Prediction program, prediction method, and information processing apparatus

Abstract

To follow a simulation up to the lead time required to execute an intervention.SOLUTION: A prediction program causes a computer to execute processing of: measuring a feature amount related to a plurality of options and a people flow to be reproduced on the basis of the plurality of options; executing a simulation of the people flow using the measured feature amount; identifying a model that determines a behavior of the people flow in accordance with an input of the measured feature amount on the basis of the executed simulation of the people flow and the measured people flow; and predicting the people flow on the basis of the identified model.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present invention relates to a prediction program, a prediction method, and an information processing device. [Background technology]

[0002] In situations where there is a large flow of people, such as during events or when evacuating after a disaster, congestion can occur, leading to accidents caused by overcrowding and human damage caused by delayed evacuation. Therefore, if we can predict possible congestion situations in advance and consider and intervene in measures such as evacuation guidance before an incident occurs, it will lead to the avoidance of accidents and human damage and the reduction of risks. Therefore, predicting future human flows is becoming important.

[0003] Agent-based simulation, which models the movements of individual people, is used to predict people flow. Agent-based simulation predicts people flow by simulating the movement and interference of each pedestrian based on assumptions about their behavior. However, agent-based simulation is a so-called what-if scenario analysis, and since it is not known whether the situation in the simulation will actually occur, it does not predict actual situations.

[0004] From this perspective, there are studies underway to apply a method of improving the accuracy of simulations such as weather forecasts by incorporating observational data (sometimes called "data assimilation") to predicting people flows.

[0005] The following conventional technology has been proposed as an example of a technique that applies data assimilation to predicting people flows. For example, conventional technology assumes that people flows are a superposition of actions based on multiple behavioral intentions. Under this assumption, conventional technology predicts people flows that are reproduced by a combination of different behavior modes, and when observed data is available, evaluates the predicted results based on their degree of agreement with the observed data, and estimates and updates the behavioral intentions of the crowd based on the predicted results. This allows for more accurate simulation results to be obtained as a superposition of plausible simulation sets. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent Publication No. 2023-90405 Summary of the Invention [Problem to be solved by the invention]

[0007] However, in the above-mentioned conventional technology, if trends in crowd behavior change after the simulation has been adjusted to fit observed data through data assimilation, it is difficult for the simulation to follow reality up to the lead time required to implement intervention.

[0008] In one aspect, the present invention aims to provide a prediction program, a prediction method, and an information processing device that can track a simulation up to the lead time required for implementing an intervention. [Means for solving the problem]

[0009] A prediction program in one embodiment causes a computer to execute a process of measuring features associated with a plurality of options and a target pedestrian flow to be reproduced based on the plurality of options, simulating the pedestrian flow using the measured features, identifying a model that determines the behavior of the pedestrian flow in response to input of the measured features based on the implemented simulation of the pedestrian flow and the measured pedestrian flow, and predicting the pedestrian flow based on the identified model. Effect of the Invention

[0010] Simulations can be made to track the lead time required to implement an intervention. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of a prediction system. [Diagram 2] FIG. 2 is a diagram illustrating one aspect of the problem. [Diagram 3] FIG. 3 is a diagram showing a problem-solving approach of the prior art. [Figure 4] FIG. 4 is a diagram illustrating a problem-solving approach according to the first embodiment. [Diagram 5] FIG. 5 is a block diagram illustrating an example of a functional configuration of the server device. [Figure 6] FIG. 6 is a diagram showing the people flow to be reproduced and predicted. [Figure 7] FIG. 7 is a diagram illustrating an example of the observation data. [Figure 8] FIG. 8 is a diagram illustrating an example of the feature amount. [Figure 9] FIG. 9 is a schematic diagram showing an example of data assimilation. [Figure 10] FIG. 10 is a diagram showing an example of a change in the total number of people over time. [Figure 11] FIG. 11 is a diagram showing an example of a prediction result in the conventional technology. [Figure 12] FIG. 12 is a diagram illustrating an example of a prediction result in the first embodiment. [Figure 13] FIG. 13 is a diagram illustrating one aspect of the prediction error. [Figure 14] FIG. 14 is a flowchart showing the procedure of the data assimilation process. [Figure 15] FIG. 15 is a schematic diagram showing an application example of data assimilation. [Figure 16] FIG. 16 is a diagram illustrating an example of a hardware configuration. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, with reference to the attached drawings, an embodiment of the prediction program, prediction method, and information processing device according to the present application will be described. Each embodiment is merely an example, and such examples do not limit the configuration, structure, functions, numerical values ​​used to specify these, usage scenes, etc. Furthermore, each embodiment can be adaptively combined within a range that does not cause contradiction in the processing contents.

[0013] <Example 1> <System configuration> Fig. 1 is a diagram showing an example of the configuration of a prediction system. The prediction system 1 shown in Fig. 1 estimates a behavior model that reproduces observed data by data assimilation, which incorporates observed data related to people flow into a simulation, and provides a prediction function that predicts people flow based on the estimated behavior model. For example, the prediction system 1 constructs a digital twin that predicts people flow.

[0014] 1, the prediction system 1 may include a server device 10, observation devices 30A-30N, and a client terminal 50. Hereinafter, when there is no need to distinguish between the individual observation devices 30A-30N, the observation devices 30A-30N may be referred to as "observation device 30."

[0015] The server device 10, the observation device 30, and the client terminal 50 may be communicatively connected via any network NW. For example, the network NW may be realized by any technology, whether wired or wireless, such as an intranet, the Internet, or a low-power wireless communication standard for the Internet of Things (IoT). The network NW may not be a single network, but may be an intranet and the Internet connected via a network device such as a gateway.

