Disaster prediction support method and disaster prediction support device

Abstract

To provide a disaster prediction support method and a disaster prediction support device capable of effectively supporting prediction of a disaster situation by a behavioral analysis.SOLUTION: A disaster prediction support method is a method for supporting prediction of a disaster situation of an object including a structure and natural objects, and includes: comparing a behavioral analysis result acquired on the basis of input information including object information and disaster information by behavioral analysis obtained by modeling mechanical behaviors of the object during a disaster, with a disaster evaluation result acquired by processing the input information by a disaster evaluation system undergoing machine learning; and correcting the behavioral analysis result in accordance with a comparison result between the behavioral analysis result and the disaster evaluation result.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a method and an apparatus for assisting in predicting the damage state of structures, natural objects, and other objects in the event of a disaster. [Background technology]

[0002] For structures such as reservoirs, levees, and buildings, as well as cliffs, land masses in mountainous areas, and other natural objects, various behavior analyses are conducted that model the mechanical behavior of the structures or natural objects in question in order to predict damage conditions in the event of a disaster such as an earthquake or heavy rain.

[0003] As an example, Non-Patent Document 1 describes an irrigation pond disaster prevention support system targeted at irrigation ponds among reservoirs. This irrigation pond disaster prevention support system calculates the amount of embankment subsidence from earthquake information and behavior analysis of the embankment of the irrigation pond, and judges the risk level against the allowable subsidence set for each irrigation pond in each area. In the behavior analysis used here, the so-called SIP Newmark-D method is used to calculate the amount of embankment subsidence for multiple earthquake motions with different maximum accelerations, and the relationship between the maximum earthquake acceleration and the amount of embankment subsidence is obtained. The above-mentioned SIP Newmark-D method is based on the Newmark method, which is one of the slope stability analyses (circular slide analysis) of embankments, and uses a model of the strength reduction of the embankment soil during an earthquake to make it easy to grasp the reduction in the strength of the embankment soil (see Non-Patent Document 2).

[0004] Incidentally, examples of disaster-related technologies include those described in Patent Documents 1 to 3.

[0005] Patent Document 1 describes a disaster prevention project planning support system that has an information input device, an information storage device, an information calculation device, and an information output device, and calculates a landslide occurrence limit line, an evacuation reference line, or a warning reference line (hereinafter, these may be collectively referred to as CL) using rainfall data and actual information on the occurrence or non-occurrence of landslides in each region, and the information input device is capable of inputting rainfall data and actual information on the occurrence or non-occurrence of landslides in each region (hereinafter, the rainfall data and the actual information on the occurrence or non-occurrence of landslides may be referred to as analysis data), and analysis parameters and analysis models for calculating the CL, into the information storage device. The information calculation device is provided with an analysis parameter setting section which reads out the analysis model from the information storage device and inputs to the analysis model the analysis parameters read out from the information storage device or the analysis parameters input from the information input device, and a discriminant boundary surface analysis section which calculates the CL by reading out analysis data from the information storage device or using the analysis data input from the information input device, and the information output device is provided with a means capable of outputting at least one piece of information out of the analysis data, the analysis parameters, the analysis model, and the CL.

[0006] Patent Document 2 describes a disaster prevention project planning support system that has an information input device, an information storage device, an information calculation device, and an information output device, and calculates the risk of disaster occurrence in a certain area based on a discrimination boundary line or discrimination boundary surface (hereinafter, the discrimination boundary line is collectively referred to as the discrimination boundary surface) that separates occurrence and non-occurrence of a disaster obtained by using factor data related to the factors of disaster occurrence in each area and actual data on the occurrence and non-occurrence of disasters, the information input device is a means capable of inputting data on factors of disaster occurrence in the certain area and boundary data constituting the discrimination boundary surface to the information storage device, and the information calculation device reads the boundary data from the information storage device. the factor data for the certain area read from the information storage device or the factor data for the certain area input from the information input device as coordinates on a two or more dimensional space (hereinafter referred to as multidimensional space) in which the boundary data is configured; and a potential risk calculation unit that calculates the distance from the discrimination boundary surface to the coordinates of the factor data for the certain area as the potential risk of the disaster occurring, wherein the information output device is a means capable of outputting at least one piece of information out of the factor data for the certain area, the boundary data, and the potential risk.

[0007] Patent Document 3 describes a "flood prediction device that predicts the presence or absence of flooding in a specified area, comprising: input means for inputting flood area data showing data on the flood area at the time of past flooding of the specified area together with map data of the specified area, rainfall data at the time of the past flooding, and considered rainfall data; parameter setting means for setting mesh division conditions and flood determination criteria; flood determination means for dividing the flood area data into meshes based on the mesh division conditions, determining the presence or absence of flooding for each mesh area using the flood determination criteria, and creating flood presence or absence data; machine learning means for applying a machine learning model to the flood presence or absence data and the rainfall data, and creating flood prediction standard data that predicts the presence or absence of flooding in each mesh area from the rainfall; prediction calculation means for creating flood prediction data that predicts the presence or absence of flooding in each mesh area at the rainfall amount of the considered rainfall data, based on the considered rainfall data and the flood prediction standard data; and display means capable of displaying the flood prediction data together with the map data of the specified area." [Prior art documents] [Patent documents]

