Water quantity prediction device, water quantity prediction method, and program

Abstract

To provide a water volume prediction device, a water volume prediction method, and a program that can more accurately predict the water volume without having specialized knowledge. 【Solution means】The water volume prediction device according to the embodiment includes a reception unit that receives an input of information including inflow data to a target point, an estimation unit that estimates parameters of a predetermined physical model that uses a predetermined large language model and takes the inflow data as an input and outputs outflow data, and a prediction unit that inputs the information including the inflow data and the parameters into the physical model to predict the outflow data at the target point. The predetermined physical model includes a tank model.

Description

Technical Field

[0001] The present invention relates to a water volume prediction device, a water volume prediction method, and a program.

Background Art

[0002] In recent years, for the analysis and prediction of natural disasters, the utilization of deep learning using a large amount of data accumulated in the past has been promoted. In this connection, a technique for predicting the water level of a river caused by a meteorological disaster by a prediction model of an ANN (Artificial Neural Network) using deep learning is known (see, for example, Non-Patent Document 1). Further, in recent years, a technique for evaluating a flood risk or the like by image analysis via generative AI (Generative Artificial Intelligence) is known (see, for example, Non-Patent Document 2).

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Non-Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, the flow rate data and the like used for learning the flood prediction model include the content of human judgment (judgment based on experience and intuition), such as the release adjustment of the dam, the operation of the floodgate, and the presence or absence of the operation of the drainage pump. Since deep learning can only make predictions based on the patterns contained in the data, it is difficult to patternize the data with variations caused by the content of human judgment. In addition, for the content to be queried by the generative AI, methods and techniques are required to obtain the optimal answer, and there is a need to have specialized knowledge to judge the reliability of the answer. Therefore, conventionally, there has been a possibility that accurate water volume prediction could not be made.

[0005] Aspects of the present invention have been made in consideration of such circumstances, and it is an object to provide a water volume prediction device, a water volume prediction method, and a program that can more accurately predict the water volume without having specialized knowledge.

Means for Solving the Problems

[0006] The water volume prediction device, the water volume prediction method, and the program according to the present invention adopt the following configurations. A water volume prediction device according to a first aspect of the present invention includes a reception unit that receives an input of information including inflow data to a target location, an estimation unit that estimates parameters of a predetermined physical model that uses a predetermined large language model and takes the inflow data as an input and outputs outflow data, and a prediction unit that inputs the information including the inflow data and the parameters into the physical model to predict the outflow data at the target location. The predetermined physical model includes a tank model and is a water volume prediction device.

[0007] A water volume prediction device according to a second aspect of the present invention further includes a determination unit that determines whether an error between the outflow data predicted by the prediction unit and predetermined outflow data is within a predetermined range. The estimation unit updates the parameters when the determination unit determines that the error is not within the predetermined range, and the prediction unit predicts the outflow data again using the updated parameters.

[0008] In the water volume prediction device according to the third aspect of the present invention, further, the prediction unit predicts at least one of a flood situation and a flooding situation based on the outflow data.

[0009] The water volume prediction device according to the fourth aspect of the present invention further includes a providing unit that generates information indicating the result predicted by the prediction unit and provides it to the user.

[0010] In the water volume prediction device according to the fifth aspect of the present invention, further, the estimation unit refers to a data set in which the inflow data, the outflow data, and the parameters stored in advance are associated, and extracts a data set having a similarity with the inflow data received by the reception unit equal to or higher than a threshold value, and estimates the parameters.

[0011] In the water volume prediction method according to the sixth aspect of the present invention, a computer receives an input of information including inflow data to a target location, estimates parameters of a predetermined physical model that uses a predetermined large language model with the inflow data as an input and outflow data as an output, inputs the information including the inflow data and the parameters into the physical model, predicts the outflow data at the target location, and the predetermined physical model includes a tank model.

[0012] The program according to the seventh aspect of the present invention causes a computer to receive an input of information including inflow data to a target location, estimate parameters of a predetermined physical model that uses a predetermined large language model with the inflow data as an input and outflow data as an output, input the information including the inflow data and the parameters into the physical model, and predict the outflow data at the target location, and the predetermined physical model includes a tank model.

Advantages of the Invention

[0013] According to the aspect of the present invention described above, it is possible to more accurately predict the water volume without having specialized knowledge.

Brief Description of the Drawings

[0014]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Embodiments for Carrying Out the Invention

[0015] Hereinafter, with reference to the drawings, embodiments of the water volume prediction device, water volume prediction method, and program of the present invention will be described. In the following, for example, a water volume prediction device that predicts outflow data for inflow data at a target point to be predicted, or predicts the influence (for example, flood situation, inundation situation, etc.) in the vicinity (peripheral area) of the target point based on the outflow data will be described. The target point is, for example, a river, a dam, a water storage tank, etc., but is not limited thereto. The inflow data is, for example, the inflow volume of water such as precipitation, and the outflow data is, for example, information related to the water volume such as the outflow volume of water and the water level, but is not limited thereto. Also, the water volume may include not only the mere amount of water but also the amount of sediment, etc.

