Evacuation assistance device and evacuation assistance method

Abstract

This evacuation assistance device contributes to improvement of the prediction accuracy related to evacuation by having an estimation unit that is configured to: derive, on the basis of information obtained by classifying the number of damage caused by a disaster in certain areas into a plurality of degrees related to the magnitude of the damage, a first probability distribution having an index related to the magnitude of the damage as a probability variable; derive, on the basis of the first probability distribution, of the magnitude of a factor of the disaster, of a second probability distribution having an average value different from that of the first probability distribution, and of the magnitude of the factor corresponding to the second probability distribution, a third probability distribution related to an index corresponding to a state without the factor and an increment in the index when the factor has increased by a unit amount; and derive, on the basis of the first probability distribution and of the actual result of the number of evacuees due to the disaster, an index-related threshold for initiating an evacuation.

Description

Evacuation support device and evacuation support method

[0001] The present invention relates to an evacuation assistance device and an evacuation assistance method.

[0002] A technology has been developed that generates a sequential logit model for predicting the timing and number of residents who will begin evacuating when a disaster occurs by obtaining model parameters from past disasters (Non-Patent Document 1).

[0003] Furthermore, based on Non-Patent Document 1, a technology for quantifying disaster risk has been developed that uses the magnitude of the risk of damage to a home from past disasters as an index to predict when individual residents will begin evacuation (Non-Patent Document 2).

[0004] Fu, H., & Wilmot, C.G., "A Sequential Logit Dynamic Travel Demand Model for Hurricane Evacuation," TRB 2004 Annual Meeting, 2004. Urata et al., "Evaluating the Influence of Other Evacuations and Human Networks on Evacuation Initiation Selection in Heavy Rain Disasters," Journal of the Japan Society of Civil Engineers, D3, Vol. 73, No. 1, pp. 24-39, March 20, 2017.

[0005] According to the technology of Non-Patent Document 1, it is possible to predict the number of disaster victims who will evacuate at each time when a disaster occurs, but because the model parameters depend on the time interval, the model needs to be re-estimated when the time interval changes. Furthermore, due to this characteristic, it is difficult to interpret the meaning of the parameters.

[0006] The technique of Non-Patent Document 2 has difficulty in quantitatively measuring the disaster risk on an interval scale, and therefore the prediction error may be large.

[0007] The present invention has been made in consideration of the above points, and aims to contribute to improving the accuracy of predictions regarding evacuation.

[0008] Therefore, in order to solve the above problem, the evacuation support device has an estimation unit that is configured to derive a first probability distribution in which an index related to the magnitude of the damage is a random variable based on information classifying the number of damages caused by a disaster in a certain area into multiple degrees related to the magnitude of the damage, derive a third probability distribution for the index corresponding to a state in which the factor is absent and an increase in the index when the factor increases by a unit amount based on the first probability distribution, the magnitude of the factor, a second probability distribution whose mean value differs from that of the first probability distribution, and the magnitude of the factor corresponding to the second probability distribution, and derive a threshold at which evacuation should be started for the index based on the first probability distribution and the actual number of evacuees due to the disaster.

[0009] This can contribute to improving the accuracy of evacuation predictions.

