WO2022172442A1

WO2022172442A1 - Flooding prediction program, flooding prediction device, and machine-learning method

Info

Publication number: WO2022172442A1
Application number: PCT/JP2021/005495
Authority: WO
Inventors: 裕介大石; 孝政新出
Original assignee: 富士通株式会社
Priority date: 2021-02-15
Filing date: 2021-02-15
Publication date: 2022-08-18
Also published as: JPWO2022172442A1

Abstract

This flooding prediction program causes a computer to execute an acquisition process, a generation process, and an output process. The acquisition process acquires observation information pertaining to a target district. The generation process generates a first image representing a prediction result of a flooding situation in the target district by inputting the observation information to a first machine-learning model, and generates a second image of a higher resolution than the first image by inputting the first image to a second machine-learning model. The output process outputs the second image as the flooding prediction result of the target district.

Description

Inundation prediction program, inundation prediction device and machine learning method

Embodiments of the present invention relate to inundation prediction technology.

Conventionally, there is a known technique for predicting subsequent inundation based on observational information such as precipitation and typhoon conditions. As a conventional technology related to this inundation prediction, we build a mathematical model that derives optimal parameters by machine learning past rainfall and water level data. Then, using the constructed mathematical model, future water levels can be accurately predicted based on rainfall and water level data up to the present, as well as weather (precipitation forecast) data several hours ahead delivered to each local government from meteorological organizations. there is a technology to In addition, there is a conventional technology that predicts in real time the planar inundation status of urban water areas and presents evaluation results when analysis grids are standard 5-m and 10-m meshes.

However, with the conventional technology described above, when trying to perform highly accurate inundation prediction using a machine learning model with a finer mesh (analysis grid) and high resolution across the surface, the amount of data required for processing becomes large, and the required memory and memory are required. There is a problem that the amount of calculation becomes enormous.

One aspect aims to reduce the computational cost of inundation prediction.

In one proposal, the inundation prediction program causes a computer to execute acquisition processing, generation processing, and output processing. The acquisition process acquires observation information about the target area. In the generating process, the observation information is input to the first machine learning model to generate the first image showing the prediction result of the inundation situation in the target area, and the first image is applied to the second machine learning model. The input produces a second image with higher resolution than the first image. In the outputting process, the second image is output as the inundation prediction result of the target area.

　The calculation cost of inundation prediction can be reduced.

FIG. 1 is an explanatory diagram illustrating an overview of inundation prediction (at the time of machine learning) according to the embodiment. FIG. 2 is an explanatory diagram for explaining downsampling and super-resolution. FIG. 3 is an explanatory diagram for explaining an outline of inundation prediction (at the time of prediction) according to the embodiment. FIG. 4 is a block diagram of a functional configuration example of the information processing apparatus according to the embodiment; FIG. 5 is a flowchart illustrating an operation example of the information processing apparatus according to the embodiment during machine learning. FIG. 6 is an explanatory diagram for explaining an overview of downsampling. 7 is a flowchart illustrating an operation example of the information processing apparatus according to the embodiment during machine learning; FIG. FIG. 8 is a flowchart illustrating an operation example of the information processing apparatus according to the embodiment during prediction. FIG. 9 is a block diagram showing an example of a computer configuration.

The inundation prediction program, inundation prediction device, and machine learning method according to the embodiment will be described below with reference to the drawings. Configurations having the same functions in the embodiments are denoted by the same reference numerals, and overlapping descriptions are omitted. Note that the inundation prediction program, the inundation prediction device, and the machine learning method described in the following embodiments are merely examples, and do not limit the embodiments. Moreover, each of the following embodiments may be appropriately combined within a non-contradictory range.

In this embodiment, a prediction map showing the predicted value of the inundation situation at each point using a machine learning model based on observational information related to inundation in the target area such as precipitation and water level. Ask for Note that the target area for inundation prediction is a specific area (for example, a city area, a town area) or the like that is demarcated by administration or the like. Observation information includes the water level, flow rate, inflow, and discharge of rivers and sea areas related to the target area, and the amount of precipitation in the area including the target area.

Specifically, in the embodiment, based on observation information about the target area and a prediction map corresponding to this observation information, machine learning is performed so that the prediction map is output as the correct answer for the observation information input. , to create a machine learning model. Then, at the time of prediction, by inputting actual observation information about the target area into the machine learning model, a prediction map is obtained from the output of the machine learning model.

