JP7388242B2 - Road condition prediction weight parameter setting device, road condition prediction weight parameter setting method, and road condition prediction weight parameter setting program - Google Patents

Description

The present invention relates to a technique for creating weight parameters used for road condition prediction.

Conventionally, various road condition prediction methods have been disclosed. For example, the traffic jam prediction system disclosed in Patent Document 1 uses a traffic jam prediction model that is machine-learned using traffic information and information indicating whether a traffic jam has occurred as training data.

Such conventional congestion prediction systems set a congestion prediction model for each route.

JP 2019-169028 Publication

Conventional congestion prediction systems use a congestion prediction model for each route, so each time a route changes, a congestion prediction model must be set individually. For this reason, the weighting parameters used in the congestion prediction model had to be set individually for each route. As a result, the setting of the traffic congestion prediction model has been complicated and troublesome.

Therefore, an object of the present invention is to provide a weight parameter setting technique that allows easy setting of a congestion prediction model.

In order to achieve the above-mentioned object, the congestion prediction weight parameter setting device of the present invention is configured as follows.

The prediction model setting unit sets a prediction model based on the connections between the plurality of sensors arranged on the route, using weight parameters depending on the type of connection between the routes. The convolution calculation unit updates sensing data for each of the plurality of sensors using a prediction model to which weight parameters are applied. The prediction error calculation unit calculates a prediction error of the updated sensing data for each of the plurality of sensors. The optimization processing unit optimizes weight parameters from the prediction error.

In this configuration, one type of weight parameter is set depending on the type of route. This eliminates the need to newly set weight parameters even if the shape of the route differs. In other words, if the routes are of the same type, traffic congestion can be predicted using one type of weight parameter regardless of the shape of the route.

In addition, the prediction model setting unit includes, in the weight parameters, a forward weight parameter that sets an influence on sensors adjacent to the downstream side of the route, and a backward weight parameter that sets an influence on sensors adjacent to the upstream side of the route. good. The optimization processing unit may optimize the forward weight parameter and the backward weight parameter.

Further, the optimization processing unit may optimize the weight parameters so that the prediction errors of the plurality of sensors on the route are minimized.

Further, the sensing data may include the number of vehicles, the number of high vehicles, and the occupancy rate of the route.

Further, the prediction model setting unit may include, in the weight parameters, an auxiliary weight parameter that sets the influence from a type of road different from the route. The optimization processing unit may optimize the auxiliary weight parameters.

According to this invention, it is possible to set weight parameters that are easy to set for a congestion prediction model.

FIG. 1 is a functional block diagram showing the configuration of a weight parameter setting device. FIG. 2 is a flowchart showing traffic congestion prediction according to the embodiment of the present invention. FIG. 3 is a flowchart illustrating a method for setting weight parameters for traffic jam prediction according to an embodiment of the present invention. FIG. 4(A) is a schematic diagram of a route, and FIG. 4(B) is a diagram showing a graph of a congestion prediction model set from the route shown in FIG. 4(A). FIG. 5(A) is a schematic diagram of a route, and FIG. 5(B) is a diagram showing a graph of a congestion prediction model set from the route shown in FIG. 5(A). FIG. 6 is a functional block diagram showing the configuration of the traffic congestion prediction device. FIG. 7 is a flowchart showing a traffic jam prediction method. FIG. 8(A) is a schematic diagram of a route, and FIG. 8(B) is a diagram showing a graph of a congestion prediction model set from the route shown in FIG. 8(A). FIG. 9 is a flowchart showing a congestion prediction method.

Embodiments of the present invention will be described below with reference to the drawings.

- Application Example FIG. 3 is a flowchart showing a method for setting weight parameters for traffic congestion prediction according to an embodiment of the present invention.

The traffic congestion prediction weight parameter setting device (hereinafter referred to as weight parameter setting device) acquires sensing data for each of a plurality of sensors installed on the route (S21). The weight parameter setting device performs a predetermined calculation, for example a convolution calculation, on the sensing data of the plurality of sensors, and updates the sensing data (S22). At this time, the weight parameter setting device uses the concept of GraphCNN, assumes that the plurality of sensors arranged on the route have the same relationship, and sets the weight parameter between the plurality of sensors. are common (for example, three types) (weight parameter W).

The weight parameter setting device calculates a prediction error of the updated sensing data for each of the plurality of sensors (S23). The weight parameter setting device optimizes the weight parameter W from the prediction error (S24).