[0016] <Water volume prediction system> FIG. 1 is a diagram showing an example of the schematic configuration of a water volume prediction system 1 including a water volume prediction device 100 according to an embodiment. The water volume prediction system 1 includes, for example, a water volume prediction device 100 and at least one or more terminal devices 200-1 to 200-n. Hereinafter, unless otherwise distinguished and described for each of the terminal devices 200-1 to 200-n, they will be simply referred to as "terminal device 200" for description. The water volume prediction device 100 and the terminal device 200 communicate with each other via, for example, a network NW. The network NW includes, for example, a cellular network, a Wi-Fi network, Bluetooth (registered trademark), the Internet, a WAN (Wide Area Network), a LAN (Local Area Network), a public line, a provider device, a dedicated line, a wireless base station, etc.

[0017] The water volume prediction device 100 predicts the water volume at the target point based on input data from a user who uses the terminal device 200. In addition, the water volume prediction device 100 generates information (for example, an image, etc.) for providing the prediction result to the user, and displays it on the display unit of the water volume prediction device 100 or transmits it to the terminal device 200 to provide it to the user. The water volume prediction device 100 may be, for example, a server device or a PC (Personal Computer), or may be a cloud server configured by cloud computing including one or more processing devices.

[0018] The terminal device 200 is a terminal used by a user. For example, the terminal device 200 receives input of input data including a target location for predicting water volume and inflow data, etc., transmits the received input data to the water volume prediction device 100, and receives prediction results such as outflow data obtained from the water volume prediction device 100 and provides them to the user. The terminal device 200 may be, for example, a PC, a tablet terminal, etc., or a communication terminal such as a smartphone.

[0019] Next, the functional configurations of the water volume prediction device 100 and the terminal device 200 will be specifically described. <Water volume prediction device 100> The water volume prediction device 100 includes, for example, a communication unit 110, a reception unit 120, an output unit 130, an estimation unit 140, a prediction unit 150, a determination unit 160, a provision unit 170, a control unit 180, and a storage unit 190. One or more of the estimation unit 140, the prediction unit 150, the determination unit 160, the provision unit 170, and the control unit 180 are realized, for example, by a hardware processor such as a CPU (Central Processing Unit) executing a program (software). One or more of these components may be realized by hardware (including a circuit unit; circuitry) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), GPU (Graphics Processing Unit), or may be realized by the cooperation of software and hardware. The program may be stored in advance in a storage device (a storage device having a non-transitory storage medium) such as an HDD (Hard Disk Drive) or a flash memory of the water volume prediction device 100, or may be stored in a removable storage medium such as a DVD, a CD-ROM, or a memory card, and may be installed in the HDD or flash memory of the water volume prediction device 100 when the storage medium (non-transitory storage medium) is mounted on a drive device.

[0020] The memory unit 190 is implemented by, for example, an HDD, a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), a ROM (Read Only Memory), or a RAM (Random Access Memory), etc. The memory unit 190 stores, for example, a generation AI model 192, a physical model 194, a dataset 196, programs, and various other information.

[0021] The generation AI model 192 is a model for autonomously generating and outputting, for example, parameters (such as setting conditions) to be input to the physical model 194 in response to the input of predetermined input information (for example, input information corresponding to the type of the object). The generation AI model may include, for example, large language models (LLMs) specialized in natural language processing (NLP; Natural Language Processing), and may also include dialogue AI models specialized for enabling natural conversations (dialogues) with people (users) by applying large language models. In the dialogue AI model, for example, various question items for collecting elements (information) necessary for estimating parameters, etc. based on the dialogue with the user are stored in advance, the next question is selected according to the user's answer to the question, and necessary elements are collected by repeating the dialogue including the question and the answer.

[0022] In addition, the generative AI model 192 has an LLM or a diffusion model for image generation, and is a multi-modal model that can handle text data, image data, etc. as input data. In the embodiment, using this generative AI model 192, for example, a physical model (hydraulic model) 194 that provides physical information such as outflow data is operated without manual intervention (for example, adjusting parameters described later). The generative AI model 192 may use, for example, ChatGPT (registered trademark), or may use other models or systems (for example, Gemini (registered trademark), OpenAI (registered trademark), BingAI). The generative AI model 192 may be stored in the storage unit 190, or may be provided in an external device (for example, a generative AI server, etc.) via the network NW. FIG. 1 shows an example where the generative AI model 192 is stored in the storage unit 190.

[0023] The physical model 194 is a model that predicts outflow data, etc. for inputs such as inflow data and parameters. The physical model 194 is, for example, a tank model. The tank model replaces the target location with a container (tank) having an outflow hole according to the characteristics (shape, size, geology, etc.) of that location, and derives the water outflow amount for the inflow data and parameters. In the tank model, for example, since the relationship between rainfall and flow rate in the target area is obtained using parameters, the setting of parameters is very important.