FIG. 1 is an explanatory diagram explaining an overview of inundation prediction (during machine learning) according to the embodiment. As shown in FIG. 1, in S1, the information processing apparatus according to the embodiment prepares a set of training data used for machine learning.

Specifically, in S1, the information processing device performs a simulation or the like (S1a) based on the observation information D1 regarding the target area, and obtains a planar high-resolution prediction (for example, a 5m mesh) obtained by making the mesh finer. Obtain high-resolution inundation prediction data D2, which is a map.

The high-resolution inundation prediction data D2 is the correct answer corresponding to the observation information D1, which indicates the predicted value of the inundation situation at each point corresponding to each mesh (water level from the ground, underfloor inundation, presence of above-floor inundation, etc.) is a forecast map of Also, the pixel value of each pixel in the high-resolution inundation prediction data D2 corresponds to the prediction value of the inundation situation at each point in the target area.

The method of obtaining high-resolution inundation prediction data D2 by an information processing device is not limited to obtaining by simulation from observation information D1. For example, in S1a, the information processing device may acquire high-resolution inundation prediction data D2 based on observed values (water level at each point, etc.) obtained by actual observation.

Next, in S1, the information processing device down-samples the prediction map in the high-resolution inundation prediction data D2 (S1b), thereby obtaining low-resolution inundation prediction data D3 having a resolution lower than that of the high-resolution inundation prediction data D2. .

FIG. 2 is an explanatory diagram explaining downsampling and super-resolution. As shown in FIG. 2, for example, the information processing device obtains low-resolution inundation prediction data D3 of 50 m mesh by down-sampling the high-resolution prediction inundation data D2 of 5 m mesh.

In S1, the information processing device prepares data sets of high-resolution inundation prediction data D2 and low-resolution inundation prediction data D3 corresponding to observation information D1 from observation information D1 in various cases as training data.

Next, the information processing device performs known machine learning using a regression prediction technique such as a neural network based on the observation information D1 in the prepared training data and the low-resolution inundation prediction data D3. A machine learning model M1 is generated (S2). Specifically, the information processing device is configured so that the output from the first machine learning model M1 when the observation information D1 is input to the first machine learning model M1 is the low-resolution inundation prediction data D3 that is the correct answer. , parameters of the first machine learning model M1 are set using known methods such as the gradient method and the error backpropagation method.

In addition, the information processing device uses the low-resolution inundation prediction data D3 and the high-resolution inundation prediction data D2 in the prepared training data to create a model in a known super-resolution technology using a CNN (Convolutional Neural Network) or the like. By performing machine learning, a second machine learning model M2 is generated (S3). Specifically, the second machine learning model M2 is a CNN that provides a single image super-resolution method for obtaining a higher-resolution image by increasing the resolution (super-resolution) from one image. . The information processing device has a gradient such that the output from the second machine learning model M2 when the low-resolution inundation prediction data D3 is input to the second machine learning model M2 is the low-resolution inundation prediction data D3 that is correct. The parameters of the second machine learning model M2 are set using a known method such as the method or error backpropagation method.

In general super-resolution methods, various images (with different sub-grid scale structures) are targeted for high resolution, so machine learning is performed using a large number of training data (examples). On the other hand, in the super-resolution method of the embodiment, the predicted value (pixel value) of each point in the target area (same topography and building shape within the sub-grid scale) is the object of high resolution. Therefore, less training data (examples) than general super-resolution is sufficient for generating the second machine learning model M2.

FIG. 3 is an explanatory diagram for explaining an overview of inundation prediction (at the time of prediction) according to the embodiment. As shown in FIG. 3, at the time of prediction (S4), the information processing apparatus of the embodiment collects information distributed from meteorological organizations and the like and measurement data from measurement devices, and generates observation information D10 regarding the target area. get.

Next, the information processing device inputs the obtained observation information D10 to the first machine learning model M1 to generate low-resolution inundation prediction data D11 indicating the prediction result (prediction map) of the inundation situation in the target area.

