JP7466969B1 - AI earthquake damage prediction system, AI earthquake damage prediction method, and AI earthquake damage prediction program - Google Patents

Abstract

[Problem] The present invention aims to provide a new technology for predicting earthquake damage to pipelines when an expected earthquake occurs using a mathematical model. [Solution] An AI earthquake damage prediction system for predicting earthquake damage to buried facilities including existing pipelines, comprising a storage unit and a prediction unit, the storage unit stores pipeline attribute information and a mathematical model for predicting earthquake damage or parameters of the mathematical model, pipeline attribute information related to the pipeline, and earthquake information including an earthquake index of an expected earthquake, the mathematical model is a model that has been trained using teacher data based on the pipeline attribute information, event information including earthquake impact information related to the impact of earthquakes on events, and earthquake history related to earthquakes that have occurred in the past, and the prediction unit uses the mathematical model to predict damage to the pipeline when an earthquake with a specific earthquake index occurs based on the pipeline attribute information and the earthquake information, and outputs the prediction result. [Selected Figure] Figure 1

Description

The present invention relates to an AI earthquake damage prediction system, an AI earthquake damage prediction method, and an AI earthquake damage prediction program.

In recent years, as the threat of natural disasters has become more serious, there has been a push to make buried pipelines, such as water pipes, sewer pipes, gas pipes, and fiber optics, earthquake-proof. However, from a cost perspective, it is difficult to implement earthquake-proofing measures for all pipelines. Therefore, if it were possible to identify pipelines that are likely to leak or burst in the event of an earthquake, it would be possible to make pipelines earthquake-proof efficiently and effectively.

Patent Publication No. 7288229

The invention described in Patent Document 1 is an invention for accurately estimating the vulnerability of pipelines even in situations where earthquake indicators are not available, and it is not possible to easily predict damage based on information about earthquakes, such as the seismic intensity of earthquakes that have already occurred or are expected to occur.

The objective of the present invention is to provide a new technology that uses mathematical models to predict earthquake damage to pipelines in the event of a predicted earthquake.

In order to solve the above problems, the present invention provides an AI earthquake damage prediction system for predicting earthquake damage to buried facilities including existing pipelines, comprising:
The apparatus includes a storage unit and a prediction unit,
The storage unit stores pipeline attribute information, a mathematical model for predicting earthquake damage or parameters of the mathematical model, pipeline attribute information related to the pipeline, and earthquake information including an earthquake index of an expected earthquake;
The mathematical model is a model that has been trained using teacher data based on pipeline attribute information, event information including earthquake impact information regarding the impact of earthquakes on events, and earthquake history regarding earthquakes that have occurred in the past,
The prediction unit uses the mathematical model to predict damage to a pipeline in the event of an earthquake with a specific earthquake index, based on pipeline attribute information and the earthquake information, and outputs the prediction result.

With this configuration, it is possible to predict damage to pipelines in the event of an anticipated earthquake by inputting information about the earthquake, including earthquake indicators such as the anticipated seismic intensity. Even if event information is acquired about events such as leakage caused by factors other than earthquakes, such as corrosion deterioration or electrolytic corrosion leakage, it is possible to predict pipeline deterioration using a model trained on event information that takes into account the effects of earthquakes based on earthquake impact information.

In a preferred embodiment of the present invention, the event information is information about pipelines for which event investigations have been conducted and/or about pipelines in which an event has suddenly occurred.

This configuration makes it possible to predict earthquake damage to pipelines using a model trained using event information related to event investigations such as water leak investigations and sudden events.

In a preferred embodiment of the present invention, the mathematical model is a model that has been trained using training data generated based on event information regarding events that have occurred due to the effects of an earthquake identified based on the earthquake impact information.

This configuration makes it possible to predict earthquake damage using a highly accurate trained model, without using event information about events that occurred due to causes other than earthquakes that are identified based on earthquake impact information.

In a preferred embodiment of the present invention, the AI earthquake damage prediction system further includes a learning unit,
The memory unit stores earthquake history regarding earthquakes that have occurred in the past,
The data processing unit generates teacher data based on the earthquake history, the pipeline attribute information, and the event information,
The learning unit is a model that has been trained using the generated teacher data.

With this configuration, the AI earthquake damage prediction system can learn based on information about the pipeline, including pipeline attribute information and event information, and information about earthquakes, including earthquake history, to improve the accuracy of the mathematical model.

In a preferred embodiment of the present invention, the AI earthquake damage prediction system further includes a reception unit and a display processing unit,
The storage unit stores pipeline attribute information to which position information relating to a pipeline position is assigned, and map information including position information;
The reception unit receives a selection of an earthquake index,
The display processing unit displays a map showing the earthquake damage prediction result based on the pipeline position information, the map information, and the earthquake damage prediction result for a specified earthquake index.

With this configuration, once an earthquake index such as seismic intensity is specified, the results of the earthquake damage forecast for the specified earthquake index can be easily confirmed on a map.

In a preferred embodiment of the present invention, the display processing unit displays a map showing the earthquake damage prediction results, in which pipelines are filled in based on the earthquake damage prediction results.

With this configuration, the pipelines displayed on the map can be filled in based on the earthquake damage prediction results, making it easy to visually confirm the results of the earthquake damage prediction.

In a preferred embodiment of the present invention, the AI earthquake damage prediction system further includes a display processing unit and a reception unit,
The storage unit stores pipeline attribute information to which pipeline position information is assigned and map information including a map of a location where the pipeline is laid,
The display processing unit processes display of the pipeline on a map based on the map information and the position information of the pipeline,
The reception unit receives a designation of a pipeline for which a damage prediction is to be displayed from among pipelines displayed on a map,
The display processing unit processes and displays the damage prediction results for a specified pipeline.