[0024] The physical model 194 may be used individually or in combination with the generative AI model 192. Also, the physical model 194 may be stored in the storage unit 190, or may be provided in an external device via the network NW. FIG. 1 shows an example where the physical model 194 is stored in the storage unit 190. The above-described generative AI model 192 and physical model 194 may be updated at any time.

[0025] The dataset 196 is a group of data consisting of, for example, precipitation data, water level data, text data in which numerical values of precipitation and water level are marked down in tabular form, etc. accompanying changes over time. Note that the dataset 196 may be stored for each target location and may include information regarding parameters used in the physical model 194. The dataset 196 is used, for example, when estimating parameters to be input to the physical model 194. The dataset 196 includes, for example, measured data such as past flood data. By using this data, it becomes possible to achieve highly accurate flood prediction and the like.

[0026] The communication unit 110 communicates with the terminal device 200, other external devices, etc. via the network NW, for example. For example, the communication unit 110 transmits information processed by each component in the water volume prediction device 100 to the terminal device 200, other external devices, etc. Further, the communication unit 110 outputs information received from the terminal device 200, other external devices, etc. to a predetermined component in the water volume prediction device 100. The above-described transmission and reception processing may be controlled by the control unit 180 or may be controlled by other components.

[0027] The reception unit 120 receives input of various information from the administrator of the water volume prediction device 100 (administrator of the water volume prediction system 1), the user of the terminal device 200, etc. The various information is, for example, various input information (target location, inflow data, etc.) for obtaining prediction results such as outflow at the target location. Note that the above administrator is also an example of a "user" who uses this system.

[0028] The output unit 130 has, for example, a display unit and a speaker, etc., and outputs predetermined information to each of them. The display unit is, for example, an LCD (Liquid Crystal Display), an organic EL (Electro Luminescence) display, or the like. The display unit displays a screen showing various types of information (text and images) in the embodiment. The speaker outputs predetermined sound. Note that the reception unit 120 may be configured integrally with the display unit as a touch panel. For example, the output unit 130 causes, for example, information received by the reception unit 120, information generated by the provision unit 170, etc. to be output.

[0029] The estimation unit 140 estimates parameters for input to the physical model 194 that outputs (predicts) outflow data for the inflow data using the generation AI model 192. Details of the functions of the estimation unit 140 will be described later.

[0030] The prediction unit 150 inputs the input data including the inflow data and the parameters estimated by the estimation unit 140 into the physical model 194 to predict the outflow data at the target location. Further, the prediction unit 150 predicts at least one of the flood situation and the inundation situation as an influence in the vicinity (surrounding area) of the target location based on the predicted outflow data. Details of the functions of the prediction unit 150 will be described later.

[0031] The determination unit 160 determines whether the error between the result predicted by the prediction unit 150 (for example, the predicted outflow volume) and a predetermined value (comparative outflow volume) is within a predetermined range. The comparative outflow volume is, for example, a reference value of a value that can be guaranteed by physical laws, and may be a value set based on past measured data at the same location, or may be a predetermined standard value. The comparative outflow volume may be stored in the storage unit 190 in advance, or may be acquired from an external device via the network NW. When it is determined that the error is not within the predetermined range (exceeds the predetermined range), the estimation unit 140 is requested to update the parameters. The estimation unit 140 receives the update request from the determination unit 160 and updates the parameters. At this time, the estimation unit 140 may adjust the amount of change in the parameters based on the magnitude of the error and the magnitude relationship between the predicted outflow volume and the comparative outflow volume, or may adjust based on a predetermined fixed amount of change. Then, the prediction unit 150 predicts the outflow data at the target location again using the updated parameters. By repeatedly executing the above processing until the error is within the predetermined range, a more accurate prediction result can be obtained.

[0032] The providing unit 170 generates various screens for allowing the user to input input data such as the target area and inflow data, and transmits the generated various screens to the terminal device 200 or displays them on the display unit of the output unit 130 or the like, and provides them to the user (including the system administrator) and the like. Further, the providing unit 170 generates information (provided information) such as an image or voice indicating the result predicted by the prediction unit 150, and outputs the generated provided information to the output unit 130 or transmits it to the terminal device 200 to provide it to the user.

[0033] The control unit 180 controls the entire functional configuration of the water volume prediction device 100. For example, the control unit 180 controls the transmission and reception of information (data) in the communication unit 110, causes the estimation unit 140 to estimate parameters or the like based on the information received by the reception unit 120, and causes the prediction unit 150 to execute prediction processing. Further, the control unit 180 causes the determination unit 160 to execute determination processing, causes the providing unit 170 to generate provided information, and provides it to the user.

[0034] <Terminal device 200> FIG. 2 is a diagram showing an example of the functional configuration of the terminal device 200. The terminal device 200 includes, for example, a terminal-side communication unit 210, an input unit 220, an output unit 230, a control unit 240, an app execution unit 250, and a terminal-side storage unit 260. Part or all of the control unit 240 and the app execution unit 250 are realized, for example, by a hardware processor such as a CPU executing a program (software). Also, part or all of these components may be realized by hardware (including a circuit unit; circuitry) such as an LSI, an ASIC, an FPGA, or a GPU, or may be realized by the cooperation of software and hardware. The program may be stored in advance in a storage device (a storage device having a non-transitory storage medium) such as an HDD or a flash memory, or may be stored in a removable storage medium (a non-transitory storage medium) such as a DVD, a CD-ROM, or a memory card, and may be installed in the storage device when the storage medium is mounted in a drive device or a card slot of the terminal device 200.