Next, the information processing device inputs the generated low-resolution inundation prediction data D11 to the second machine learning model M2 to generate high-resolution inundation prediction data D12 having a higher resolution than the low-resolution inundation prediction data D11. Specifically, as shown in FIG. 2, the information processing device obtains high-resolution inundation prediction data D12 by increasing the resolution (super-resolution) of the low-resolution inundation prediction data D11 using the second machine learning model M2. . As a result, the information processing apparatus can obtain the high-resolution inundation prediction data D12, which is a prediction map of the inundation situation with high resolution (for example, 5m mesh).

When attempting to directly obtain high-resolution inundation prediction data D12, which is a high-resolution prediction map, from observation information D10 using a machine learning model (Task A), the amount of data required for processing becomes large, requiring a large amount of memory and computation. become a thing. The reason for this is that, for example, if the predicted value is y and the observed value included in the observation information D10 is x, the number of data elements in y is orders of magnitude larger than that of x in y=f(x). becomes large, and machine learning requires a huge amount of data.

Therefore, in the information processing apparatus of the embodiment, low-resolution inundation prediction data D11, which is a low-resolution prediction map with coarser resolution than high-resolution inundation prediction data D12, is obtained from the observation information D10 using the first machine learning model M1 (task B). Next, the information processing apparatus of the embodiment obtains high-resolution inundation prediction data D12 by increasing the resolution of the low-resolution inundation prediction data D11 using the second machine learning model M2 (task C).

As an example, let N_A be the data volume (number of cases) during machine learning in task (A) to directly obtain high-resolution inundation prediction data D12, which is a high-resolution prediction map, from observation information D10, and the data size for each case be S_A. Also, let N_B be the data amount (number of cases) during machine learning in task (B) to obtain low-resolution inundation prediction data D11, which is a high-resolution prediction map, from observation information D10, and S_B be the data size for each case. and Let N_C be the data amount (the number of cases) during machine learning in task (C) for obtaining high-resolution prediction flood data D12 from low-resolution prediction flood data D11, and S_C be the data size for each case.

Since task (B) performs machine learning on how flooding occurs from various observation information D10, the number of cases is about the same as task (A) (N_A ≒ N_B). However, since the amount of data for each case used for machine learning is the low-resolution inundation prediction data D3 down-sampled from the high-resolution inundation prediction data D2, S_A>>S_B.

Task (C) obtains high-resolution inundation prediction data D12 by increasing the resolution of low-resolution inundation prediction data D11, and differences in observation information D1 do not matter during machine learning. In addition, in the machine learning related to task (C), the predicted values (pixel values) of each point in the same target area (same terrain and building shape within the sub-grid scale) are targeted for high resolution, so the number of cases is less (N_A>>N_C).

For this reason, comparing the data size (number of cases x data amount for each case) during machine learning in task (B) + task (C) with the data size during machine learning in task (A) is as follows. be.
N_A*S_A>>N_B*S_B+N_C*S_C

That is, task (B) + task (C) has a smaller data size than task (A). Specifically, a difference of about 100 times may occur. Therefore, in the information processing apparatus of the embodiment, it is possible to suppress the amount of data during machine learning and prevent the amount of calculation from increasing.

FIG. 4 is a block diagram showing a functional configuration example of the information processing device according to the embodiment. As shown in FIG. 4 , the information processing apparatus 1 has a communication section 10 , a display section 11 , an operation section 12 , an input/output section 13 , a storage section 14 and a control section 15 . This information processing device 1 is an example of a flood prediction device, and for example, a PC (Personal Computer) can be applied.

The communication unit 10 is realized by, for example, a NIC (Network Interface Card) or the like. The communication unit 10 is a communication interface that is wired or wirelessly connected to another information processing apparatus via a network (not shown) and controls information communication with the other information processing apparatus.

The display unit 11 is a display device for displaying various information. The display unit 11 is implemented by, for example, a liquid crystal display as a display device. The display unit 11 displays various screens such as a display screen input from the control unit 15 .

The operation unit 12 is an input device that receives various operations from the user of the information processing device 1 . The operation unit 12 is realized by, for example, a keyboard, a mouse, etc. as an input device. The operation unit 12 outputs the operation input by the user to the control unit 15 as operation information. The operation unit 12 may be realized by a touch panel or the like as an input device, and the display device of the display unit 11 and the input device of the operation unit 12 may be integrated.