With this configuration, you can easily check the earthquake damage prediction results for a pipeline by specifying the pipeline displayed on the map.

The present invention can provide a new technology that uses mathematical models to predict earthquake damage to pipelines in the event of a predicted earthquake.

Block diagram showing the overall system configuration System hardware configuration diagram Example of data structure of pipeline attribute information, event information, and earthquake history Example of data structure of training data Screen display example Screen display example Processing Flowchart

The present invention will now be described in more detail with reference to the accompanying drawings. The drawings show a preferred embodiment, but the present invention can be implemented in different forms and is not limited to the embodiment described in this specification. In this embodiment, the configuration and operation of an earthquake damage prediction device and a learning device will be described, but similar effects can also be achieved by a method executed by the devices, a computer program, etc. The computer program may be provided as a non-transitory recording medium that is readable by a computer.

In the embodiment described below, the AI earthquake damage prediction system predicts damage caused by an earthquake to pipelines that supply fluids, and predicts damage to existing pipelines that supply liquids, such as water supply pipes. However, if the damage prediction is related to pipelines, it may also predict damage caused by an earthquake to buried facilities and pipelines, such as gas pipes.

The AI earthquake damage prediction system is also used by businesses, such as local governments that manage pipelines, to predict earthquake damage to the pipelines they manage, but the scope of the pipelines for which damage predictions are made is not limited, and earthquake damage predictions, as described below, may be made to existing pipelines laid throughout the country.

<Definition of terms>
The definitions of terms and phrases used in this specification are as follows:
The raw data refers to data related to pipelines stored in a pipeline database (pipe DB) provided in a system for pipeline management such as a GIS.
Processed data is data generated by transforming raw data obtained from a pipeline database, etc., by adjusting the format, units, and number of digits, or by combining multiple data, and includes prediction data and training data.
The prediction data is data input to the earthquake damage prediction device 1 for earthquake damage prediction, and is processed data generated based on pipeline attribute information and earthquake information, which is provisional data regarding anticipated earthquakes.
Teacher data is data used to learn a mathematical model that predicts earthquake damage, and is processed data that is generated based on pipeline attribute information, event information related to events that have occurred in the pipeline, and earthquake history, which is the history of earthquakes that have occurred in the location where the pipeline is installed.

For raw data, conversion processing such as replacing, adding, and deleting information may be performed as necessary to make it usable within the system. For example, the data that has undergone this conversion processing becomes processed data related to the pipeline. In this embodiment, the raw data is data acquired from a pipeline DB, which is a database related to pipelines managed by a business operator and includes pipeline attribute information and event information, and the processed data is prediction data that has been preprocessed for use in earthquake damage prediction and teacher data that has been preprocessed for use in learning a mathematical model. If no processing is required, the raw data is used as is as processed data. In this embodiment, processed data that has undergone various conversion processing on various information including pipeline attribute information and event information for use in learning a mathematical model is used as teacher data. In addition, processed data that has undergone various conversion processing on various information including pipeline attribute information for earthquake damage prediction using a mathematical model is used as prediction data. In addition, if conversion is not required for information related to various pipelines including pipeline attribute information and event information, data that combines pipeline attribute information and event information may be used as teacher data. In addition, if conversion is not required for pipeline attribute information and event information, data that combines pipeline attribute information and event information may be used as prediction data.

The raw data is data related to pipelines, and includes pipeline attribute information indicating the characteristics (attributes) of the pipeline and event information related to events that have occurred in the pipeline. In this embodiment, the pipeline attribute information related to the pipeline and event information related to events such as water leakage that have occurred in the pipeline are linked to the pipeline attribute information by a unique pipeline ID included in the pipeline attribute information. In this embodiment, the AI earthquake damage prediction system 0 predicts earthquake damage to the pipeline using a mathematical model that has been trained based on earthquake information assigned based on the earthquake history related to earthquakes that have occurred in an area managed by a specific business operator, such as an area managed by a specific local government, and the pipeline attribute information related to the pipeline.

<System Configuration>
1 is a block diagram showing a system configuration of an embodiment of the present invention. As shown in FIG. 1, the AI earthquake damage prediction system 0 includes an earthquake damage prediction device 1 and a learning device 2.

The earthquake damage prediction device 1 is a device for predicting earthquake damage, which is a prediction of damage caused by an anticipated earthquake, and is installed at the base of a business operator that manages pipelines in a local government, etc.

The earthquake damage prediction device 1 is a device equipped with functions such as a GIS (Geographic Information System) that allows operators to manage information related to pipelines within their bases. The earthquake damage prediction device 1 manages pipeline information, including pipeline attribute information and event information, for pipelines managed by operators in a pipeline database (pipe DB). Note that the pipeline attribute information stored in the pipeline DB, or the pipeline information including pipeline attribute information and event information, is assumed to have one or more pieces of location information added thereto, such as addresses, codes such as coordinates and EPSG codes, hashed location information such as Geohash and Quadkey, mesh codes such as standard regional meshes, etc. (collectively referred to as location information), and location information indicating the location where the pipeline and event occurred.

The learning device 2 is a device that learns a mathematical model for predicting earthquake damage. In this embodiment, the learning device 2 learns the mathematical model based on information about the pipeline, including pipeline attribute information and event information, which are managed in the earthquake damage prediction device 1, and the earthquake history of earthquakes that have occurred in the past.

The terminal device 3 may be used for inputting and outputting information, such as transferring raw data such as pipeline information to the earthquake damage prediction device 1. As the terminal device 3, devices such as a personal computer, a smartphone, a tablet terminal, and a personal computer may be used. Multiple terminal devices 3 may be used, or only one may be used.

