1,x2,...x M} is acquired.

[0042] The learning unit 11 uses the movement amount obtained by sampling the movement amount sequence along the time series based on the lengths of the movement vectors of the plurality of vehicles (communication terminals 2) acquired by the acquisition unit 10, based on the observation period composed of a plurality of time zones, to construct a multivariate probability model, and estimates the parameters of the constructed multivariate probability model, thereby generating a plurality of matrices each corresponding to a plurality of observation periods along the time series, the matrices representing conditional independence (dependency relationship) between the movement amounts of the plurality of vehicles. More specifically, the learning unit 11 estimates the parameters of a multivariate normal distribution model with a sample mean of 0 for multivariate data having the movement amounts of a plurality of vehicles as variables by maximum likelihood estimation, and generates the precision matrix of the multivariate normal distribution model as a matrix indicating conditional independence between the movement amounts of the plurality of vehicles.

[0043] The learning unit 11, based on the movement amount sequence acquired by the acquisition unit 10, generates an observation value group D whose elements are the movement amounts of the M vehicles at time t k k ={x1,x2,...x​M The data is sampled based on an observation period (e.g., 5 hours) consisting of multiple time periods (e.g., 1 minute). Specifically, the learning unit 11 samples the observed data group D k Using observations obtained by sampling at a 1-minute interval, a data set D consisting of N=300 observations corresponding to a 5-hour observation period is defined as D={x (1) ,x (2) ,···,x (N) It is generated as}. Here, each observed value x (n) (n=1,...,N) is an M-dimensional observation whose elements are the amount of movement (length of the movement vector) of M vehicles.

[0044] The learning unit 11 generates multiple accuracy matrices along a time series, including the first matrix Λ, which is the accuracy matrix for the first observation period, and the second matrix Λ, which is the accuracy matrix for the second observation period following the first observation period. The observation period is a monitoring period for traffic congestion prediction, and the first observation period is, for example, the first 5 hours, and the second observation period is the 5 hours immediately following the first observation period. The observation period is set to the length of time necessary to detect the occurrence of traffic congestion, and can be set appropriately in the range of, for example, several minutes to several hours. In addition, the first observation period and the second observation period can overlap at least partially, and the first matrix, the second matrix, and so on can be generated sequentially by sliding the time window. For example, the learning unit 11 obtains multiple (for example, 100) accuracy matrices Λ by sequentially generating the accuracy matrix Λ by sliding the start time in 1-minute increments.

[0045] As described above, the learning unit 11 learns the essential dependencies between variables, using the movement amounts of multiple vehicles as variables. Here, Figures 5(a) and (b) show the directed graph used by the learning unit 11. Each node a, b, and c represents a variable, and each node represents the movement amount of each vehicle. Each edge represents a direct probabilistic dependency. In the directed graph of Figure 5(a), the value of node c is unobserved. At this time, the joint probability distribution ρ(a,b,c) of variables a, b, and c is expressed by the following equation (2).

number

[0046] Marginalizing the variable c, it can be expressed by the following equation (3).

number

[0047] In equation (3) above, the two variables a and b are not independent, as they cannot generally be expressed as ρ(a)ρ(b). On the other hand, in the directed graph of Figure 2(b), the variable c is observed. The joint probability ρ(a, b|c) when the variable c is observed is given by equation (4).

number

[0048] In equation (4) above, considering the common condition of the observed variable c, we can see that variables a and b do not influence each other, that is, they are independent. This is called conditional independence. By considering conditional independence, it becomes possible to extract the essential or true relationship between variables. For example, even if the movement of one vehicle appears to be related to the movement of another vehicle, if we consider the movement of yet another vehicle, this falls under the category of conditional independence when these vehicle movements do not influence each other.