[0010] 1 is a diagram illustrating an example of a hardware configuration of the evacuation support device 10 according to an embodiment of the present invention. FIG. 1 is a diagram illustrating an example of a functional configuration of the evacuation support device 10 in the estimation phase. FIG. 2 is a flowchart illustrating an example of a processing procedure executed by the evacuation support device 10 in the estimation phase. FIG. 2 is a diagram illustrating an example of a configuration of a residential damage DB 121. FIG. 3 is a diagram illustrating an example of a configuration of an evacuation shelter DB 122. FIG. 4 is a diagram illustrating an example of a configuration of a rainfall DB 123. FIG. 4 is a diagram illustrating the derivation of the probability distribution of residential damage level for each region at the time of a disaster and the criteria for determining residential damage classification. FIG. 5 is a diagram illustrating the derivation of the probability distribution of residential damage level at the critical time. FIG. 6 is a diagram illustrating the derivation of the average value of the probability distribution of residential damage level at the initial time. FIG. 7 is a diagram illustrating the relationship between the probability distribution of residential damage level and movement speed at the time of a disaster, at the critical time, and at the initial time. FIG. 8 is a diagram illustrating a method for estimating an evacuation start threshold. FIG. 9 is a diagram illustrating an example of parameter values ​​estimated for a residential damage / evacuation start model. FIG. 10 is a diagram illustrating an example of parameter values ​​estimated for a residential damage reference model. FIG. 11 is a diagram illustrating an example of a functional configuration of the evacuation support device 10 in the prediction phase. 1 is a flowchart illustrating an example of a processing procedure executed by the evacuation support device 10 in the prediction phase. FIG. 2 is a diagram illustrating an example of the results of generating initial values ​​of the degree of residential damage for each resident in each region. FIG. 3 is a diagram illustrating an example of predicted cumulative rainfall for each region. FIG. 4 is a diagram illustrating a method for acquiring the degree of residential damage for each resident in each region at a certain time and whether or not evacuation has started. FIG. 5 is a diagram illustrating an example of the results of tallying the number of cases of each residential damage classification for each region at each time. FIG. 6 is a diagram illustrating an example of the results of tallying the number of evacuees for each region at each time.

[0011] In this embodiment, we address the issues in Non-Patent Document 1 by assuming a threshold (for example, uniform) for the magnitude of the disaster risk when residents of a certain area begin to evacuate, and introducing a model of evacuation behavior in which, when the magnitude of the disaster risk of a residence at a certain time is given, only residents whose risk is equal to or exceeds the threshold will evacuate.

[0012] Furthermore, in this embodiment, it is assumed that the magnitude of the damage risk of houses in a certain area follows (for example) a normal distribution, and (for example) the "Damage Assessment Standards" (URL: https: / / www.bousai.go.jp / taisaku / unyou.html), which is an ordinal scale common throughout Japan, is converted into an interval scale based on past data and used, thereby addressing the issues in Non-Patent Document 2.

[0013] The extent of damage to homes in disaster-stricken areas varies from household to household. It is considered appropriate for residents whose homes are not damaged to stay at home rather than go to evacuation shelters. Therefore, it is assumed that the criteria for determining whether residents should begin evacuation (for each region) are determined by the degree of risk of damage to their homes. In this embodiment, a common index (hereinafter referred to as "home damage level") is introduced for both the criteria and the degree of risk of damage, making it possible to predict how many residents will begin evacuation based on a certain amount of rainfall.

[0014] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0015] 1 is a diagram showing an example of the hardware configuration of an evacuation support device 10 according to an embodiment of the present invention. The evacuation support device 10 in FIG. 1 includes a drive device 100, an auxiliary storage device 102, a memory device 103, a processor 104, and an interface device 105, all of which are interconnected via a bus B.

[0016] The program that realizes the processing in the evacuation support device 10 is provided by a recording medium 101 such as a CD-ROM. When the recording medium 101 storing the program is set in the drive device 100, the program is installed from the recording medium 101 to the auxiliary storage device 102 via the drive device 100. However, the program does not necessarily have to be installed from the recording medium 101, but may be downloaded from another computer via a network. The auxiliary storage device 102 stores the installed program as well as necessary files, data, etc.

[0017] When an instruction to start a program is received, the memory device 103 reads and stores the program from the auxiliary storage device 102. The processor 104 is a CPU or a GPU (Graphics Processing Unit), or a CPU and a GPU, and executes functions related to the evacuation support device 10 in accordance with the program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

[0018] In this embodiment, the evacuation support device 10 executes two phases: an estimation phase in which parameters of a model for predicting a damage situation and an evacuation situation (hereinafter referred to as a "prediction model") are estimated from past disasters, and a prediction phase in which the estimated prediction model is used to predict a damage situation and an evacuation situation. Each phase will be described below.