[0035] The terminal-side storage unit 260 may be realized by the above-described various storage devices, or by an EEPROM, a ROM, a RAM, or the like. In the terminal-side storage unit 260, for example, an application 262, a program, and other various types of information are stored.

[0036]

[0037] The terminal-side communication unit 210 communicates with the water volume prediction device 100 and other external devices via, for example, a network NW. For example, the terminal-side communication unit 210 transmits information processed by each component in the terminal device 200 to the water volume prediction device 100 and other external devices. Also, the terminal-side communication unit 210 outputs information received from the water volume prediction device 100 and other external devices to a predetermined component in the water volume prediction device 100. The above-described transmission and reception processing may be controlled by the control unit 240 or may be controlled by other components.The input unit 220 receives, for example, user input through operations of various keys, buttons, etc. Further, the input unit 220 may include a microphone that receives user voice input.

[0038] The output unit 230 includes, for example, a display unit 232 and a speaker 234. The output unit 230 outputs predetermined information to the display unit 232 and the speaker 234. The display unit 232 is, for example, an LCD, an organic EL display, or the like. The input unit 220 may be integrally configured with the display unit 232 as a touch panel. The display unit 232 displays various types of information in the embodiment. The speaker 234 outputs predetermined sound. For example, the output unit 230 outputs a screen or sound corresponding to the information provided by the water volume prediction device 100, or outputs the information input by the input unit 220.

[0039] The control unit 240 controls the overall functions of the terminal device 200. For example, the control unit 240 controls the transmission and reception of information (data) in the terminal-side communication unit 210, controls the input and output of data by the input unit 220 and the output unit 230, and controls the execution of the application 262 by the app execution unit 250.

[0040] The app execution unit 250 is realized by executing the application 262 stored in the terminal-side storage unit 260. The application 262 is, for example, an application downloaded from an external device via the network NW and installed in the terminal device 200. Further, the application 262 is an application program (software) provided by the water volume prediction device 100, and may perform a process of accessing a predetermined address (URL; Uniform Resource Locator) via, for example, a browser. The application 262 transmits input data such as the target location and inflow data input from the terminal device 200 by the user to the water volume prediction device 100 to execute the water volume prediction process and obtains the execution result. When the application 262 transmits the information input by the input unit 220 to the water volume prediction device 100, the application 262 may perform a process of including identification information for identifying the terminal device 200 and identification information for identifying the user in the information to be transmitted.

[0041] <Regarding water volume prediction> Next, the water volume prediction in the embodiment will be described with reference to the drawings. FIG. 3 is a diagram for explaining an example of water volume prediction in the embodiment. In the water volume prediction system 1 of the present embodiment, a system (CoT simulator) is constructed in which the generation AI model 192 and the simulator using the physical model 194 are linked via CoT (Chain of Thought) prompting that can generate the thinking process performed by a person. For example, in the embodiment, a CoT simulator is constructed that targets flood prediction caused by climate change, accepts input of inflow data such as precipitation at the target location, and predicts the outflow volume of rivers and the like. As an example of the physical model 194 (a computer that outputs physical information) used in this simulator, a tank model will be described below. Further, in the following, a simple one-stage tank model will be used, but a multi-stage model may be used according to the characteristics of the target location. For example, when the water depth (or the flow rate in the tank) of a one-stage tank model is η, the outflow volume (Q) is given by the following equation (1). Q = (a) × η …(1)

[0042] Here, (a) shown in formula (1) is a proportional constant related to the outflow rate and is an example of a parameter (internal parameter). The estimation unit 140 estimates (automatically adjusts) the parameters input to this tank model using the generation AI model 192. The prediction unit 150 predicts outflow data and the like using the parameters estimated by the estimation unit 140. The providing unit 170 generates providing information including the result predicted by the prediction unit 150 and provides it to the user.

[0043] More specifically, for example, as shown in FIG. 3, the user inputs input data (prompt) including inflow data such as text data and image data via the terminal device 200 or the like, and makes an inquiry about the prediction result of the outflow data at the target location to the water volume prediction device 100 side. The input data (prompt) includes information such as "When there is a rainfall of 100 [mm] per hour in the river at location A, what is the flood situation of the river after 2 hours?" The reception unit 120 or the control unit 180 performs known language analysis processing on the above-mentioned text, and obtains from the analysis result, as the content of the input data, the target location "location A", the inflow data "rainfall of 100 [mm] per hour", the inquiry (prediction target) "flood situation of the river after 2 hours", and the like.

[0044] On the water volume prediction device 100 side, in addition to inputting the above-mentioned target location (location A) and inflow data (rainfall of 100 [mm] per hour) into the tank model which is the physical model 194, it refers to the data set 196 to search for data similar to the content of the input data, and estimates the parameters (for constructing the tank model) to be input to the tank model from the search results.