[0016] The server device 10 is an example of an information processing device that provides the above prediction function. For example, the server device 10 can provide a prediction service corresponding to the above prediction function as a cloud service by being realized as a PaaS (Platform as a Service) type or SaaS (Software as a Service) type application. In addition, the server device 10 may be realized as a server that provides the above prediction function on-premise.

[0017] The observation device 30 is a device for observing the flow of people within a predetermined area. For example, the observation device 30 may be realized by a camera device. In this case, an example of observation data is an image captured by the observation device 30, or a density map of the flow of people (people / m 2 ) is obtained. Alternatively, the observation device 30 may be realized by a GPS receiver or the like. In this case, position information of people flow is obtained as an example of observation data.

[0018] The client terminal 50 is a terminal device used by a client who receives the above-mentioned prediction function. For example, the client terminal 50 may be realized by any computer, such as a personal computer, a smartphone, a tablet terminal, or a wearable terminal.

[0019] 1 shows an example in which a plurality of observation devices 30 are included in the prediction system 1, but the prediction system 1 may include only one observation device 30. Also, while FIG. 1 shows an example in which one client terminal 50 is connected to the server device 10, a plurality of client terminals 50 may be connected to the server device 10.

[0020] <One aspect of the issue> As explained in the Background section, with the above-mentioned conventional technologies, if trends in crowd behavior change after the simulation has been adjusted to fit observed data through data assimilation, it is difficult for the simulation to follow the actual situation up to the lead time required to implement intervention.

[0021] Fig. 2 is a diagram showing one aspect of the problem. Fig. 2 shows, as just one example of a use case, an example in which an agent-based simulation is applied to an exit selection problem in which a crowd entering from the lower left of the figure selects Exit 1 or Exit 2.

[0022] In this case, the above-mentioned conventional technology focuses on the fact that it is often sufficient to predict macro phenomena such as congestion at exits, rather than estimating the positions of individual pedestrians, and assimilates the observed behavior on the assumption that it is a superposition of several simple behavioral intentions (modes). That is, in the example shown in Figure 2, data assimilation is performed on the assumption that it is a superposition of behavioral intention Θ1 to head toward exit 1 and behavioral intention Θ2 to head toward exit 2.

[0023] According to such conventional techniques, the prediction accuracy after estimation is high when the tendency of behavior does not change. On the other hand, when the apparent behavior changes, it is difficult for the simulation to follow the reality if the behavior changes after the simulation is adjusted to the observed data.

[0024] This is because the behavioral tendency is not necessarily constant in the exit selection problem shown in Figure 2. In other words, the selection ratio of Exit 1 and Exit 2 differs when the exits are empty and when they are crowded. For example, if Exit 1 close to the entrance position is empty, Exit 1 will be selected, whereas if Exit 1 close to the entrance position is crowded, Exit 2 will be selected.

[0025] As described above, with the conventional technology described above, it is difficult to predict the situation when Exit 1 is congested or when Exit 1 is empty, based on observation data when Exit 1 is empty and observation data when Exit 1 is crowded.

[0026] <One aspect of the problem-solving approach> In order to solve such problems, the prediction function according to this embodiment estimates a behavior model that reproduces observed data through data assimilation, and predicts people flow based on a simulation including the estimated behavior model.

[0027] Fig. 3 is a diagram showing a problem-solving approach of the conventional technology. As shown in Fig. 3, in the above conventional technology, the overlap of the behavioral intention Θ1 toward the exit 1 and the behavioral intention Θ2 toward the exit 2 is estimated to be 7:3 by data assimilation.

[0028] In this way, the data assimilation according to the above-mentioned conventional technology is limited to predictions based on the assumption that "since things are like this now, this trend will likely continue in the future." Therefore, when Exit 1 becomes available, the selection ratio of Exit 1 and Exit 2 will be 10:0, so the prediction based on the overlap of the behavioral modes of Behavioral Intention Θ1 and Behavioral Intention Θ2, "7:3," will be incorrect.

[0029] On the other hand, in this embodiment, the method is developed into data assimilation that estimates a behavioral model itself, such as "in this situation, this decision will be made."

[0030] Fig. 4 is a diagram showing a problem-solving approach according to the first embodiment. As shown in Fig. 4, in this embodiment, data assimilation is performed to estimate the decision-making logic or model parameters, such as how many people will choose to head to exit 2 when the congestion level at exit 1 is about this level. Therefore, since it is possible to accurately predict the flow of people even when exit 1 is empty, it is possible to make the simulation track reality even if the tendency of crowd behavior changes after the simulation is adjusted to the observed data by data assimilation.

[0031] Therefore, the prediction function according to the present embodiment allows the simulation to track up to the lead time required for implementing the intervention.

[0032] As a problem-solving approach, the entrance selection problem may be used instead of the exit selection problem. Agent-based simulation can be applied to the entrance selection problem in which a crowd selects entrance 1 or entrance 2.

[0033] <Configuration of Server Device 10> Next, a functional configuration example of the server device 10 according to the present embodiment will be described. Fig. 5 is a block diagram showing a functional configuration example of the server device 10. Fig. 5 shows a block diagram of the server device 10 related to a prediction function.

[0034] As shown in Fig. 5, the server device 10 includes a communication control unit 11, a storage unit 13, and a control unit 15. Note that Fig. 5 merely illustrates functional units related to the above-described prediction function, and the server device 10 may include functional units other than those illustrated.

[0035] The communication control unit 11 is a functional unit that controls communication with other devices such as the observation device 30 and the client terminal 50. As just one example, the communication control unit 11 can be realized by a network interface card. As one aspect, the communication control unit 11 can receive observation data from the observation device 30 and output instructions and settings for uploading the observation data to the observation device 30. As another aspect, the communication control unit 11 can receive a request for a simulation prediction result from the client terminal 50 and output the simulation prediction result to the client terminal 50.