The input/output unit 13 is, for example, a memory card R/W (Reader/Writer). The input/output unit 13 may read the observation information 141 or the like stored in the memory card and store it in the storage unit 14 instead of the observation information 141 or the like received by the communication unit 10 . Also, the input/output unit 13 stores, for example, the prediction result output from the control unit 15 in a memory card. As the memory card, for example, an SD memory card or the like can be used.

The storage unit 14 is realized by, for example, semiconductor memory devices such as RAM (Random Access Memory) and flash memory, and storage devices such as hard disks and optical disks. The storage unit 14 stores observation information 141, training data 142, prediction data 143, first machine learning model information 144, second machine learning model information 145, and the like.

The observation information 141 is observation information about the target area obtained from a server or the like, and corresponds to the observation information D1 and D10 described above.

The training data 142 is data used for machine learning of the first machine learning model M1 and the second machine learning model M2. Specifically, the training data 142 is a set of the observation information D1 described above, the high-resolution inundation prediction data D2 corresponding to the observation information D1, and the low-resolution inundation prediction data D3 for each case used for machine learning. be.

The prediction data 143 is data indicating prediction results from the observation information D10. Specifically, the prediction data 143 is the low-resolution inundation prediction data D11 obtained by inputting the observation information D10 into the first machine learning model M1, and the low-resolution inundation prediction data D11 obtained by the second machine learning model M2. corresponds to the high-resolution inundation prediction data D12 obtained by inputting to .

The first machine learning model information 144 is information related to the first machine learning model M1, and includes parameters and the like for constructing the first machine learning model M1 such as a neural network. The second machine learning model information 145 is information about the second machine learning model M2, and includes parameters and the like for constructing the second machine learning model M2 such as CNN.

The control unit 15 is a processing unit that controls the operation of the information processing device 1 . The control unit 15 has an acquisition unit 151 , a training data generation unit 152 , a first machine learning unit 153 , a second machine learning unit 154 , an estimation unit 155 and an output unit 156 . The control unit 15 can be realized by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like. The control unit 15 can also be realized by hardwired logic such as ASIC (Application Specific Integrated Circuit) and FPGA (Field Programmable Gate Array).

The acquisition unit 151 is a processing unit that acquires the observation information 141. For example, the acquisition unit 151 acquires the observation information 141 by communicating via the communication unit 10 with a specific server that provides the observation information 141 regarding the target area. The acquisition unit 151 stores the acquired observation information 141 in the storage unit 14 .

The training data creation unit 152 is a processing unit that creates the training data 142 used for machine learning of the first machine learning model M1 and the second machine learning model M2. Specifically, the training data creation unit 152 performs a simulation or the like from the observation information D1 (141) to obtain high-resolution inundation prediction data D2 corresponding to the observation information D1. Note that the training data creation unit 152 may obtain the high-resolution inundation prediction data D2 by directly applying the observation values included in the observation information D1.

Next, the training data creation unit 152 acquires low-resolution inundation prediction data D3 with a resolution lower than that of the high-resolution inundation prediction data D2 by down-sampling the prediction map in the high-resolution inundation prediction data D2. Next, the training data creation unit 152 stores in the storage unit 14 training data 142 that is a set of the observation information D1 and the high-resolution inundation prediction data D2 and low-resolution inundation prediction data D3 corresponding to the observation information D1. The training data creation unit 152 creates training data 142 for the number of cases by performing the above process for the number of cases used for machine learning.

The first machine learning unit 153 is a processing unit that generates the first machine learning model M1 based on the training data 142. Specifically, the first machine learning unit 153 reads the observation information D1 and the low-resolution inundation prediction data D3 for each case included in the training data 142 . Next, the first machine learning unit 153 inputs the read observation information D1 to the first machine learning model M1, and the low-resolution inundation prediction data D3 that the output from the first machine learning model M1 is correct. The parameters of the first machine learning model M1 are set (adjusted) so that

Next, after the first machine learning unit 153 performs the above processing on each of the predetermined number of cases prepared as the training data 142, the parameters of the first machine learning model M1 are set to the first machine learning model information 144 , and stored in the storage unit 14 .

