JP5070574B2 - Local traffic prediction program generation device, local traffic prediction device, local traffic prediction program generation method, local traffic prediction method and program - Google Patents

Description

  The present invention relates to a local traffic volume prediction technique for predicting the traffic volume of an arbitrary designated section in a road network using past traffic volume information of its neighboring sections, and in particular, information related to the correlation of the traffic volume of each section. The present invention relates to a local traffic volume prediction technique that can flexibly perform prediction even if it is unknown, and that can also perform prediction of a plurality of future times in a plurality of designated sections in a short time.

  In recent years, VICS (Vehicle Information and Communication System), ATIS (Advanced Traffic Information Service), car navigation systems, etc. are widely used to alleviate traffic congestion and to accurately guide destinations. In addition to grasping and transmitting the current traffic volume on the road, it is required to accurately predict the traffic volume on each road. Therefore, various technologies for traffic prediction are being researched and developed.

  As conventional traffic prediction technology, for example, those described in Patent Documents 1 to 3 are known.

  The traffic situation estimation system described in Patent Document 1 predicts future traffic volume as accurately as possible by combining traffic volume prediction based on past and current traffic volume measurements and traffic volume prediction based on traffic flow simulation. It is what.

  In this traffic situation estimation system, first of all, the future traffic is measured by pattern matching, auto-regression, Kalman filter, neural network, etc. Calculate the predicted value of the quantity. On the other hand, a traffic flow simulation is performed using road network data, signal data on the road network, and OD traffic volume (traffic volume from the origin to the destination), and traffic volume for each road link. Calculate In the traffic flow simulation, a macro traffic flow model such as a block density method or an input / output method, or a micro traffic flow model based on a tracking model is used. Then, a target value is set for each road link in order to match the calculated value of the traffic volume with the observed value of the current traffic volume and the predicted value of the traffic volume obtained by pattern matching or the like. Based on the calculated traffic volume and the target value, the traffic volume is predicted by correcting the OD traffic volume.

  In Patent Document 2, a road network is represented by a finite-state machine (finite-state machine), and a quasi-optimal finite-state machine is created for the traffic prediction of a road network by an evolutionary method. The traffic volume is predicted using the created finite state machine.

  FIG. 18 is a diagram showing an example of a finite state machine used in Patent Document 2 and the structure of its gene. A finite state machine refers to an ordered five-term set M = (I, O, S, f, φ). I and O are input sets and output sets, S is a state set, f is a function that maps set S × I to set S, and φ is a function that maps set S × I to set I. As shown in the state transition table of FIG. 18A, the finite state machine M determines the transition destination of the next state S from the current state S and the input value I. The output value O corresponds to each state S. The input value I and the output value O are values obtained by quantizing the average speed of the vehicle traveling on the road link into three stages. FIG. 18B shows the state transition table of FIG. 18A as a state transition diagram.

For example, when the current state is S _{1 and} 0 is input as the input value, the next state is shifted to S ₂ and 0 is output as the predicted average speed of the vehicle traveling on the road link. When the initial state is S1 and the average speed time-series data 0, 0, 1, 2 is input as the input, the state transitions as S ₁ → S ₂ → S ₃ → S ₄ and the road link 2 is output as the predicted average speed of the vehicle traveling on the road.

  In order to semi-optimize the finite state machine shown in FIGS. 18A and 18B by an evolutionary algorithm, the finite state machine is expressed in a gene form as shown in FIG. Then, using the traffic flow data group measured in the past several months or years as the training data group, the evolutionary algorithm repeatedly evaluates the genetic form of the finite state machine population, and performs mutation and mutation. Create an optimal finite state machine.

  The evaluation of each finite state machine is given 1 point when the training data is given to the genetic form finite state machine (hereinafter referred to as “gene”), and 1 point is given when the machine output matches the training data. The operation of giving points is performed N times, and each gene is evaluated based on the final evaluation points. Acupuncture means to increase the number of genes with a high score and reduce the number of genes with a low score by roulette strategy or the like in a gene group. Mutation means that a part of one gene is randomly changed.

  Then, the estimated average speed of the vehicle traveling on each road link is estimated using a sub-optimal finite state machine obtained as a result of this evolutionary algorithm.

Also in Patent Document 3, a neural network that predicts future traffic data from past traffic data and current traffic data is adaptively generated using an evolutionary algorithm, and the generated neural network is also disclosed. Although a traffic situation prediction apparatus that predicts traffic volume using a network is described, this document does not describe a specific calculation method.
JP 2004-118735 A JP 2000-57482 A Japanese Patent Laid-Open No. 11-328570 Hirasawa et al., “Performance comparison between Genetic Network Programming and Genetic Programming in the evolution of ant behavior”, The Institute of Electronics, Information and Communication Engineers, IEICE Theory, Vol. 121, No. 6, 2001, pp.1001-1009.

  In the traffic situation estimation system described in Patent Document 1, it is necessary to model traffic flow in a road network when using traffic flow simulation. In this modeling, some feature quantities that inevitably affect the traffic flow are extracted, and calculation is performed using the feature quantities as feature parameters.

  However, the road network has a complex structure, the speed and travel route of each vehicle on the road is diverse, traffic parameters such as traffic lights and roads have various parameters that affect traffic volume, In general, the temporal fluctuation of traffic volume is extremely complicated due to various external factors that affect traffic volume such as time zone, weather, and events. Therefore, it is extremely difficult to extract an accurate feature parameter, and there is a problem that the traffic volume prediction accuracy cannot be improved unless an accurate feature parameter can be extracted.

  In addition, when performing road simulation, it is necessary to model the road network individually for each road network, so it is not easy to actually introduce the road network.

  On the other hand, in the method using the evolutionary algorithm described in Patent Document 2, it is not necessary to extract feature parameters as long as there is information on past and present traffic, so that more adaptive prediction with respect to environmental changes is possible. Is possible. In addition, since the evolutionary algorithm does not select information by specific modeling, it is possible to predict traffic volume by efficiently using information included in past traffic volume data. Therefore, this method can be said to be a method suitable for a problem in which the number of parameters such as traffic volume prediction in a road network is large, fluctuates depending on the environment, the object is difficult to grasp, and the search space is enormous.

  In Patent Document 2, a finite state machine is generated as a traffic volume prediction program that predicts traffic volume of road links (average speed of vehicles traveling on road links) by an evolutionary algorithm. A finite state machine is an automaton with a finite number of distinct internal states (a collective term for black boxes that can obtain discrete information from the output when discrete information is input to the input). is there. And the state quantity of each node of the finite state machine (the node represented by ◯ in FIG. 18B) does not have a specific physical meaning, and theoretically, any prediction model is adaptive. Can express. Further, in the evolution process of the evolutionary algorithm, the number of nodes in the finite state machine and the connection relation of transitions vary due to mutations and the like.

  In the case of traffic volume prediction of a road network with relatively few road links, it is not difficult to create a traffic volume prediction program by an evolutionary algorithm using such a general-purpose finite state machine. However, in general, there are an infinite number of finite state machine node combinations, so when applied to traffic prediction in a road network with many road links, the number of nodes in the finite state machine broadens during the evolution process. There is a problem that the degree of freedom increases so that it becomes difficult to converge to an optimal traffic prediction program. That is, when the number of nodes of a finite state machine increases, the number of combinations of finite state machines increases dramatically, and the probability of approaching an optimal solution by mutation or crossover decreases, so that convergence is also reduced.

  Therefore, an object of the present invention is to easily converge even in a complex road network and predict a more accurate sub-optimal traffic volume when predicting the local traffic volume of a road network using an evolutionary algorithm. It is to provide a local traffic prediction technology that can do this.