[0008] [Patent Document 1] JP 2005-350972 A [Patent Document 2] JP 2007-035017 A [Patent Document 3] JP 2010-054266 A [Non-patent literature]

[0009] [Non-Patent Document 1] Toshikazu Hori et al., “Development of Disaster Prevention Support System for Irrigation Pond (DPSIP),” Journal of Disaster Research, March 1, 2019, Vol.14, No.2, p.303-314 [Non-Patent Document 2] Kazuhiro Ueno and 4 others, "Simple Estimation Method of Undrained Shear Strength of Saturated Soils Subjected to Undrained Cyclic Loading", Geotechnical Journal, Geotechnical Society of Japan, 2020, Vol. 15, No. 2, pp. 241-256 Summary of the Invention [Problem to be solved by the invention]

[0010] In accordance with the concept of safety design, structures are designed to ensure high safety in the event of a disaster, particularly to prevent threats to human life and property. The above-mentioned analysis of the behavior of a structure during a disaster is often based on the calculation method used in the design of the structure. In addition, when taking measures against landslides and other disasters, analysis of the behavior of natural objects can be performed to set conditions that anticipate sufficient safety.

[0011] In this case, the above behavior analysis simulates the damage to be more severe than it actually is, and the analysis results are far removed from reality. For example, the above-mentioned irrigation pond disaster prevention support system tends to overestimate the amount of subsidence of the embankment of an irrigation pond, and as a result, the risk is sometimes judged to be high, which is different from the actual situation.

[0012] Under these circumstances, it is desirable to be able to obtain highly accurate analysis results that closely resemble actual damage conditions when analyzing the behavior of structures and natural objects.

[0013] The present invention has an object to solve the above-mentioned problems, and an object of the present invention is to provide a damage prediction support method and a damage prediction support device that can effectively support prediction of a damage situation through behavior analysis. [Means for solving the problem]

[0014] The damage prediction support method of the present invention is a method for supporting prediction of a damage situation of an object including a structure and a natural object, Disaster prediction support device, A behavior analysis that models the mechanical behavior of the object during a disaster based on input information including object information and disaster information. Do a simulation The behavior analysis results obtained, A damage assessment result regarding a damage state of the object at the time of a disaster based on the disaster information. The input information The learning model is built in and comparing the behavior analysis result with a damage assessment result obtained by processing the behavior analysis result in a damage assessment system that has undergone machine learning, and correcting the behavior analysis result in accordance with a comparison result between the behavior analysis result and the damage assessment result. the behavior analysis result includes a behavior analysis value obtained by the behavior analysis and expressing the mechanical behavior of the object at the time of the disaster, and a judgment result regarding the magnitude relationship of the behavior analysis value with respect to an allowable reference value; and in correcting the behavior analysis result, a correction coefficient is obtained using the allowable reference value and the behavior analysis value, and the behavior analysis value is corrected by the correction coefficient. It is something.

[0015] The damage prediction support method of the present invention may include processing the input information with a damage assessment system that has undergone machine learning to obtain the damage assessment result.

[0016] It is preferable that the damage assessment system is one that has been trained using teacher data regarding object information during an actual disaster and the relationship between disaster information and the damage situation.

[0017] In the damage prediction support method of the present invention, the behavior analysis result and the damage assessment result each include a graded indicator of the degree of damage, and when the behavior analysis result and the damage assessment result indicate different graded indicators, the behavior analysis result can be corrected.

[0018] In addition, in the damage prediction support method of the present invention, it is preferable that the input information includes multiple disaster information having different disaster levels, and for each disaster information, the behavior analysis result is compared with the damage assessment result, and the behavior analysis result is corrected.

[0019] In this case, the behavior analysis result is ,before It is preferable to compare the behavior analysis results with the damage assessment results in order of disaster information with a greater disaster level, and when a difference occurs between the behavior analysis result and the damage assessment result in disaster information, a correction coefficient is calculated using the acceptable standard value and the behavior analysis value to correct the behavior analysis result.

[0020] In correcting the behavior analysis results, each behavior analysis value in each disaster information item can be corrected by the correction coefficient.

[0021] In the disaster prediction support method of the present invention, the disaster may be an earthquake and the target object may be a reservoir.

[0022] The damage prediction support device of the present invention includes a processing unit that executes a process for supporting prediction of a damage state of an object including a structure and a natural object, and the processing unit performs a behavior analysis in which a mechanical behavior of the object during a disaster is modeled, based on input information including object information and disaster information. Do a simulation The behavior analysis results obtained, A damage assessment result regarding a damage state of the object at the time of a disaster based on the disaster information, The input information The learning model is built in A damage assessment result obtained by processing the damage assessment result in a damage assessment system for which machine learning has been performed is compared, and the behavior analysis result is corrected according to a comparison result between the behavior analysis result and the damage assessment result. The behavior analysis result includes a behavior analysis value obtained by the behavior analysis and representing the mechanical behavior of the object at the time of the disaster, and a determination result regarding a magnitude relationship of the behavior analysis value with respect to an allowable reference value, and the processing unit, in correcting the behavior analysis result, determines a correction coefficient using the allowable reference value and the behavior analysis value, and corrects the behavior analysis value with the correction coefficient. It is something.