The second machine learning unit 154 is a processing unit that generates a second machine learning model M2 based on the training data 142. Specifically, the second machine learning unit 154 reads the low resolution inundation prediction data D3 and the high resolution inundation prediction data D2 for each case included in the training data 142 . Next, the second machine learning unit 154 inputs the read low-resolution flood prediction data D3 to the second machine-learning model M2, and the high-resolution flood prediction that the output from the second machine-learning model M2 is correct. The parameters of the second machine learning model M2 are set (adjusted) so as to become the data D2.

Next, the second machine learning unit 154 performs the above processing on each of the predetermined number of cases prepared as the training data 142, and then converts the parameters of the second machine learning model M2 into the second machine learning model information 145. , and stored in the storage unit 14 .

The estimation unit 155 is a processing unit that estimates high-resolution inundation prediction data D12, which is a prediction map of the flood situation in the target area, based on the observation information 141 (D10) of the target area. Specifically, the estimation unit 155 constructs the first machine learning model M1 based on the first machine learning model information 144 read from the storage unit 14 . Next, the estimation unit 155 generates low-resolution inundation prediction data D11 by inputting the observation information D10 to the first machine learning model M1. The estimation unit 155 stores the generated low-resolution inundation prediction data D<b>11 as prediction data 143 in the storage unit 14 .

Next, the estimation unit 155 constructs the second machine learning model M2 based on the second machine learning model information 145 read from the storage unit 14. Next, the estimation unit 155 generates high-resolution flood prediction data D12 by inputting the low-resolution prediction flood data D11 to the second machine learning model M2. The estimation unit 155 stores the generated high-resolution inundation prediction data D<b>12 as prediction data 143 in the storage unit 14 .

The output unit 156 is a processing unit that outputs the estimation result of the estimation unit 155 . Specifically, the output unit 156 reads the high-resolution inundation prediction data D12 included in the prediction data 143 from the storage unit 14, and based on the high-resolution inundation prediction data D12, predicts the inundation situation at each point in the target area. Generate a map display screen. Next, the output unit 156 outputs the data of the generated display screen to the display unit 11 to display the prediction map on the display device.

The output destination of the output unit 156 is not limited to the display unit 11. For example, the output unit 156 may output the high resolution flood prediction data D12 read from the storage unit 14 to the memory card via the input/output unit 13 .

FIG. 5 is a flowchart showing an operation example of the information processing apparatus 1 according to the embodiment during machine learning. In the flowchart shown in FIG. 5, it is assumed that the training data 142 is created by simulation based on the observation information D1.

As shown in FIG. 5, when the process is started, the training data creation unit 152 performs various types of simulations (for example, different weather conditions) N times based on the training data 142 (D1). (S10) to acquire N sets of high-resolution flood prediction data D2.

Next, the training data creation unit 152 down-samples the N sets of high-resolution inundation prediction data D2 to acquire low-resolution inundation prediction data D3 from the high-resolution inundation prediction data D2 (S11).

FIG. 6 is an explanatory diagram explaining an outline of downsampling. As shown in FIG. 6, the training data creation unit 152 acquires low-resolution inundation prediction data D3 from high-resolution inundation prediction data D2 by performing downsampling on a predetermined scale (for example, 5 m mesh→50 m mesh). .

Here, when down-sampling pixels included in a specific area of the high-resolution inundation prediction data D2, the training data creation unit 152 samples pixels with the worst inundation prediction evaluation among the pixels included in the area. By doing so, the low-resolution inundation prediction data D3 is acquired. Here, the good or bad inundation evaluation means the height of the water level from the ground, and the higher the water level, the worse the evaluation (the evaluation is worse for the inundation above the floor than for the inundation under the floor). As a result, the training data creation unit 152 prevents pixels with the worst inundation prediction evaluation from being removed by downsampling.

Specifically, in case C1, the high-resolution inundation prediction data D2a of the specific region includes a right diagonal shaded portion (predicted value a) and a left diagonal shaded portion (predicted value b). , and the predicted value b has a lower evaluation than the predicted value a. In this case, the training data creation unit 152 samples the predicted value b with a poor evaluation. That is, the training data creation unit 152 down-samples the high-resolution flood prediction data D2a to the low-resolution flood prediction data D3a with the prediction value b for one pixel.