The configuration of the present invention relating to the local traffic volume prediction program generating device is that an intersection node n _i corresponding to each intersection i (i∈V _c : V _c is a set of all intersections), an intersection i to an intersection j (i, j∈) V _c ), road links l _{i, j} corresponding to each road (i, j) heading to V _c ), and traffic v _{i, j} (t) measured on the road (i, j) at each time t Traffic at a predicted time t ₀ + n (n = 1,..., N _t : N _t is a predicted time range) after a certain time t ₀ of a specific road link l _{k, l} A local traffic volume prediction program generating device for automatically generating a local traffic volume prediction program for calculating a predicted value v ^ _{k, l} (t ₀ + n) of an amount v _{k, l} (t ₀ + n),
A traffic volume v _{i, j} (t ₀ −m) (t ₀ -m) of the road link l _{i, j} before the time t ₀ is provided corresponding to each road link l _{i, j} (i, j∈V _c ). m = 0, 1,..., M: M is a traffic volume time range used for prediction) It is determined to which of a plurality of classes it belongs according to its size, and transition is made according to each class of the determined traffic volume. A decision node that branches the destination, and
Corresponds to all or part of the predicted values v ^ _{k, l} (t ₀ + n) of the predicted times t ₀ + n of the specific road link l _{k, l} (k, lεV _c ) to be predicted. A processing node provided,
A start node that is the start position of the transition, and
One directional transition branch from the start node to any one of the determination nodes, N _c from the determination node to any one of the determination nodes or any of the processing nodes according to each class of the traffic volume A directional transition branch of the book (N _c is the number of traffic classes) and a directional transition branch from each processing node to the determination node;
In the configuration, each processing node, the individual multiple storing individual data of the local traffic prediction program the determination node and identification label on the total number N _v number of each node of each starting node is represented as a labeled directed graph Data storage means;
Traffic data storage means for storing the traffic v _{i, j} (t) measured at each time t on each road (i, j);
An initial individual constructing means for generating a predetermined number N ₀ of individual data of the local traffic prediction program of the initial generation and storing it in the individual data storage means;
For each of the individual data stored in the individual data storage means, fitness calculation means for calculating the fitness F of the individual data by the fitness function of (Equation 1);
Among the individual data, a predetermined number N ₁ is selected based on the fitness F, and the rest of the individual data means for hesitation,
Wherein for each of the selected predetermined number N ₁ pieces of said individual data, and a solid reproducing means for reproducing the individual data in response to the matching degree F of the individual data,
Among the regenerated individual data, crossover processing is performed on the number of nodes in proportion to the crossover probability of the two individual data selected by a predetermined method and the directional transition branch starting from this, and new solid data is obtained. A crossover means for generating and storing in the individual data storage means as the individual data of a new generation;
Among the regenerated individual data, a connection mutation process is performed on the number of directed transition branches in proportion to the connection mutation probability in a predetermined number of individual data to generate new solid data to generate a new generation of the data A connection mutation means for storing in the individual data storage means as individual data;
Among the regenerated individual data, node mutation processing is performed on a number of nodes in proportion to the node mutation probability in a predetermined number of individual data, and new solid data is generated to generate a new generation of the individual data. Node mutation means for storing in the individual data storage means;
An end determination means for determining whether or not a predetermined end condition is satisfied after the process of generating the individual data of a new generation is performed by the crossover means, the connection mutation means, and the node mutation means;
When it is determined that the termination condition is satisfied by the termination determination unit, the fitness level F of the individual data is calculated by the fitness level calculation unit, and the individual data having the smallest calculated fitness level is used as the final local traffic. A sub-optimal solution selection means for outputting as a quantity prediction program;
With
The crossover means converts two pieces of individual data into a parent-child data 1 and a parent individual data 2 by random numbers from the individual data group consisting of the set of individual data having the _Nv number of nodes with identification labels. Select as
Selecting two or more nodes from the respective nodes of the parent individual data 1 by random numbers;
The crossover process is performed by exchanging the selected node of the parent individual data 1 and the node of the parent individual data 2 having the same identification label as the node selected by the parent individual node 1 And
The connection mutation means selects the directional transition branch for changing the connection from the directional transition branches of the individual data to be subjected to the connection mutation process,
From each node of the individual data other than the node to which the selected directional transition branch selected in the individual data is connected, a new connection destination node is selected by a random number,
The connection mutation process is performed by changing the transition destination of the selected directional transition branch to the selected new connection destination node.
The node mutation means selects a change node whose contents are changed by a random number from the nodes other than the start node of the individual data to be subjected to the connection mutation process,
If the change node is the determination node, the node is changed to another type of the determination node, and if the change node is the processing node, the node mutation process is performed by changing to another type of the processing node. Is,
When it is determined by the termination determination means that the termination condition is not satisfied, the fitness calculation means, the individual selection means, the crossover means, the connection mutation means, the node mutation means, and the termination determination The means is characterized in that the same processing is repeatedly executed for each individual data stored in the individual data storage means.

  According to this configuration, the local traffic volume prediction program is represented as a directed graph in which a start node, a determination node, and a processing node are connected in a network shape with directed transition branches. A large number of individual data of this local traffic prediction program is generated, and an evolutionary algorithm by repeating reproduction, crossover, mutation (connection mutation and node mutation), and trapping (in the case of the present invention, “genetic network programming ( Genetic network programming (GNP) ”(see Non-Patent Document 1)), the individual data group is adapted to the prediction model, and the most suitable individual by the suboptimal solution selection means from the finally obtained individual data group By selecting the data, a sub-optimal local traffic prediction program can be obtained. At this time, there is no increase or decrease in the number of nodes in the process of evolution. Further, each determination node and each processing node correspond to a road link, and the meaning represented by the node is clear. Therefore, useless nodes with unknown meanings do not occur. Therefore, compared to the method of Non-Patent Document 2, it is possible to efficiently search a solution space in which no broad or the like occurs.

The first configuration of the present invention relating to the local traffic volume prediction device is a set of a plurality of intersections {i: i∈V _c )} and each road from the intersection i to the intersection j (i, j∈V _c ). In a road network consisting of a set {(i, j)}, a specific prediction target road (k) based on the traffic volume v _{i, j} (t) measured on each road (i, j) at each time t. , L) is a local traffic prediction device for predicting future traffic,
Prediction program storage means in which individual data of a local traffic prediction program generated by the local traffic prediction program generation device according to claim 1 is stored on the basis of traffic measured in the past in the road network;
At each time t ₀ -m (m = 0, 1,..., M) before the current time t ₀ , the traffic volume measurement values v _{i, j} (t ₀ -m) on each road (i, j) are stored. Traffic data storage means,
By executing the local traffic volume prediction program stored in the prediction program storage means and inputting the traffic data before the current time t ₀ stored in the traffic data storage means, the prediction target road (k, prediction program execution means for calculating a predicted value v ^ _{k, l} (t ₀ + n) of traffic volume at time t ₀ + n (n = 1,..., N _t ) prior to the current time in l);
It is provided with.

According to a second configuration of the present invention relating to a local traffic volume prediction device, the prediction program storage means includes a plurality of environmental conditions that affect traffic based on traffic volume measured in the past in the road network. A plurality of local traffic volume prediction programs generated by the local traffic volume prediction program generation device according to claim 1 are stored in association with the environmental conditions,
Environmental condition input means for inputting the environmental condition at the current time t ₀ for predicting local traffic volume;
Program selection means for searching and extracting a local traffic volume prediction program stored in the prediction program storage means corresponding to the environmental conditions input by the environmental condition input means;
With
The prediction program execution means executes the local traffic volume prediction program selected by the program selection means, and inputs the traffic volume data before the current time t ₀ stored in the traffic volume data storage means. A time t ₀ + n (n = 1,..., N _t : N _t is a predicted time range) ahead of the current time t ₀ on the prediction target road (k, l) is calculated.