[0023] The damage prediction support device of the present invention may include a damage assessment system that performs machine learning and calculates the damage assessment result from the input information.

[0024] It is preferable that the damage assessment system is one that has been trained using teacher data regarding object information during an actual disaster and the relationship between disaster information and the damage situation.

[0025] In the damage prediction support device of the present invention, the behavior analysis result and the damage assessment result each include a graded indicator of the degree of damage, and when the behavior analysis result and the damage assessment result indicate different graded indicators, the processing unit can correct the behavior analysis result.

[0026] It is preferable that the input information includes a plurality of pieces of disaster information having different disaster levels, and the processing unit compares the behavior analysis result with the damage assessment result for each piece of disaster information, and corrects the behavior analysis result.

[0027] In the disaster prediction support device of the present invention, ,before It is preferable that the processing unit compares the behavior analysis result with the damage assessment result in order of disaster information of a larger disaster level, and when a different result occurs between the behavior analysis result and the damage assessment result in disaster information, a correction coefficient is calculated using the acceptable standard value and the behavior analysis value to correct the behavior analysis result.

[0028] In this case, the processing unit can correct the behavior analysis results by correcting each of the behavior analysis values ​​in each piece of disaster information with the correction coefficient.

[0029] In the disaster prediction support device of the present invention, the disaster may be an earthquake and the target object may be a reservoir. Effect of the Invention

[0030] According to the damage prediction support method and damage prediction support device of the present invention, the behavior analysis results are compared with the damage assessment results, and the behavior analysis results are corrected according to the comparison results, thereby effectively supporting the prediction of damage conditions through behavior analysis. [Brief description of the drawings]

[0031] [Figure 1] 2 is a flowchart showing each step of a disaster prediction support method according to one embodiment of the present invention. [Diagram 2]1 is a block diagram showing a disaster prediction support device according to an embodiment of the present invention; [Diagram 3] 3 is a graph showing an example of a relationship between a disaster level and a behavior analysis value of a behavior analysis result that can be used in the damage prediction support method of FIG. 1 or the damage prediction support device of FIG. 2. [Figure 4] 1 is a graph showing an example of the relationship between the maximum acceleration of an earthquake and the amount of embankment subsidence when behavior analysis of a reservoir embankment is performed using the irrigation pond disaster prevention support system. [Diagram 5] 13 is a flowchart showing each step of a disaster prediction support method according to another embodiment. [Figure 6] 6 is a flowchart showing the steps when the damage prediction support method of FIG. 5 is applied to damage prediction of a reservoir during an earthquake. [Figure 7] 1 is a photograph showing a centrifugal model experiment of an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0032] Hereinafter, an embodiment of the present invention will be described in detail. A damage prediction support method according to one embodiment of the present invention is a method for supporting prediction of a damage state of an object including a structure and a natural object. This embodiment includes, for example, steps S1 to S4 shown in FIG.

[0033] As shown in Fig. 1, first, a behavior analysis result and a damage assessment result are respectively obtained (step S1). The behavior analysis result is obtained based on input information including object information and disaster information by a behavior analysis that models the mechanical behavior of an object during a disaster. The damage assessment result is obtained by processing the above input information with a damage assessment system that has undergone machine learning.

[0034] Next, the behavior analysis result is compared with the damage assessment result (step S2), and it is determined whether or not these results are the same (step S3). If the behavior analysis result and the damage assessment result show the same result, no correction is made and the process ends. On the other hand, if the behavior analysis result and the damage assessment result show different results, the behavior analysis result is corrected (step S4) and then the process ends. In other words, the behavior analysis result is corrected according to the comparison result between the behavior analysis result and the damage assessment result. The corrected behavior analysis result can be used to predict the damage status of the target object.

[0035] However, in the past, behavior analysis that models the mechanical behavior of an object often resulted in behavior analysis results that indicated that the damage would be more severe than the actual damage situation, for example, due to an excessive safety factor being set to ensure sufficient safety against disasters. In this case, the actual situation was not properly simulated, and such behavior analysis results could not be used. In contrast, in this embodiment, as described above, the behavior analysis results are compared with the damage assessment results obtained by a damage assessment system that has been subjected to machine learning (step S2), and if the results differ, the behavior analysis results are corrected (steps S3 and S4). As a result, the results of the behavior analysis that were constructed with an expectation of excessive safety are corrected, so that the prediction accuracy of the damage situation in the behavior analysis results can be improved.

[0036] A damage prediction support device according to an embodiment of the present invention includes a processing unit that executes a process for supporting prediction of a damage state of an object including a structure and a natural object, The processing unit compares the behavior analysis result with a damage assessment result, and corrects the behavior analysis result in accordance with a comparison result between the behavior analysis result and the damage assessment result.