Similarly, in case C2, the high-resolution inundation prediction data D2b of the specific area includes a right diagonal shaded portion (predicted value a) and a left diagonal shaded portion (predicted value b). Therefore, the training data generation unit 152 down-samples the high-resolution flood prediction data D2b to the low-resolution flood prediction data D3b in which the prediction value b is one pixel.

In case C3, the high-resolution inundation prediction data D2c of the specific region includes the diagonally right shaded portion (predicted value a), but does not include the diagonally left shaded portion (predicted value b). Therefore, the training data generation unit 152 down-samples the high-resolution flood prediction data D2c to the low-resolution flood prediction data D3c in which the prediction value a is one pixel.

Next, the training data creation unit 152 prepares N sets of observation information D1 and low-resolution inundation prediction data D3 as training data 142 (S12). Next, the first machine learning unit 153 divides the N sets of training data 142 into N=N_train+N_test (S13).

Next, the first machine learning unit 153 performs machine learning for the first machine learning model M1 using a set of the N_train observation information D1 and the low-resolution flood prediction data D3. The first machine learning unit 153 also validates the first machine learning model M1 after machine learning using a set of the N_test observation information D1 and the low-resolution inundation prediction data D3. The first machine learning unit 153 adopts the first machine learning model M1 whose prediction error is confirmed to be within a predetermined range by validation, and constructs the first machine learning model M1 (S14).

Further, following S11, the training data creation unit 152 prepares M sets of high-resolution inundation prediction data D2 and low-resolution inundation prediction data D3 as training data 142 (S15). Next, the second machine learning unit 154 divides the M sets of training data 142 into M=M_train+M_test (S16). Note that the number of sets is M<<N.

Next, the second machine learning unit 154 performs machine learning of the second machine learning model M2 using a set of the M_train high-resolution inundation prediction data D2 and low-resolution inundation prediction data D3. Also, the second machine learning unit 154 validates the second machine learning model M2 after machine learning with a set of the high-resolution inundation prediction data D2 of M_test and the low-resolution inundation prediction data D3. The second machine learning unit 154 adopts the second machine learning model M2 whose prediction error is confirmed to be within a predetermined range by validation, and constructs the second machine learning model M2 (S17).

Following S14 and S17, the control unit 15 stores the first machine learning model information 144 and the second machine learning model information 145 regarding the constructed first machine learning model M1 and second machine learning model M2 in the storage unit 14. , and terminate the process. The processing of S13 and S14 by the first machine learning unit 153 and the processing of S16 and S17 by the second machine learning unit 154 may be performed in parallel.

FIG. 7 is a flowchart showing an operation example of the information processing apparatus 1 according to the embodiment during machine learning. Note that in the flowchart shown in FIG. 7, the training data 142 is created by adopting the observation data included in the observation information D1 without performing a simulation.

As shown in FIG. 7, when the process is started, the training data creation unit 152 prepares N pieces of observation information D1 of various types (S10a). Here, it is assumed that the observation information D1 also includes high-resolution inundation observation data (actually measured inundation situation map) corresponding to the high-resolution inundation prediction data D2.

The training data creation unit 152 down-samples the low-resolution inundation observation data from the high-resolution inundation observation data for the N sets of observation information D1 (S11a). Next, the training data generator 152 prepares N sets of observation information D1 and low-resolution inundation observation data as training data 142 (S12a). After that, the first machine learning unit 153 performs the same processing as S13 and S14 described above based on the prepared training data 142 to build the first machine learning model M1.

Further, following S11a, the training data creation unit 152 prepares M sets of high-resolution inundation observation data and low-resolution inundation observation data as training data 142 (S15a). After that, the second machine learning unit 154 performs the same processing as S16 and S17 described above based on the prepared training data 142 to build the second machine learning model M2.

FIG. 8 is a flowchart showing an operation example of the information processing apparatus 1 according to the embodiment during prediction. As shown in FIG. 8, when the process is started, the acquisition unit 151 acquires observation information D10 at the time of disaster occurrence from a specific server or the like that provides observation information on the target area (S20).

Next, the estimation unit 155 constructs the first machine learning model M1 from the first machine learning model information 144. Next, the estimation unit 155 acquires low-resolution inundation prediction data D11 by inputting the acquired observation information D10 into the constructed first machine learning model M1 (S21).

