Machine learning device and power outage section prediction device

Abstract

To provide a machine learning device capable of constructing a learning model for efficiently predicting with a higher degree of accuracy a power outage occurrence section caused by the adhesion of salt to an insulating member of an electric power system, and a power outage occurrence section prediction device using the learning model.SOLUTION: A machine learning device 1 includes an input data acquisition unit 11 that acquires, as input data, a data group including positional information indicating the respective positions of a plurality of insulating members of an electric power system in an area A, topographical information on the area A, meteorological information on the area A from a point before a base time to the base time, and zero-phase voltage information on the power system from the point before the base time to the base time, a label acquisition unit 12 that acquires, as labels, a section in which a power outage occurs due to adhesion of salt to the insulating members in the area A after the lapse of a predetermined period from the base time, and a learning model construction unit 14 that constructs a learning model for predicting a power outage occurrence section by performing supervised learning using pairs of input data and labels as teacher data.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a machine learning device and a power outage occurrence section prediction device. [Background technology]

[0002] Conventionally, there are known techniques for predicting the impact of typhoons, strong winds, and the like on power grids. Examples of these types of techniques are described in Patent Documents 1 and 2. Patent Document 1 describes a technique for predicting the state of contamination of insulators after a typhoon has passed, based on insulator position information, typhoon information, and the like. Patent Document 2 describes a technique for creating a relational equation between resistance information of distribution equipment against salt damage, etc., past weather information, and damage information, and predicting damage due to salt damage from the resistance information, future weather information, and the relational equation. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2008-123805 A [Patent Document 2] JP 2015-75828 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, when predicting the section where a power outage will occur due to the adhesion of salt to insulating members of a power system, it takes time and effort to predict the section where a power outage will occur because it is difficult to visually check the adhesion of salt and because stormy weather such as typhoons and strong winds occurs over a wide area. Although the techniques of Patent Documents 1 and 2 can predict the section where a power outage will occur using meteorological information, there is room for improvement in terms of prediction accuracy.

[0005] The present invention aims to provide a machine learning device that can construct a learning model for efficiently predicting with higher accuracy the occurrence of power outage sections caused by the adhesion of salt to insulating components in a power system, and a power outage occurrence section prediction device that uses the learning model. [Means for solving the problem]

[0006] The present invention relates to a machine learning device including: an input data acquisition unit that acquires as input data a group of data including location information indicating the location of each of a plurality of insulating members that electrically insulate electric wires of a power system from the ground within a specified area; topographical information of the specified area; weather information for the specified area from a time earlier than a reference time to the reference time; and zero-phase voltage information of the power system from a time earlier than the reference time to the reference time; a label acquisition unit that acquires as labels an area in which a power outage has occurred due to adhesion of salt to the insulating members in the specified area after a specified period of time has elapsed from the reference time; and a learning model construction unit that constructs a learning model for predicting the area in which the power outage will occur by performing supervised learning using a pair of the input data and the label as teacher data.

[0007] The input data acquisition unit further acquires, as the input data, insulation performance information of each of the plurality of insulating members.

[0008] The present invention also relates to a power outage occurrence section prediction device that uses the learning model constructed by the machine learning device, and includes a prediction data acquisition unit that acquires the location information, the topographical information, weather information for the specified area from a specified time in the past to the present time, and zero-phase voltage information of the power system from the specified time in the past to the present time as a prediction data group, and a power outage occurrence section prediction unit that predicts the power outage occurrence section that will newly occur after a specified period of time has elapsed from the present time based on the prediction data group and the learning model. Effect of the Invention

[0009] According to the present invention, it is possible to provide a machine learning device that can construct a learning model for efficiently predicting with higher accuracy the occurrence of sections where power outages will occur due to the adhesion of salt to insulating components in a power system, and a power outage occurrence section prediction device that uses the learning model. [Brief description of the drawings]