  The configuration of the present invention relating to the local traffic volume prediction program generation program is characterized by causing the computer to function as the local traffic volume prediction program generation device of the first configuration by being read and executed by a computer.

  The configuration of the present invention relating to the local traffic volume prediction program is characterized by causing the computer to function as the local traffic volume prediction apparatus having the second or third configuration by being read and executed by a computer.

The configuration of the present invention relating to the method for generating a local traffic volume prediction program is that an intersection node n _i corresponding to each intersection i (i∈V _c : V _c is a set of all intersections), an intersection i to an intersection j (i, j∈) V _c ), road links l _{i, j} corresponding to each road (i, j) heading to V _c ), and traffic v _{i, j} (t) measured on the road (i, j) at each time t Traffic at a predicted time t ₀ + n (n = 1,..., N _t : N _t is a predicted time range) after a certain time t ₀ of a specific road link l _{k, l} A local traffic prediction program generating method for automatically generating a local traffic prediction program for calculating a predicted value v ^ _{k, l} (t ₀ + n) of a quantity v _{k, l} (t ₀ + n),
A traffic volume v _{i, j} (t ₀ −m) (t ₀ -m) of the road link l _{i, j} before the time t ₀ is provided corresponding to each road link l _{i, j} (i, j∈V _c ). m = 0, 1,..., M: M is a traffic volume time range used for prediction) It is determined to which of a plurality of classes it belongs according to its size, and transition is made according to each class of the determined traffic volume. A decision node that branches the destination, and
Corresponds to all or part of the predicted values v ^ _{k, l} (t ₀ + n) of the predicted times t ₀ + n of the specific road link l _{k, l} (k, lεV _c ) to be predicted. A processing node provided,
A start node that is the start position of the transition, and
One directional transition branch from the start node to any one of the determination nodes, N _c from the determination node to any one of the determination nodes or any of the processing nodes according to each class of the traffic volume A directional transition branch of the book (N _c is the number of traffic classes) and a directional transition branch from each processing node to the determination node;
In the configuration, each processing node, the individual multiple storing individual data of the local traffic prediction program the determination node and identification label on the total number N _v number of each node of each starting node is represented as a labeled directed graph Data storage means;
Traffic data storage means for storing the traffic v _{i, j} (t) measured at each time t on each road (i, j);
In a computer with
An initial individual configuration step of generating a predetermined number N ₀ of individual data of the local traffic prediction program of the initial generation and storing it in the individual data storage means;
A fitness calculation step for calculating the fitness F of the individual data for each of the individual data stored in the individual data storage means by a fitness function of (Equation 1);
An individual selection step of selecting a predetermined number N _{1 from the} individual data based on the fitness F, and the rest of the individual data;
Wherein for each of the selected predetermined number N ₁ pieces of said individual data, and solid reproduction step of reproducing the individual data in response to the matching degree F of the individual data,
Among the regenerated individual data, crossover processing is performed on the number of nodes in proportion to the crossover probability of the two individual data selected by a predetermined method and the directional transition branch starting from this, and new solid data is obtained. Generating a new generation of the individual data and storing it in the individual data storage means,
Among the regenerated individual data, a connection mutation process is performed on the number of directed transition branches in proportion to the connection mutation probability in a predetermined number of individual data to generate new solid data to generate a new generation of the data A connection mutation step for storing in the individual data storage means as individual data;
Among the regenerated individual data, node mutation processing is performed on a number of nodes in proportion to the node mutation probability in a predetermined number of individual data, and new solid data is generated to generate a new generation of the individual data. A node mutation step to be stored in the individual data storage means;
An end determination step for determining whether or not a predetermined end condition is satisfied after the process of generating the individual data of a new generation is performed in the crossover step, the connection mutation step, and the node mutation step;
When it is determined in the termination determination step that the termination condition is satisfied, the fitness F of each individual data is calculated by executing the fitness calculation step, and the individual data with the smallest calculated fitness is A suboptimal solution selection step that is output as a typical local traffic volume prediction program;
Have
In the crossover step, two individual data are converted into a parent-child data 1 and a parent-individual data from a group of individual data consisting of a set of the individual data having _Nv number of nodes with identification labels. Select as 2,
Selecting two or more nodes from the respective nodes of the parent individual data 1 by random numbers;
The crossover process is performed by exchanging the selected node of the parent individual data 1 and the node of the parent individual data 2 having the same identification label as the node selected by the parent individual node 1 And
In the connection mutation step, the directional transition branch for changing the connection is selected from the directional transition branches of the individual data to be subjected to the connection mutation process,
From each node of the individual data other than the node to which the selected directional transition branch selected in the individual data is connected, a new connection destination node is selected by a random number,
The connection mutation process is performed by changing the transition destination of the selected directional transition branch to the selected new connection destination node.
In the node mutation step, a change node for changing the content by a random number is selected from the nodes other than the start node of the individual data to be subjected to the connection mutation process,
If the change node is the determination node, the node is changed to another type of the determination node, and if the change node is the processing node, the node mutation process is performed by changing to another type of the processing node. Is,
When it is determined in the termination determination step that the termination condition is not satisfied, the fitness calculation step, the individual selection step, the crossover step, the connection mutation step, the node mutation step, and the termination determination The step is repeatedly executed for each individual data stored in the individual data storage means.

The configuration of the present invention related to the local traffic volume prediction method includes a set of a plurality of intersections {i: i∈V _c )} and a set of roads {(i, j∈V _c ) from the intersection i {( i, j)} and a specific prediction target road (k, l) based on the traffic volume v _{i, j} (t) measured at each road (i, j) at each time t. A local traffic prediction method for predicting future traffic in Japan,
Prediction program storage means in which individual data of a local traffic prediction program generated by the local traffic prediction program generation device according to claim 1 is stored on the basis of traffic measured in the past in the road network;
At each time t ₀ -m (m = 0, 1,..., M) before the current time t ₀ , the traffic volume measurement values v _{i, j} (t ₀ -m) on each road (i, j) are stored. Traffic data storage means,
In a computer with
By executing the local traffic volume prediction program stored in the prediction program storage means and inputting the traffic data before the current time t ₀ stored in the traffic data storage means, the prediction target road (k, It is characterized in that a predicted value v ^ _{k, l} (t ₀ + n) of traffic volume at a time t ₀ + n (n = 1,..., N _t ) prior to the current time _{in l} ) is calculated.

  As described above, according to the present invention, when predicting the local traffic volume of a road network using an evolutionary algorithm, it is easily converged even in a complex road network, and a more accurate sub-optimal traffic volume is obtained. It is possible to make a prediction.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

  FIG. 1 is a diagram illustrating a hardware configuration of a local traffic volume prediction program generation device 1 according to the first embodiment of the present invention. The local traffic volume prediction program generation device 1 includes an input device 2, a CPU 3, an output device 4, and a storage device 5. These are the same as the configuration of a general computer, but the local traffic volume prediction program generation device 1 of the present invention is stored in the storage device 5 as a program. At the time of execution, this program is loaded onto the CPU and executed. Thus, the computer is caused to function as the local traffic volume prediction program generation device 1.

  FIG. 2 is a diagram illustrating a functional configuration of the local traffic volume prediction program generation device 1 according to the first embodiment of the present invention. The local traffic volume prediction program generating apparatus 1 includes a road network storage unit 11, a traffic volume data storage unit 12, an individual data storage unit 13, an initial individual component unit 14, a fitness calculation unit 15, an individual cage unit 16, and a solid reproduction unit 17. A crossover means 18, a connection mutation means 19, a node mutation means 20, an end determination means 21, and a sub-optimal solution selection means 22.