In this embodiment, data exchange between the earthquake damage prediction device 1, the learning device 2, and the terminal device 3 is performed using portable storage media such as USB memory and CD-ROM in consideration of the risk of the data being handled being leaked. On the other hand, some or all of these businesses may be configured to be able to communicate with the earthquake damage prediction device 1 and the terminal device 3 connected to the learning device 2 via any network such as a VPN (Virtual Private Network) on an IP (Internet Protocol) network (not shown), and data may be exchanged via the network.

<Hardware Configuration>
Fig. 2 is a hardware configuration diagram. As shown in Fig. 2(a), the information processing device 10 (earthquake damage prediction device 1, terminal device 3) has a processing unit 101, a storage unit 102, a communication unit 103, an input unit 104, and an output unit 105, and is used to perform the functions of each unit and each process. Also, as shown in Fig. 2(b), the learning device 2 has a processing unit 201, a storage unit 202, and a communication unit 203, and is used to perform the functions of each unit and each process.

As the earthquake damage prediction device 1, the learning device 2, and the terminal device 3, one or more computer devices such as a general-purpose server or a personal computer can be used. In addition, the earthquake damage prediction device 1 and the learning device 2 have the functional configuration described below, but part of the functional configuration of the earthquake damage prediction device 1 and the learning device 2 may be placed in a separate device configured to be able to communicate with the earthquake damage prediction device 1 and the learning device 2.

The processing unit 101 and the processing unit 201 each have a processor such as a CPU (Central Processing Unit) capable of executing an instruction set, and execute an OS (Operating System) as well as a learning program (in the case of the learning device 2), an AI earthquake damage prediction program (in the case of the earthquake damage prediction device 1), an earthquake damage prediction device utilization program, or a learning device utilization program (in the case of the terminal device 3).

The storage unit 102 and the storage unit 202 have a volatile memory such as a RAM (Random Access Memory) capable of storing an instruction set, and a non-volatile recording medium such as a HDD (Hard Disk Drive) or SSD (Solid State Drive) capable of recording an OS, etc. The storage unit 102 of the earthquake damage prediction device 1 stores an earthquake damage prediction program including a learning program, and a learned mathematical model for earthquake damage prediction or parameters of the learned mathematical model. The storage unit 202 of the learning device 2 stores a learning program. The storage unit 102 of the terminal device 3 stores a learning device utilization program or an earthquake damage prediction device utilization program, etc. The storage unit 102 of the earthquake damage prediction device 1 and the storage unit 202 of the learning device 2 store various data necessary for earthquake damage prediction or learning of a mathematical model, including pipeline attribute information, event information, earthquake information, earthquake history, and map information.

The communication unit 103 and the communication unit 203 have an interface for connecting to a network, and execute communication control with the network to communicate with other information processing devices 10 and the learning device 2. Furthermore, when connection to a network is undesirable from the standpoint of risk management or the like, the information processing device 10 and the learning device 2 may not be provided with the communication unit 103 and the communication unit 203.

The input unit 104 has an operation input device capable of input processing, such as a touch panel or a keyboard, and an audio input device capable of audio input, such as a microphone. The output unit 105 has a display device capable of display processing, such as a display, and an audio output device, such as a speaker.

<Data structure>
Hereinafter, examples of data configurations stored in the storage unit 102 of the earthquake damage prediction device 1 or the storage unit 202 of the learning device 2 in this embodiment will be shown with reference to FIGS. 3 and 4. FIG.

Figure 3 (a) shows an example of pipeline attribute information that indicates the attributes of a pipeline. Pipeline attribute information is information that indicates the characteristics (attributes) of a pipeline, and is raw data obtained from a database (pipeline DB) of a business entity such as a local government that manages the pipeline, but some or all of it may be data obtained from an external source such as an external database.

The pipeline attribute information shown in FIG. 3(a) is information about the pipeline, and is data in which feature quantities about the pipeline, including the year of installation relating to the year the pipeline was installed, the pipe type, which is information about the material of the pipeline such as ductile cast iron pipe (DIP) or ordinary cast iron pipe (CIP), the diameter of the pipeline, and the presence or absence of a leak investigation, which is information about the presence or absence of event information such as whether a leak investigation has been conducted, are linked with a unique ID that identifies the pipeline.

In addition, although not shown in FIG. 3(a), in this embodiment, the pipeline attribute information includes environmental information related to the environment, such as the ground and climate, at the location where the pipeline is laid. The ground information included in the environmental information is information related to the ground at the location where the pipeline is laid, and includes the average S-wave velocity (AVS), the surface ground amplification factor, the type of topography, and geology. The environmental information related to the environment at the location where the pipeline is laid includes information such as the average temperature at the laying location, the average precipitation, and the distance from major roads. In addition, other information, such as the installation location of the pipeline, may be included as the pipeline attribute information as long as it is a feature related to the deterioration of the pipeline.

An example of event information is shown in FIG. 3B. The event information is information about events such as water leakage and damage that have occurred in the pipeline. In this embodiment, the event information includes information about events such as water leakage discovered by an event investigation such as a water leakage investigation conducted by a business operator who manages the pipeline, and information about water leakage in the pipeline that has been discovered due to a sudden water leakage accident. In this embodiment, the event information is information about events that occur in water pipes, such as water leakage, but it may also be information about other events such as damage or rupture of gas pipes, so long as it is information about events that occur in the pipeline. In this embodiment, the event investigation is an investigation of water pipes, such as a water leakage investigation conducted on pipelines managed by a specific business operator, but it may also be an investigation of other pipelines, such as gas pipes. In addition, an event-investigated pipeline is a pipeline for which an event investigation such as a water leakage investigation has already been conducted, and it is known whether an event has occurred or not.