[0037] Typically, the disaster prediction support device can include, in addition to the above-mentioned processing unit, an input unit into which a user inputs predetermined information, an output unit that outputs the results of processing in the processing unit, and a storage unit that stores information, as exemplified in Fig. 2. The processing unit is configured to be capable of transmitting and receiving data or signals via a bus with the input unit, output unit, storage unit, etc. The processing unit often includes a processor, RAM (Random Access Memory), ROM (Read Only Memory), etc., and performs the above-mentioned processing by executing a program recorded in the ROM or storage unit. The storage unit may be composed of an HDD (Hard Disk Drive), flash memory, etc.

[0038] (Object) The objects to be predicted for damage include structures and natural objects. Among these, the structures include, for example, civil engineering structures such as reservoirs including irrigation ponds, dams, levees, artificial rivers, bridges, roads, railways, tunnels, and slopes formed by cutting and filling, as well as buildings and other architectural structures including various buildings and facilities. The natural objects include slopes formed by earth and sand such as cliffs and land masses, oceans, natural rivers and ponds, etc.

[0039] (disaster) Disasters are often caused by natural phenomena, and specific examples include heavy rain, heavy snow, floods, cliff collapses, debris flows, storm surges, earthquakes, tsunamis, landslides, etc. The embodiments of the present invention can be used to support prediction of damage conditions of the above-mentioned objects due to such various disasters.

[0040] Specific examples of damage include the subsidence or collapse of structures such as reservoirs, levees, and buildings during earthquakes, the collapse of natural soil and sand slopes, landslides, and debris flows, etc. There are also floods caused by the overflowing of artificial or natural rivers during heavy rains.

[0041] (Input information) The input information is used for input to the behavior analysis to obtain behavior analysis results, and for input to the damage assessment system to obtain damage assessment results, and includes object information and disaster information.

[0042] Among these, the object information may include various conditions of the object. For example, if the object is a reservoir embankment, a river embankment, a dam, or the like, which is a structure, the object information may include at least one piece of information selected from the group consisting of the crest width, embankment height, embankment crest length, upstream slope gradient, downstream slope gradient, embankment cross-sectional area, embankment volume, construction date, soil type, and embankment shape. Alternatively, if the object is a natural soil slope or an artificially constructed embankment or cut, the object information may include the slope or slope gradient, soil type, etc.

[0043] The disaster information includes the conditions of the disaster for which the damage situation is to be predicted, specifically, the seismic intensity and / or the maximum acceleration of the earthquake if the disaster is an earthquake, and the amount of rainfall per unit time and / or the total amount of rainfall if the disaster is heavy rain.

[0044] (Acquisition of behavior analysis results) The damage caused to an object during a disaster may be predicted by simulating the behavior analysis that models the structural mechanics, soil mechanics, material mechanics, hydraulics including hydraulics, and other mechanical behavior of the object. In many cases, behavior analysis is performed by calculation using a computer.

[0045] In such behavior analysis, a simulation is performed by inputting input information including object information and disaster information, and a behavior analysis result is obtained as a result. The behavior analysis result tends to indicate that the damage will be more severe than the actual damage situation. Therefore, in this embodiment, the behavior analysis result is corrected to be closer to the actual damage situation, as will be described in detail later.

[0046] The behavior analysis result may include a behavior analysis value that is directly obtained by the behavior analysis and represents the mechanical behavior of the object. Furthermore, the behavior analysis result may include a determination result as to whether the behavior analysis value is greater than, equal to, or smaller than a predetermined allowable reference value. In addition, when multiple pieces of disaster information with different disaster levels are input as input information in the behavior analysis, a behavior analysis result may be obtained for each of the multiple pieces of disaster information.

[0047] For example, Fig. 3 shows an example in which multiple pieces of disaster information with disaster levels 0 to 5 are input in behavior analysis, and behavior analysis values ​​corresponding to each piece of disaster information are obtained. Here, the behavior analysis value exceeds a preset allowable reference value when the disaster level is large, 3 to 5. In this example, the behavior analysis result includes a determination result that when the disaster level is small, such as 0 to 2, the behavior analysis value is smaller than the allowable reference value, but when the disaster level is large, such as 3 to 5, the behavior analysis value is larger than the allowable analysis value.

[0048] The behavior analysis result may include a graded indicator of the degree of damage, such as "heavy damage" when the behavior analysis value is larger than the allowable reference value, and "minor damage" when the behavior analysis value is equal to or smaller than the allowable reference value. In the example shown in FIG. 3, for disaster levels 3 to 5, the behavior analysis value is larger than the allowable reference value, so the indicator can be "heavy damage", and for disaster levels 0 to 2, the behavior analysis value is equal to or smaller than the allowable reference value, so the indicator can be "minor damage". The graded indicator is not limited to two levels, such as "heavy damage" or "minor damage", and may be set to three or more levels.