[0019] [Estimation Phase] Fig. 2 is a diagram showing an example of the functional configuration of the evacuation support device 10 in the estimation phase. In Fig. 2, the evacuation support device 10 has an estimation unit 11. The estimation unit 11 is realized by processing in which one or more programs installed in the evacuation support device 10 are executed by the processor 104. The evacuation support device 10 also uses databases (storage units) such as a residential damage DB 121, an evacuation shelter DB 122, and a rainfall DB 123. Each of these databases can be realized using, for example, the auxiliary storage device 102 or a storage device connectable to the evacuation support device 10 via a network.

[0020] FIG. 3 is a flowchart illustrating an example of a processing procedure executed by the evacuation support device 10 in the estimation phase.

[0021] In step S101 , the estimation unit 11 acquires information stored in each of the residential damage DB 121 , the evacuation shelter DB 122 , and the rainfall DB 123 .

[0022] FIG. 4 is a diagram illustrating an example of the configuration of the residential damage DB 121. As shown in FIG. 4, the residential damage DB 121 stores, for each region, information on the actual number of cases (number of households) of certain disasters (floods, landslides, etc.) that have occurred in the past due to heavy rain in that region, classified into multiple levels of residential damage, as well as the total number of households. The multiple levels of residential damage are classified into six levels, in descending order of severity: total destruction, large-scale partial destruction, partial destruction, semi-partial destruction, partial damage, and damage below the standard. This classification (hereinafter referred to as "residential damage classification") is based on damage assessment standards. Note that the heavy rain in each region may be the same heavy rain, or may be heavy rain at different times (related to different disasters).

[0023] Fig. 5 is a diagram showing an example of the configuration of the evacuation shelter DB 122. As shown in Fig. 5, the actual values ​​of the number of evacuees and total population by region related to the disaster corresponding to the residential damage DB 121 are stored. An evacuee is someone who has moved from their home to an evacuation shelter.

[0024] Fig. 6 is a diagram showing an example of the configuration of the rainfall DB 123. As shown in Fig. 6, actual values ​​of disaster-time cumulative rainfall and critical cumulative rainfall for disasters corresponding to the residential damage DB 121 are stored by region.

[0025] The cumulative rainfall during a disaster is the cumulative rainfall from the start of the rainfall that caused the disaster until the rainfall completely ends.

[0026] Strictly speaking (theoretically), the critical cumulative rainfall is the maximum cumulative rainfall that does not cause damage (i.e., the rainfall immediately before the occurrence of a state in which at least one house exceeds the "standard" level) (hereinafter referred to as the "critical time"). However, determining the critical cumulative rainfall requires the government to make a determination after the disaster, and it is difficult to accurately determine when damage to houses occurred. Therefore, the maximum cumulative rainfall of past heavy rains that did not cause a disaster on a day other than the disaster day may be used as the critical cumulative rainfall.

[0027] Next, the estimation unit 11 derives (estimates) values ​​of model parameters of the prediction model based on the information acquired in step S101 (S102). As shown in Fig. 2, the prediction model estimated by the estimation unit 11 includes a residential damage reference model and a residential damage / evacuation start model.

[0028] The residential damage standard model includes values ​​of the residential damage degree that are the boundaries of the residential damage classifications (hereinafter referred to as "determination criteria for residential damage classifications"). In this embodiment, since there are six residential damage classifications (see Figure 4), there are five determination criteria for the residential damage classifications.

[0029] The residential damage and evacuation initiation model includes, for each region, a probability distribution (probability distribution of the magnitude of residential damage risk) in which the residential damage level at each of the timings of "disaster," "critical time," and "initial time" is a random variable, the increase in residential damage level when rainfall increases by a unit amount (1 mm) (hereinafter referred to as "movement speed"), and an evacuation initiation threshold. In this embodiment, the probability distribution of residential damage level is assumed to follow a normal distribution. Therefore, calculating the probability distribution of residential damage level is equivalent to calculating the mean value and standard deviation of residential damage level.

[0030] The term "at the time of disaster" refers to the time when the damage situation is confirmed. Therefore, the estimation unit 11 derives (estimates) the probability distribution of the degree of damage to a house at the time of a disaster so as to be consistent with the house damage DB 121 (FIG. 4).