[0036] The storage unit 13 is a functional unit that stores various data. As an example, the storage unit 13 is realized by an internal, external, or auxiliary storage of the server device 10. For example, the storage unit 13 stores a first behavior model 13A and a second behavior model 13B. The first behavior model 13A and the second behavior model 13B will be described together when reference, generation, or registration is performed.

[0037] The control unit 15 is a functional unit that performs overall control of the server device 10. For example, the control unit 15 can be realized by a hardware processor. As shown in Fig. 5, the control unit 15 has a measurement unit 16, a data assimilation unit 17, and a prediction unit 18. The control unit 15 may be realized by hardwired logic or the like.

[0038] The measurement unit 16 is a processing unit that measures a feature amount related to the multiple options and a target pedestrian flow to be reproduced based on the multiple options. As one aspect, the measurement unit 16 acquires observation data uploaded from the observation device 30. As an example of such observation data, the measurement unit 16 acquires an image captured by the observation device 30, or a density map of pedestrian flow generated from the image. Here, the density map may be generated by the observation device 30, or may be generated by the measurement unit 16 from an image captured by the observation device 30. The measurement unit 16 generates observation data related to pedestrian flow by analyzing an image of an area in which structures (entrances and exits) corresponding to the multiple options are arranged, which is captured by a camera device, for example. As another aspect, the measurement unit 16 measures a feature amount corresponding to an option of behavior determined by an agent in a virtual space to be reproduced from the observation data. Note that the multiple options are, for example, a gate indicating either an entrance for a person to enter a predetermined place or an exit for a person to exit a predetermined place. In addition, the feature amount is, for example, a distance to a gate or a degree of congestion at a gate.

[0039] Fig. 6 is a diagram showing the people flow to be reproduced and predicted. Fig. 6 shows, as merely one example of a use case, an application of the agent-based simulation to an exit selection problem in which a crowd entering from the lower left of the figure selects exit 1 or exit 2. In this use case example, the measurement unit 16 can obtain a density map 20 shown in Fig. 7 from the observation device 30 at any period, for example, at a period of 5 seconds.

[0040] FIG. 7 is a diagram showing an example of observation data. In FIG. 7, a density map 20 is illustrated as an example of the observation data. As shown in FIG. 7, the population density per area is mapped on the density map 20. For example, the population density (people / m 2 ) can be calculated by dividing the image captured by the observation device 30 into a mesh and dividing each grid cell included in the mesh by the area of ​​the grid cell, by the number of people observed in that grid cell.

[0041] FIG. 8 is a diagram showing an example of a feature quantity. In FIG. 8, as an example of an agent's action options, the options of exit 1 and exit 2 corresponding to a destination or a route are given as an example. In this case, the distance to the exit can be measured as one of the feature quantities. For example, as shown in FIG. 8, the distance L1 from the position of each agent to the exit 1 can be calculated. Here, FIG. 8 illustrates the distance L1 to the exit 1, but it goes without saying that the distance L2 from the position of each agent to the exit 2 can also be measured in the same way. As another feature quantity, the congestion degree of each exit can be measured. For example, as shown in FIG. 8, the congestion degree D1 of the exit 1 can be calculated. Here, FIG. 8 illustrates the congestion degree D1 of the exit 1, but it goes without saying that the congestion degree D2 of the exit 2 can also be measured in the same way.

[0042] The data assimilation unit 17 is a processing unit that executes data assimilation to incorporate observation data related to people flow into a simulation. An example of a simulation used for such people flow prediction is an agent-based simulation. Hereinafter, agent-based simulation may be abbreviated as "ABS."

[0043] As just one example, the data assimilation unit 17 compares the results of the simulation of people flow with the people flow obtained as observed data, and estimates a behavioral model that reproduces the observed data from among the behavioral models that determine the agent's behavior.

[0044] Hereinafter, as an example only, a case where a particle filter is used for estimating a behavior model in data assimilation will be given. As shown in Fig. 5, the data assimilation unit 17 has an implementation unit 17A and an estimation unit 17B.

[0045] The implementation unit 17A is a processing unit that implements a people flow simulation. For example, in the case of an ABS model, the simulation continues by incrementing the time step Δt, for example, one second, from the initial value of the ABS time t. At this time, when the initial value of the time t is set, a virtual space corresponding to the environment of the exit selection problem shown in FIG. 6 is modeled via GIS (Geographical Information System) software or the like. Agents having position information of two-dimensional or three-dimensional coordinates are mapped into such a virtual space.

[0046] More specifically, the implementation unit 17A generates a group of simulations by executing a plurality of ABS models in parallel at the start of data assimilation. A different behavior model is incorporated into each particle included in such a group of simulations. For example, the implementation unit 17A assigns a different parameter set to the first behavior model 13A stored in the storage unit 13 for each particle and incorporates it into the ABS model. It goes without saying that the implementation unit 17A can generate any number N (natural number) of particles.

[0047] As an example of such a first behavioral model 13A, a discrete choice model will be taken as an example. The discrete choice model has a deterministic term V i and the error term ε i A utility function U i Here, "i" refers to the index of the option. For example, the determinant term V i is the explanatory variable x i For example, the choice of Exit 1 is i=1, the choice of Exit 2 is i=2, the congestion degree of Exit 1 is x1, and the congestion degree of Exit 2 is x2. Then, the deterministic term V i can be defined by the following formula (1) and formula (2), which are merely examples. Note that, although an example in which the congestion degree is included as an example of an explanatory variable has been given here, the distance to each exit may also be included.