[0049] As a more specific example, as mentioned above, an irrigation pond disaster prevention support system is used as a behavior analysis to predict the damage situation of an irrigation pond during an earthquake. According to this irrigation pond disaster prevention support system, as shown in FIG. 4, the relationship between the maximum acceleration of an earthquake included in the disaster information of the input information and the behavior analysis value of the embankment subsidence of the irrigation pond by the behavior analysis is obtained. The allowable subsidence of the embankment is set in advance for each irrigation pond. In this case, the behavior analysis result may include the behavior analysis value of the embankment subsidence (subsidence calculated by earthquake analysis), the allowable subsidence of the embankment, and the judgment result of the magnitude relationship between them. In addition, the behavior analysis result may include a gradation index such as "large damage" when the behavior analysis value of the embankment subsidence is larger than the allowable subsidence, and "small damage" when the behavior analysis value of the embankment subsidence is equal to or smaller than the allowable subsidence. In addition, in the irrigation pond disaster prevention support system, when the behavior analysis value of the embankment subsidence is larger than the allowable subsidence, it is indicated as "large damage" and the risk of collapse is judged to be high, but the result may be that the risk is high even if the embankment does not actually collapse.

[0050] Other examples include the real-time damage prediction website cmap and a real-time tsunami inundation and damage estimation system.

[0051] The above-mentioned behavior analysis result may be obtained by performing behavior analysis in the processing unit of the damage prediction support device or the damage prediction support method of this embodiment, but it is also possible to separately obtain a result that has already been obtained by performing behavior analysis externally (step S1). Therefore, performing behavior analysis itself is not essential.

[0052] (Acquisition of damage assessment results) The damage assessment results used for comparison with the behavior analysis results are obtained by processing input information including object information and disaster information with a damage assessment system with machine learning. By using the damage assessment results from the damage assessment system with machine learning for comparison with the behavior analysis results, behavior analysis results that tend to be overly cautious can be effectively revised.

[0053] The damage assessment system is not particularly limited as long as it can learn through experience and output a damage assessment result on the damage state of an object when a disaster occurs due to the disaster information, in response to input information including object information and disaster information. Such damage assessment systems include those using k-means, decision trees, support vector machines (SVM), naive Bayes, ensemble methods, linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), k-nearest neighbor analysis (k-NN), etc. Among them, support vector machines are preferable because they have good discrimination accuracy even when there is a lot of object information, and parameters can be easily set.

[0054] The damage assessment system can also be one that has undergone deep learning using a Convolutional Neural Network (CNN). In a Convolutional Neural Network, features are extracted using multiple convolution layers, and classification is performed using multiple fully connected layers. This makes it possible to predict changes in the behavior of an object in response to a disaster based on more detailed damage conditions (for example, damage conditions that include quantitative indicators of the location of damage and the extent of damage of the object, rather than general indicators such as "heavy damage" or "minor damage"), making it possible to assess the damage to the object more specifically and with higher accuracy. Machine learning here also includes deep learning.

[0055] It is preferable that the above-mentioned damage assessment system is trained using teacher data on the relationship between information on objects that have actually been affected by disasters in the past, disaster information at that time, and the actual damage situation. In this way, by comparing and correcting the behavior analysis results using the damage assessment results of the damage assessment system, the accuracy of predicting the damage situation by behavior analysis can be further improved.

[0056] For example, the object information and disaster information of the training data can be the object conditions (top width, bank height, etc. when the object is a reservoir embankment, etc.) and disaster conditions similar to those of the object information and disaster information of the input information described above. The damage situation linked to such object information and disaster information can be a graded index of the degree of damage. Damage to reservoirs during earthquakes can be expressed in two graded indices, with "major damage" being the occurrence of major deformation or collapse of the reservoir embankment, and "minor damage" being the occurrence of no damage, minor cracks, subsidence, or damage to associated facilities. However, when expressing the damage situation using such graded indices, it is also possible to use three or more graded indices.

[0057] When the damage assessment system is machine-learned using the teacher data as described above, it is preferable to verify the accuracy of the damage assessment result by the machine-learned damage assessment system. For this verification, verification data on the relationship between the object information at the time of the actual disaster and the disaster information and the damage situation can be used. For example, a large amount of data including object information, disaster information, and the damage situation of the object that was actually damaged can be obtained, and machine learning of the damage assessment system can be performed using a part of it as teacher data, and the remaining part can be used as verification data to verify the machine-learned damage assessment system. In the verification using the verification data, it is preferable that the accuracy rate is 80% or more. The specific conditions of the verification at this time will be described later in the section on examples. If a high accuracy rate is obtained by the verification, it can be said that the reliability of the damage assessment result by the damage assessment system is high, and accordingly, the prediction accuracy of the damage situation in the behavior analysis can be greatly improved.

[0058] In order to facilitate comparison with the behavior analysis results described below, it is preferable that the damage assessment results obtained by the machine learning-based damage assessment system include an index of the degree of damage similar to that of the behavior analysis results. For example, when the behavior analysis results include a two or more stage indicator such as "heavy damage" or "little damage" as described above, it is preferable that the damage assessment results also include a similar two or more stage indicator.