[0031] Figure 7 is a diagram illustrating the derivation of the probability distribution of residential damage levels for each region during a disaster and the criteria for determining residential damage classification. In Figure 7, the horizontal axis represents residential damage levels, and each normal distribution curve (probability density function) represents the probability distribution of residential damage levels for each region. The dashed lines parallel to the Y axis represent the residential damage levels that serve as the boundaries between residential damage classifications (i.e., the criteria for determining residential damage classifications). While there is a "near-destroyed" category between "partially damaged" and "partially damaged" (see Figure 4), the number of "near-destroyed" cases is small, so the dashed line indicating the boundary between "near-destroyed" and "near-destroyed" (the criteria for determining residential damage classifications) appears to overlap with the boundary between "near-destroyed" and "partially damaged."

[0032] Specifically, the estimation unit 11 first uses the mean value and standard deviation of the probability distribution of the degree of residential damage in each region and the criteria for determining the residential damage classification (the degree of residential damage that is the boundary between each residential damage classification) as unknowns, and performs maximum likelihood estimation of each unknown so that the number (probability) of each residential damage classification based on the probability distribution and the criteria matches the actual results (Figure 4) as closely as possible. As a result, the probability distribution (mean value and standard deviation) for each region during the "disaster" and the criteria for determining the residential damage classification are all simultaneously determined.

[0033] In other words, by using the average value and standard deviation of the residential damage level for each region and the criteria for determining residential damage classification as parameters, it is possible to calculate the probability that residential homes in a certain region will be classified into a certain residential damage classification during a disaster. Assuming that multiple residential homes in a certain region are independent events, it is possible to calculate the probability that a certain number of residential damage classifications will occur. Therefore, when a certain number of residential damage classifications are observed, the likelihood of the observation can be expressed as a parameter formula. In this embodiment, this formula assumes a normal distribution and is therefore differentiable, so it is possible to determine the parameter values ​​that maximize the likelihood using an iterative calculation procedure such as a gradient method. In this case, the estimation unit 11 performs maximum likelihood estimation by fixing the criteria for determining residential damage classifications common to all regions: the criterion distinguishing between "below standard" and "partial damage" at 0, and the criterion distinguishing between "partial damage" and "semi-destroyed" at 1. In this way, the scale of the degree of residential damage (the scale on the horizontal axis in Figure 4) is determined, and the average value and standard deviation of the degree of residential damage at the time of the disaster and the criteria for determining the residential damage classification can be obtained.

[0034] Once the probability distribution of residential damage levels at the time of the disaster is calculated, the estimation unit 11 derives a probability distribution of residential damage levels at the "limit time," which has a different mean value from the probability distribution at the time of the disaster, by shifting the probability distribution at the time of the disaster to the left (toward a smaller residential damage level) (i.e., by decreasing the mean value of the probability distribution). Specifically, the estimation unit 11 determines the probability distribution at the limit time as the probability distribution when (almost) everything is just barely "below the standard" (a probability distribution that is "below the standard" with almost 100% probability). In this embodiment, the standard deviation of the probability distribution of residential damage levels is assumed to be constant (unchanging) at each timing ("at the time of the disaster," "at the limit time," and "initial time"). Therefore, by changing only the mean value of the probability distribution at the time of the disaster, probability distributions at other timings can be derived.

[0035] 8 is a diagram for explaining the derivation of the probability distribution of the degree of residential damage at the "critical point." Fig. 8 shows an example in which the probability distribution (average value) in which, for example, less than 0.1% of the total exceeds "below the standard" is derived as the probability distribution (average value) of the degree of residential damage at the critical point.

[0036] Although the probability density of a normal distribution is not strictly zero in the real number domain, the estimation unit 11 determines the average value of the degree of residential damage so that the number of damaged homes is essentially (as close as possible to) zero. For example, in an area with a total of 500 homes, if the percentage of damaged homes is less than 0.1%, the expected number of damaged homes is less than 0.5, which can be considered to be essentially as close as possible to zero. Figure 8 shows an example in which the probability distribution in this case is derived as a "limit" probability distribution.