[0010] [Figure 1] 1 is a diagram showing an overall configuration of a power outage occurrence section prediction system according to an embodiment of the present invention. [Diagram 2] 1 is a schematic diagram showing an example of a region in which a power outage section is predicted by a power outage occurrence section prediction system according to an embodiment of the present invention. FIG. [Diagram 3] FIG. 1 is a functional block diagram of a machine learning device according to an embodiment of the present invention. [Figure 4] 1 is a functional block diagram of a power outage occurrence section prediction device according to an embodiment of the present invention. [Diagram 5] FIG. 2 is a flowchart illustrating the operation of a machine learning device according to an embodiment of the present invention. [Figure 6] 1 is a diagram showing a flowchart of the operation of a power outage occurrence section prediction device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. Fig. 1 is a diagram showing the overall configuration of a power outage occurrence section prediction system 100. Fig. 2 is a schematic diagram showing an example of an area A in which a power outage occurrence section is predicted by the power outage occurrence section prediction system 100 according to this embodiment.

[0012] The following describes the configuration of the power outage occurrence section prediction system 100. As shown in FIG.

[0013] Here, the machine learning device 1 and the power outage occurrence section prediction device 2 are paired one-to-one and connected to be able to communicate with each other. Although not shown in FIG. 1, the machine learning device 1 and the power outage occurrence section prediction device 2 may be connected to each other via a network. The network may be, for example, a local area network (LAN), the Internet, a public telephone network, or a combination of these. There are no particular limitations on the specific communication method in the network, whether it is a wired connection or a wireless connection, etc. Alternatively, the machine learning device 1 and the power outage occurrence section prediction device 2 may be directly connected via a connector, rather than communicating using a network.

[0014] The power outage occurrence section prediction system 100 is used to predict sections where power outages will occur due to the adhesion of salt to insulating members of a power system. Insulating members are placed in power plants, substations, transmission lines, distribution lines, etc. of a power system, and electrically insulate electric wires from the ground. Examples of insulating members include insulators that are attached to electric wires and utility poles or steel towers that support the electric wires.

[0015] When salt contained in sea air adheres to insulating materials such as insulators attached to utility poles near the coastline, the insulating performance deteriorates, causing a ground fault and a power outage. The power outage occurrence section prediction system 100 predicts a section in a specified area A where a power outage will occur due to the adhesion of salt to insulating materials by using various information described later.

[0016] There is no particular limitation on the region A that is the target of prediction of the section where a power outage will occur. For example, as shown in FIG. 2, the region A may be the entire Chugoku region including Okayama, Hiroshima, Yamaguchi, Shimane, and Tottori prefectures in Japan, or may be a range up to a predetermined distance from the coastline in the Chugoku region. Alternatively, the region A may be within one or more cities, towns, or villages, or within an area to which power is supplied from one or more substations. In this embodiment, the region A is the range up to a predetermined distance inland from the coastline in the Chugoku region shown in FIG. 2.

[0017] The machine learning device 1 of the power outage occurrence section prediction system 100 will be described with reference to Fig. 3. Fig. 3 is a diagram showing functional blocks of the machine learning device 1.

[0018] The machine learning device 1 constructs a learning model by supervised learning for use in the power outage occurrence section prediction device 2. The learning model constructed by the machine learning device 1 is used to predict a section in which a power outage will occur due to adhesion of salt to insulating members in area A. As shown in FIG. 3 , the machine learning device 1 includes an input data acquisition unit 11, a label acquisition unit 12, a memory unit 13, and a learning model construction unit 14.

[0019] The input data acquisition unit 11 acquires various types of information on the area A as input data. The input data includes a data group including topographical information on the area A, positional information indicating the positions of each of a plurality of insulating members in the area A, insulation performance information on each of a plurality of insulating members in the area A, meteorological information on the area A, and phase-to-phase voltage information of the power system in the area A.