The road network storage means 11 stores a road network database. The road network database is a database that holds information on prediction target roads for predicting traffic volume, roads around the prediction target roads, and their connection relations. In general, the road network includes a set of intersections i {i | i∈V _c : V _c is a set of all intersections} and a road (i, j) from the intersection i to the intersection j (i, j∈V _c ). It is represented by the set {(i, j) | i, jεV _c }. The road (i, j) has a direction. In the road network database, each intersection i is stored as an intersection node n _i and each road (i, j) is stored as a road link l _{i, j} in the form of a directed graph.

For example, if the road network is as shown in FIG. 3, the road network database stored in the road network storage means 11 is a table as shown in FIG. In FIG. _{4, l i,} the road link _{j l i,} an identification code identifying the _j (ID) is assigned an identification code identifying the intersection node _{n i} is the _{n i} (ID) is assigned.

The traffic volume data storage means 12 stores the traffic volume v _{i, j} (t) measured at each past time t of each road link l _{i, j} as a traffic volume database.

Here, “traffic volume” refers to the traffic status of the road link, such as the average time it takes for the vehicle to move on the road link, the average speed of the vehicle moving on the road link, and the number of vehicles existing on the road link. The quantity to be characterized. The time t represents a time zone having a certain time width (for example, 10 minutes). The traffic volume v _{i, j} (t) is a value obtained by quantizing the traffic volume measured in the time zone into several classes. In this embodiment, it is assumed that the traffic volume of each road link is quantized into _Nc- stage classes.

For example, for a road network as shown in FIG. 3, the traffic volume v _{i, j} (t) is stored as a table as shown in FIG.

Moreover, the expression method of the traffic is as shown in FIG. When the maximum value of traffic volume on each road link is v _Nc and the minimum value is v ₀ , the traffic volume is discretized in advance using threshold values such as v ₁ and v ₂ , and the traffic volume is changed from V ₁ to V _Nc N _Expressed as one value of class _c . Therefore, the value of the traffic volume v _{i, j} (t) at each time of each road link and the predicted value thereof are represented by one of the values V ₁ to V _Nc (V _a ). . The number of configuration classes in discretization is assumed that the predetermined threshold value for setting the V _a may be different by the road link.

  The individual data storage means 13 stores individual data of the local traffic volume prediction program in a gene format. Details of the data format of the individual data will be described later.

Here, the “individual data” of the local traffic volume prediction program is data representing a program for predicting the traffic volume on a specific road link in the road network, and includes one start node, a plurality of determination nodes, and a plurality of processes. Expressed as a directed graph consisting of a node, a directed transition branch connecting each node (start node, decision node, and processing node; the same applies hereinafter) and an identification label attached to each node. . In this embodiment, the total number of nodes (processing nodes, determination nodes, and start nodes) is N _v .

The “judgment node” is a node provided corresponding to each road link l _{i, j} (i, j∈V _c ), and before the time t _{0 of the} road link l _{i, j.} Of traffic v _{i, j} (t ₀ -m) (m = 0, 1,..., M: time range of traffic used for prediction) at time t ₀ -m It is a node that determines which class it belongs to and branches the transition destination according to each class of the determined traffic volume. Time t ₀ represents the time (the current time) at the time to make predictions. “Processing node” means each predicted value v ^ _{k, l} (t ₀ ) at each predicted time t ₀ + n of a specific road link l _{k, l} (k, lεV _c ) to be predicted. + N) is a node provided corresponding to all or part of the above. The “Start node” is a node that represents the start position of the transition when the local traffic volume prediction program is executed. “Directed transition branch” refers to a branch representing a transition from one node to another node. One directional transition branch from the start node to any one of the determination nodes, and the traffic volume from the determination node. _Nc directional transition branches ( _Nc is the number of traffic volume classes) heading to any decision node or any processing node according to each class, and directed from each processing node to the decision node There are transition branches. The "identification label" is a label attached to allow for the identification N _v number of each node. Details of the determination node and the processing node will be described later.

The initial individual constructing unit 14 generates a predetermined number N ₀ of individual data of the local traffic prediction program of the initial generation in genetic network programming (GNP) and stores it in the individual data storage unit 13. Details of GNP will be described later.

  The fitness calculation unit 15 performs a process of calculating the fitness F of the individual data by using the fitness function of the equation (1) for each individual data stored in the individual data storage unit 13.

Individual selection unit 16, among the individual data, selects _one predetermined number N from those fit F is small, the rest performs a process of selection.

Solid reproducing unit 17, for each of the selected predetermined number N ₁ pieces of individual data, performs a process of reproducing the individual data in accordance with the fitness F of the individual data.

  The crossover means 18 performs a crossover process on the number of nodes in proportion to the crossover probability of two pieces of individual data selected by a predetermined method and the directional transition branch starting from this. To generate new solid data. These new solid data are stored in the individual data storage means 13 as individual data of a new generation.

  The connection mutation means 19 performs connection mutation processing on the number of directional transition branches in proportion to the connection mutation probability in a predetermined number of individual data among the regenerated individual data, and generates new solid data. To do. These new solid data are stored in the individual data storage means 13 as individual data of a new generation.

  The node mutation means 20 performs node mutation processing on the number of nodes in proportion to the node mutation probability among a predetermined number of individual data among the reproduced individual data, and generates new solid data. These new solid data are stored in the individual data storage means 13 as the individual data of a new generation.

  The end determination means 21 determines whether or not a predetermined end condition is satisfied after the process of generating the individual data of the new generation is performed by the crossover means 18, the connection mutation means 19 and the node mutation means 20. To do. Here, various conditions can be considered as the “predetermined end condition”. For example, when an evolutionary generation reaches a certain generation, it may be determined that “the end condition is satisfied”.

  The semi-optimal solution selection means 22 calculates the fitness F of each individual data by the fitness calculation means 15 when it is determined by the termination determination means 21 that the termination condition is satisfied. Then, the individual data having the smallest calculated fitness is output to the output device 4 as a final local traffic volume prediction program.

  With respect to the local traffic volume prediction program generation apparatus 1 of the present embodiment configured as described above, the local traffic volume prediction program generation method will be described below.

(1) Outline As described above, the local traffic volume prediction program generation device 1 generates a local traffic volume prediction program by using genetic network programming (GNP).

For example, it is assumed that the specific road link (prediction target road link) to be predicted is lp _{, q} in FIG. p is the prediction target road link l _p, the start intersection node of _q, q is a prediction target road link l _p, the end point intersection node of _q. The directly connected intersection node in the intersection node p j _m, the intersection node to connect two sections in shortest intersection node p and i _n (see FIG. 6). FIG. 7 is an example using the traffic volume moving toward the intersection node p in the traffic volume prediction of the prediction target road links l _{p and q} . Similarly, the road links from the intersection node q to other intersection nodes _{Alternatively,} the traffic volume of the road link in the reverse direction on the same road expressed as l _{s, r} with respect to the road link l _{r, s} may be used.

In the local traffic volume prediction program generation device 1, when the traffic volume at the time t of the prediction target road link l _{p, q} is v _{p, q} (t) and the predicted value is v ^ _{p, q} (t), the time {v _{j, p} (t−m ₁ ) | j∈B (p)} and {v _{r, s} (t−m ₂ ) | r, s∈B (p)}, {V _{i, j} (t−m ₃ ) | i∈B (j), j∈B (p)}, {v _{r, s} (tm ₄ ) | r, s∈B (j), j∈ B ^ _{p, q} (t + n) is determined by genetic network programming (GNP) using B (p)} or the like. Here, m ₁ , m ₂ , m ₃ , m ₄ , n are positive integers, and B (k) represents a set of intersection nodes at the start point of the road link toward the intersection node k.

  The local traffic volume prediction program generation device 1 acquires, by learning, individual data (GNP individual) that is optimal for traffic volume prediction in a specific section, using past traffic volume data accumulated in the traffic volume data storage unit 12 as training data. . In actual traffic volume prediction, the individual data acquired by learning is used for the main process in the traffic volume prediction process, so that the traffic volume in a specific section is predicted and output from the input new traffic volume information. The optimum individual data is prepared in consideration of the environmental conditions such as weather and time zone, and can be selected and used for prediction. Can also be predicted using. Furthermore, it is possible to make a comprehensive use taking into account prediction results for neighboring sections.