In addition, because event information is information about events that occur in a pipeline, if multiple events occur in one pipeline, multiple pieces of event information are linked. By linking multiple pieces of event information to one pipeline in this way, information about all events that have occurred in each pipeline can be used for learning. In addition, the longer an operator manages a pipeline, the more event information related to the number of event investigations and sudden events will be accumulated. Therefore, the amount of event information linked to pipeline attribute information will also increase.

In this embodiment, the event information is registered by specifying a pipeline ID, and is linked to the pipeline attribute information. However, if the pipeline attribute information and the event information each contain location information, the event information may be registered without specifying a pipeline ID. In this case, the event information may be registered linked to the pipeline attribute information of the pipeline that exists near the location where the event occurred, based on the pipeline attribute information and the location information contained in the event information. When information about a pipeline leak that was discovered due to a sudden water leakage accident is registered as event information, the event information may be registered without being linked to the pipeline attribute information. In addition, when the event information and the pipeline attribute information are registered without being linked, the data processing unit 12 of the earthquake damage prediction device 1 or the learning device 2 may perform a process of linking the pipeline attribute information and the event information based on the pipeline attribute information and the location information added to the event information.

The event information shown in FIG. 3(b) includes a water leakage data ID, which is a unique ID for identifying an event, and timing information, which includes the event occurrence year (water leakage year) related to the year in which the event occurred, and earthquake impact information related to the presence or absence of the impact of the earthquake. The timing information is information related to the timing of the event occurring in the pipeline. In the data shown in FIG. 3(b), the timing information includes data related to the year in which the event occurred, but the timing information may also include the time until the event occurred, obtained from the year in which the pipeline was laid and the year in which the event occurred. In addition, if an event has occurred multiple times in the pipeline, the timing information may include the time from the previous event occurrence to the current event occurrence. In this embodiment, the event information and the timing information are managed in association with the same water leakage data ID. In addition, the event information may include the event type, such as corrosion deterioration, electrolytic corrosion leakage, and vibration leakage in the water pipe.

Causes of leakage discovered by event investigations such as leakage investigations include causes other than earthquakes, such as leakage due to electrical corrosion, leakage due to vibration, and corrosion deterioration. Earthquake impact information is information for identifying event information related to events that clearly occurred due to the influence of an earthquake, such as leakage accidents that occurred immediately after the occurrence of an earthquake or events discovered by event investigations such as leakage investigations conducted after an earthquake, and is information related to the influence of an earthquake on the occurrence of an event. Therefore, even if an event such as leakage discovered by an event investigation occurred in a pipeline, if the event occurred without being affected by the earthquake, such as leakage due to electrical corrosion or corrosion deterioration, it is learned as a pipeline without leakage due to an earthquake. In this embodiment, event information including earthquake impact information is information registered by a water utility or the like that discovered the event, such as by conducting a leakage investigation or repairing the pipeline, but it may also be information acquired by a sensor or the like placed in the pipeline or its vicinity.

In this embodiment, the pipeline attribute information is registered in association with the pipeline attribute information, but if no event such as a leak has occurred, the pipeline attribute information may not be associated with the pipeline. In such a case, the data processing unit 12 performs processing to associate the event information indicating that there is no leak with the pipeline attribute information.

Figure 3 (c) shows an example of an earthquake history showing the history of earthquakes that have occurred and have affected the pipeline. In this embodiment, the earthquake history is a history of earthquakes that have occurred in the past within the scope of management of a specific business operator such as a local government (for example, a specific city, ward, town, or village), but it may also be information on the history of earthquakes that have occurred nationwide, and may include location information on the location of the earthquake. In this embodiment, the earthquake history is a history of earthquakes of a predetermined seismic intensity or higher, but it may also be a history of earthquakes within a range of scales or magnitudes set based on other earthquake indicators. Since it is difficult to divide earthquake information or earthquake history for each business operator, in this embodiment, one earthquake information and earthquake history as shown in Figure 3 (c) are registered for each business operator. Therefore, all pipelines in each pipeline segment that is within the scope of management of a specific business operator such as a local government refer to one earthquake history and earthquake information for that pipeline segment.

The earthquake history shown in FIG. 3(c) includes an earthquake ID, which is a unique ID indicating an earthquake, the year of occurrence, and an earthquake index indicating the magnitude of the earthquake. In this embodiment, the earthquake index is seismic intensity, which is relatively easy to obtain, but one or more of seismic intensity, magnitude, peak ground acceleration, peak ground velocity, SI value (spectrum intensity), etc. may also be used as the earthquake index. In addition, the earthquake index may be an earthquake index such as seismic intensity obtained from a database such as a public database.

In this embodiment, the earthquake damage prediction process is performed using earthquake information, which is data on provisional earthquakes that are expected to occur. The earthquake information may have a data structure including the year of occurrence and earthquake index, similar to the earthquake history shown in FIG. 3(c), but if the year for the damage prediction is fixed, such as when predicting earthquake damage in the year in which the prediction was made, only earthquake indexes such as seismic intensity may be used. The earthquake damage prediction device 1 predicts earthquake damage to the pipeline using prediction data generated based on earthquake information, which is provisional data on earthquakes that are expected to occur. The earthquake information used for earthquake damage prediction may be data on provisional earthquakes input by the operator at the terminal device 3. In addition, the present invention preferably uses earthquake indexes such as seismic intensity and magnitude so that the user can easily recognize them intuitively, but earthquake indexes such as acceleration that can be converted into earthquake indexes familiar to users, such as seismic intensity and magnitude, may also be used.

Figure 4(a) shows an example of training data that is generated based on the pipeline attribute information, event information, and earthquake history shown in Figure 3 and is used in the learning process. The data shown in Figure 4(a) is data related to pipeline damage caused by earthquakes, and is data that is generated by the data processing unit 12, which will be described later, performing preprocessing on the pipeline attribute information, event information, and earthquake history.