[0049] Here, as mentioned above, the amount of movement of each vehicle is the M-dimensional observed value x = (x1, x2, ..., x M It is expressed as ). A data set D consisting of N observed values ​​x is given by D = {x (1) ,x (2) ,···,x (N) If expressed as}, this multivariate normal distribution model can be represented by the following equation (5).

number

[0050] In equation (5) above, μ represents the mean, Σ represents the covariance matrix, and |·| represents the determinant. The learning unit 11 performs maximum likelihood estimation to find μ and Σ, which are parameters of the multivariate normal distribution model, from the data set D of observed data. The log-likelihood L(μ,Σ|D) of the data set D is expressed by the following equation (6).

number

[0051] Substituting equation (5) into equation (6) above yields equation (7).

number

[0052] The parameters μ and Σ that maximize the log-likelihood L(μ,Σ|D) in equation (7) above are estimated as the maximum likelihood solution. For the maximum likelihood solution of parameters μ and Σ, μ and Σ are estimated respectively. -1 If we differentiate with respect to zero, the maximum likelihood solutions for the parameter μ and Σ are given by equations (8) and (9), respectively.

number

number

[0053] The inverse matrix Σ of the parameter Σ -1 This is called the precision matrix Λ. The precision matrix Λ can be obtained by finding the inverse of equation (9) above. In the case of maximum likelihood estimation as shown in equations (8) and (9) above, if regularization is ineffective and there is a risk of overfitting, the learning unit 11 can also perform estimation by applying maximum posterior probability estimation (MAP estimation) to the parameters μ and Σ.

[0054] Here, the correlation structure in a multivariate normal distribution is represented by a graph model. Such a graph model is called a Gaussian graphic model. Below, we will explain how to calculate conditional probabilities when applying a multivariate normal distribution model to graph theory. In the multivariate normal distribution model of equation (5) above, the inverse matrix Σ of the parameter Σ -1 Using the precision matrix Λ and setting the parameter μ to 0, the following equation (10) is given.

number

[0055] The ρ(x) in equation (10) above is called a Gaussian graphic model. Under a multivariate normal distribution, the conditional probability ρ(x1,x2|x3,···,x) of variables x1 and x2 is given by... M ) is expressed by the following equation (11).

number

[0056] The functions of the variables x1 and x2 in equation (11) above are ρ(x) in equation (10) above, i.e., N(x|0,Λ -1 Since it is proportional to ), if we pick out all the parts in equation (10) that relate to the variables x1 and x2, we obtain the relationship in equation (12).

number

[0057] Here, conditionally independent ρ(x1|x3,···,x M )ρ(x2|x3,···,x M The conditions for satisfying ) are given by the following equation (13).

number

[0058] Equation (13) above indicates that the values ​​of the 1st and 2nd components of the precision matrix Λ are 0, that is, there is no edge between variables x1 and x2, as shown by variables a and b in the directed graph of Figure 5. Figures 6 and 7 (a) and (b) are schematic diagrams showing the relationship between the correlation between variables on the graph of the Gaussian graphic model and the precision matrix Λ, respectively. The nodes in Figure 6 (a) and Figure 7 (a) correspond to vehicles from 1 to M units and represent the amount of movement of each vehicle. Furthermore, Figure 6 (a) shows the correlation between variables based on the amount of movement of each vehicle, which is the observed value for each time period in the first observation period, and Figure 7 (a) shows the correlation between variables based on the amount of movement of each vehicle, which is the observed value for each time period in the second observation period following the first observation period.

[0059] Figure 6(a) shows solid edges between nodes where conditional dependencies exist during the first observation period, indicating a direct relationship between the movement amounts of each vehicle corresponding to those nodes. Furthermore, in the precision matrix Λ shown in Figure 6(b), in addition to the diagonal components, the components corresponding to the nodes with solid edges in Figure 6(a) are non-zero. As an example, the second and ninth components Λ of the precision matrix corresponding to the second and ninth nodes are shown. 2,9 This value can be non-zero and is shown exemplarily as 0.8. In Figure 6(b) and Figure 7(b), non-zero components of the precision matrix Λ are represented by "*".

[0060] Figure 5(a) shows the conditional dependency in the second observation period following the first observation period. Between the second node and the ninth node, a solid line edge exists, similar to the first observation period, corresponding to the second and ninth components Λ of the precision matrix Λ. 2,9 The value remains non-zero at 0.8 even in the second observation period. On the other hand, the dependency relationship between the third node and the thirteenth node that existed in the first observation period disappears in the second observation period, and the third and thirteenth components Λ of the corresponding precision matrix Λ 3,13 It changes from a non-zero value to a zero value. Furthermore, a new dependency arises between the third node and the M-3 node during the second observation period, and correspondingly, the third and M-3 components Λ of the precision matrix Λ change. 3,M-3 It is changing from a zero value to a non-zero value.