[0037] This allows us to determine the probability distribution (average value) of the degree of damage to housing at the "critical point."

[0038] Once the probability distribution of residential damage levels at the time of the disaster and at the critical time has been calculated for each region, the estimation unit 11 derives (calibrates) the average value of the probability distribution of residential damage levels at the initial time and the movement speed for each region based on the average value of each probability distribution and the cumulative rainfall at the time of the disaster and the critical cumulative rainfall for that region. The initial time refers to the time when the cumulative rainfall is 0 (the time before the disaster has started).

[0039] FIG. 9 is a diagram illustrating the derivation of the average value of the probability distribution of residential damage levels at the "initial time." The estimation unit 11 calculates the average value of the probability distribution of residential damage levels at the "initial time" (when the cumulative rainfall is 0) by assuming that the residential damage level is a linear function of cumulative rainfall (i.e., that the residential damage level changes in proportion to the cumulative rainfall). Specifically, the estimation unit 11 calculates the average value of the probability distribution of residential damage levels at the "critical time" (when the cumulative rainfall is 0) by assuming that X1 is the average value of the probability distribution of residential damage levels at the "critical time," Y1 is the critical cumulative rainfall (corresponding to the probability distribution of residential damage levels at the "critical time"), X2 is the average value of the probability distribution of residential damage levels at the "disaster time," and Y2 is the cumulative disaster rainfall (corresponding to the probability distribution of residential damage levels at the "disaster time"). The estimation unit 11 then calculates the value (X3) of the Yx intercept of the line passing through the two points (X1, Y1) and (X2, Y2) as the average value of the probability distribution of residential damage levels at the "initial time." This average value serves as the initial value of residential damage levels. As described above, the standard deviation of the probability distribution of the degree of residential damage at the "initial time" is the same as at the "disaster time" and the "critical time."

[0040] The movement speed is the increase in the residential damage level when the cumulative rainfall increases by 1 mm. The estimation unit 11 calculates the movement speed by dividing (average residential damage level at the time of the disaster - average residential damage level at the critical time) by (cumulative precipitation at the time of the disaster - critical cumulative precipitation).

[0041] This embodiment is based on an example where there is one disaster record for each region (FIGS. 4 to 6), but if there are records of two or more disasters as shown in FIG. 4 to 6, the two timings of "at the limit" and "at the time of disaster" in this embodiment may be replaced with two disaster times. In other words, in this case, "at the limit" does not need to be used.

[0042] That is, Figure 9 shows the procedure for determining the slope and intercept of a linear function, assuming that the residential damage level is a linear function of cumulative rainfall. Generally, (though there are exceptions) it is possible to determine the slope and intercept of a linear function if two or more data points are available. In this embodiment, a total of two data points are derived from one record of damage and one record of no damage. One of these data points, the critical rainfall, is the "amount of rainfall above which a disaster would occur." However, since no disaster actually occurred, it is considered difficult to determine whether this is truly the critical rainfall. Therefore, if there are two or more records of damage, improved accuracy can be expected by using the cumulative rainfall at each disaster and the average value of the probability distribution as two data points.

[0043] From the above, the probability distribution of the degree of damage to housing at the "time of disaster," "time of limit," and "time of initial stage" and the movement speed are determined. Figure 10 shows the relationship between the probability distribution of the degree of damage to housing at the "time of disaster," "time of limit," and "time of initial stage" and the movement speed. In Figure 10, the probability distributions for "time of initial stage," "time of limit," and "time of disaster" are shown from the left. In other words, the larger the disaster, the more the probability distribution shifts to the right (in the direction in which the degree of damage to housing increases). The magnitude of the shift per unit of rainfall (per 1 mm in this embodiment) is the movement speed.

[0044] The estimation unit 11 further estimates an evacuation start threshold for each region based on the probability distribution of the degree of damage to houses in the event of a disaster, so as to be consistent with the actual results of the evacuation shelter DB 122 (FIG. 5). The evacuation start threshold is a value indicating the degree of damage to houses at which residents should begin evacuating.