With such a configuration, when predicting traffic jams, it is possible to use the optimized weight parameter W, regardless of the shape of the route, as long as the type of route is the same. This makes it easy to set up a traffic congestion prediction model, and traffic congestion prediction is performed with a predetermined accuracy.

・Configuration example (first embodiment)
A congestion prediction weight parameter setting technique according to a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a functional block diagram showing the configuration of a weight parameter setting device. FIG. 2 is a flowchart of traffic jam prediction performed using the technology of the present invention.

(Simple processing flow up to traffic congestion prediction)
As shown in FIG. 2, the weight parameter setting device first sets a weight parameter W according to the type of route (S11). At this time, the weight parameter setting device uses the GraphCNN technology, with the plurality of sensors as nodes, the sensing data as the feature amount of each node, and the connections between the plurality of sensors by lines as edges. At this time, the weight parameter W is set depending on the type of route. In other words, the weight parameter W is determined regardless of the configuration of the route, as long as there is no change in the type of the route (for example, a change from an automobile-only road to a general road, or a change from a general road to an automobile-only road). ), one type is set. That is, the weight parameter W is common to each route connection.

Once the weight parameter W is set, the traffic jam prediction device predicts traffic jam using the set weight parameter W (S12). More specifically, the congestion prediction device sets a congestion prediction model using a graph that reflects the route and the arrangement of multiple sensors, and applies the weight parameter W set as described above to this congestion prediction model. do. The traffic congestion prediction device uses the concept of GraphCNN to estimate predicted values of sensing data from a plurality of sensors from this traffic congestion prediction model, and performs traffic congestion prediction.

By performing such congestion prediction, the weighting parameter W becomes common regardless of the route configuration, and congestion prediction can be realized with a single model that is easy to manage.

When using such a traffic congestion prediction model, the setting accuracy of the weight parameter W affects the accuracy of traffic congestion prediction. For this reason, the weight parameter setting device 10 has the configuration shown below, and sets the weight parameter W by using the method shown below.

(Configuration of weight parameter setting device)
The weight parameter setting device 10 is realized by, for example, an arithmetic processing device such as a computer. As shown in FIG. 1, the weight parameter setting device 10 includes a CPU 11, a memory 12, a storage section 13, an operation section 14, a display section 15, an IF 16, a GPU 17, and a data bus 100. The CPU 11, memory 12, storage section 13, operation section 14, display section 15, and IF 16 are connected via a data bus 100.

The CPU 11 and the GPU 17 read the weight parameter setting program and set the weight parameter W using the memory 12, for example. Note that a specific method for setting the weight parameter W will be described later.

The memory 12 is, for example, a volatile memory or the like, and is a temporary storage area, and is used when the CPU 11 executes the weight parameter setting program. The storage unit 13 is realized by a magnetic storage device (HDD), a semiconductor storage device (SSD), or the like. The storage unit 13 stores a weight parameter setting program. The weight parameter setting program includes settings of a prediction model, convolution calculation, prediction error calculation, weight parameter optimization, and the like.

The operation unit 14 is realized by a keyboard, a mouse, etc., and receives operation input from an operator. The display unit 15 displays, for example, the operation details, the execution status of the weight parameter setting program, and the set weight parameters. The IF 16 is an interface that performs data communication between the weight parameter setting device 10 and an external device.

(How to set weight parameters)
The weight parameter setting device 10 has the above-described configuration and sets the weight parameter W as shown below. As mentioned above, FIG. 3 is a flowchart illustrating a method for setting weight parameters for traffic congestion prediction according to an embodiment of the present invention. FIG. 4(A) is a schematic diagram of a route, and FIG. 4(B) is a diagram showing a graph of a congestion prediction model set from the route shown in FIG. 4(A).

The weight parameter setting device 10 uses a plurality of sensors (for example, sensors 910 to 915 in FIG. ), sensing data is acquired from (S21). Note that the route used for setting the weight parameter W does not have to be the route 90 to be predicted.