(2) Individual data As described above, the individual data of the local traffic volume prediction program refers to one start node, a plurality of determination nodes, a plurality of processing nodes, and each node (a start node, a determination node, and a processing node). (Same)) expressed as a directed graph composed of directional transition branches connecting each other and identification labels attached to the respective nodes.

The determination node is expressed as shown in FIG. The first neighboring road link determination nodes (FIGS. 8A and 8B) are road links (first neighboring roads) whose intersection node on the end point side is the intersection node at the start point of the prediction target road link l _{p, q.} Link) l _{j, p} (j∈B (p)) (or l _{r, s} (r, s∈B (p))) which traffic volume at time t−m ₁ (or t−m ₂ ) It is a node that determines whether it belongs to a class (note that road links and intersection nodes are links and nodes in the road network, and that the determination node is a node in a directed graph representing individual data). The secondary neighboring road link determination nodes (FIGS. 8 (c) and 8 (d)) are road links l _{i, j} whose end node is the starting node of the primary neighboring road link l _{j, p.} (IεB (j), jεB (p)) (or l _{r, s} (r, sεB (j), jεB (p))) at time t−m ₃ (or t− This is a node for determining to which class the traffic volume of m ₄ ) belongs. Each determination node determines to which class the traffic volume of the corresponding road link belongs based on the determination contents given in advance to the determination node. Specifically, for example, in the determination node of FIG. 8A _, the traffic volume v _{j, p} (t−m ₁ ) of the corresponding road link l _{j, p} is any of the classes from V ₁ to V _Nc. To determine whether it is included. Further, a transition is performed using a directed transition branch that is a branch destination of a determination node corresponding to these classes. Each determination node includes a parameter m related to the magnitude of the past time from the prediction time t, and is updated to an optimal value through an evolution process.

On the other hand, the processing node is represented as shown in FIG. In the processing node, the value V _a of v ^ _{p, q} (t + n) is output as a predicted value. Each processing node includes a parameter n relating to a future time size from the prediction time t, and is updated to an optimal value through an evolution process.

In the individual data, one directional transition branch comes out from the start node, and transitions to any one of the determination nodes. From each determination node, there are _Nc directed transition branches corresponding to each class of traffic volume, and each transitions to any other determination node or any processing node. One directional transition branch comes out from each processing node, and transitions to any one of the determination nodes.

FIG. 10 is a diagram illustrating an example of one piece of individual data. The individual data has a directed graph structure in which the start node S, the determination nodes J _{1 to} J ₅ , and the processing nodes P _{1 to} P ₃ are connected by directed transition branches. FIG. 10 is simplified for the sake of explanation. In practice, one piece of individual data is composed of one start node, 50 to 100 decision nodes, and about 10 processing nodes. Each node corresponds to a “unit (fragment)” of the program. When the traffic volume is predicted, a transition (execution of calculation) is started from the start node S. The determination node J _i branches depending on the determination contents and transitions to the next node (that is, shifts to execution of the next program unit). Further, the processing node P _j has one connection destination, and transitions to that node after processing. Therefore, until the transition end condition is satisfied from the start node, work on the requested task is continued while transitioning between the determination node J _i and the processing node P _j . In the present embodiment, the determination node continues to determine traffic information and the processing node continues to predict future traffic.

  As can be seen from FIG. 10, when the transition is continued according to the directed graph representing the individual data, an infinite loop is formed. Therefore, the processing is terminated under certain conditions (for example, whether the number of transitions and the number of processing nodes that have passed have reached a predetermined value).

(3) Local Traffic Prediction Program Generation Processing FIG. 11 is a flowchart showing the operation of the local traffic prediction program generation device 1 according to this embodiment.

(Step S1) Generation of Initial Generation Individual Data First, in step S1, the initial individual configuration means 14 generates initial generation individual data by random numbers and stores them in the individual data storage means 13 as gene-format data.

FIG. 12 shows a data structure of individual data in one gene format as an example. For example, in the case of individual data represented by a directed graph as shown in FIG. 12A, the individual data in the gene format is as shown in FIG. In FIG. 12A, each node is given an identification label (0 to 4). At this time, the identification label of the start node S is always 0. Hereinafter, each node is called by this identification label. For example, the processing node P _{1 with} the identification label 1 is called “node 1”, and the determination node J _{1 with the} identification label 2 is called “node 2”. As shown in FIG. 12B, the individual data in the gene format is composed of a data set table arranged in the order of the identification labels of the respective nodes. The data structure of the data set of each node is as shown in FIG. First, the top three numbers K _i , ID _i , and v _i are the node i type number (0: start node, 1: determination node, 2: processing node), and the road link corresponding to node i, respectively. The identification number represents the output value of the node i (in this case, only the processing node has a quantized traffic value as the output value. Other than the processing node is 0). These three numerical sets are called “node genes”. The numerical value set following the node gene represents the identification label of the node to which the directed transition branch starting from this node i is connected. These are called “connection genes”. A set of node genes and connection genes corresponding to each road link is stored in a gene library as shown in FIG. 12D, and is read out from this gene library and used as necessary. The gene library is stored in the individual data storage means 13 in FIG.

  In the local traffic volume prediction program generation processing using normal GNP, individual data (GNP individual) as shown in FIG. 10 is randomly generated for about 200 to 500 individuals in one generation. Therefore, individual individual data differs in connection destination, determination / processing contents, and parameters of each node.

(Step S2) Evaluation of Individual Data In step S2, the fitness calculator 15 evaluates and ranks the generated individual data by calculating the fitness F of Expression (1). For the calculation of the fitness, the predicted value of the traffic volume at each future time of the prediction target road link output from the processing node is obtained by executing a program by making a transition from the start node for each individual data. Then, based on the predicted value of the traffic volume and the measured value of the traffic volume stored at the same time stored in the traffic volume data storage unit 12, the fitness F of the equation (1) is sequentially calculated. In order to avoid an infinite loop, the transition is aborted when the number of processing nodes that have passed in the transition process reaches a predetermined number. Thereby, the value of the fitness F is determined.

FIG. 13 shows an example of program execution by state transition. In FIG. 13, a part of the individual data is extracted and displayed. In this individual, it is assumed that a transition occurs as indicated by a thick line arrow in FIG. In this case, the past traffic volume data (v _2,8 (t-3) = V ₃ ∧v _3,9 (t-2) = V ₃ ∧v _2,7 (t-2) = V ₄ ∧v _{4, 8} (t−1) = V ₃ ), it can be considered that v ^ _1,2 (t + 2) = V ₄ is predicted. Further, if a transition as indicated by a double line arrow, past traffic amount data _{(v 2,8 (t-3)} = V 3 ∧v 3,9 (t-2) = V 3 ∧v _{2, 7} (t−2) = V ₃ ∧v _5,6 (t−1) = V ₃ ), it can be considered that v ^ _1,2 (t + 3) = V ₃ is predicted. As can be seen from these examples, it can be considered that the transition of the connection of a plurality of decision nodes and the prediction at the processing node are one unit of work. Further, even after the prediction at the processing node, the transition toward the next prediction can be continued.

(Step S3) Individual data trapping In step S3, the individual trapping means 16 performs trapping of individual data stored in the individual data storage means 13. This wrinkle processing is performed based on the fitness F. For example, fitness F is selecting a predetermined number N ₁ pieces of individual data from the small, A number discard the rest (culling) process and, from the individual data groups stored in the individual data storage means 13, randomly There is a method (tournament selection) of selecting individual data of (about 10) and selecting the individual data having the smallest fitness F among them by repeating N ₁ times. Normally, _{N 1} is about 100 to 300 pieces in a single generation.