[0045] 11 is a diagram for explaining a method for estimating the evacuation start threshold, which shows the probability distribution by region shown in FIG.

[0046] The estimation unit 11 performs maximum likelihood estimation of the evacuation start threshold for each region so that the number of evacuees when the evacuation start threshold is set for the probability distribution of the degree of damage to houses in the "event of a disaster" matches the actual results in the evacuation shelter DB 122 (Figure 5). Multiple vertical lines in Figure 11 indicate evacuation start thresholds for regions related to probability distributions of the same line type.

[0047] By using the cumulative distribution function Φ(z) of the normal distribution, it is possible to calculate the probability Φ(z) of a value equal to or greater than a certain value z. Conversely, it is also possible to calculate the threshold z when the proportion Φ(z) is given. Assuming that in a certain area, all residents whose residential damage level is equal to or greater than the threshold z will evacuate, when the proportion of evacuated residents (number of evacuees ÷ total population in Figure 5) is Φ(z), it is possible to determine the threshold z so that the number of evacuees is consistent. The estimation unit 11 sets z calculated in this way for each area as the evacuation start threshold for that area.

[0048] The probability distribution is a distribution based on the number of households, whereas the evacuees in the evacuation shelter DB 122 are measured in units of numbers. However, this embodiment is based on the assumption that the number of people in each household is approximately the same.

[0049] Following step S102, the estimation unit 11 stores the model parameter values ​​estimated in step S102, for example, in the auxiliary storage device 102 (S103). Figure 12 shows an example of parameter values ​​estimated for the residential damage / evacuation start model, and Figure 13 shows an example of parameter values ​​estimated for the residential damage reference model.

[0050] 14 is a diagram showing an example of the functional configuration of the evacuation support device 10 in the prediction phase. In Fig. 14, the evacuation support device 10 has an initial value generation unit 12 and a prediction unit 13. These units are realized by processing in which one or more programs installed in the evacuation support device 10 are executed by the processor 104.

[0051] The evacuation support device 10 may be configured using different computers for the estimation phase and the prediction phase.

[0052] FIG. 15 is a flowchart illustrating an example of a processing procedure executed by the evacuation support device 10 in the prediction phase.

[0053] In step S201, the initial value generation unit 12 generates an initial value of the residential damage level (residential damage level when cumulative rainfall is 0) for each resident of each region based on the probability distribution (initial probability distribution) relating to the mean value and standard deviation at the "initial time" of the residential damage / evacuation start model ( FIG. 12 ) for that region. Specifically, a random number based on the initial probability distribution is generated as the initial value. FIG. 16 shows an example of the generated initial value of the residential damage level for each resident of each region.

[0054] Next, the prediction unit 13 acquires a time-series predicted value of the cumulative rainfall for each region (hereinafter referred to as "predicted cumulative rainfall") (S202).

[0055] Fig. 17 is a diagram showing an example of predicted cumulative rainfall for each region. As shown in Fig. 17, predicted values ​​of cumulative rainfall for each time period are obtained. The predicted cumulative rainfall may be obtained, for example, from an actual weather forecast. Alternatively, when the processing procedure of Fig. 15 is executed as a simulation, the predicted cumulative rainfall may be a value set based on some kind of estimation.

[0056] Next, the prediction unit 13 obtains the degree of residential damage and whether or not evacuation has begun for each resident in each area in chronological order (at each time point) based on the initial values ​​of the predicted cumulative rainfall and residential damage degree, the residential damage / evacuation start model, and the residential damage standard model (S203).

[0057] 18 is a diagram illustrating a method for acquiring the degree of residential damage and whether or not evacuation has started for each resident of each area at a certain time. For a certain time (hereinafter referred to as "time t1"), the initial value, movement speed, cumulative rainfall, residential damage degree, residential damage classification, evacuation start threshold, and whether or not evacuation has started are shown for each resident of each area.