[0061] Figures 4(a) and 4(b) show examples of the precision matrix Λ during congestion and in non-congestion situations, respectively, for the first and second observation periods. For multiple vehicles traveling in the same direction on a main road, in a congested section, many vehicles stop or travel at low speeds, and the distance between vehicles is compressed, resulting in the movement of each vehicle being homogenized under common constraints. Therefore, the conditional dependency formed between the movement amounts of multiple vehicles in a congested section remains stable and does not change significantly over time (lower precision matrix Λ in Figures 4(a) and 4(b)). 1(渋滞) , Λ 2(渋滞) ).

[0062] Correspondingly, the precision matrix Λ (multiple matrices including the first matrix and the second matrix) generated by the learning unit 11 for each observation period is the precision matrix Λ for the first observation period in Figure 4(a). 1(渋滞) Λ in the second observation period of the same figure (b) 2(渋滞) As shown, the composition of non-zero components is maintained to be the same or within a range that can be considered the same, and the values ​​of the corresponding components can continue to match within a predetermined tolerance range.

[0063] On the other hand, in sections where congestion does not occur, each vehicle travels at a different speed, resulting in variability in the amount of travel. Therefore, the conditional dependencies between vehicles tend to change depending on the observation period, and the composition of the non-zero components and / or component values ​​of the precision matrix Λ often do not match between multiple observation periods (the upper precision matrix Λ in Figures 4(a) and (b)). 1(非渋滞) , Λ 2(非渋滞) ). In particular, when no congestion occurs, the precision matrix Λ for the first observation period in Figure 4(a) is shown. 1(非渋滞) Λ in the second observation period of the same figure (b) 2(非渋滞) As shown, the component values ​​of the precision matrix are not uniform and show large variability. Therefore, when the observation period is shifted, the structure and component values ​​of the non-zero components tend to change.

[0064] Furthermore, in order to make the above equation (10) available, the learning unit 11 may construct a multivariate normal distribution model after setting the sample mean μ to 0. In this case, the learning unit 11 considers the nth element x of the data set D to be... (n) The mth element (m=1···M) of (n=1···N) (observed value) x m (n) After applying the following standardization transformation to (the amount of movement of each vehicle) as shown in equation (14), the observed value x (n) A multivariate normal distribution model may be constructed using the sample.

number

number

[0065] Returning to Figure 1, the determination unit 12 determines whether or not congestion is occurring on the road based on a comparison of values ​​indicating the conditional dependency between the movement amounts of multiple vehicles in each of the multiple precision matrices generated by the learning unit 11. More specifically, the determination unit 12 determines that congestion is occurring on the main road if, in the multiple precision matrices along the time series, there is a certain number of components whose values ​​are non-zero and whose value changes are within an acceptable range, indicating the conditional dependency between the movement amounts of multiple vehicles.

[0066] The determination unit 12, for example, compares the values ​​of each component of the precision matrix Λ for the first observation period with the values ​​of each component of the precision matrix Λ for the second observation period. If there are a certain number of component values ​​that maintain a non-zero and identical range, the determination unit 12 can determine that congestion is occurring on the main road. More specifically, the determination unit 12 compares, for example, multiple (e.g., 100) precision matrices Λ generated by the learning unit 11 by sliding the time window of the observation period. If the configuration of non-zero components matches consecutively or between a predetermined number or more precision matrices, and the values ​​of each component match within a predetermined range, the determination unit 12 determines that congestion is occurring.

[0067] The notification unit 13 notifies the system of congestion when the determination unit 12 determines that congestion has occurred on a main road. The notification unit 13 notifies an external road traffic information system (not shown) via the network NW of information indicating the occurrence of congestion. The notification unit 13 can notify the system of the time information when the congestion occurred and the information of the corresponding section of the main road.

[0068] The memory unit 14 stores the parameters μ and Σ of the multivariate normal distribution model estimated by learning by the learning unit 11, and the precision matrix Λ generated by the learning unit 11.

[0069] [Hardware configuration of traffic congestion prediction system] Next, an example of a hardware configuration for realizing the traffic congestion prediction device 1 having the functions described above will be explained using Figure 8.