[0059] In step S1, the damage assessment result obtained by another device such as a computer incorporating the above-mentioned damage assessment system can be obtained, for example, by wireless or wired communication or a storage device. Alternatively, it is also possible to obtain the damage assessment result by operating the damage assessment system in the embodiment of the present invention. In the latter case, the damage prediction support method of this embodiment further includes processing the input information with a damage assessment system that has undergone machine learning to obtain a damage assessment result. In this case, the damage prediction support device of this embodiment is equipped with a damage assessment system that has undergone machine learning and calculates a damage assessment result from the input information.

[0060] (Comparison and correction of results) The behavior analysis result and the damage assessment result as described above are compared to determine whether or not they represent the same level of damage (step S3). If the behavior analysis result and the damage assessment result differ, the behavior analysis result is corrected (step S4), and if the behavior analysis result and the damage assessment result are the same, the correction is not performed.

[0061] There is no particular restriction on the method of comparison, but as mentioned above, if the behavior analysis result and the damage assessment result each include a stage indicator for the degree of damage, it is possible to compare whether the stage indicator is the same or different between the behavior analysis result and the damage assessment result. For example, if the stage indicator is a two-stage indicator of "heavy damage" or "little damage", it is confirmed whether the behavior analysis result and the damage assessment result are both "heavy damage" or "little damage". When the behavior analysis result and the damage assessment result show different stage indicators, the behavior analysis result can be corrected.

[0062] Here, when the input information includes multiple disaster information with different disaster levels, the behavior analysis result can be compared with the damage assessment result for each of those disaster levels, and the behavior analysis result can be corrected according to the comparison result. When the behavior analysis result includes the above-mentioned behavior analysis value, the comparison between the behavior analysis result and the damage assessment result and the correction of the behavior analysis result can be specifically performed as shown in FIG. 5.

[0063] In FIG. 5, the behavior analysis result is compared with the damage assessment result in order of the disaster information of the largest disaster level. For example, when the disaster level is three levels, namely, "large", "medium" and "small", the behavior analysis result is compared with the damage assessment result for the disaster information of the "large" disaster level (step S11). Here, when the behavior analysis result and the damage assessment result differ from each other (for example, when the behavior analysis result is "large damage" and the damage assessment result is "small damage"), the result is "No" in step S12, and in step S13, the behavior analysis value and the allowable standard value for the disaster information of the "large" disaster level are used to calculate a correction coefficient. More specifically, the correction coefficient can be obtained by, for example, dividing the allowable standard value for the disaster information of the "large" disaster level by the behavior analysis value for the disaster information of the same disaster level of "large". After that, in step S14, the behavior analysis result (more specifically, the behavior analysis value) of each disaster information of the "large", "medium" and "small" disaster levels is corrected using the above correction coefficient, and the process ends.

[0064] Alternatively, if the behavior analysis result and the damage assessment result are the same for disaster information with a disaster level of "large," the behavior analysis result and the damage assessment result are compared for disaster information with a disaster level of "medium" (step S15). At this time, if the behavior analysis result and the damage assessment result differ (step S16), a correction coefficient is calculated using the behavior analysis value and the allowable standard value for disaster information with a disaster level of "medium" (step S17). Then, the behavior analysis results for disaster information with disaster levels of "large," "medium," and "small" are corrected with the correction coefficient (step S18), and the process ends.

[0065] If the behavior analysis result and the damage assessment result are the same for disaster information with disaster levels of "large" and "medium", in step S19, the behavior analysis result and the damage assessment result are compared for disaster information with a disaster level of "small". If the behavior analysis result and the damage assessment result are different for disaster information with a disaster level of "small" in step S20, a correction coefficient is calculated using the behavior analysis value and the allowable standard value for disaster information with a disaster level of "small" (step S21), and each behavior analysis result of disaster information with disaster levels of "large", "medium", and "small" is corrected with the correction coefficient (step S22), and the process ends.

[0066] In any of the disaster information with a disaster level of "large," "medium," or "small," if the behavior analysis result and the damage assessment result are the same, no correction is made and the process ends.

[0067] When such processing is applied to damage prediction of a reservoir during an earthquake, the flow is as shown in Fig. 6. In the example shown in Fig. 6, if it is assumed that the behavior analysis result (earthquake analysis) and the damage assessment result (machine learning) show different results ("small damage", "heavy damage") for earthquake information of seismic intensity 6, as shown in Fig. 4, the allowable subsidence of 2.0 m can be divided by the embankment subsidence of 4.0 m, which is the behavior analysis value at an earthquake acceleration equivalent to seismic intensity 6, to calculate a correction coefficient of 0.5. Then, by multiplying the behavior analysis value of the embankment subsidence at other earthquake accelerations by this correction coefficient of 0.5, the behavior analysis value can be corrected as shown as "after correction" in Fig. 4.

[0068] As described above, when the behavior analysis value of each disaster level is corrected using the correction coefficient calculated by dividing the allowable reference value by the behavior analysis value, the corrected behavior analysis value may become too small. In this case, the value calculated by dividing the allowable reference value by the behavior analysis value may be set as the upper limit, and the correction coefficient may be set by gradually decreasing the upper limit. However, the calculation of the correction coefficient is not limited to the method described above, and may be changed as appropriate according to the type of object or disaster, behavior analysis method, etc.