[0045] FIG. 4 is a diagram for explaining an example of the content of the dataset 196. In the dataset 196 shown in FIG. 4, precipitation data, water level data, situation description text, parameters, etc. associated with time changes are associated with each other. More specifically, in the example of FIG. 4, in order to make the prediction target easy to understand, the precipitation data is parametric time-series data that can be expressed by a Gaussian distribution. Also, the precipitation data stores various distributions using three parameters: mean, variance, and total precipitation. In the example of FIG. 4, the precipitation data stores a total of 576 datasets including multiple points, and the water level data, which is the output result when these datasets are input into the tank model, is stored in a time-series data format with a time lag. In the embodiment, a one-stage tank model is used, three types of internal parameters included in the tank model are set, and 576×3 sets of water level data are stored. The internal parameter is, for example, a parameter used to predict (calculate) the outflow data using the physical model 194, which is the proportionality constant (a) described above. Note that the type, data volume, type, and number of parameters of the dataset 196 in the embodiment are not limited to the example of FIG. 4, and are appropriately adjusted according to the type of the physical model 194, the number of parameters, the number of target locations to be stored, etc. Also, the dataset 196 may be associated with text data (text data showing the relationship between precipitation and water level) showing a situation description when the water level indicated by the water level data, parameters other than the internal parameters, image data such as graphs, etc.

[0046] For example, the estimation unit 140 causes the dataset 196 to be referred to based on the target location (location A) and the inflow data via the above-described generation AI model 192 (LLM), and extracts data that is close to the target location input by the user (including the same location) and has a high similarity to the inflow data (for example, precipitation data with a similarity equal to or higher than a threshold). When the input data includes information such as a situation description from the user, data with a high situation similarity may be extracted from the dataset 196 using that information.

[0047] Also, when the input data includes image data, the estimation unit 140 may perform image analysis or the like via the generative AI model 192 (or other image analysis models). For image analysis, a known standard image analysis model may be used, or a model such as CLIP (Contrastive Language-Image Pre-training) may be used. For example, in a standard image analysis model, a pre-trained model in which an image feature extractor and a linear classifier are learned in a pair is used to estimate text data corresponding to the features of an image. On the other hand, in CLIP, an image encoder and a text encoder are learned in a pair to estimate correct combinations for a series of image and text learning. CLIP is a model pre-trained to have a short distance after mapping for pairs of text and images that are in a corresponding relationship. Therefore, by using CLIP, text data such as precipitation data and water level data corresponding to an image can be extracted more appropriately than with a standard image analysis model.

[0048] Then, the generative AI model 192 acquires water level data, internal parameters, etc. associated with precipitation data having high similarity. Note that when the generative AI model 192 acquires three types of internal parameters from the dataset 196 as shown in FIG. 4, any one of the three types of internal parameters may be selected, a plurality of parameters may be combined to estimate one parameter, or the parameter may be estimated in correspondence with other information (input data) or the like. Furthermore, the generative AI model 192 may select (or estimate) a parameter for the user through interaction with the user or the like.

[0049] Next, the prediction unit 150 inputs the inflow data (rainfall of 100 [mm] per hour) and the estimated parameters into the tank model (physical model 194), operates the tank model, and predicts the flood situation of the river two hours later corresponding to the inquiry content. At this time, the prediction unit 150 constructs a tank model corresponding to the estimated parameters, inputs the inflow data into the constructed tank model to calculate the water level (outflow rate) for a predetermined time (for example, one hour), and predicts the water level two hours later based on the calculated water level. Further, the prediction unit 150 predicts the flood situation of the target river (near point A which is the target location) from the water level two hours later. The flood situation may be, for example, the situation of the water level or flow rate at the target location, or the situation of the scale of flood damage caused by the occurrence of a flood. Also, the flood situation may be the situation of the flood risk obtained by comparing the height information of surrounding levees etc. acquired from map information with the predicted water level, or by comparing the predicted water level or flow rate with the water level or flow rate value at the time of past disaster occurrence. Also, the flood situation may include information regarding the situation (scale) of flood warnings and evacuation warnings.

[0050] Note that in the embodiment, the prediction unit 150 may predict the inundation situation in addition to (or instead of) the flood situation. The inundation situation includes, for example, when there are buildings such as houses near the target river, the depth of inundation (the height from the ground to the water surface in the inundated area), and the inundation range estimated based on it. Regarding the terrain and buildings near the target river in the inundation situation, for example, map information may be referred to and acquired, or it may be standard terrain and buildings assumed as a simulation. The above-mentioned map information may be stored in the storage unit 190, or may be acquired from an external device via the network NW. Also, the prediction results of the flood situation and the inundation situation may include image data in which each situation is mapped or image data in which it is graphed (for example, flood waveforms, inundation distributions, etc.).