(Step S4) Reproduction of individual data In step S4, the solid reproduction means 17 reproduces individual data for each of the selected N ₁ individual data according to the fitness F of the individual data. The relationship between the fitness F and the number of individuals F to be reproduced in this reproduction process is not particularly limited. For example, the smaller the fitness F, the larger the number of individuals may be reproduced. The reproduced individual data is stored in the individual data storage means 13 as next generation individual data. Individual data that is the basis of the reproduction is discarded.

(Step S5) Crossover / mutation process In step S5, the crossover process by the crossover means 18, the connection mutation process by the connection mutation means 19, and the node mutation means 20 are performed on the generated next generation individual data group. Node mutation processing is performed. In this case, the next-generation individual data group includes an individual data group that does not perform crossover and mutation, an individual data group that performs crossover, an individual data group that performs connection mutation, and an individual that performs node mutation at a predetermined ratio. Divided into data groups, crossover and mutation processes are performed in each individual data group.

(A) Crossover process In the crossover process, first, the crossover means 18 selects two pieces of individual data as parent-child body data 1 and parent individual data 2 from the next-generation individual data group using random numbers. Next, two or more nodes are selected by random numbers from each node of the parent individual data 1. Then, the crossover process is completed by exchanging the node of the selected parent individual data 1 and the node of the parent individual data 2 with the same identification label as the node selected in the parent individual data 1.

FIG. 14 shows an example of the mutation process. In FIG. 14, the nodes J ₂ and P ₃ of the parent individual data 1 (parent 1) are exchanged with the nodes J ₇ and P _{6 of} the parent individual data 2 (parent 2), thereby performing crossover.

(B) Connection Mutation Processing In connection mutation processing, first, the connection mutation means 19 connects each next-generation individual data to be subjected to connection mutation processing from each directed transition branch in the individual data. Select the directional transition branch to be changed. There may be one or more directed transition branches for changing the connection. For example, you may make it select the directional transition branch which changes a connection stochastically about each directional transition branch in the said individual data. Next, a new connection destination node is selected from each node in the individual data other than the node to which the selected directional transition branch in the individual data is connected. Then, the connection mutation process is completed by changing the transition destination of the selected directional transition branch to the selected new connection destination node.

FIG. 15 shows an example of connection mutation processing. In this example, (shown by a thick line arrow) directed transition branch from decision node J ₃ parent individuals data to the determination node J ₄ has been selected in the directed transition branches to change the connection. Then, processing node P ₃ are selected as a new connection destination of the effective transition branch.

(C) Node mutation processing The node mutation processing is performed by using random numbers from the nodes other than the start node S for each next-generation individual data for which the node mutation means 20 performs node mutation processing. Select the change node to change. There may be one or more change nodes for changing the contents. For example, you may make it select the change node which changes the content stochastically about each node in the said individual data. Then, if the change node is a determination node, the node mutation process is completed by changing to another type of determination node, and if the change node is a processing node, changing to another type of processing node.

FIG. 16 shows an example of node mutation processing. In this example, the determination node J ₁ and processing node P ₃ parent individuals data is selected to change the node. Then, the determination node J ₁ as a new kind of change nodes in determining a node J _6, processing node P ₃ is changed to processing node P _4.

(Step S6) Determination of End Condition In step S6, the end determination means 21 determines whether or not a predetermined end condition is satisfied. In this case, various conditions can be considered as the “end condition”. For example, whether or not the time elapsed since the start of GNP processing has reached a predetermined time, whether or not the number of loop iterations has reached a predetermined number, and the minimum fitness F among the fitness F of each individual data It can be considered whether or not a certain threshold value or less.

  If the end condition is not satisfied, the process returns to step S2. At this time, in order to take in information by a wider range of data, the process of steps S2 to S6 is repeated after the reference time t for performing traffic prediction is advanced by 1.

  On the other hand, if the end condition is satisfied, the process proceeds to the next step S7.

(Step S7) Evaluation of Individual Data In step S7, the fitness calculation means 15 performs step S2 on each individual data of the generation stored in the individual data storage means 13 when the end condition is satisfied. In the same manner as described above, evaluation and ranking are performed by calculating the fitness F of Equation (1).

(Step S8) Sub-optimal individual data selection In step S8, the sub-optimal solution selection means 22 uses the individual data with the smallest fitness F calculated in step S7 as the final local traffic volume prediction program as an output device. 4 is output.

  As described above, according to the local traffic volume prediction program generating apparatus 1 of the present embodiment, the genetic network programming method is applied to the creation of the traffic volume prediction program, and the increase or decrease in the number of nodes of individual data used for prediction is increased. By eliminating it, it is possible to easily converge even in a complex road network and to create a sub-optimal traffic prediction program with higher accuracy than the conventional method. Further, the finally obtained traffic volume prediction program obtains a plurality of predicted values of the future traffic volume on the prediction target road link at the time t when the traffic volume prediction is performed at times t + 1, t + 2,. Therefore, it is possible to obtain the predicted occurrence time of traffic jam and the expected time of traffic jam elimination on the road link. In addition, when a traffic volume prediction program is created by the local traffic volume prediction program generation device 1 at the same time using a plurality of road links as prediction target road links, the prediction results of the created traffic volume prediction program can be integrated. It is possible to predict the temporal change in traffic volume of a dynamic road network.

  FIG. 17 is a diagram illustrating a functional configuration of the local traffic volume prediction device 30 according to the second embodiment of the present invention. The local traffic volume prediction device 30 of this embodiment includes a traffic volume data input device 31, a traffic volume data storage device 32, a traffic volume data storage means 33, a local traffic volume prediction program generation apparatus 34, a prediction program storage means 35, and environmental conditions. Input means 36, program selection means 37, prediction program execution means 38, and local traffic volume prediction display device 39 are provided.

  The traffic data input device 31 is a device that inputs traffic data of each road link at the current time in real time. As the traffic volume data input device 31, for example, traffic volume data of each road link transmitted from a traffic volume sensor installed on the road via a line is received, and the traffic volume is quantized with a predetermined threshold value. Thereafter, a device that outputs the traffic data at the current time t of each road link is used.

  The traffic volume data storage device 32 is a device for storing / accumulating the traffic volume data input by the traffic volume data input device 31 in the traffic volume data storage means 33 at each time. The accumulated traffic volume data is used as past traffic volume data for generation of a traffic volume prediction program and prediction of future traffic volume.

  As the local traffic volume prediction program generation device 34, the local traffic volume prediction program generation device 1 described in the first embodiment is used.

  The prediction program storage means 35 stores the local traffic volume prediction program created by the local traffic volume prediction program generation device 34 in a linked manner with the environmental conditions of the traffic volume data used for the prediction.

  The environmental condition input means 36 is an apparatus for inputting environmental conditions at a time when traffic volume is predicted. Here, “environmental condition” refers to an environmental condition that affects traffic, for example, conditions such as day of the week, time of day, presence of event, occurrence of traffic accident, and the like.

  The program selection unit 37 searches for and extracts a local traffic volume prediction program stored in the prediction program storage unit 35 corresponding to the environmental condition input by the environmental condition input unit 36.

The prediction program execution means 38 executes the local traffic volume prediction program selected by the program selection means 37, and inputs the traffic volume data before the current time t stored in the traffic volume data storage means 33, so that the prediction target Time t + n (n = 1,..., N _t : N _t is a predicted time range) ahead of the current time t on the road (p, q) is calculated.

  The local traffic volume prediction display device 39 displays the traffic volume prediction on the prediction target road calculated by the prediction program execution means 38. As the local traffic volume prediction display device 39, for example, a traffic information display board installed on a road, a display screen of a car navigation system, or the like is used.