The sensed data includes, for example, number of vehicles, high number of vehicles, and occupancy rate. The number of vehicles is the number of vehicles detected by the sensing data, and the number of large vehicles is the number of large vehicles detected by the sensor. The occupancy rate is the occupancy rate (area ratio) of cars on the road surface in the area detected by the sensor. Note that the sensing data (feature amount) is not limited to these, and may be other data as long as it can be used to predict traffic jams. However, since the number of vehicles, the high number of vehicles, and the occupancy rate are greatly related to traffic congestion, including the number of vehicles, the high number of vehicles, and the occupancy rate in the sensing data leads to an improvement in the accuracy of the weighting parameter W, and as a result, This will lead to improved accuracy in predicting traffic congestion.

The weight parameter setting device 10 sets the traffic congestion prediction model shown in FIG. 4(B). Specifically, the weight parameter setting device 10 uses the plurality of sensors 910-915 as nodes and the portions of the route 90 that sequentially connect the plurality of sensors 910-915 as edges, and performs the traffic congestion prediction shown in FIG. 4(B). Set up the model.

Specifically, in the example of FIG. 4B, the weight parameter setting device 10 sets the sensor 910 as the node N0, and sets the sensing data of the sensor 910 as the feature h0 of the node N0. Similarly, the weight parameter setting device 10 connects the sensors 911, 912, 913, 914, and 915 to nodes N1, N2, N3, N4, and N5, respectively. shall be. The weight parameter setting device 10 sets the sensing data of the sensor 910, the sensing data of the sensor 911, the sensing data of the sensor 912, the sensing data of the sensor 913, the sensing data of the sensor 914, and the sensing data of the sensor 915, respectively. The feature amount h0 of the node N0, the feature amount h1 of the node N1, the feature amount h2 of the node N2, the feature amount h3 of the node N3, the feature amount h4 of the node N4, and the feature amount h5 of the node N5 are set. For example, the feature quantity h0 of the sensor 910 (node N0) is the number of passing vehicles for 5 minutes, the high number of passing vehicles for 5 minutes, and the vehicle occupancy rate for 5 minutes, which are detected by the sensor 910 at 6 points in time ( One group consists of 30 minutes). Note that this feature amount is an example, and the type of data and the number of time points can be set as appropriate.

The weight parameter setting device 10 sets a forward weight parameter W1 as an index that affects the route 90 from upstream to downstream. This forward weight parameter W1 is common to all edges for route 90.

The weight parameter setting device 10 sets the reverse direction weight parameter W2 as an index that affects the route 90 from downstream to upstream. This backward weight parameter W2 is common to all edges for route 90.

In this way, the weight parameter setting device 10 sets the forward weight parameter W1 and the backward weight parameter W2 depending only on the connections (relationships) between a plurality of sensors (nodes). That is, if the relationship is the same, the forward weight parameter W1 and the backward weight parameter W2 are set to unique parameters regardless of the spacing between multiple sensors (nodes) and their positions on the route 90. .

The weight parameter setting device 10 sets a self-weight parameter W0 for convolution to itself by each node N0-N5. The self-weight parameter W0 is also common to all nodes N0-N5. Alternatively, non-adjacent sensors may be virtually connected with edges, and the update may be performed using the feature amounts of the connected sensors.

Note that the initial values of the forward weighting parameter W1, the backward weighting parameter W2, and the self-weighting parameter W0 are set to arbitrary values.

The weight parameter setting device 10 updates (convolves) the sensed data of each sensor 910-915 (each node N0-N5) using the GraphCNN technology for such a traffic congestion prediction model (S22). Specifically, the weight parameter setting device 10 updates the sensing data using the feature amount of a sensor adjacent to the sensor to be updated and its own feature amount. For example, in the case of the sensor 913 shown in FIG. 4A, the sensing data is updated using the sensor 912 adjacent to the upstream side and the sensor 914 adjacent to the downstream side. The update method uses the known GraphCNN technology, and its explanation will be omitted.

The weight parameter setting device 10 calculates the prediction error of the updated sensing data in each sensor 910-915 (each node N0-N5) (S23). Specifically, the weight parameter setting device 10 calculates the difference between the updated sensing data and the actually measured sensing data at a time corresponding to the updated sensing data as a prediction error.

The weight parameter setting device 10 optimizes the forward weight parameter W1, the backward weight parameter W2, and the self-weight parameter W0 from the prediction error (S24). Specifically, the weight parameter setting device 10 optimizes the forward weight parameter W1, the backward weight parameter W2, and the self-weight parameter W0 so that the sum of prediction errors of the sensors 910-915 is minimized. do.