[0048] V1 = θ1 + θ2 x 1 (1) V2 = 0 + θ2 x 2 (2)

[0049] For example, the implementation unit 17A changes a parameter set including a parameter θ1 and a parameter θ2 of a discrete choice model incorporated in a simulation corresponding to each of N particles. As an example, the implementation unit 17A sets an initial value of the parameter θ1 according to a uniform random number U(-5,5) and sets an initial value of the parameter θ2 according to a uniform random number U(-5,0).

[0050] In this manner, in this embodiment, even if the discrete choice models are the same, discrete choice models to which different parameter sets are assigned are identified as different behavioral models.

[0051] Then, the implementation unit 17A performs a simulation at time t for each particle to generate a simulation result. For example, the implementation unit 17A updates the state of the agent, such as the position and speed, based on the behavior, such as the selection result of an exit, output from a behavior model to which the feature amount, such as the congestion degree of each exit, measured by the measurement unit 16 for each agent is input. This allows the simulation result at time t to be obtained.

[0052] The estimation unit 17B is a processing unit that estimates a behavior model that reproduces the observation data from among a group of behavior models corresponding to a group of simulations. Note that the word "estimation" described in this embodiment may be replaced with "identification." For example, the "estimation unit 17B" may be replaced with the "identification unit 17B."

[0053] More specifically, when new observation data is acquired, the estimation unit 17B can start processing. That is, the estimation unit 17B compares the result of a simulation performed for each particle by the implementation unit 17A with the observation data, and calculates the degree of agreement between the simulation result corresponding to each particle and the observation data.

[0054] At this time, if the formats of the simulation results and the observation data are different, the simulation results are converted into the same format as the observation data. For example, in the ABS model, the position information of each agent is obtained for each time t, while a density map 20 shown in FIG. 7 is obtained as the observation data. In this case, the simulation results are converted into a density map. As an example, such a conversion can be realized by counting the number of agents whose position information falls within the range of each grid cell of the density map.

[0055] Fig. 9 is a schematic diagram showing an example of data assimilation. Simulation results RA to RN corresponding to N particles are shown in Fig. 9. As shown in Fig. 9, the simulation results RA to RN are the results of executing a simulation incorporating behavioral models A to N in which different parameter sets A to N are assigned to parameters θ1 and θ2 of a discrete choice model.

[0056] These simulation results RA to RN are converted into density maps 40A to 40N to match the format of the observation data 20. Then, the degree of agreement between the density maps 40A to 40N corresponding to each particle and the density map 20 corresponding to the observation data is calculated.

[0057] Such a degree of match may be, for example, the inverse of the error between the two density maps. For example, the estimation unit 17B calculates the total amount of error between the two density maps by tallying up the difference in population density of each grid cell between the two density maps as an error. Then, the estimation unit 17B calculates the likelihood by taking the inverse of the total amount of error between the two density maps. The likelihood obtained in this way can be used as the degree of match.

[0058] Then, the estimation unit 17B evaluates the simulation results RA to RN based on the degree of match between the simulation results RA to RN and the observation data 20. For example, the degree of match (likelihood) increases as the error between the simulation results 40 and the observation data 20 decreases, whereas the degree of match (likelihood) decreases as the error between the two increases. For this reason, it can be evaluated that the particle with a higher degree of match (likelihood) has a higher degree of reproducibility of the observation data.

[0059] As an example only, the estimation unit 17B estimates a behavior model that reproduces the observation data 20 from among the behavior models of N particles corresponding to the simulation results RA to RN. For example, the estimation unit 17B can estimate a behavior model of particles corresponding to a simulation result having the highest degree of match with the observation data 20, and the simulation result. In addition, the estimation unit 17B can estimate a behavior model of a particle group corresponding to a simulation result having a degree of match with the observation data 20 equal to or greater than a threshold value Th1, and a superposition of the simulation results of the particle group. In addition, the estimation unit 17B can estimate a behavior model of a particle group corresponding to a simulation result having a degree of match with the observation data 20 that is in the top specific number, and a superposition of the simulation results of the particle group.

[0060] After estimating such a behavior model, the estimation unit 17B reconstructs a group of simulations based on the evaluation results of the simulation results RA to RN. As one aspect, the estimation unit 17B replicates a simulation in which the behavior model obtained as an estimation result is incorporated. For example, the number of replicates of the simulation in which the behavior model of the estimation result is incorporated can be set to be large as the degree of matching with the observation data 20 increases. At this time, the parameter set of the behavior model does not necessarily have to be completely the same between the simulations, and uniform random numbers can be set within a range close to the parameters θ1 and θ2 included in the parameter set of the behavior model of the estimation result. As another aspect, the estimation unit 17B can delete particles corresponding to a simulation result in which the degree of matching with the observation data 20 is less than a threshold value Th2.

[0061] Thereafter, the estimation unit 17B outputs the reconstructed simulation group and the agent's state, such as its position and speed, to the implementation unit 17A, and causes the implementation unit 17A to execute a simulation at time t + time step Δt. At this time, the estimation unit 17B can output, as the agent's state, a superposition of simulation results whose matching degree with the observation data 20 satisfies a specific condition, or a correction result in which the superposition of the simulation results is corrected based on the observation data.

[0062] Such data assimilation is repeated over a predetermined period of time, and the estimation result of the behavior model obtained is stored in the storage unit 13 as the second behavior model 13B.

[0063] The prediction unit 18 is a processing unit that predicts a pedestrian flow based on the behavior model estimated by the estimation unit 17B. As an example only, the prediction unit 18 executes a simulation based on the second behavior model 13B stored in the storage unit 13 to perform a future prediction of a pedestrian flow.