  Thus, in the local traffic volume prediction apparatus 30 of the present embodiment, the traffic volume prediction is performed using the traffic volume prediction program created by the local traffic volume prediction program generation apparatus 1 of the first embodiment. Since a plurality of predicted values of the future traffic volume on the prediction target road link at the time t to be obtained are obtained for the times t + 1, t + 2,..., T + n, the traffic jam occurrence predicted time and the traffic jam resolution predicted time are also obtained. Obtainable. In addition, when a traffic volume prediction program is created by the local traffic volume prediction program generation device 1 at the same time using a plurality of road links as prediction target road links, the prediction results of the created traffic volume prediction program can be integrated. It is possible to predict the temporal change in traffic volume of a dynamic road network.

It is a figure showing the hardware constitutions of the local traffic prediction program production | generation apparatus which concerns on Example 1 of this invention. It is a figure showing the function structure of the local traffic prediction program production | generation apparatus which concerns on Example 1 of this invention. It is a figure which shows an example of a road network. FIG. 4 is a diagram illustrating a road network database table for the road network of FIG. 3. It is a figure showing the traffic volume database with respect to the road network of FIG. It is a figure explaining the local traffic prediction method. This is an example using the traffic volume moving toward the intersection node p in the traffic volume prediction of the prediction target road links l _{p, q} . It is a figure showing the determination node in individual data. It is a figure showing the processing node in individual data. It is a figure which shows the example of one individual data. It is a flowchart showing operation | movement of the local traffic prediction program production | generation apparatus 1 of a present Example. It is an example of the data structure of the individual data of a gene format. It is a figure which shows the example of execution of the program by a state transition. It is a figure which shows an example of a mutation process. It is a figure which shows an example of a connection mutation process. It is a figure which shows an example of a node mutation process. It is a figure showing the function structure of the local traffic prediction apparatus 30 which concerns on Example 2 of this invention. It is a figure showing an example of the finite state machine used in patent document 2, and the structure of the gene.

Explanation of symbols

1 Local Traffic Prediction Program Generation Device 2 Input Device 3 CPU
4 Output device 5 Storage device 11 Road network storage means 12 Traffic volume data storage means 13 Individual data storage means 14 Initial individual construction means 15 Fitness calculation means 16 Individual trap means 17 Solid regeneration means 18 Crossover means 19 Connection mutation means 20 Node Mutation means 21 Termination judgment means 22 Sub-optimal solution selection means 30 Local traffic volume prediction device 31 Traffic volume data input device 32 Traffic volume data storage device 33 Traffic volume data storage means 34 Local traffic volume prediction program generation device 35 Prediction program storage means 36 environmental condition input means 37 program selection means 38 prediction program execution means 39 local traffic volume prediction display device

Claims (7)

Intersection node n _i corresponding to each intersection i (i∈V _c : V _c is a set of all intersections), corresponding to each road (i, j) from intersection i to intersection j (i, j∈V _c ) The specific road link l based on the road link l _{i, j} and the road network database provided with the traffic volume v _{i, j} (t) measured on the road (i, j) at each time t. Predicted value v ^ of traffic volume v _{k, l} (t ₀ + n) at predicted time t ₀ + n (n = 1,..., N _t : N _t is a predicted time range) after time t ₀ at which _{k and l} are present. _A local traffic volume prediction program generating device for automatically generating a local traffic volume prediction program for calculating _{k, l} (t ₀ + n),
A traffic volume v _{i, j} (t ₀ −m) (t ₀ -m) of the road link l _{i, j} before the time t ₀ is provided corresponding to each road link l _{i, j} (i, j∈V _c ). m = 0, 1,..., M: M is a traffic volume time range used for prediction) It is determined to which of a plurality of classes it belongs according to its size, and transition is made according to each class of the determined traffic volume. A decision node that branches the destination, and
Corresponds to all or part of the predicted values v ^ _{k, l} (t ₀ + n) of the predicted times t ₀ + n of the specific road link l _{k, l} (k, lεV _c ) to be predicted. A processing node provided,
A start node that is the start position of the transition, and
One directional transition branch from the start node to any one of the determination nodes, N _c from the determination node to any one of the determination nodes or any of the processing nodes according to each class of the traffic volume A directional transition branch of the book (N _c is the number of traffic classes) and a directional transition branch from each processing node to the determination node;
In the configuration, each processing node, the individual multiple storing individual data of the local traffic prediction program the determination node and identification label on the total number N _v number of each node of each starting node is represented as a labeled directed graph Data storage means;
Traffic data storage means for storing the traffic v _{i, j} (t) measured at each time t on each road (i, j);
An initial individual constructing means for generating a predetermined number N ₀ of individual data of the local traffic prediction program of the initial generation and storing it in the individual data storage means;
For each of the individual data stored in the individual data storage means, fitness calculation means for calculating the fitness F of the individual data by the fitness function of (Equation 1);
Among the individual data, a predetermined number N ₁ is selected based on the fitness F, and the rest of the individual data means for hesitation,
Wherein for each of the selected predetermined number N ₁ pieces of said individual data, and a solid reproducing means for reproducing the individual data in response to the matching degree F of the individual data,
Among the regenerated individual data, crossover processing is performed on the number of nodes in proportion to the crossover probability of the two individual data selected by a predetermined method and the directional transition branch starting from this, and new solid data is obtained. A crossover means for generating and storing in the individual data storage means as the individual data of a new generation;
Among the regenerated individual data, a connection mutation process is performed on the number of directed transition branches in proportion to the connection mutation probability in a predetermined number of individual data to generate new solid data to generate a new generation of the data A connection mutation means for storing in the individual data storage means as individual data;
Among the regenerated individual data, node mutation processing is performed on a number of nodes in proportion to the node mutation probability in a predetermined number of individual data, and new solid data is generated to generate a new generation of the individual data. Node mutation means for storing in the individual data storage means;
An end determination means for determining whether or not a predetermined end condition is satisfied after the process of generating the individual data of a new generation is performed by the crossover means, the connection mutation means, and the node mutation means;
When it is determined that the termination condition is satisfied by the termination determination unit, the fitness level F of the individual data is calculated by the fitness level calculation unit, and the individual data having the smallest calculated fitness level is used as the final local traffic. A sub-optimal solution selection means for outputting as a quantity prediction program;
With
The crossover means converts two pieces of individual data into a parent-child data 1 and a parent individual data 2 by random numbers from the individual data group consisting of the set of individual data having the _Nv number of nodes with identification labels. Select as
Selecting two or more nodes from the respective nodes of the parent individual data 1 by random numbers;
The crossover process is performed by exchanging the selected node of the parent individual data 1 and the node of the parent individual data 2 having the same identification label as the node selected by the parent individual node 1 And
The connection mutation means selects the directional transition branch for changing the connection from the directional transition branches of the individual data to be subjected to the connection mutation process,
From each node of the individual data other than the node to which the selected directional transition branch selected in the individual data is connected, a new connection destination node is selected by a random number,
The connection mutation process is performed by changing the transition destination of the selected directional transition branch to the selected new connection destination node.
The node mutation means selects a change node whose contents are changed by a random number from the nodes other than the start node of the individual data to be subjected to the connection mutation process,
If the change node is the determination node, the node is changed to another type of the determination node, and if the change node is the processing node, the node mutation process is performed by changing to another type of the processing node. Is,
When it is determined by the termination determination means that the termination condition is not satisfied, the fitness calculation means, the individual selection means, the crossover means, the connection mutation means, the node mutation means, and the termination determination The means repeatedly executes the same process for each individual data stored in the individual data storage means.