By performing such processing, the forward weighting parameter W1, the backward weighting parameter W2, and the self-weighting parameter W0 are determined based on the arrangement pattern of the sensors 910-915 and the distance between adjacent sensors of the sensors 910-915. It depends on the type of connection of the route 90, not on the distance. Therefore, if the type of connection of the route 90 does not change, it is possible to accurately predict traffic congestion using the forward weight parameter W1, the backward weight parameter W2, and the self-weight parameter W0. That is, as long as the type of connection between the routes 90 does not change, even if the configuration of the routes 90 changes, it is possible to accurately predict traffic jams using the forward weight parameter W1, the backward weight parameter W2, and the self-weight parameter W0. It can be carried out.

For example, even if the forward weight parameter W1, backward weight parameter W2, and self-weight parameter W0 are set from the configuration in FIGS. 4(A) and 4(B), these parameters are ), it can be used to predict traffic congestion on a route configured as shown in FIG. 5(B). FIG. 5(A) is a schematic diagram of a route, and FIG. 5(B) is a diagram showing a graph of a congestion prediction model set from the route shown in FIG. 5(A).

In FIG. 5A, a route 91 joins a route 90, and a route 92 branches off. Route 91 and route 92 are of the same type as route 90. For example, if route 90 is an automobile-only road, then route 91 and route 92 are also automobile-only roads.

In such a case, since the route configuration changes, conventionally it was necessary to set the weight parameters between individual sensors (nodes) again. Specifically, when changing from FIG. 4(A) and FIG. 4(B) to FIG. 5(A) and FIG. 5(B), the sensor 916 (node N6) and the sensor 917 (node N7) has increased, so it was necessary to take these sensors (nodes) into consideration when setting weight parameters between individual sensors (nodes).

However, the weight parameter setting device 10 of the present invention does not depend on the number and arrangement of sensors, so the weight parameters that have already been set (forward weight parameter W1 and backward weight parameter W2) are can be used to accurately predict traffic jams. For example, in the cases of FIGS. 5A and 5B, forward weight parameters W1 and backward weight parameters W2 are set for nodes N0 to N5 and each edge corresponding to sensors 910 to 915. The edge between the node N3 corresponding to the sensor 913 and the node N6 corresponding to the sensor 916, which is the change point, and the edge between the node N4 corresponding to the sensor 914 and the node N7 corresponding to the sensor 917 The forward weighting parameter W1 and the backward weighting parameter W2 can also be set for the edges in between. Therefore, a congestion prediction model can be easily set.

(Configuration of traffic jam prediction device)
Using the weight parameters set as described above, congestion prediction is achieved, for example, as follows.

The traffic jam prediction device 20 is realized by, for example, an arithmetic processing device such as a PC. FIG. 6 is a functional block diagram showing the configuration of the traffic congestion prediction device. As shown in FIG. 6, the traffic jam prediction device 20 includes a CPU 21, a GPU 27, a memory 22, a storage section 23, an operation section 24, a display section 25, an IF 26, and a data bus 200. The CPU 21, memory 22, storage section 23, operation section 24, display section 25, and IF 26 are connected via a data bus 200.

The CPU 21 and the GPU 27 read out the traffic congestion prediction program and use the memory 22, for example, to predict traffic congestion. Note that a specific method for setting traffic congestion prediction will be described later.

The memory 22 is made of, for example, a volatile memory, and is a temporary storage area, and is used when the CPU 21 executes a traffic congestion prediction program. The storage unit 23 is realized by a magnetic storage device (HDD), a semiconductor storage device (SSD), or the like. The storage unit 23 stores a traffic congestion prediction program.

The congestion prediction program includes settings for a congestion prediction model, weight parameter settings, input of sensed data, congestion prediction using convolution calculations, and the like.

The operation unit 24 is realized by a keyboard, a mouse, etc., and receives operation input from an operator. The display unit 25 displays, for example, the operation details, the execution status of the traffic congestion prediction program, and the traffic congestion prediction results. The IF 26 is an interface that performs data communication between the traffic congestion prediction device 20 and an external device.

(Congestion prediction method)
The traffic congestion prediction device 20 has the above-described configuration and performs traffic congestion prediction as described below. FIG. 7 is a flowchart showing a traffic jam prediction method.

The traffic congestion prediction device 20 sets a traffic congestion prediction model through operation input via the operation unit 24 (S31). For example, the congestion prediction model shown in FIG. 4(B) or FIG. 5(B) described above is set.