The teacher data is data used to learn the mathematical model, and in this embodiment, includes the pipeline ID, installation year, number of earthquakes experienced, experienced seismic intensity, and whether or not there is earthquake leakage. Although not shown in FIG. 4, the teacher data also includes ground information and environmental information included in the pipeline attribute information. The number of earthquakes experienced is the number of earthquakes that have occurred in the area where the pipeline is installed, which is identified based on the earthquake history. The experienced seismic intensity is the seismic intensity of earthquakes that have occurred up until the damage prediction in the area where the pipeline is installed, which is identified based on the earthquake history. In this embodiment, the experienced seismic intensity is the largest seismic intensity of earthquakes experienced by the pipeline, but may include the seismic intensity of all earthquakes experienced by the pipeline. The presence or absence of earthquake leakage is the presence or absence of an event that has occurred in the pipeline due to the effects of an earthquake, which is identified based on the earthquake impact information included in the event information.

Figure 4(b) shows an example of prediction data, which is processed data generated based on pipeline attribute information and earthquake information. In this embodiment, the prediction data includes pipeline attribute information such as pipeline ID and installation year, and earthquake information such as the number of earthquakes experienced and earthquake index, and is data with a uniform format. In this embodiment, the number of earthquakes experienced included in the prediction data is the number of earthquakes experienced by the pipeline from installation to the time damage prediction is performed, which is obtained by the data processing unit 12 based on the earthquake history shown in Figure 3(c) and the installation year included in the pipeline attribute information.

The map information is information about a map of the pipeline, and includes image information used when displaying the map and location information about the position on the map. In this embodiment, the location information is latitude and longitude, but it may be coordinate information on the map, etc. In this embodiment, the map information is linked to the pipeline attribute information based on the location information.

Although not shown in Figures 4(a) and 4(b), in this embodiment, the teacher data and prediction data include data obtained by processing environmental information at the location where the pipeline is installed, which is included in the pipeline attribute information, and ground information that is highly relevant to damage caused by earthquakes. The teacher data and prediction data include data related to the environment at the location where the pipeline is installed, such as the average temperature, minimum temperature, maximum temperature, average precipitation, traffic volume, building density, and distance from major roads at the installation location. The teacher data and prediction data also include data related to the ground, such as the average S-wave velocity (AVS), surface ground amplification factor, type of terrain and geology, slope, and elevation.

<Functional configuration>
As shown in Fig. 1, the earthquake damage prediction device 1 includes a reception unit 11, a data processing unit 12, a prediction unit 13, and a display processing unit 14. The learning device 2 includes an acquisition unit 21 and a learning unit 22. This is information processing by software (a program stored transiently or non-transiently in a storage unit or the like) specifically realized by hardware (a processing unit or the like).

<Functional configuration of earthquake damage prediction device 1>
The reception unit 11 of the earthquake damage prediction device 1 performs processing to receive necessary information from various information processing devices including the terminal device 3. In this embodiment, the reception unit 11 receives earthquake information and earthquake history from the terminal device 3, and receives pipeline information including pipeline attribute information and event information from the pipeline DB. The reception unit 11 may also perform processing to receive pipeline attribute information and event information from the terminal device 3. The reception unit 11 may also perform processing to receive data related to pipelines provided via a portable storage medium such as a USB.

The reception unit 11 also receives input from the user via the input unit 104, including the specification of the pipeline for which earthquake damage prediction is to be performed and the earthquake index expected when performing earthquake damage prediction, and performs processing to store the input in the memory unit 102 of the earthquake damage prediction device 1.

The data processing unit 12 in the earthquake damage prediction device 1 generates prediction data based on the pipeline attribute information and earthquake information. The prediction data generated by the data processing unit 12 is processed data generated by converting the pipeline attribute information and earthquake information into an easy-to-handle form. In this embodiment, the prediction data is some or all of the attributes stored in the pipeline DB as raw data, and is data that includes pipeline attribute information including the pipeline ID and installation year, and earthquake information including seismic intensity. In addition, the prediction data may include any data related to the pipeline, and in this embodiment, it includes environmental information and ground information included in the pipeline attribute information.

In this embodiment, the data processing unit 12 generates prediction data based on the pipeline attribute information and earthquake information related to the expected earthquake. In this embodiment, the prediction data is data generated by combining some or all of the pipeline attribute information with some or all of the earthquake information, and is data with a converted format, unit, number of digits, etc. The data processing unit 12 may also perform conversion processing such as calculating a value suitable for the format of the prediction data from multiple pieces of information such as pipeline attribute information, earthquake information, and earthquake history, such as calculating the time from laying to the occurrence of an earthquake based on the year of laying and the year of the earthquake. The data processing unit 12 may also perform processing to calculate the number of earthquakes experienced by the pipeline before the prediction is made based on the year of laying the pipeline, earthquake history, and earthquake information, and set this as the number of experienced earthquakes included in the prediction data.

The data processing unit 12 performs preprocessing of raw data used for earthquake damage prediction, converting the format, number of digits, units, etc. of the raw data into easy-to-handle processed data in the AI earthquake damage prediction system 0 as preprocessing for various information used for earthquake damage prediction. In addition, in this embodiment, the data processing unit 12 is provided in both the earthquake damage prediction device 1 and the learning device 2, but may be provided in either the earthquake damage prediction device 1 or the learning device 2.

In addition, it is expected that information related to pipelines, such as pipeline attribute information and event information, will have different formats for each business operator and some information may be missing. The data processing unit 12 may perform processing such as unifying the data format of the pipeline information, including pipeline attribute information and event information, so that it is easy to use in the AI earthquake damage prediction system, or complementing the data.