Next, the estimation unit 155 constructs a second machine learning model M2 from the second machine learning model information 145. Next, the estimating unit 155 acquires high-resolution inundation prediction data D12 by inputting the low-resolution inundation prediction data D11 acquired in S21 to the constructed second machine learning model M2 (S22). The output unit 156 outputs the high-resolution inundation prediction data D12 acquired in S22 to the display unit 11 (S23), and displays a prediction map based on the observation information D10 when a disaster occurs.

As described above, the information processing device 1 acquires the observation information D10 regarding the target area. The information processing device 1 inputs the obtained observation information D10 to the first machine learning model M1 to generate a first image (low-resolution inundation prediction data D11) showing the prediction result of the inundation situation in the target area. . The information processing device 1 inputs the first image to the second machine learning model M2 to generate a second image (high-resolution flood prediction data D12) having a resolution higher than that of the first image. The information processing device 1 outputs a second image generated as a flood prediction result for the target area.

As a result, the information processing device 1 can perform highly accurate inundation prediction using a machine learning model at high resolution while suppressing an increase in the amount of memory and calculation required for processing. In this way, the information processing device 1 can assist in increasing the accuracy of flood prediction.

The first image generated by the information processing device 1 is a map image showing the flooding situation at each point in the target area. As a result, the information processing apparatus 1 can obtain a higher-definition map image of the inundation situation at each point of the target area.

The information processing device 1 also down-samples the observation information D1 about the target area and the corresponding inundation prediction image (high-resolution inundation prediction data D2). The information processing device 1 generates a first machine learning model M1 based on the down-sampled inundation prediction image (low-resolution inundation prediction data D3) and the observation information D1. The information processing device 1 generates a first machine learning model M1 based on the downsampled inundation prediction image (low-resolution inundation prediction data D3) and the original inundation prediction image (high-resolution inundation prediction data D2). A second machine learning model M2 is generated that takes the output as an input and outputs an image obtained by increasing the resolution of the input image.

As a result, the information processing device 1 can create a machine learning model that performs highly accurate inundation prediction at high resolution while suppressing increases in memory and computational complexity required for processing. In this way, the information processing device 1 can assist in increasing the accuracy of inundation prediction.

Further, when down-sampling pixels included in a specific region of the predicted flood image (high-resolution predicted flood data D2), the information processing device 1 performs the most inundation prediction evaluation among pixels included in the region. Sample bad pixels. Thereby, in the information processing device 1, it is possible to prevent pixels with the worst inundation prediction evaluation from being removed by downsampling. In addition, the information processing device 1 creates the first machine learning model M1 using the downsampled inundation prediction image (low-resolution inundation prediction data D3) in this way, thereby obtaining the first machine learning model M1 as It becomes possible to accurately predict the worst case in which the evaluation is the worst in the inundation prediction used.

It should be noted that each component of each illustrated device does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution/integration of each device is not limited to the illustrated one, and all or part of them can be functionally or physically distributed/integrated in arbitrary units according to various loads and usage conditions. Can be integrated and configured. For example, the functional configuration related to machine learning in the information processing device 1 (e.g., training data creation unit 152, first machine learning unit 153 and second machine learning unit 154) and the functional configuration related to prediction (e.g., estimating unit 155 and The output unit 156) may be a separate configuration, or may be realized by an independent device configuration.

In addition, the various processing functions performed by the information processing device 1 are executed in whole or in part on a CPU (or a microcomputer such as an MPU or MCU (Micro Controller Unit)) or a GPU (Graphics Processing Unit). You may make it Also, all or any part of the various processing functions should be executed on a program analyzed and executed by the CPU (or microcomputer such as MPU or MCU) or GPU, or on hardware based on wired logic. It goes without saying that Further, various processing functions performed by the information processing apparatus 1 may be performed in collaboration with a plurality of computers by cloud computing.

By the way, the various processes described in the above embodiments can be realized by executing a prepared program on a computer. Therefore, an example of a computer (hardware) that executes a program having functions similar to those of the above embodiments will be described below. FIG. 9 is a block diagram showing an example of a computer configuration.