[0070] As shown in Figure 8, the traffic congestion prediction device 1 can be implemented, for example, by a computer equipped with a processor 102, main memory 103, communication interface 104, auxiliary storage 105, and input / output I / O 106 connected via a bus 101, and a program that controls these hardware resources. The traffic congestion prediction device 1 also includes a display device 107 connected via the bus 101.

[0071] The processor 102 is a circuit or device that performs arithmetic processing, and can be implemented by, for example, a general-purpose central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), etc. Alternatively, some or all of these may be combined to form the processor.

[0072] The main memory 103 is composed of, for example, volatile random access memory (RAM), and contains pre-stored programs for the processor 102 to perform various controls and calculations. The processor 102 and the main memory 103 work together to realize the various functions of the traffic congestion prediction device 1, such as the acquisition unit 10, learning unit 11, determination unit 12, and notification unit 13 shown in Figure 1.

[0073] The communication interface 104 is an interface circuit for networking the traffic congestion prediction device 1 with various external electronic devices.

[0074] The auxiliary storage device 105 consists of a read / write storage medium and a drive device for reading and writing various information such as programs and data to the storage medium. The auxiliary storage device 105 can use non-volatile storage such as a hard disk or flash memory as the storage medium.

[0075] The auxiliary storage device 105 has a program storage area for storing the traffic congestion prediction program. The auxiliary storage device 105 also has a program storage area for storing the learning program executed by the traffic congestion prediction device 1. The storage unit 14 described in Figure 1 is realized through the auxiliary storage device 105. Furthermore, it may have, for example, a backup area for backing up the aforementioned data and programs.

[0076] The I / O106 is an input / output device that accepts signals from external devices and outputs signals to external devices.

[0077] The display device 107 is composed of an organic EL display or a liquid crystal display, etc. The display device 107 can display the traffic congestion status of main roads on its screen.

[0078] [Operation of the traffic congestion prediction system] Next, the operation of the traffic congestion prediction device 1 having the above-described configuration will be explained with reference to the flowchart in Figure 9.

[0079] First, as shown in Figure 9, the acquisition unit 10 acquires GPS location information for each of the M vehicles (communication terminals 2) traveling on the main road to be predicted, which is attached to the location registration request signal transmitted by each of the M vehicles at regular intervals (e.g., 1 [ms]) from the UDR42 (step S1). Next, based on the GPS location information for each of the M vehicles acquired in step S1 at regular intervals, the acquisition unit 10 calculates the length of the movement vector for each vehicle at regular intervals and acquires a series of movement amounts along the time series (step S2). In step S2, the acquisition unit 10 calculates the length of the movement vector representing the amount of vehicle movement from the difference in GPS position at regular intervals, and at each time t k Observational data set D of the movement series of M vehicles (k=1,...,K) k ={x1,x2,...x M Get}.

[0080] Next, the learning unit 11 processes each time t obtained in step S2. k Observational data set D of the movement series of M vehicles (k=1,...,K) k ={x1,x2,...x M For example, by sampling at time intervals of 1 minute, we obtain a data set D = {x} consisting of N observed values ​​x over an observation period (e.g., 5 [h]). (1) ,x (2) ,···,x (N)Create} (Step S3). In Step S3, for example, using observed values ​​obtained by sampling each vehicle at a 1-minute period, observed values ​​x, which are the amount of movement (length of the movement vector) for each N=300 vehicle corresponding to a 5-hour observation period, are obtained.

[0081] Next, the learning unit 11 estimates the parameters μ and Σ of a multivariate normal distribution model with a sample mean of 0 for the data set D, which is multivariate data with the movement amounts of multiple vehicles as variables, by maximum likelihood estimation (step S4). In step S4, the learning unit 11 uses equation (7) above to estimate the parameters μ and Σ that maximize the log-likelihood L(μ,Σ|D) as the maximum likelihood solution (equations (8) and (9) above).

[0082] Next, the learning unit 11 generates the precision matrix Λ of the multivariate normal distribution model as a matrix that shows the conditional independence between the movement amounts of multiple vehicles (step S5). In step S5, the learning unit 11 generates the inverse matrix Σ of the parameters Σ estimated in step S4. -1 The precision matrix Λ is generated. The processes from step S1 to step S5 are performed for each observation period, and the learning unit 11 sequentially generates a precision matrix Λ for each time window by sequentially shifting the start time of the time window defining the observation period by a predetermined slide amount (e.g., 1 minute), thereby obtaining multiple (e.g., 100) precision matrices Λ.