[0064] More specifically, when data assimilation is performed for a specific period of time, the prediction unit 18 generates in parallel simulations incorporating each behavior model estimated as the second behavior model 13B. At this time, the state of the agent mapped in the virtual space in each simulation may be set as an initial value to a superposition of the simulation results associated with the second behavior model 13B, or a correction result obtained by correcting the superposition of the simulation results based on the latest observation data. The group of simulations generated in this way sets the current time as a point in time when a specific period has elapsed since the start of data assimilation, and performs future prediction of people flow from that point in time for a period of time equal to or greater than the lead time required for implementing intervention.

[0065] Thereafter, the prediction unit 18 can output the result of the future prediction of the people flow to the client terminal 50. For example, the prediction unit 18 can output to the client terminal 50 the superposition of the simulation results at each time when the simulation group is executed, or a density map generated from the superposition of the simulation results. In addition, the prediction unit 18 can also output an alert to the client terminal 50 when the congestion degree of a specific area calculated from the superposition of the simulation results or the density map, for example, the surrounding area of ​​the exit 1 or the surrounding area of ​​the exit 2, is equal to or higher than a threshold value Th3. Note that, although the client terminal 50 is given as an example of an output destination here, the output destination is not limited to this. For example, the result of the future prediction of the people flow may be output to a device other than the client terminal 50, or may be output to a service or application executed in the backend.

[0066] <An example of a numerical experiment> Next, the results of predicting people flow in the exit selection problem shown in Fig. 6 are compared between the above-mentioned conventional technology and this embodiment. As an example only, in the above-mentioned conventional technology and this embodiment, the simulation is carried out under the following conditions.

[0067] First, we assume that a density map with a 10m mesh is obtained as observation data. Second, we assume that data assimilation is performed based on observation data acquired at 5-second intervals. Third, we assume that the ground truth data is generated with parameters θ1 and θ2 of the discrete choice model set to 4.5±0.45(σ)-2.5±0.25(σ).

[0068] Under such conditions, in the above-mentioned conventional technology, data assimilation is performed to estimate a parameter Θ that determines the distribution of a behavioral intention Θ1 toward exit 1 and a behavioral intention Θ2 toward exit 2. On the other hand, in the present embodiment, data assimilation is performed to estimate parameters θ1 and θ2 of a discrete choice model.

[0069] FIG. 10 is a diagram showing an example of a change in the total number of people over time. FIG. 10 shows a time series of the cumulative number of people flowing into the environment having Exit 1 and Exit 2 based on the correct answer data. The vertical axis of the graph shown in FIG. 10 indicates the total number of people, and the horizontal axis indicates the time of the simulation. As shown in FIG. 10, according to the correct answer data, it can be seen that a cumulative total of about 5,500 people flow into the environment having Exit 1 and Exit 2. Here, the slope in the section from 0 seconds to about 370 seconds in the simulation time and the slope in the section from about 1,350 seconds to about 1,750 seconds in the simulation time are larger than the other parts. These sections correspond to the first and second wave congestion where the number of pedestrians increases due to events such as getting on and off trains.

[0070] FIG. 11 is a diagram showing an example of a prediction result in the conventional technology. FIG. 12 is a diagram showing an example of a prediction result in Example 1. The vertical axis of the graphs shown in FIG. 11 and FIG. 12 indicates the number of people, and the horizontal axis indicates the time of the simulation. Furthermore, in FIG. 11 and FIG. 12, the number of people around exit 1 is plotted with a thick line, and the number of people around exit 2 is plotted with a thin line. Furthermore, in FIG. 11, the number of people around exit 1 and exit 2 corresponding to the prediction result in the conventional technology are plotted with a solid line, while the number of people around exit 1 and exit 2 corresponding to the correct answer data are plotted with a dashed line.

[0071] 11 and 12, it is clear that the simulation based on parameters θ1 and θ2 of the discrete choice model estimated by data assimilation in this embodiment is better able to track the correct data than the simulation based on parameter Θ estimated by data assimilation in the conventional technology.

[0072] For example, as shown in Fig. 11, in the above conventional technology, there are sections in which the predicted number of people around Exit 1 during the first wave of congestion cannot track the correct data, and in the second wave of congestion, the predicted numbers of people around Exit 1 and Exit 2 both deviate from the correct data. On the other hand, as shown in Fig. 12, in this embodiment, it can be seen that the predicted numbers of people around Exit 1 and Exit 2 during both the first and second waves can track the correct data.

[0073] Fig. 13 is a diagram showing one aspect of prediction error. Fig. 13 shows prediction errors of congestion peaks at each exit of the second wave of people flow shown in Figs. 11 and 12. As shown in Fig. 13, the prediction errors in this embodiment are smaller than the prediction errors in the conventional technology for both the congestion peak at exit 1 of the second wave and the congestion peak at exit 2 of the second wave. For example, the prediction error of the congestion peak at exit 1 of the second wave can be suppressed to about one-twentieth of the conventional technology, and the prediction error of the congestion peak at exit 2 of the second wave can be suppressed to about one-third or less.

[0074] <Processing flow> 14 is a flowchart showing the procedure of the data assimilation process. This process is merely an example and can be started when a request to start data assimilation is received or at any timing when observation data is available.

[0075] 14, the implementation unit 17A generates a simulation group by implementing multiple ABS models in parallel at the start of data assimilation (step S101). Each particle included in such a simulation group is incorporated with a different behavior model.

[0076] Next, the data assimilation unit 17 executes loop process 1, which repeats the processes from step S102 to step S108 described below for the number M of time steps corresponding to the data assimilation interval.