As shown in FIG. 9, a computer 200 includes a CPU 201 that executes various types of arithmetic processing, a GPU 201a that specializes in predetermined arithmetic processing such as image processing and machine learning processing, an input device 202 that receives data input, and a monitor 203. , and a speaker 204 . The computer 200 also has a medium reading device 205 for reading a program or the like from a storage medium, an interface device 206 for connecting with various devices, and a communication device 207 for communicating with an external device by wire or wirelessly. The computer 200 also has a RAM 208 that temporarily stores various information, and a hard disk device 209 . Each unit ( 201 to 209 ) in computer 200 is connected to bus 210 .

The hard disk device 209 includes the acquisition unit 151, the training data creation unit 152, the first machine learning unit 153, the second machine learning unit 154, the estimation unit 155, and the output unit 156 in the control unit 15 described in the above embodiment. A program 211 for executing various kinds of processing in, etc. is stored. The hard disk device 209 also stores various data 212 such as observation information that the program 211 refers to. The input device 202 receives input of operation information from an operator, for example. The monitor 203 displays, for example, various screens operated by the operator. The interface device 206 is connected with, for example, a printing device. The communication device 207 is connected to a communication network such as a LAN (Local Area Network), and exchanges various information with external devices via the communication network.

The CPU 201 or the GPU 201a reads out the program 211 stored in the hard disk device 209, develops it in the RAM 208, and executes it to obtain the acquisition unit 151, the training data creation unit 152, the first machine learning unit 153, the second machine Various processes related to the learning unit 154, the estimation unit 155, the output unit 156, and the like are performed. Note that the program 211 does not have to be stored in the hard disk device 209 . For example, the computer 200 may read and execute the program 211 stored in a readable storage medium. Storage media readable by the computer 200 are, for example, portable recording media such as CD-ROMs, DVD discs, USB (Universal Serial Bus) memories, semiconductor memories such as flash memories, hard disk drives, and the like. Alternatively, the program 211 may be stored in a device connected to a public line, the Internet, a LAN, or the like, and the computer 200 may read the program 211 from these devices and execute it.

Reference Signs List 1 information processing device 10 communication unit 11 display unit 12 operation unit 13 input/output unit 14 storage unit 15 control unit 141 observation information 142 training data 143 prediction data 144 first machine learning model Information 145 Second machine learning model information 151 Acquisition unit 152 Training data creation unit 153 First machine learning unit 154 Second machine learning unit 155 Estimation unit 156 Output unit 200 Computer 201 CPU
201a GPU
202... Input device 203... Monitor 204... Speaker 205... Medium reading device 206... Interface device 207... Communication device 208... RAM
209 Hard disk device 210 Bus 211 Program 212 Various data C1-C3 Cases D1, D10 Observation information D2, D2a-D2c, D12 High-resolution inundation prediction data D3, D3a-D3c, D11 Low-resolution inundation prediction Data M1... First machine learning model M2... Second machine learning model

Claims

Acquire observation information about the target area,
By inputting the observation information into the first machine learning model, generating a first image showing a prediction result of the inundation situation in the target area,
generating a second image having a higher resolution than the first image by inputting the first image into a second machine learning model;
Outputting the second image as a result of inundation prediction for the target area;
An inundation prediction program characterized by causing a computer to execute processing.
The first image is a map image showing the flood situation at each point overlooking the target area,
The inundation prediction program according to claim 1, characterized by:
Acquire observation information about the target area,
By inputting the observation information into the first machine learning model, generating a first image showing a prediction result of the inundation situation in the target area,
generating a second image having a higher resolution than the first image by inputting the first image into a second machine learning model;
Outputting the second image as a result of inundation prediction for the target area;
A flood prediction device comprising a control unit that executes processing.
The first image is a map image showing the flood situation at each point overlooking the target area,
The inundation prediction device according to claim 3, characterized by:
Downsample the observation information and the corresponding inundation prediction image for the target area,
generating a first machine learning model based on the downsampled predicted inundation image and the observation information;
Second machine learning for outputting an image obtained by increasing the resolution of the input image based on the downsampled predicted inundation image and the predicted inundation image, with the output of the first machine learning model as input. generate the model,
A machine learning method characterized in that processing is executed by a computer.
In the downsampling process, when downsampling pixels included in a specific region of the predicted inundation image, pixels with the worst evaluation of inundation prediction among the pixels included in the region are sampled. The machine learning method according to claim 5, wherein