[0083] Next, the determination unit 12 compares the precision matrices Λ of multiple observation periods generated in step S5, and determines that congestion has occurred on the main road if the corresponding component values ​​are non-zero and a certain number of component values ​​remain unchanged within the acceptable range (step S6). The number of precision matrices Λ to compare, the certain number of non-zero and unchanging components within the acceptable range, and the acceptable range in step S6 are appropriately set according to the target road, traffic volume, length of the observation period, slide amount, etc., so as to balance the false detection rate and detection delay within the desired range.

[0084] Next, if the notification unit 13 determines in step S6 that congestion has occurred, it notifies an external road traffic information provision system (not shown) via the network NW of information indicating the occurrence of congestion (step S7). In step S7, the notification unit 13 can provide information on the time the congestion occurred and information on the corresponding section of the main road.

[0085] Subsequently, the road traffic information system updates the traffic condition information for the corresponding section of the main road based on the traffic congestion information associated with the received section and time of the main road, and distributes or displays the traffic congestion information to road users.

[0086] As described above, the traffic congestion prediction device 1 according to this embodiment uses location information attached to a location registration request signal transmitted by a communication terminal 2 mounted on a vehicle in a mobile communication network to predict traffic congestion based on the location distribution of multiple vehicles. Therefore, without installing dedicated sensors on the road or mounting special detection devices on vehicles, it is possible to stably grasp traffic congestion conditions in real time and with high accuracy using existing mobile communication infrastructure.

[0087] Furthermore, according to the traffic congestion prediction device 1 of this embodiment, a multivariate probability model is constructed using the lengths of movement vectors, which are the movement amounts of multiple vehicles. By estimating the parameters of the constructed multivariate probability model, multiple time-series precision matrices Λ are generated that represent the conditional dependency between the movement amounts of multiple vehicles. If the values ​​of the corresponding components of the multiple precision matrices Λ are non-zero and there are a certain number of components whose value changes within an acceptable range, it is determined that traffic congestion has occurred on a main road. Therefore, traffic congestion can be grasped with a simpler configuration.

[0088] Furthermore, according to the traffic congestion prediction device 1 of this embodiment, by using a precision matrix Λ obtained from a multivariate probability model in which the movement amounts of multiple vehicles (communication terminals 2) traveling on a main road are variables, it is possible to extract the essential relationships between vehicles based on conditional dependencies that take into account the movement of other vehicles, rather than simple correlations. Moreover, by comparing multiple precision matrices Λ generated by sliding the time window and determining the congestion state based on whether the composition of non-zero components and the agreement of component values ​​continue for a predetermined number or more, it is possible to grasp the congestion situation with high accuracy and in real time with a simpler configuration while suppressing misjudgments caused by temporary deceleration and stopping.

[0089] Furthermore, according to the traffic congestion prediction device 1 of this embodiment, a precision matrix Λ obtained from a multivariate probability model with the movement amounts of multiple vehicles (communication terminals 2) as variables is used to determine the congestion state based on the configuration of non-zero components in the precision matrix Λ generated for each observation period and / or the temporal maintenance of component values. As a result, even before congestion becomes apparent and vehicles stop or travel at extremely low speeds, it is possible to capture the stabilization of conditional dependencies between vehicles that occur when the movement of a group of vehicles begins to be influenced by preceding vehicles, etc., thus enabling early prediction of congestion.

[0090] In the embodiment described, a traffic congestion prediction system compatible with a 5G mobile communication network was explained, but the communication network may also be 3G, LTE / 4G, or 6G.

[0091] Although embodiments of the traffic congestion prediction device and traffic congestion prediction method of the present invention have been described above, the present invention is not limited to the embodiments described above, and various modifications that a person skilled in the art can envision are possible within the scope of the invention described in the claims. [Explanation of Symbols]

[0092] 1...Traffic congestion prediction device, 2...Communication terminal, 20...SIM, 10...Acquisition unit, 11...Learning unit, 12...Determination unit, 13...Notification unit, 14...Storage unit, 101...Bus, 102...Processor, 103...Main memory, 104...Communication interface, 105...Auxiliary memory, 106...Input / output I / O, 107...Display device, NW...Network.