The prediction unit 13 performs earthquake damage prediction using a mathematical model based on the prediction data generated by the data processing unit 12. In this embodiment, the prediction unit 13 outputs an earthquake damage occurrence rate, which is the probability that an event such as a water leak will occur in the pipeline due to the effects of an earthquake, as a pipeline earthquake damage prediction based on the prediction data generated based on the pipeline attribute information and earthquake information related to the expected earthquake. The prediction unit 13 also performs pipeline earthquake damage prediction based on earthquake indices related to earthquakes that will occur in the pipeline, but multiple earthquake damage predictions based on multiple expected earthquake indices such as seismic intensity 5 and seismic intensity 6 may also be performed for one pipeline.

The prediction unit 13 performs a simulation of damage to the pipeline in the event of an earthquake. In this embodiment, the prediction unit 13 predicts earthquake damage to the pipeline based on prediction data generated from earthquake information, which is dummy data related to an earthquake of an expected seismic intensity, and pipeline attribute information related to the pipeline, and stores the obtained prediction result in the memory unit 102 of the earthquake damage prediction device 1 in association with the pipeline attribute information. In this embodiment, the prediction unit 13 calculates the probability of damage to the pipeline in an earthquake of a specific seismic intensity as a prediction result of earthquake damage, but the degree of damage (damage) to the pipeline due to the earthquake may also be used. In addition, the prediction unit 13 may predict damage caused by an earthquake that has already occurred based on earthquake information registered based on information related to the earthquake that has occurred, such as earthquake alerts, for pipelines that have experienced an earthquake but have not yet been identified as having leaks.

In this embodiment, the prediction unit 13 predicts earthquake damage at a specific seismic intensity. In this embodiment, the earthquake index used by the prediction unit 13 is seismic intensity, for which data can be easily collected, but earthquake damage to pipelines may also be predicted using other earthquake indexes, such as earthquake acceleration.

The mathematical model used by the prediction unit 13 to predict earthquake damage to pipelines is described below. The mathematical model used by the prediction unit 13 is a trained model that predicts earthquake damage to pipelines based on prediction data generated based on pipeline attribute information about the pipelines and event information about damage caused by earthquakes that have occurred in the pipelines.

In this embodiment, the mathematical model is a model that uses the gradient boosting technique, but any mathematical model formed by a machine learning technique may be used, such as gradient descent, boosting, decision trees, neural networks, logistic regression, k-nearest neighbors, or a technique that combines multiple techniques. The prediction unit 13 may predict earthquake damage to pipelines using a model that combines multiple techniques, or multiple models that are each formed by different techniques.

The display processing unit 14 processes and displays various necessary information including the prediction results by the prediction unit 13. In this embodiment, various screens processed and displayed by the display processing unit 14 are displayed on the earthquake damage prediction device 1, but may also be displayed on other devices such as the terminal device 3. Below, examples of screens processed and displayed by the display processing unit 14 are shown using drawings.

Figure 5 shows an example of a pipeline information display screen that displays information about a pipeline selected by a user. The pipeline information display screen is displayed by pressing and selecting a pipeline displayed on a map, and displays information about the pipeline including pipeline attribute information and the results of earthquake damage prediction. When a pipeline on the map is pressed, the pipeline associated with the pipeline is selected based on the position information at that position, and the reception unit 11 accepts the selection of the pipeline on the map. In this embodiment, the pipeline information display screen displays, as information about the selected pipeline, pipeline attribute information including the pipe category, installation year, pipe type, and diameter, and the results of earthquake damage prediction including the probability of leakage of the pipeline at each seismic intensity. The pipeline information display screen as shown in Figure 5 may be displayed simultaneously with the map related to the pipeline, or may be displayed on a different screen.

Also, FIG. 6 shows a prediction result map display screen, which is a map that displays the results of earthquake damage prediction at a specific seismic intensity. The prediction result map display screen is a screen that displays the results of earthquake damage prediction, and shows the occurrence probability of an event due to the influence of an earthquake, which is the earthquake damage prediction result when an earthquake with a predetermined earthquake index (in the example shown in FIG. 6, a seismic intensity of 5+) occurs in each pipeline. In the example shown in FIG. 6, the pipelines on the map are colored with different chromatic colors for each occurrence probability. In this embodiment, the pipelines displayed on the map are colored with different chromatic colors based on the occurrence probability of each event in the pipeline, but as long as the method shows the occurrence probability of each event in the pipeline, the display may be performed in a configuration different from the example shown in FIG. 6, such as filling the pipelines with different shades of color or different patterns based on the occurrence probability of the event. In addition, the pipeline information display screen shown in FIG. 5 may be displayed by pressing and selecting a pipeline displayed on the prediction result map display screen.

Furthermore, in this embodiment, the earthquake damage prediction device 1 includes a classification unit. In this embodiment, the classification unit performs classification based on the estimated values, which are the provisional earthquake damage prediction results obtained by the prediction unit 13, and the pipeline attribute information, and classifies pipelines with similar pipeline attribute information and estimated values into the same classification.

When the earthquake damage prediction device 1 includes a classification unit, the prediction unit 13 performs earthquake damage prediction using the pipeline attribute information and a mathematical model, and obtains an estimated value that is a provisional prediction result. The estimated value is a provisional prediction result based on the pipeline attribute information, and in this embodiment, it is the probability of earthquake damage occurring, but it may be the time from earthquake damage to pipeline damage, etc. The classification unit performs classification based on the estimated value obtained by the prediction unit 13 and the pipeline attribute information. The prediction unit 13 determines a final prediction result based on the estimated value and the classification result by the classification unit. In this embodiment, the prediction unit 13 determines a final earthquake damage prediction result by weighting the estimated value of the pipeline for which earthquake damage prediction is performed and the estimated value of pipelines classified in the same category as that pipeline, based on the classification result.