[0069] Furthermore, even when predicting damage to an object during a disaster other than that of an irrigation pond during an earthquake, a corrected behavior analysis result can be obtained by calculating the correction coefficient as described above (see Figure 3). As described above, the allowable standard value is divided by the behavior analysis value to obtain the correction coefficient, and each behavior analysis value is corrected by multiplying it by the correction coefficient, so that the behavior analysis value of the behavior analysis result is reduced as shown in the figure. This makes the behavior analysis result more approximate to the actual damage situation. By using such corrected behavior analysis result, it is possible to improve the accuracy of predicting the damage situation during a disaster using behavior analysis. EXAMPLES

[0070] Next, a damage prediction support method and a damage prediction support device according to the present invention were experimentally implemented, which will be described below. However, the description here is merely for illustrative purposes and is not intended to be limiting.

[0071] The target of damage prediction was reservoirs, and the aforementioned irrigation pond disaster prevention support system was used to obtain behavior analysis results. In addition, a damage assessment system was constructed to evaluate the damage situation of reservoirs during earthquakes in order to obtain damage assessment results.

[0072] The damage assessment system uses a support vector machine (SVM) as its algorithm and outputs a damage assessment result of either "major damage" or "minor damage" for input information including object information and disaster information.

[0073] In order to machine-learn the damage assessment system and to verify it, we collected and organized data on damage to irrigation ponds during past earthquakes. The total number of data was 1,655. The object information of the data includes the crest width, levee height, levee crest length, upstream slope gradient, downstream slope gradient, levee cross-sectional area, levee volume, construction date, soil type and levee shape, while the disaster information includes seismic intensity and maximum acceleration. In organizing the data, for example, the soil information was categorized as "1: Clay soil," "2: Sandy soil," or "3: Gravelly soil," the levee shape information as "1: Uniform type," "2: Surface water-proof," "3: Central core," or "4: Front bladed gold soil," and the construction date as "Before the Edo period," "Meiji," "Taisho," or "Showa," etc. Regarding the damage status, "minor damage" was used for no damage, or damage to water intake facilities, spillways, or revetment blocks, or minor cracks or subsidence, while "major damage" was used for collapse or damage to the embankment (major subsidence or deformation). Of the 1,655 pieces of data, 1,000 were "minor damage" and 655 were "major damage."

[0074] Seventy percent of the above data was used for machine learning of the damage assessment system, and the remaining 30% was used to verify the accuracy of the damage assessment results by the machine-learned damage assessment system. In the verification, feature scaling: standardization, boundary model: RBF kernel, grid search: c: 0.1 to 1000, γ: 0.0001 to 10. Then, the test was predicted using cross-validation and best_estimatоr, and verified with a confusion matrix. The results are shown in Table 1. From Table 1, it can be seen that the accuracy rate was 82%, and the damage assessment system can obtain damage assessment results with sufficiently high accuracy.

[0075] [Table 1]

[0076] Next, for 71 existing irrigation ponds, we targeted specific earthquakes that had actually occurred in the past, and compared the behavior analysis results from the irrigation pond disaster prevention support system with the damage assessment results from the above-mentioned damage assessment system, as described above using Figures 4 and 6, and corrected the behavior analysis values. Some of the results are shown in Table 2. Furthermore, it was found from field surveys that none of the 71 irrigation ponds suffered significant damage in this earthquake.

[0077] [Table 2]

[0078] From Table 2, it can be seen that in many irrigation ponds, the predicted subsidence amount has decreased due to the correction of the behavior analysis value after comparing the behavior analysis result with the damage assessment result. In addition, in the irrigation pond disaster prevention support system, the markings are displayed in the order of blue, yellow, and red, indicating increasing danger, according to the magnitude relationship of the subsidence amount of the behavior analysis value relative to the allowable subsidence amount. However, after the correction of the behavior analysis value, as shown in Table 2, the markings changed from yellow to blue (irrigation pond No. 3) and from red to yellow (irrigation ponds No. 1, 7, 12, and 15). From this, it can be seen that in the behavior analysis results before correction, many irrigation ponds that were not actually severely damaged were marked with a high danger level such as red, whereas in the behavior analysis results after correction, the markings are corrected to show a more realistic damage situation. Note that the red markings mean "danger," the yellow markings mean "caution," and the blue markings mean "safety."

[0079] Furthermore, a centrifugal model experiment was conducted simulating a specified reservoir to confirm whether the risk of earthquakes would be underestimated by using the above damage assessment system. The experiment is shown in the photograph in Figure 7. In this experiment, the assumed earthquake had a seismic intensity of 6, a maximum acceleration of 500 gal, and an input waveform of 5 Hz. The settlement of the model embankment after the earthquake in the experiment was 3 m. Using the above experimental conditions, we performed a behavior analysis of the irrigation pond disaster prevention support system and a damage assessment of the above damage assessment system in the same manner as described above, and compared the behavior analysis results with the damage assessment results and performed a correction process.The risk level indication after correction was still red, just as it was before the correction.This shows that the risk level was not underestimated by the damage assessment system.