The traffic jam prediction device 20 receives the forward weight parameter W1, the backward weight parameter W2, and the self-weight parameter W0 set by the weight parameter setting device 10, and sets them in the traffic congestion prediction model (S32).

The traffic jam prediction device 20 receives input of sensed data, and sets the sensed data in the traffic congestion prediction model as a feature amount of a node of the traffic congestion prediction model (S33).

The traffic jam prediction device 20 estimates the predicted value at each sensor (node) using the traffic congestion prediction model, and performs traffic congestion prediction from this estimated value (S34).

(Second embodiment)
A congestion prediction weight parameter setting technique according to a second embodiment of the present invention will be described with reference to the drawings.

FIG. 8(A) is a schematic diagram of a route, and FIG. 8(B) is a diagram showing a graph of a congestion prediction model set from the route shown in FIG. 8(A).

As shown in FIGS. 8A and 8B, the weight parameter setting method according to the second embodiment shows a case where a route 901 and a route 902 are connected to a route 90. Route 901 and route 902 are different types of routes from route 90. For example, route 90 is a road exclusively for automobiles, while route 901 and route 902 are public roads. For example, as shown in FIG. 8(A), there is a gate G1 at the connection part (merging part) between route 901 and route 90, and gate G1 at the connection part (divergent part) between route 902 and route 90. There is G2.

In such a case, the weight parameter setting device sets the confluence weight parameters W3, W5 and the branch weight parameters W4, W6, as well as the forward weight parameter W1, the backward weight parameter W2, and the self weight parameter W0, and sets these. Optimize.

By using such a method, even if route 901 and route 902 are connected, forward weight parameter W1, backward weight parameter W2, and self-weight parameter W0 for route 90 are the same as those of sensors 910-915. Parameters that are common to adjacent sensors (edges) can be set regardless of positional relationships, etc. This makes it easy to set up a traffic congestion prediction model.

Furthermore, with this method, merging to the route 90 and branching from the route 90 can each be set with one weight parameter (merging weight parameter W3 and branching weight parameter W4). Therefore, even if there are merging from different types of routes and branching to different types of routes, and these aspects are different, congestion prediction can be performed using these merging weight parameter W3 and branching weight parameter W4. As a result, even when there is merging from different types of routes or branching to different types of routes, a congestion prediction model can be easily set regardless of the mode of merging or branching.

In such a case, traffic congestion prediction may be performed as follows. FIG. 9 is a flowchart showing a congestion prediction method.

The traffic congestion prediction device 20 sets a traffic congestion prediction model through operation input via the operation unit 24 (S31A). For example, the congestion prediction model shown in FIG. 8(B) described above is set.

The traffic jam prediction device 20 receives the forward weight parameter W1, the backward weight parameter W2, and the self-weight parameter W0 set by the weight parameter setting device 10, and sets them in the traffic congestion prediction model (S32).

The traffic jam prediction device 20 receives the merging weight parameter W3 and the diversion weight parameter W4 set by the weight parameter setting device 10, and sets them in the traffic congestion prediction model (S41).

The traffic jam prediction device 20 receives input of sensed data, and sets the sensed data in the traffic congestion prediction model as a feature amount of a node of the traffic congestion prediction model (S33).

The traffic jam prediction device 20 estimates the predicted value at each sensor (node) using the traffic congestion prediction model, and performs traffic congestion prediction from this estimated value (S34).

Note that, although the above description shows a mode in which the forward weighting parameter W1 and the backward weighting parameter W2 are used, it is also possible to use only the forward weighting parameter W1. In this case, the weight parameters can be more easily set, and the congestion prediction model can also be more easily set.

Further, in the above description, traffic congestion prediction was explained as an example, but the configuration and method of the present invention can be applied not only to traffic congestion prediction but also to road condition prediction including road traffic volume, for example.

<Additional notes>
・A weight parameter setting device for traffic congestion prediction that sets weight parameters for predicting traffic congestion on a route, and uses weight parameters according to the type of route to set a prediction model from the connection of multiple sensors placed on the route. a convolution calculation unit that updates sensing data for each of multiple sensors using a prediction model to which weight parameters are applied; A weight parameter setting device for traffic congestion prediction, comprising a prediction error calculation unit that calculates each prediction error, and an optimization processing unit that optimizes weight parameters from the prediction errors.