Claims

1. A first acquisition unit configured to acquire location information of each of the multiple vehicles attached to the location registration request signal transmitted from each of the multiple vehicles traveling on the road, A second acquisition unit is configured to calculate the length of the movement vector of each of the multiple vehicles based on the position information of each of the multiple vehicles acquired by the first acquisition unit, and to acquire it as a series of movement amounts along a time series, A learning unit is configured to construct a multivariate probability model using the movement amounts obtained by sampling the movement amount series of each of the plurality of vehicles acquired by the second acquisition unit based on observation periods consisting of multiple time periods, and to estimate the parameters of the constructed multivariate probability model, thereby generating a precision matrix representing the conditional dependency between the movement amounts of each of the plurality of vehicles, the precision matrix corresponding to each of the multiple observation periods along the time series, A determination unit is configured to determine that congestion is occurring on the road if, in each of the plurality of precision matrices generated by the learning unit, there is a certain number of corresponding components whose values ​​are non-zero and whose value changes within an acceptable range, each representing the conditional dependency between the respective amounts of movement of each of the plurality of vehicles. A traffic congestion prediction device equipped with the following features.

2. In the traffic congestion prediction device according to claim 1, The learning unit estimates the parameters of a multivariate normal distribution model with a sample mean of 0 for multivariate data where the amount of movement of the plurality of vehicles for each time period of the observation period is the variable, by maximum likelihood estimation, thereby generating the precision matrix of the multivariate normal distribution model as the precision matrix that shows the conditional dependency between the amount of movement of the plurality of vehicles. A traffic congestion prediction device characterized by the following features.

3. In the traffic congestion prediction device according to claim 1, The first acquisition unit acquires location information for each of the plurality of vehicles, which is associated with the location registration request signal transmitted from each of the plurality of vehicles and is stored in the data repository, which is included in the core network and comprehensively manages subscriber information. A traffic congestion prediction device characterized by the following features.

4. In the traffic congestion prediction device according to claim 1, Furthermore, the system includes a notification unit configured to notify the system of the occurrence of traffic congestion when the determination unit determines that traffic congestion has occurred on the road. A traffic congestion prediction device characterized by the following features.

5. A first acquisition step of acquiring location information of each of the multiple vehicles attached to the location registration request signal transmitted from each of the multiple vehicles traveling on the road, A second acquisition step involves calculating the length of the movement vector of each of the multiple vehicles based on the position information of each of the multiple vehicles acquired in the first acquisition step, and acquiring it as a series of movement amounts along the time series, A learning step which involves constructing a multivariate probability model using the movement amounts obtained by sampling the movement amount series of each of the multiple vehicles acquired in the second acquisition step based on an observation period consisting of multiple time periods, and estimating the parameters of the constructed multivariate probability model to generate a precision matrix that represents the conditional dependency between the movement amounts of each of the multiple vehicles, and which corresponds to each of the multiple observation periods along the time series, A determination step in which congestion is determined to occur on the road if, in each of the plurality of precision matrices generated in the learning step, there is a certain number of corresponding components whose values ​​are non-zero and whose value changes within an acceptable range, indicating the conditional dependency between the amount of movement of each of the plurality of vehicles, and A traffic congestion prediction method that includes the following features.

6. In the traffic congestion prediction method described in claim 5, The learning step involves estimating the parameters of a multivariate normal distribution model with a sample mean of 0 for multivariate data where the amount of movement of the multiple vehicles for each time period of the observation period is the variable, by maximum likelihood estimation, thereby generating the precision matrix of the multivariate normal distribution model as the precision matrix that shows the conditional dependency between the amounts of movement of the multiple vehicles. A method for predicting traffic congestion characterized by the following features.

7. In the traffic congestion prediction method described in claim 5, The first acquisition step involves acquiring location information for each of the plurality of vehicles, which is associated with the location registration request signals transmitted from each of the plurality of vehicles and is stored in the data repository, from a data repository included in the core network that comprehensively manages subscriber information. A method for predicting traffic congestion characterized by the following features.

8. In the traffic congestion prediction method described in claim 5, Furthermore, if the determination step determines that traffic congestion has occurred on the road, the system includes a notification step to notify the system of the occurrence of traffic congestion. A method for predicting traffic congestion characterized by the following features.

Images