[0058] The initial values ​​for each resident are the initial values ​​(Figure 16) generated for that resident in step S201. The movement speed and evacuation start threshold for each resident are the movement speed and evacuation start threshold included in the residential damage / evacuation start model (Figure 12) for the area to which the resident belongs. Therefore, the initial values, movement speed, and evacuation start threshold are invariant with time. The cumulative rainfall is the cumulative rainfall at time t1 among the predicted cumulative rainfall obtained in step S202. Based on these, the prediction unit 13 obtains (derives) the residential damage level, residential damage classification, and whether or not evacuation has started at time t1 for each resident in each area.

[0059] Specifically, the prediction unit 13 obtains the residential damage level for a certain resident at time t1 by calculating the initial value for that resident plus the velocity plus the cumulative rainfall. The prediction unit 13 also determines the residential damage classification by applying the residential damage level to a residential damage reference model (FIG. 13). The prediction unit 13 further determines whether or not evacuation should be initiated by comparing the residential damage level with an evacuation start threshold. That is, if the residential damage level is equal to or greater than the evacuation start threshold, the prediction unit 13 determines whether or not evacuation should be initiated as "yes" (evacuation will be initiated), and if the residential damage level is less than the evacuation start threshold, the prediction unit 13 determines whether or not evacuation should be initiated as "no" (evacuation will not be initiated).

[0060] By performing the above for each resident and each time, the degree of damage to the home and whether or not evacuation has begun for each resident in each area can be obtained for each time. As a result, the timing of the start of evacuation can be predicted for each resident in each area.

[0061] Next, the prediction unit 13 tally up the number of residential damage classifications by region at each time based on the results of step S203 (FIG. 16) (S204). For example, if the number of residential damage classifications for each region at time t1 is to be counted, the number of residential damage classifications can be counted based on the "Residential Damage Classification" column in FIG. 16. FIG. 19 shows an example of the counting results of the number of residential damage classifications by region at each time. Note that while FIG. 16 shows data by resident, FIG. 19 shows data by household. For example, data by household can be derived from data by resident based on the correspondence between residents and households.

[0062] Next, the prediction unit 13 tallies the number of evacuees for each region at each time based on the results of step S203 (FIG. 16) (S205). For example, the number of evacuees for each region at time t1 can be obtained by calculating the sum of the number of residents whose "Evacuation Started" field is "Yes" for each time before time t1. In other words, the number of evacuees refers to the total (cumulative) number of residents who have evacuated. FIG. 20 shows an example of the results of tallying the number of evacuees for each region at each time.

[0063] Next, the prediction unit 13 outputs the counting results of steps S204 and S205 (S206). The user may create an evacuation plan by referring to the output counting results.

[0064] As described above, according to this embodiment, the problem of Non-Patent Document 1 can be solved, and the possibility of solving the problem of Non-Patent Document 2 can be increased.

[0065] In other words, the model parameters are independent of time intervals and are based on residential damage, making it easier to interpret the meaning. Also, by converting the disaster risk into an interval scale, prediction errors can be reduced. This can contribute to improving the accuracy of evacuation predictions. For example, by improving the accuracy of predicting the timing and number of residents who will begin evacuating when a disaster occurs, this can support the development of evacuation shelters and evacuation plans.

[0066] Note that interval scales and ordinal scales are concepts related to statistical variables, and generally speaking, if statistical processing is not performed in accordance with the concept of the variables, problems such as a decrease in accuracy will occur. In this embodiment, the five types of damage assessment criteria commonly used in Japan ("total destruction," "large-scale partial destruction," "partial destruction," "semi-partial destruction," and "partial damage") are ordinal scales, so if they are simply mechanically associated with numbers from 1 to 5 (interval scale) and subjected to statistical processing, there is a possibility of a decrease in accuracy and an increase in prediction error. In this embodiment, such mechanical association is not performed, and therefore, based on general theory, it is possible to avoid a decrease in accuracy and an increase in prediction error that would otherwise occur if an ordinal scale were mechanically associated with a numerical value.