- The prediction model setting unit includes, as weight parameters, a forward weight parameter that sets an influence on a sensor adjacent to the downstream side of the route, and a reverse weight parameter that sets an influence on a sensor adjacent to the upstream side of the route, The optimization processing unit optimizes forward weight parameters and backward weight parameters.

- The optimization processing unit optimizes the weight parameters so that the prediction errors of the plurality of sensors on the route are minimized.

・Sensed data includes the number of vehicles, high number of vehicles, and route occupancy rate.

- The prediction model setting unit includes an auxiliary weight parameter for setting the influence from a type of road different from the route in the weight parameter, and the optimization processing unit optimizes the auxiliary weight parameter.

10: Parameter setting device 11, 21: CPU
12, 22: Memory 13, 23: Storage section 14, 24: Operation section 15, 25: Display section 16, 26: IF
17, 27: GPU
20: Traffic congestion prediction device 90, 91, 92, 901, 902: Route 100, 200: Data bus 910, 911, 912, 913, 914, 915, 916, 917: Sensor G1, G2: Gate h0, h1, h2 , h3, h4, h5: Feature amount N0, N1, N2, N3, N4, N5, N6, N7: Node

Claims (6)

A weight parameter setting device for road condition prediction that sets weight parameters for predicting road conditions for each type of route, the device comprising:
A plurality of sensors arranged on the route are connected according to the arrangement arranged on the route, and for each of the plurality of sensors, a pre-setting weight parameter according to the type of the route on which the sensor is arranged. a prediction model setting section that sets a prediction model in which the
a convolution calculation unit that updates sensing data for each of the plurality of sensors using the prediction model set by the prediction model setting unit ;
a prediction error calculation unit that calculates a prediction error of the updated sensing data for each of the plurality of sensors;
an optimization processing unit that obtains a weight parameter that minimizes the sum of the prediction errors calculated for each of the plurality of sensors as an optimized weight parameter for each type of route;
A weight parameter setting device for road condition prediction, comprising: The prediction model setting section includes:
In the pre- setting weight parameter,
a forward weight parameter that sets an influence on downstream adjacent sensors of the line;
a reverse weight parameter that sets an influence on a sensor adjacent upstream of the route;
including;
The optimization processing unit includes:
obtaining the optimized forward weight parameters and the optimized backward weight parameters;
The weight parameter setting device for road condition prediction according to claim 1. The sensed data includes the number of vehicles, the number of high vehicles, and the occupancy rate of the route;
The weight parameter setting device for road condition prediction according to claim 1 or 2 . The prediction model setting section includes:
In the weight parameter,
including a self-weighting parameter that sets the influence on the sensor itself;
The optimization processing unit includes:
obtaining the optimized self-weight parameters;
The weight parameter setting device for road condition prediction according to any one of claims 1 to 3 . A method for setting weight parameters for road condition prediction, which sets weight parameters for each route type, the method comprising:
A plurality of sensors arranged on the route are connected according to the arrangement arranged on the route, and for each of the plurality of sensors, a pre-setting weight parameter according to the type of the route on which the sensor is arranged. a prediction model setting step of setting a prediction model with the
a convolution calculation step of updating sensing data for each of the plurality of sensors using the prediction model set in the prediction model setting step ;
a prediction error calculation step of calculating a prediction error of the updated sensing data for each of the plurality of sensors;
an optimization processing step of obtaining a weight parameter that minimizes the sum of the prediction errors calculated for each of the plurality of sensors as an optimized weight parameter for each type of route;
A method for setting weight parameters for road condition prediction that is performed by a computer . A weight parameter setting program for road condition prediction that sets weight parameters used for predicting road conditions for each type of route, the program comprising:
A plurality of sensors arranged on the route are connected according to the arrangement arranged on the route, and for each of the plurality of sensors, a pre-setting weight parameter according to the type of the route on which the sensor is arranged. a prediction model setting step of setting a prediction model with the
a convolution calculation step of updating sensing data for each of the plurality of sensors using the prediction model set in the prediction model setting step ;
a prediction error calculation step of calculating a prediction error of the updated sensing data for each of the plurality of sensors;
an optimization processing step of obtaining a weight parameter that minimizes the sum of the prediction errors calculated for each of the plurality of sensors as an optimized weight parameter for each type of route;
A weight parameter setting program for road condition prediction that allows a computer to execute the following.

Images