<Functional configuration of learning device 2>
The functional configuration of the learning device 2 shown in Fig. 1 will be described below. The learning device 2 includes an acquisition unit 21, a data processing unit 12, and a learning unit 22.

The acquisition unit 21 performs processing to acquire various information necessary for performing earthquake damage prediction, including pipeline attribute information and event information. The acquisition unit 21 also acquires various information related to pipelines, including pipeline attribute information and event information, which are managed in the earthquake damage prediction device 1. However, if it is undesirable for the earthquake damage prediction device 1 and the learning device 2 to be connected to a network from a security standpoint, the acquisition unit 2 may also perform processing to acquire data transferred via a portable storage medium such as a USB. If the acquired data is not suitable for use as is, the acquisition unit 21 may also perform processing to adjust the format of the acquired data as preprocessing.

The acquisition unit 21 may also perform processing to acquire information necessary for earthquake damage prediction directly from external systems, such as a weather information providing system or a map information providing system, that are communicatively connected via a network, etc., using various methods, such as using an API.

The acquisition unit 21 converts data acquired from an external system, such as an external system, into a format that can be used by the system for use in learning the mathematical model, and stores the converted data in the storage unit 102. The conversion process in the acquisition unit 21 is a process of converting data acquired from an external system into a predetermined number of digits or units, or converting the data into a format that can be handled by the earthquake damage prediction device 1 or the learning device 2. The acquisition unit 21 collects external data by transmitting location information to an API such as a web API (Application Programming Interface) provided by the external system, and performs a predetermined conversion. At that time, the acquisition unit 21 performs a process of linking the acquired data to pipeline attribute information or event information, but may also identify the pipeline to be linked based on the location information linked to the pipeline attribute information or event information, and perform the linking process. The reception unit 11 of the earthquake damage prediction device 1 may also perform the same process as the acquisition unit 21, such as the acquisition process of external data.

The data processing unit 12 in the learning device 2 generates training data used to learn the mathematical model based on the pipeline attribute information, event information, and earthquake history.

In this embodiment, the data processing unit 12 generates the teacher data shown in FIG. 4 based on the pipeline attribute information, event information, and earthquake history shown in FIG. 3. The data processing unit 12 generates the teacher data based on the acquired raw data including the pipeline attribute information and event information, and the earthquake history. The data processing unit 12 generates the teacher data by changing the format, unit, number of digits, etc. of the raw data, but may also perform processing to obtain data with units and number of digits suitable for learning from multiple data, such as calculating the number of earthquakes experienced by the pipeline before earthquake damage prediction based on the year of installation and the year of occurrence of earthquakes included in the earthquake history, and using this as the number of experienced earthquakes included in the teacher data.

In this embodiment, the data processing unit 12 acquires pipeline attribute information and event information related to pipelines in an area managed by a specific business operator (e.g., a specific city), and generates teacher data based on the earthquake history that has occurred in the area managed by the specific business operator. Furthermore, when the data processing unit 12 determines that an earthquake has influenced the occurrence of an event based on the earthquake impact information, it identifies the earthquake that influenced the occurrence of the event based on the timing information (the year in which the event occurred) and the year in which the earthquake occurred included in the earthquake history, and uses this to generate teacher data.

The learning unit 22 learns a mathematical model for predicting earthquake damage using a combination of pipeline attribute information, event information, and earthquake history as teacher data. In this embodiment, the learning unit 22 learns using teacher data having the data structure shown in FIG. 4. The learning unit 22 learns the mathematical model using processed data preprocessed by the data processing unit 12 as teacher data.

In this embodiment, the learning unit 22 performs learning using teacher data generated based on pipeline attribute information and event information related to pipelines managed by the operator where the learning unit 22 is located, and earthquake history related to earthquakes that occurred within the scope managed by the operator. However, the learning unit 22 may perform learning using teacher data generated based on pipeline attribute information, event information, and earthquake history of each operator or nationwide earthquake history of multiple operators. In this embodiment, the learning unit 22 of the learning device 2 located at the administrator's base performs learning based on processed data acquired from the base of each operator, but may perform learning using processed data generated by processing raw data acquired from the base of each operator. The learning unit 22 may also generate a national model, which is a mathematical model that has been learned based on teacher data generated based on data related to pipelines nationwide. The learning unit 22 may further perform learning of the national model again based on teacher data generated based on pipeline attribute information, event information, and earthquake history managed by each operator, and generate a base model, which is a mathematical model suitable for each operator. The learning unit 22 may also perform re-learning on the national model once learned, using data acquired from the previous learning to the current learning. The method of generating a nationwide model is not limited to the above, and the learning unit 22 may also perform a process of integrating multiple mathematical models obtained by combining the parameters of the mathematical models, etc., to generate a nationwide model.

<Processing flow>
An example of a process flow for earthquake damage prediction and learning of a mathematical model will be described below with reference to the drawings. The process flow described with reference to FIG. 7 is only an example, and the process may be performed in a different order from the process described below.

<Earthquake damage prediction processing flow>
7(a) shows a flow of the earthquake damage prediction process. When the reception unit 11 receives earthquake information on an expected earthquake (S101), the data processing unit 12 generates prediction data based on the acquired earthquake information and pipeline attribute information (S102). The prediction unit 13 uses the prediction data generated by the data processing unit 12 and a mathematical model to perform earthquake damage prediction (S103).

<Learning process flow>
The data processing unit 12 generates training data, which is data for learning, based on necessary information including pipeline attribute information, event information, and earthquake history (S201). In this embodiment, the data processing unit 12 generates training data shown in Fig. 4 based on the pipeline attribute information, event information, and earthquake history shown in Fig. 3.