[0067] In this embodiment, the probability distribution of the residential damage degree is assumed to be a normal distribution, but it does not necessarily have to be a normal distribution. However, by using a distribution such as a normal distribution whose cumulative distribution function (CDF) is easy to calculate, it becomes easier to estimate the model parameters.

[0068] Furthermore, in this embodiment, damage caused by disasters such as floods and landslides due to heavy rain is assumed, and cumulative rainfall is used as a measure of the magnitude of the disaster factors, and it is assumed that the degree of residential damage is determined by a linear function of cumulative rainfall. However, this embodiment may also be applied to disasters that are influenced by other factors. In that case, the other factors can be incorporated into the prediction model. For example, when other factors are incorporated, a neural network may be used to estimate a function showing the relationship between the degree of residential damage and the other factors.

[0069] Furthermore, Non-Patent Document 1 reports that more residents begin evacuating during the day than at night. In this embodiment, the number of evacuees is assumed to be a linear function of only the degree of housing damage, regardless of time. However, to increase the number of residents who begin evacuating during the day, time may be introduced into the function from the degree of housing damage to the number of evacuees. When time is introduced as an input to the function, the function may be estimated using, for example, a neural network.

[0070] Disaster damage includes not only damage to houses but also human damage. Since human damage can be classified into categories such as death, serious injury, minor injury, and injury, when applying this embodiment, human damage may be converted into a continuous index (an index for the degree of damage to houses).

[0071] In this embodiment, the probability distribution at the "time of disaster" is an example of a first probability distribution, the probability distribution at the "time of limit" is an example of a second probability distribution, and the probability distribution at the "time of initial stage" is an example of a third probability distribution.

[0072] Although the embodiments of the present invention have been described in detail above, the present invention is not limited to such specific embodiments, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

[0073] REFERENCE SIGNS LIST 10 Evacuation support device 11 Estimation unit 12 Initial value generation unit 13 Prediction unit 100 Drive device 101 Recording medium 102 Auxiliary storage device 103 Memory device 104 Processor 105 Interface device 121 House damage DB 122 Evacuation shelter DB 123 Rainfall DB B Bus

Claims

1. An evacuation support device comprising: an estimation unit configured to derive a first probability distribution in which an index related to the magnitude of the damage is a random variable based on information classifying the number of damage cases caused by a disaster in a certain area into multiple levels related to the magnitude of the damage; derive a third probability distribution for the index corresponding to a state in which the factor is absent and an increase in the index when the factor increases by a unit amount based on the first probability distribution, the magnitude of the factor, a second probability distribution having a different average value from the first probability distribution, and the magnitude of the factor corresponding to the second probability distribution; and derive a threshold value at which evacuation should begin for the index based on the first probability distribution and an actual value of the number of evacuees due to the disaster.

2. The evacuation support device as described in claim 1, characterized in having: an initial value generation unit configured to generate an initial value of the index for a resident in the area based on the third probability distribution; and a prediction unit configured to calculate a time series value for the index based on the initial value, a time series prediction value for the magnitude of the factor, and the increment, and to predict the timing at which the resident will begin evacuation based on the time series value and the threshold value.

3. An evacuation support method characterized by being executed by a computer, comprising the steps of: deriving a first probability distribution in which an index relating to the magnitude of the damage is a random variable based on information classifying the number of damage cases caused by a disaster in a certain area into multiple levels related to the magnitude of the damage; deriving a third probability distribution for the index corresponding to a state in which the factor is absent and an increase in the index when the factor increases by a unit amount based on the first probability distribution, the magnitude of the factor, a second probability distribution having a different mean value from the first probability distribution, and the magnitude of the factor corresponding to the second probability distribution; and deriving a threshold at which evacuation should begin for the index based on the first probability distribution and the actual number of evacuees due to the disaster.

4. The evacuation support method according to claim 3, characterized in that the computer executes the following steps: an initial value generation step of generating an initial value of the indicator for a resident in the area based on the third probability distribution; and a prediction step of calculating a time series value for the indicator based on the initial value, a time series prediction value for the magnitude of the factor, and the increment, and predicting the timing at which the resident will begin evacuation based on the time series value and the threshold value.

Images