In a road network comprising a set of intersections {i: i∈V _c )} and a set of roads {(i, j)} from the intersection i to the intersection j (i, j∈V _c ), Based on the traffic volume v _{i, j} (t) measured on each road (i, j) at time t, the local traffic volume prediction for predicting the future traffic volume on the specific prediction target road (k, l). A device,
Prediction program storage means in which individual data of a local traffic prediction program generated by the local traffic prediction program generation device according to claim 1 is stored on the basis of traffic measured in the past in the road network;
At each time t ₀ -m (m = 0, 1,..., M) before the current time t ₀ , the traffic volume measurement values v _{i, j} (t ₀ -m) on each road (i, j) are stored. Traffic data storage means,
By executing the local traffic volume prediction program stored in the prediction program storage means and inputting the traffic data before the current time t ₀ stored in the traffic data storage means, the prediction target road (k, prediction program execution means for calculating a predicted value v ^ _{k, l} (t ₀ + n) of traffic volume at time t ₀ + n (n = 1,..., N _t ) prior to the current time in l);
A local traffic volume prediction device characterized by comprising: The said prediction program memory | storage means was produced | generated by the local traffic prediction program production | generation apparatus of Claim 1 on the basis of the traffic volume measured in the past in the said road network on the several environmental conditions which affect traffic. Multiple local traffic prediction programs are stored in association with the environmental conditions,
Environmental condition input means for inputting the environmental condition at the current time t ₀ for predicting local traffic volume;
Program selection means for searching and extracting a local traffic volume prediction program stored in the prediction program storage means corresponding to the environmental conditions input by the environmental condition input means;
With
The prediction program execution means executes the local traffic volume prediction program selected by the program selection means, and inputs the traffic volume data before the current time t ₀ stored in the traffic volume data storage means. The time t ₀ + n (n = 1,..., N _t : N _t is a range of the predicted time) ahead of the current time t ₀ on the prediction target road (k, l) is calculated. Local traffic prediction device.   A program that causes a computer to function as the local traffic volume prediction program generation device according to claim 1 by being read and executed by a computer.   A program that causes a computer to function as the local traffic prediction device according to claim 2 by being read and executed by a computer. Intersection node n _i corresponding to each intersection i (i∈V _c : V _c is a set of all intersections), corresponding to each road (i, j) from intersection i to intersection j (i, j∈V _c ) The specific road link l based on the road link l _{i, j} and the road network database provided with the traffic volume v _{i, j} (t) measured on the road (i, j) at each time t. Predicted value v ^ of traffic volume v _{k, l} (t ₀ + n) at predicted time t ₀ + n (n = 1,..., N _t : N _t is a predicted time range) after time t ₀ at which _{k and l} are present. _A local traffic prediction program generating method for automatically generating a local traffic prediction program for calculating _{k, l} (t ₀ + n),
A traffic volume v _{i, j} (t ₀ −m) (t ₀ -m) of the road link l _{i, j} before the time t ₀ is provided corresponding to each road link l _{i, j} (i, j∈V _c ). m = 0, 1,..., M: M is a traffic volume time range used for prediction) It is determined to which of a plurality of classes it belongs according to its size, and transition is made according to each class of the determined traffic volume. A decision node that branches the destination, and
Corresponds to all or part of the predicted values v ^ _{k, l} (t ₀ + n) of the predicted times t ₀ + n of the specific road link l _{k, l} (k, lεV _c ) to be predicted. A processing node provided,
A start node that is the start position of the transition, and
One directional transition branch from the start node to any one of the determination nodes, N _c from the determination node to any one of the determination nodes or any of the processing nodes according to each class of the traffic volume A directional transition branch of the book (N _c is the number of traffic classes) and a directional transition branch from each processing node to the determination node;
In the configuration, each processing node, the individual multiple storing individual data of the local traffic prediction program the determination node and identification label on the total number N _v number of each node of each starting node is represented as a labeled directed graph Data storage means;
Traffic data storage means for storing the traffic v _{i, j} (t) measured at each time t on each road (i, j);
In a computer with
An initial individual configuration step of generating a predetermined number N ₀ of individual data of the local traffic prediction program of the initial generation and storing it in the individual data storage means;
A fitness calculation step for calculating the fitness F of the individual data for each of the individual data stored in the individual data storage means by the fitness function of (Equation 2);
An individual selection step of selecting a predetermined number N _{1 from the} individual data based on the fitness F, and the rest of the individual data;
Wherein for each of the selected predetermined number N ₁ pieces of said individual data, and solid reproduction step of reproducing the individual data in response to the matching degree F of the individual data,
Among the regenerated individual data, crossover processing is performed on the number of nodes in proportion to the crossover probability of the two individual data selected by a predetermined method and the directional transition branch starting from this, and new solid data is obtained. Generating a new generation of the individual data and storing it in the individual data storage means,
Among the regenerated individual data, a connection mutation process is performed on the number of directed transition branches in proportion to the connection mutation probability in a predetermined number of individual data to generate new solid data to generate a new generation of the data A connection mutation step for storing in the individual data storage means as individual data;
Among the regenerated individual data, node mutation processing is performed on a number of nodes in proportion to the node mutation probability in a predetermined number of individual data, and new solid data is generated to generate a new generation of the individual data. A node mutation step to be stored in the individual data storage means;
An end determination step for determining whether or not a predetermined end condition is satisfied after the process of generating the individual data of a new generation is performed in the crossover step, the connection mutation step, and the node mutation step;
When it is determined in the termination determination step that the termination condition is satisfied, the fitness F of each individual data is calculated by executing the fitness calculation step, and the individual data with the smallest calculated fitness is A suboptimal solution selection step that is output as a typical local traffic volume prediction program;
Have
In the crossover step, two individual data are converted into a parent-child data 1 and a parent-individual data from a group of individual data consisting of a set of the individual data having _Nv number of nodes with identification labels. Select as 2,
Selecting two or more nodes from the respective nodes of the parent individual data 1 by random numbers;
The crossover process is performed by exchanging the selected node of the parent individual data 1 and the node of the parent individual data 2 having the same identification label as the node selected by the parent individual node 1 And
In the connection mutation step, the directional transition branch for changing the connection is selected from the directional transition branches of the individual data to be subjected to the connection mutation process,
From each node of the individual data other than the node to which the selected directional transition branch selected in the individual data is connected, a new connection destination node is selected by a random number,
The connection mutation process is performed by changing the transition destination of the selected directional transition branch to the selected new connection destination node.
In the node mutation step, a change node for changing the content by a random number is selected from the nodes other than the start node of the individual data to be subjected to the connection mutation process,
If the change node is the determination node, the node is changed to another type of the determination node, and if the change node is the processing node, the node mutation process is performed by changing to another type of the processing node. Is,
When it is determined in the termination determination step that the termination condition is not satisfied, the fitness calculation step, the individual selection step, the crossover step, the connection mutation step, the node mutation step, and the termination determination A method for generating a local traffic volume prediction program, wherein the step is repeatedly executed for each individual data stored in the individual data storage means.

In a road network comprising a set of intersections {i: i∈V _c )} and a set of roads {(i, j)} from the intersection i to the intersection j (i, j∈V _c ), Based on the traffic volume v _{i, j} (t) measured on each road (i, j) at time t, the local traffic volume prediction for predicting the future traffic volume on the specific prediction target road (k, l). A method,
Prediction program storage means in which individual data of a local traffic prediction program generated by the local traffic prediction program generation device according to claim 1 is stored on the basis of traffic measured in the past in the road network;
At each time t ₀ -m (m = 0, 1,..., M) before the current time t ₀ , the traffic volume measurement values v _{i, j} (t ₀ -m) on each road (i, j) are stored. Traffic data storage means,
In a computer with
By executing the local traffic volume prediction program stored in the prediction program storage means and inputting the traffic data before the current time t ₀ stored in the traffic data storage means, the prediction target road (k, l) Predicting traffic volume v ^ _{k, l} (t ₀ + n) of traffic volume at time t ₀ + n (n = 1,..., N _t ) ahead of current time in l) Method.

Images