[0051] Further, in the embodiment, as shown in FIG. 4, the determination unit 160 determines whether or not the error between the predicted outflow amount (river outflow amount (water level)), which is the prediction result, and a predetermined comparison outflow amount (water level) is within a predetermined range. When the error exceeds the predetermined range, a request to update the parameters is output to the generation AI model 192. Here, the predetermined comparison outflow amount may be, for example, water level data obtained by retrieval from the dataset 196, or a value based on the water level data. Since the dataset 196 includes past measured data, by comparing with that data, it is possible to determine whether the prediction result using the physical model 194 is accurate data (whether it is an appropriate value guaranteed by physical laws or the like).

[0052] The estimation unit 140 updates the parameters input to the physical model 194 to different parameters via the generation AI model 192. The parameters to be updated may be the parameters stored in the dataset 196 (for example, the remaining unused parameters among the above-described three types of parameters (internal parameters)), or may be parameters adjusted based on the content of the error. The prediction unit 150 inputs the updated parameters into the tank model and executes the prediction again. When the determination unit 160 determines that the error is within the predetermined range, the providing unit 170 generates providing information (text data, image data, etc.) using the information of the prediction result, and outputs the generated providing information to the user (terminal device 200) as an answer to the inquiry. Note that the generation AI model 192 may be used for the generation of the providing information.

[0053] In this way, in the embodiment, since the internal parameters of the tank model necessary for predicting the outflow data can be automatically adjusted by the generation AI model 192, the user can obtain an appropriate answer guaranteed by physical laws without having specialized knowledge. Further, according to the embodiment, in the embodiment, the generation AI is relearned by adding human experience and intuition regarding natural disasters, and furthermore, mathematics and physical laws, which are the wisdom of humanity, etc., and the internal parameters are adjusted, so that, for example, a generation AI specialized in comprehensive knowledge of natural disasters can be constructed.

[0054] <Example> Several examples using the water volume prediction system 1 in the embodiment will be described below. Hereinafter, two examples regarding the cooperation part between the generative AI model 192 (LLM) and the physical model (computer) 184 will be shown.

[0055] [First Example] The first example is an example of estimating the proportional constant (a) for the outflow rate, which is the output result of the tank model, by giving the generative AI model 192 tabular data of the temporal change in precipitation. FIG. 5 is an example of the data input to the generative AI model 192 in the first example. In FIG. 5, [Table 1] shows the relationship between the inflow rate (x) and the outflow rate (y) every hour using the tank model from 1 o'clock to 12 o'clock. It is assumed that the proportional constant (a), which is a parameter of this tank model, is unknown.

[0056] Here, it is assumed that the result of predicting the flow rate with the tank model (initial tank) in which the proportional constant (a) is set to "0.2" in the prediction unit 150 is as shown in [Table 2]. In this case, the peak outflow rate is slightly larger compared to the result of [Table 1]. Therefore, the estimation unit 140 inputs the contents shown in [Table 1] and [Table 2] (data on the temporal change in precipitation) to the generative AI model 192, and also inputs inquiry information such as "What should the value of the proportional constant (a) be in order to make the result shown in [Table 2] closer to the result shown in [Table 1]?". Note that the estimation unit 140 may input image data such as a graph instead of the tabular data as shown in FIG. 5. FIG. 6 is a diagram showing an example of other input data in the first example. In the image data IMG1 shown in FIG. 6, the horizontal axis represents time (Time), and the vertical axis represents the values (Value) of the inflow rate (x), the outflow rate (y) shown in Table 5, and the outflow rate (y') shown in Table 6. Also, even when an image such as a graph is input in this way, the data can be appropriately extracted by the image analysis process of the generative AI model 192. These input data may be generated by the estimation unit 140 or may be received as a prompt by the reception unit 120.

[0057] The generative AI model 192 estimates the proportionality constant (a) while obtaining, for example, a dataset or the like, or the prediction result by the prediction unit 150, and outputs the estimation result. Further, the generative AI model 192 may estimate the proportionality constant (a) while cooperating with other models and search systems. FIG. 7 is a diagram showing an example of the estimation result of the proportionality constant (a) obtained using the generative AI model 192 of the first embodiment. The generative AI model 192 outputs text data TXT1 including explanatory information as shown in FIG. 7 for the above input. Note that the content (dialogue data) shown in FIG. 7 is merely an example and is not limited thereto, and at least a part thereof may be omitted or changed. According to the example of FIG. 7, 0.1 to 0.15 is output as the estimated value of the proportionality constant (a).

[0058] According to the first embodiment, by presenting a new outflow using the generative AI model 192 based on the tank model constructed from the input information of the inflow (precipitation) and the outflow, the proportionality constant of the tank model can be appropriately estimated. Thus, in the first embodiment, the parameter input to the physical model 194 can be estimated using the generative AI model 192, and the water volume can be predicted more accurately using this parameter.