[0077] Furthermore, the implementation unit 17A of the data assimilation unit 17 executes loop process 2, which repeats the processes of step S102 and step S103 described below for the number N of particles corresponding to the simulation group generated in step S101 or the simulation group reconstructed in step S107 described below.

[0078] Note that, although an example has been given here in which the processes in the following step S102 and the following step S103 are repeated, the processes in the following step S102 and the following step S103 may be executed in parallel for every N particles.

[0079] That is, the implementation unit 17A performs a simulation for the n-th particle at time t to generate a simulation result (steps S102 and S103).

[0080] By repeating this loop process 2, a simulation result at time t is obtained for each of N particles.

[0081] Thereafter, if the observation data is available (Yes in step S104), the estimation unit 17B converts the simulation result of each particle into the same format as the observation data (step S105).

[0082] Next, the estimation unit 17B evaluates the simulation results of each particle based on the degree of agreement between the simulation results of the N particles and the observation data (step S106).

[0083] Then, the estimation unit 17B estimates a behavior model that reproduces the observation data based on the evaluation result of the simulation result obtained in step S106, and reconstructs the simulation group based on the behavior model obtained as the estimation result (step S107).

[0084] Thereafter, the estimation unit 17B outputs the reconstructed simulation group and the state of the agent, for example, the position and speed, to the implementation unit 17A (step S108).

[0085] After the process of step S108 is executed in this manner, the simulation time t is incremented by the time step Δt until the data assimilation interval ends, and the processes from step S102 to step S108 are repeated.

[0086] By such loop process 1, a behavior model capable of reproducing the observation data acquired in the data assimilation interval is estimated.

[0087] <About Digital Twin> In the prediction function according to this embodiment, a digital twin that reproduces the observation data may be constructed, and pedestrian flow may be predicted based on the constructed digital twin.

[0088] First, the measurement unit 16 generates observation data on people flow by analyzing images of an area in which structures corresponding to multiple options are arranged, taken by a camera. Next, the data assimilation unit 17 executes a data assimilation process by incorporating the generated observation data on people flow into a simulation, and constructs a digital twin that predicts people flow in the area based on the executed data assimilation process. For example, the data assimilation unit 17 constructs a digital twin that determines the behavior of an agent. More specifically, the data assimilation unit 17 estimates a behavior model that reproduces the observation data from among behavior models that determine the behavior of an agent, and constructs a digital twin based on the estimated behavior model. Then, the prediction unit 18 predicts people flow in the area based on the constructed digital twin.

[0089] <One aspect of the effect> As described above, the server device 10 according to this embodiment estimates a behavior model that reproduces observed data by data assimilation, and predicts people flow based on a simulation including the estimated behavior model. Therefore, even if the tendency of crowd behavior changes after the simulation is adjusted to the observed data by data assimilation, the simulation can be made to follow reality. Therefore, the server device 10 according to this embodiment can make the simulation follow up to the lead time required for implementing intervention.

[0090] <Example 2> Although the embodiments of the disclosed device have been described above, the present invention may be embodied in various different forms other than the above-described embodiments. Therefore, other embodiments included in the present invention will be described below.

[0091] <Behavioral model> In the above-mentioned first embodiment, a plurality of discrete choice models to which different parameter sets are assigned are exemplified as examples of the behavioral model, but the type of behavioral model is not necessarily limited to the discrete choice model.

[0092] FIG. 15 is a schematic diagram showing an application example of data assimilation. In FIG. 15, simulation results RA to RZ corresponding to each of Z particles are shown. As shown in FIG. 15, the simulation results RA to RN are the results of a simulation in which behavioral models A to N in which different parameter sets A to N are assigned to parameters θ1 and θ2 of a discrete choice model are incorporated. In addition, the simulation results RO to RZ are the results of a simulation in which behavioral models O to Z in which different parameter sets O to Z are assigned to a neural network model that inputs features and outputs class labels corresponding to options of behavior are incorporated. For these simulation results RA to RZ, format conversion, calculation of the degree of agreement, and estimation of the behavioral model can be performed in the same manner as the example shown in FIG. 9. Note that, although FIG. 15 shows an example of a neural network, other machine learning models, such as a support vector machine or gradient boosting, may be used.

[0093] In this way, a plurality of behavioral models, each of which has different parameters, may be incorporated into the simulation group. Furthermore, while Fig. 9 and Fig. 15 show behavioral models with parameters as merely an example, the behavioral models incorporated into the simulations generated as particles may include models without parameters, such as what-if scenario analysis.

[0094] <Distribution and integration> In addition, each component of each device shown in the figure does not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc. For example, the measurement unit 16, the data assimilation unit 17, or the prediction unit 18 may be connected via a network as an external device of the server device 10. In addition, the measurement unit 16, the data assimilation unit 17, or the prediction unit 18 may be included in a separate device, and the function of the server device 10 may be realized by cooperating with the device through a network connection.

[0095] <Hardware configuration> In addition, the various processes described in the above embodiments can be realized by executing a prepared program on a computer such as a personal computer or a workstation. Therefore, an example of a computer that executes a prediction program having the same functions as those in the first and second embodiments will be described below with reference to FIG.

[0096] Fig. 16 is a diagram showing an example of a hardware configuration. As shown in Fig. 16, the computer 100 has an operation unit 110a, a speaker 110b, a camera 110c, a display 120, and a communication unit 130. Furthermore, the computer 100 has a CPU 150, a ROM 160, a HDD 170, and a RAM 180. These units 110 to 180 are connected via a bus 140.