In the above embodiment, an example in which the earthquake damage prediction device 1 and the learning device 2 are placed at the base of each business establishment has been described, but the learning device 2 may be placed only at the base of the manager of the AI earthquake damage prediction system 0, and the mathematical model trained at the manager's base may be distributed to each business operator. In this case, the learning device 2 placed at the manager's base may obtain necessary data including pipeline information and earthquake history from the business operator's base and train the mathematical model.

In this embodiment, the AI earthquake damage prediction system 0 displays the predicted earthquake damage prediction results for the pipeline on the screen, but it may also be output as a file in various formats, such as CSV (comma separated values) or Excel.

0 AI earthquake damage prediction system 1 Earthquake damage prediction device 2 Learning device 3 Terminal device

Claims (9)

An AI earthquake damage prediction system for predicting earthquake damage to buried facilities including existing pipelines,
The apparatus includes a storage unit and a prediction unit,
The storage unit stores a mathematical model for predicting earthquake damage or parameters of the mathematical model, pipeline attribute information regarding the pipeline including the year of installation of the pipeline , earthquake information including earthquake indicators of anticipated earthquakes , and an earthquake history that is a history of multiple earthquakes including the year of occurrence of the earthquake ;
The mathematical model is a model that has been trained using teacher data based on pipeline attribute information, event information including earthquake impact information regarding the impact of earthquakes on events, and earthquake history regarding earthquakes that have occurred in the past,
The teacher data includes a number of earthquakes experienced since the installation of the pipeline, the number being specified based on the installation year and the occurrence year,
The prediction unit uses the mathematical model to predict damage to the pipeline when an earthquake with a specific earthquake index occurs based on prediction data generated from the pipeline attribute information and the earthquake information, and outputs a prediction result.
The prediction data includes the number of earthquakes experienced since the installation of the pipeline, the number being specified based on the installation year and the occurrence year.
AI earthquake damage prediction system. The event information is information about a pipeline that has undergone an event investigation and/or a pipeline where an event has suddenly occurred.
The AI earthquake damage prediction system according to claim 1. The mathematical model is a model that is trained using teacher data generated based on event information regarding events that occurred due to the effects of the earthquake identified based on the earthquake effect information.
The AI earthquake damage prediction system according to claim 1. The AI earthquake damage prediction system further includes a learning unit,
The memory unit stores earthquake history regarding earthquakes that have occurred in the past,
The data processing unit generates teacher data based on the earthquake history, the pipeline attribute information, and the event information,
The learning unit is a model that has been learned using the generated teacher data.
The AI earthquake damage prediction system according to claim 1. The AI earthquake damage prediction system further includes a reception unit and a display processing unit,
The storage unit stores pipeline attribute information to which position information relating to a pipeline position is assigned, and map information including position information;
The reception unit receives a selection of an earthquake index,
The display processing unit displays a map showing the earthquake damage prediction result based on the pipeline position information, the map information, and the earthquake damage prediction result for the specified earthquake index.
The AI earthquake damage prediction system according to claim 1. The display processing unit displays a map in which pipelines are filled in based on the earthquake damage prediction result as a map showing the earthquake damage prediction result.
The AI earthquake damage prediction system according to claim 5. The AI earthquake damage prediction system further includes a display processing unit and a reception unit,
The storage unit stores pipeline attribute information to which pipeline position information is assigned and map information including a map of a location where the pipeline is laid,
The display processing unit processes display of the pipeline on a map based on the map information and the pipeline position information,
The reception unit receives a designation of a pipeline for which a damage forecast is to be displayed from among pipelines displayed on a map,
The display processing unit displays the damage prediction result of the specified pipeline.
The AI earthquake damage prediction system according to claim 1. An AI earthquake damage prediction method for predicting earthquake damage to buried facilities including existing pipelines, comprising:
one or more computers that store in a storage unit a mathematical model for predicting earthquake damage or parameters of the mathematical model, pipeline attribute information on the pipeline including the year of installation of the pipeline , earthquake information including earthquake indicators of anticipated earthquakes, and an earthquake history that is a history of multiple earthquakes including the year of occurrence of the earthquake ;
a prediction step of predicting damage to a pipeline when an earthquake with a specific earthquake index occurs based on prediction data generated from pipeline attribute information and the earthquake information using the mathematical model, and outputting the prediction result;
Equipped with
The mathematical model is a model that has been trained using teacher data based on pipeline attribute information, event information including earthquake impact information regarding the impact of earthquakes on events, and earthquake history regarding earthquakes that have occurred in the past,
The teacher data includes a number of earthquakes experienced since the installation of the pipeline, the number being specified based on the installation year and the occurrence year,
The prediction data includes the number of earthquakes experienced since the installation of the pipeline, the number being specified based on the installation year and the occurrence year.
AI earthquake damage prediction method. An AI earthquake damage prediction program for predicting earthquake damage to buried facilities including existing pipelines,
A computer that stores in a memory unit a mathematical model for predicting earthquake damage or parameters of the mathematical model, pipeline attribute information on the pipeline including the year of installation of the pipeline , earthquake information including earthquake indicators of anticipated earthquakes , and an earthquake history that is a history of multiple earthquakes including the year of occurrence of the earthquake , functions as a prediction unit,
The mathematical model is a model that has been trained using teacher data based on pipeline attribute information, event information including earthquake impact information regarding the impact of earthquakes on events, and earthquake history regarding earthquakes that have occurred in the past,
The teacher data includes a number of earthquakes experienced since the installation of the pipeline, the number being specified based on the installation year and the occurrence year,
The prediction unit uses the mathematical model to predict damage to the pipeline when an earthquake with a specific earthquake index occurs based on prediction data generated from the pipeline attribute information and the earthquake information, and outputs a prediction result.
The prediction data includes the number of earthquakes experienced since the installation of the pipeline, the number being specified based on the installation year and the occurrence year.
AI earthquake damage prediction program.