[0059] [Second Embodiment] The second embodiment is an example in which an appropriate prediction that satisfies the physical law is made even when input data regarding a simple flood event is given to the generative AI model 192. FIG. 8 is an example of the data input to the generative AI model 192 in the second embodiment. In the example of FIG. 8, the time-series relationship (tank model) between the inflow (x) when water flows into the tank every two hours from 0:00 to 24:00 and the outflow (y) proportional to the amount of the tank is shown. Further, FIG. 9 is a diagram showing an example of the inflow (x) input as inquiry information. In the second embodiment, the user inputs, as input data (prompt), a tank model (an example of inflow data) having the relationship shown in FIG. 8 and inquiry information as to what the outflow will be when the inflow (x) as shown in FIG. 9 is received from the terminal device 200.

[0060] The estimation unit 140 provides the generative AI model 192 with information (table data) showing the time-series relationship between the inflow and outflow shown in FIG. 8 as input to the tank model, and causes the generative AI model 192 to construct the tank model. In this case, as parameters, for example, data with a similarity to the data shown in FIG. 8 from the dataset 196 equal to or greater than a threshold value can be extracted, and the parameters associated with the extracted data can be used. Further, the estimation unit 140 causes the prediction unit 150 to predict the outflow in cooperation with the tank model in response to an inquiry about the time-series data of the inflow as shown in FIG. 9.

[0061] FIG. 10 is a diagram showing an example of the prediction result in the second embodiment. In the example of FIG. 10, text data TXT2 and image data IMG2 are included. The text data TXT2 is data (dialogue-type data) including explanatory information output by the generative AI model 192, but is merely an example and is not limited thereto, and at least a part thereof may be omitted or changed. Further, the image data IMG2 is data predicted by the prediction unit 150, but this content is also merely an example and is not limited thereto. In FIG. 10, x represents the inflow corresponding to the precipitation data shown in FIG. 9, y represents the outflow obtained from the tank model input in FIG. 8, and y' represents the predicted value of the outflow when the tank model constructed by the generative AI model 192 of the present embodiment is used for x. Comparing the values of y and y' shown in FIG. 10, although the timing of the outflow peak is slightly earlier, generally appropriate results are obtained.

[0062] According to the second embodiment, for example, even when the generative AI model 192 constructs a tank model from input information of time-series data regarding the inflow (precipitation) and outflow, an appropriate outflow can be predicted for new input data of precipitation. Thus, in the embodiment, it can be applied to various water volume predictions.

[0063] Note that in the above-described first and second embodiments, by also inputting information regarding the target location, more accurate estimation and prediction suitable for that location (region) can be performed. At least a part of the content shown in FIGS. 7 and 10 described above may be generated by the providing unit 170 as provided information and provided to the user via the terminal device 200.

[0064] As described above, according to the embodiment, since the cooperation method is used in which the generation AI model 192 operates the physical model 194, the water volume can be predicted more accurately without the user having specialized knowledge, and furthermore, various predictions (calculations) can be performed in real time.

[0065] <Processing flowchart> Next, an example of the process executed by the water volume prediction device 100 will be described using a flowchart. FIG. 11 is a flowchart showing an example of the process executed by the water volume prediction device 100. In the example of FIG. 11, the reception unit 120 receives input data including the inflow data of the target location (step S100). Next, the estimation unit 140 estimates the parameters to be input to the physical model 194 (for example, constructing a tank model) using the generation AI model 192 (step S110). Next, the prediction unit 150 predicts the outflow data using the physical model 194 with the input data and the parameters (step S120).

[0066] Next, the determination unit 160 performs an error determination process based on the prediction result (step S130), and determines whether the error between the outflow data obtained as the prediction result and the predetermined outflow data is within a predetermined range (step S140). If it is determined that the error is not within the predetermined range (exceeds the predetermined range), the estimation unit 140 updates the parameters using the generation AI model 192 (step S150), and returns to the process of step S120. As a result, the prediction unit 150 predicts the outflow data again using the updated parameters. Also, in the process of step S140, if it is determined that the error is within the predetermined range, the providing unit 170 generates providing information to be provided to the user (step S160), and outputs the generated providing information to the terminal device 200 or the like to provide it to the user (step S170). Thereby, the process of this flowchart ends.

[0067] <Modification Example> In the embodiment, a tank model is used as an example of the physical model 194. However, in addition to (or instead of) the tank model, other physical models may be used. In this case, as another physical model, for example, a model based on the storage function method is used. The model based on the storage function method separates the rain that has fallen into an outflow component as a flood in a short time and an outflow component that flows out over a long time due to water retention or the like, and is a model that expresses the storage (delay) at the target point (basin) for the flood outflow component. This model also requires parameters such as constants set for each target point (basin) when calculating the outflow volume. Therefore, in the embodiment, even when this model is applied, the parameters can be estimated using the generation AI model 192 in the same manner, and the outflow data can be predicted more accurately using the estimated parameters.

[0068] In addition, in the embodiment, in the generative AI model 192, Retrieval Augmented Generation (RAG) that combines a search technique with a large language model (LLM) may be used. In the generative AI model 192 using the LLM, for unknown situations without prior retraining, it is possible to make inferences by providing some auxiliary information to the pre-trained model (zero-shot inference), or to make inferences based on a previously given task (few-shot inference), etc., and thus it is possible to have a conversation with a specified format prompt input. In RAG, by utilizing the few-shot inference function and embedding available data into a template, inferences can be made using a unique database without additional training of the LLM itself. Therefore, for example, when using data of past similar phenomena in the dataset 196, by adding past data with RAG, it is possible to estimate high-precision parameters without retraining the LLM, and further, by using these parameters, more accurate water volume prediction can be realized.