[0097] 16, a prediction program 170a that exerts the same functions as the measurement unit 16, the data assimilation unit 17, and the prediction unit 18 shown in the above-mentioned first embodiment is stored in the HDD 170. This prediction program 170a may be integrated or separated, like the components of the measurement unit 16, the data assimilation unit 17, and the prediction unit 18 shown in FIG. 5. That is, the HDD 170 does not necessarily have to store all the data shown in the above-mentioned first embodiment, as long as the data used for processing is stored in the HDD 170.

[0098] Under such an environment, the CPU 150 reads out the prediction program 170a from the HDD 170 and loads it in the RAM 180. As a result, the prediction program 170a functions as a prediction process 180a as shown in FIG. 16. The prediction process 180a loads various data read out from the HDD 170 in an area of ​​the storage area of ​​the RAM 180 that is assigned to the prediction process 180a, and executes various processes using the loaded various data. For example, the process shown in FIG. 14 may be included as an example of the process executed by the prediction process 180a. Note that in the CPU 150, all of the processing units shown in the above-mentioned first embodiment do not necessarily need to operate, and it is sufficient that the processing units corresponding to the process to be executed are virtually realized.

[0099] The prediction program 170a does not necessarily have to be stored in the HDD 170 or the ROM 160 from the beginning. For example, the prediction program 170a is stored in a "portable physical medium" such as a flexible disk, a so-called FD, a CD-ROM, a DVD disk, a magneto-optical disk, or an IC card that is inserted into the computer 100. The computer 100 may acquire the prediction program 170a from the portable physical medium and execute it. The prediction program 170a may also be stored in another computer or a server device connected to the computer 100 via a public line, the Internet, a LAN, a WAN, or the like. The prediction program 170a thus stored may be downloaded to the computer 100 and then executed.

[0100] The following supplementary notes are further disclosed regarding the embodiments including the above examples.

[0101] (Appendix 1) Measure features related to a plurality of options and a target person flow to be reproduced based on the plurality of options; Using the measured feature amount, a simulation of the pedestrian flow is performed; Identifying a model that determines the behavior of the pedestrian flow in response to the input of the measured feature quantity based on the performed simulation of the pedestrian flow and the measured pedestrian flow; predicting the pedestrian flow based on the identified model; A prediction program that causes a computer to execute a process.

[0102] (Additional Note 2) The process includes a process of performing a group of people flow simulations incorporating a plurality of first models each having a same type of model but different parameter sets; The process of identifying includes a process of identifying a first model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 2. The prediction program according to claim 1,

[0103] (Appendix 3) The process includes a process of carrying out a group of people flow simulations incorporating a plurality of first models each having a first type of model with different parameter sets and a plurality of second models each having a second type of model different from the first type with different parameter sets; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 2. The prediction program according to claim 1,

[0104] (Appendix 4) The process includes carrying out a group of people flow simulations incorporating a plurality of first models to which different parameter sets are assigned in a first type model, and a second model different from the first type; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 2. The prediction program according to claim 1,

[0105] (Appendix 5) The model corresponds to a discrete choice model, The prediction program described in Appendix 1, characterized in that the identification process identifies the discrete choice model by comparing an implemented pedestrian flow simulation with a measured pedestrian flow.

[0106] (Appendix 6) The measuring process generates observation data on the pedestrian flow by analyzing an image of an area in which structures corresponding to the plurality of options are arranged, the image being taken by a camera; The process of identifying includes performing a data assimilation process by incorporating the generated observation data regarding the pedestrian flow into a simulation, and constructing a digital twin that predicts the pedestrian flow in the area based on the executed data assimilation process; The prediction program described in Appendix 1, characterized in that the prediction process predicts the flow of people in the area based on the digital twin.

[0107] (Supplementary Note 7) The plurality of options are gates indicating either an entrance for a person to enter a predetermined location or an exit for a person to exit a predetermined location, the feature amount is a distance to the gate or a congestion degree of the gate, The simulation of people flow is an agent-based simulation. 2. The prediction program according to claim 1,

[0108] (Appendix 8) Measure features related to a plurality of options and a target person flow to be reproduced based on the plurality of options; Using the measured feature amount, a simulation of the pedestrian flow is performed; Identifying a model that determines the behavior of the pedestrian flow in response to the input of the measured feature quantity based on the performed simulation of the pedestrian flow and the measured pedestrian flow; predicting the pedestrian flow based on the identified model; A prediction method characterized in that the processing is executed by a computer.

[0109] (Appendix 9) The process includes a process of performing a group of people flow simulations incorporating a plurality of first models each having a same type of model but different parameter sets; The process of identifying includes a process of identifying a first model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 9. The prediction method according to claim 8,

[0110] (Appendix 10) The process includes a process of carrying out a group of people flow simulations incorporating a plurality of first models each having a first type of model with different parameter sets and a plurality of second models each having a second type of model different from the first type with different parameter sets; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 9. The prediction method according to claim 8,

[0111] (Appendix 11) The process includes carrying out a group of people flow simulations incorporating a plurality of first models to which different parameter sets are assigned in a first type model, and a second model different from the first type; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 9. The prediction method according to claim 8,

[0112] (Appendix 12) The model corresponds to a discrete choice model, The method of claim 8, wherein the identifying process identifies the discrete choice model by comparing an implemented pedestrian flow simulation with a measured pedestrian flow.

[0113] (Appendix 13) The measuring process generates observation data on the pedestrian flow by analyzing an image of an area in which structures corresponding to the plurality of options are arranged, the image being taken by a camera; The process of identifying includes performing a data assimilation process by incorporating the generated observation data regarding the pedestrian flow into a simulation, and constructing a digital twin that predicts the pedestrian flow in the area based on the executed data assimilation process; The prediction method described in claim 8, wherein the prediction process predicts the flow of people in the area based on the digital twin.