[0080] From the above, it has been suggested that the present invention can effectively assist in predicting a damage situation through behavior analysis, and further suggests the possibility of greatly improving the accuracy of predicting a damage situation through behavior analysis of an object. [Explanation of symbols]

[0081] S1~S4, S11~S22 steps

Claims

1. A method for supporting prediction of a damage situation of an object including a structure and a natural object, comprising: Disaster prediction support device, A behavior analysis modeling the mechanical behavior of the object during a disaster is performed, and a behavior analysis result obtained by performing a simulation based on input information including object information and disaster information is compared with a damage assessment result regarding the damage state of the object during a disaster based on the disaster information, the damage assessment result being obtained by processing the input information with a damage assessment system incorporating a learning model and performing machine learning; and correcting the behavior analysis result according to a comparison result between the behavior analysis result and the damage assessment result; Including, the behavior analysis result includes a behavior analysis value obtained by the behavior analysis and representing a mechanical behavior of the object at the time of a disaster, and a determination result regarding a magnitude relationship of the behavior analysis value with respect to an allowable reference value, A damage prediction support method, comprising the steps of: determining a correction coefficient using the allowable reference value and the behavior analysis value; and correcting the behavior analysis value with the correction coefficient in correcting the behavior analysis result.

2. The method for supporting damage prediction according to claim 1 , further comprising processing the input information with a damage assessment system that has undergone machine learning to obtain the damage assessment result.

3. 3. The method for supporting disaster prediction according to claim 1, wherein the damage assessment system is trained using teacher data regarding object information during an actual disaster and a relationship between disaster information and a disaster situation.

4. The behavior analysis result and the damage assessment result each include a graded index of the degree of damage; 4. The damage prediction support method according to claim 1, further comprising the step of correcting the behavior analysis result when the behavior analysis result and the damage assessment result indicate different stage indexes.

5. the input information includes a plurality of pieces of disaster information with different disaster levels, 5. The method for supporting disaster prediction according to claim 1, further comprising the steps of: comparing the behavior analysis result with the damage assessment result for each disaster information; and correcting the behavior analysis result.

6. The comparison of the behavior analysis result and the damage assessment result is performed in order of disaster information of a large disaster level, The damage prediction support method according to claim 5, further comprising the steps of: determining a correction coefficient using the allowable standard value and the behavior analysis value to correct the behavior analysis result when a difference occurs between the behavior analysis result and the damage assessment result in disaster information;

7. The method for supporting disaster prediction according to claim 6 , wherein in correcting the behavior analysis result, each behavior analysis value in each disaster information is corrected by the correction coefficient.

8. The disaster prediction support method according to any one of claims 1 to 7, wherein the disaster is an earthquake and the object is a reservoir.

9. A damage prediction support device including a processing unit that executes a process for supporting prediction of a damage situation of an object including a structure and a natural object, The processing unit compares a behavior analysis result obtained by performing a simulation based on input information including object information and disaster information in a behavior analysis that models the mechanical behavior of the object during a disaster with a damage assessment result regarding the damage state of the object during a disaster based on the disaster information, which is obtained by processing the input information with a damage assessment system in which a learning model is incorporated and machine learning is performed, and corrects the behavior analysis result according to the comparison result between the behavior analysis result and the damage assessment result; the behavior analysis result includes a behavior analysis value obtained by the behavior analysis and representing a mechanical behavior of the object at the time of a disaster, and a determination result regarding a magnitude relationship of the behavior analysis value with respect to an allowable reference value, The processing unit, in correcting the behavior analysis result, determines a correction coefficient using the allowable reference value and the behavior analysis value, and corrects the behavior analysis value with the correction coefficient.

10. The damage prediction support device according to claim 9 , further comprising a damage assessment system that performs machine learning and calculates the damage assessment result from the input information.

11. 11. The damage prediction support device according to claim 9, wherein the damage assessment system is trained using teacher data regarding object information during an actual disaster and a relationship between disaster information and a damage situation.

12. The behavior analysis result and the damage assessment result each include a graded index of the degree of damage; 12. The damage prediction support device according to claim 9, wherein when the behavior analysis result and the damage assessment result show different stage indexes, the processing unit corrects the behavior analysis result.

13. the input information includes a plurality of pieces of disaster information with different disaster levels, 13. The disaster prediction support device according to claim 9, wherein the processing unit compares the behavior analysis result with the damage assessment result for each disaster information item, and corrects the behavior analysis result.

14. A damage prediction support device as described in claim 13, wherein the processing unit compares the behavior analysis results with the damage assessment results in order of disaster information starting from the disaster level of the largest disaster, and when a different result occurs between the behavior analysis result and the damage assessment result in disaster information, a correction coefficient is calculated using the acceptable standard value and the behavior analysis value to correct the behavior analysis result.

15. The disaster prediction support device according to claim 14 , wherein the processing unit corrects each of the behavior analysis values ​​in each disaster information item by the correction coefficient in correcting the behavior analysis result.

16. 16. The disaster prediction support device according to claim 9, wherein the disaster is an earthquake and the object is a reservoir.

Images