[0069] In addition, in the embodiment, at least a part of the functions provided in the water volume prediction device 100 may be provided in the terminal device 200. Further, for the water volume prediction device 100, the functions of the estimation unit 140 that performs processing using the generative AI model 192 and the prediction unit 150 that makes predictions using the physical model 194 may be taken on by separate devices within the system.

[0070] As described above, according to the water volume prediction device 100 of the embodiment, a reception unit that receives an input of information including inflow data to a target location, an estimation unit that uses a predetermined large language model to estimate parameters of a predetermined physical model that takes the inflow data as an input and outputs outflow data, and a prediction unit that inputs the information including the inflow data and the parameters into the physical model to predict the outflow data at the target location are provided. By including a tank model in the predetermined physical model, it is possible to more accurately predict the water volume without having specialized knowledge.

[0071] For example, in an embodiment, for flood prediction due to climate change, in order to construct a simulator (a tank model, which is a type of hydrological model) that calculates river runoff by receiving input of data such as precipitation at a specific location and automatically adjusts parameter values via a generative AI model (LLM). Then, using the tank model adjusted with the optimal parameter values, the runoff is predicted. By constructing this system, if information such as precipitation at the prediction location is input as a prompt, it is possible to predict the surrounding area's impact (e.g., flood situation, inundation situation, etc.) based on the runoff such as appropriate water level rise and flood risk. Therefore, according to the embodiment, it is possible to perform detailed risk prediction, etc. in natural disasters based on comprehensive human knowledge without having specialized knowledge.

[0072] Also, according to the embodiment, when dealing with a phenomenon where the content queried by prompt input follows physical laws, by introducing a CoT simulator that combines a physical model with a generative AI to learn the human thinking process, even a layperson with immature prompt engineering or a person lacking specialized knowledge for the answer can obtain a plausible answer.

[0073] As described above, the embodiments have been used to explain the forms for implementing the present invention. However, the present invention is not limited to such embodiments, and various modifications and substitutions can be made without departing from the gist of the present invention.

Explanation of Reference Numerals

[0074] 1... Water volume prediction system, 100... Water volume prediction device, 110... Communication unit, 120... Reception unit, 130, 230... Output unit, 140... Estimation unit, 150... Prediction unit, 160... Judgment unit, 170... Provision unit, 180, 240... Control unit, 190... Storage unit, 200... Terminal device, 210... Terminal-side communication unit, 220... Input unit, 232... Display unit, 234... Speaker, 250... Application execution unit, 260... Terminal-side storage unit, 262... Application

Claims

1. A reception unit that receives input of information including inflow data into a target point; an estimation unit that estimates parameters of a predetermined physical model using a predetermined large-scale language model, the input of which is the inflow data and the output of which is the outflow data; a prediction unit that inputs information including the inflow data and the parameters into the physical model to predict the outflow data at the target point; the predetermined physical model includes a tank model; the estimation unit refers to a data set in which inflow data, outflow data, and parameters are associated with each other and extracts a data set having a similarity to the inflow data accepted by the acceptance unit equal to or greater than a threshold value, and estimates the parameters; Water volume prediction device.

2. A determination unit that determines whether an error between the outflow data predicted by the prediction unit and predetermined outflow data is within a predetermined range, The estimation unit updates the parameter when the determination unit determines that the error is not within a predetermined range; The prediction unit predicts the runoff data again using the updated parameters. The water volume prediction device according to claim 1 .

3. The prediction unit predicts at least one of a flood situation and a flooding situation based on the runoff data. The water volume prediction device according to claim 1 .

4. A providing unit that generates information indicating the result predicted by the prediction unit and provides the information to a user. The water volume prediction device according to claim 1 .

5. The computer Accept input of information including inflow data to the target location; Using a predetermined large-scale language model, parameters of a predetermined physical model are estimated, the input of which is the inflow data and the output of which is the outflow data; inputting information including the inflow data and the parameters into the physical model to predict the outflow data at the target location; the predetermined physical model includes a tank model; referring to a data set in which inflow data, outflow data, and parameters are associated with each other and a data set having a similarity to the received inflow data equal to or greater than a threshold is extracted, and the parameter is estimated; Water volume prediction methods.

6. On the computer, Accepting input of information including inflow data to a target location; Using a predetermined large-scale language model, parameters of a predetermined physical model are estimated, the input of which is the inflow data and the output of which is the outflow data; inputting information including the inflow data and the parameters into the physical model to predict the outflow data at the target location; the predetermined physical model includes a tank model; referring to a data set in which inflow data, outflow data, and parameters are associated with each other and a data set having a similarity to the received inflow data equal to or greater than a threshold is extracted, and the parameter is estimated; program.

Images