[0114] (Supplementary Note 14) The plurality of options are gates indicating either an entrance for a person to enter a predetermined location or an exit for a person to exit a predetermined location; the feature amount is a distance to the gate or a congestion degree of the gate, The simulation of people flow is an agent-based simulation. 9. The prediction method according to claim 8,

[0115] (Appendix 15) Measure features related to a plurality of options and a target person flow to be reproduced based on the plurality of options; Using the measured feature amount, a simulation of the pedestrian flow is performed; Identifying a model that determines the behavior of the pedestrian flow in response to the input of the measured feature quantity based on the performed simulation of the pedestrian flow and the measured pedestrian flow; predicting the pedestrian flow based on the identified model; An information processing device having a control unit that executes processing.

[0116] (Appendix 16) The process includes a process of performing a group of people flow simulations incorporating a plurality of first models each having a different parameter set of the same type of model; The process of identifying includes a process of identifying a first model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 16. The information processing device according to claim 15,

[0117] (Appendix 17) The process includes a process of carrying out a group of people flow simulations incorporating a plurality of first models each having a first type of model with different parameter sets and a plurality of second models each having a second type of model different from the first type with different parameter sets; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 16. The information processing device according to claim 15,

[0118] (Appendix 18) The process includes carrying out a group of people flow simulations incorporating a plurality of first models to which different parameter sets are assigned in a first type model, and a second model different from the first type; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the simulations of the pedestrian flow; 16. The information processing device according to claim 15,

[0119] (Appendix 19) The model corresponds to a discrete choice model, The information processing device according to claim 15, wherein the identifying process identifies the discrete choice model by comparing an implemented pedestrian flow simulation with a measured pedestrian flow.

[0120] (Appendix 20) The measuring process generates observation data on the pedestrian flow by analyzing an image of an area in which structures corresponding to the plurality of options are arranged, the image being taken by a camera; The process of identifying includes performing a data assimilation process by incorporating the generated observation data regarding the pedestrian flow into a simulation, and constructing a digital twin that predicts the pedestrian flow in the area based on the executed data assimilation process; The information processing device described in Appendix 15, characterized in that the prediction process predicts the flow of people in the area based on the digital twin. [Explanation of symbols]

[0121] 1. Prediction System 10. Server device 11 Communication control section 13 Storage section 13A First Behavioral Model 13B Second Behavioral Model 15 Control section 16 Measuring part 17 Data Assimilation Department 17A Implementation Department 17B Estimation part 18 Prediction Department 30 Observation Equipment 50 client terminals

Claims

1. measuring features associated with a plurality of options and a target person flow to be reproduced based on the plurality of options; Using the measured feature amount, a simulation of the pedestrian flow is performed; Identifying a model that determines the behavior of the pedestrian flow in response to the input of the measured feature quantity based on the performed simulation of the pedestrian flow and the measured pedestrian flow; predicting the pedestrian flow based on the identified model; A prediction program that causes a computer to execute a process.

2. The performing process includes performing a group of people flow simulations incorporating a plurality of first models each having a different parameter set assigned to the same model; The process of identifying includes a process of identifying a first model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the group of simulations of the pedestrian flow; 2. The prediction program according to claim 1 .

3. The performing process includes performing a group of people flow simulations incorporating a plurality of first models, each of which is a first type of model and has different parameter sets assigned thereto, and a plurality of second models, each of which is a second type of model different from the first type and has different parameter sets assigned thereto; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the group of simulations of the pedestrian flow; 2. The prediction program according to claim 1 .

4. The process includes carrying out a group of people flow simulations incorporating a plurality of first models, each of which is a first type model with different parameter sets, and a second model different from the first type; The process of identifying includes a process of identifying a first model and / or a second model incorporated in a simulation of a pedestrian flow whose degree of match with the measured pedestrian flow satisfies a specific condition, from among the group of simulations of the pedestrian flow; 2. The prediction program according to claim 1 .

5. The model corresponds to a discrete choice model, The prediction program according to claim 1 , wherein the identifying process identifies the discrete choice model by comparing an implemented pedestrian flow simulation with a measured pedestrian flow.

6. The measuring process includes generating observation data on the pedestrian flow by analyzing an image of an area in which structures corresponding to the plurality of options are arranged, the image being taken by a camera; The process of identifying includes performing a data assimilation process by incorporating the generated observation data regarding the pedestrian flow into a simulation, and constructing a digital twin that predicts the pedestrian flow in the area based on the executed data assimilation process; The prediction program according to claim 1, characterized in that the prediction process predicts the flow of people in the area based on the digital twin.

7. the plurality of options are gateways indicating either an entrance for a person to enter a predetermined location or an exit for a person to exit a predetermined location; the feature amount is a distance to the gate or a congestion degree of the gate, The simulation of people flow is an agent-based simulation.

7. The prediction program according to claim 1, wherein the prediction program is a program for predicting a prediction result based on the prediction result.

8. measuring features associated with a plurality of options and a target person flow to be reproduced based on the plurality of options; Using the measured feature amount, a simulation of the pedestrian flow is performed; Identifying a model that determines the behavior of the pedestrian flow in response to the input of the measured feature quantity based on the performed simulation of the pedestrian flow and the measured pedestrian flow; predicting the pedestrian flow based on the identified model; A prediction method characterized in that the processing is executed by a computer.

9. measuring features associated with a plurality of options and a target person flow to be reproduced based on the plurality of options; Using the measured feature amount, a simulation of the pedestrian flow is performed; Identifying a model that determines the behavior of the pedestrian flow in response to the input of the measured feature quantity based on the performed simulation of the pedestrian flow and the measured pedestrian flow; predicting the pedestrian flow based on the identified model; An information processing device having a control unit that executes processing.

Images