[0020] The topographical information is information on the topography of region A. Examples of the topographical information include types of topography such as alluvial fans, locations of river mouths and rivers, locations of trees and fields, and information on structures other than power systems located near coastlines. Examples of the information on structures include information on the height, shape, and location of structures. Such topographical information may be obtained, for example, from satellite images or the like, or from a system that manages topographical information.

[0021] The location information is information indicating the location of each of a plurality of insulating members provided in the power system in the area A. Examples of the location information include information such as the longitude, latitude, and altitude of the insulating members, and the arrangement of the insulating members on utility poles, steel towers, etc. This location information may be acquired, for example, from a power system management system or the like.

[0022] By combining the topographical information and the positional information, it is possible to specify information about the environment surrounding each of the insulating members. Examples of the information about the environment surrounding the insulating members include various information such as the distance from the insulating member to the coastline, the distance from the insulating member to trees and structures located in the vicinity, the height of trees and structures located between the insulating member and the coastline, the type of topography where the insulating member is located, and the presence or absence of a river or estuary.

[0023] The insulation performance information is information that indicates the insulation performance of the insulating member against adhesion of salt. Examples of the insulation performance information include the length and shape of the insulating member, whether or not a water-repellent substance is applied to the surface of the insulating member, and the cleaning history of the insulating member. For example, by lengthening the insulating member, the insulation distance between the electric wire and the utility pole is extended, thereby improving the insulation performance. The cleaning history is information such as the date and time when the insulating member was cleaned to remove the adhering salt. The insulation performance information may be acquired, for example, from a power system management system or the like.

[0024] The weather information is information about the weather in region A from a time point earlier than the reference time to the reference time. The reference time is any time point in the past. The length of the period from a time point earlier than the reference time to the reference time is not particularly limited. For example, it may be less than one week or more than one week.

[0025] Examples of meteorological information include weather-related information, tide-related information, etc. Examples of weather-related information include air pressure, wind speed, wind direction, temperature, precipitation, amount of sunlight, the presence or absence of a typhoon, the path of the typhoon, and the weather after the typhoon has passed. Examples of tide-related information include tide levels, times of high and low tides, etc. These meteorological information may be values ​​announced by meteorological observation stations, or may be values ​​obtained from measuring instruments that measure air pressure, wind speed, wind direction, temperature, precipitation, amount of sunlight, etc.

[0026] The zero-sequence voltage information is information about the zero-sequence voltage of the power grid in area A from a time point earlier than a reference time to the reference time. Examples of the zero-sequence voltage information include the magnitude, direction, and frequency of the zero-sequence voltage. The zero-sequence voltage information may be, for example, a measurement result of the zero-sequence voltage that is constantly measured by a power grid management system or the like.

[0027] The zero-phase voltage is determined by the change in capacitance due to the distance between the electric wire and the ground or trees, the combination of connection phases to which the transformer is connected, and the like, and does not change depending on the current flowing through the electric wire, the season, or the weather.

[0028] The zero-sequence voltage is adjusted to a small value under normal conditions when no earth fault occurs. However, if the potential difference between any of the phases of the three-phase AC and the ground decreases, the magnitude and phase of the three-phase voltages to ground become uneven, and the zero-sequence voltage increases. If the zero-sequence voltage exceeds a certain threshold and an earth fault occurs, the section where the earth fault occurs is isolated from the power grid by the earth fault protective relay, causing a power outage.

[0029] When the insulating performance of insulating materials is degraded due to salt adhesion, etc., a micro-ground fault occurs in which the magnitude of the zero-phase sequence voltage fluctuates continuously to an extent that does not exceed a threshold. A micro-ground fault does not cause a power outage because it does not exceed the zero-phase sequence voltage threshold at which the ground fault protective relay operates, but it can be used as information that indicates a sign of a power outage occurring.

[0030] The label acquisition unit 12 acquires, as a label, a section in the region A where a power outage has occurred due to adhesion of salt to an insulating member after a predetermined period of time has elapsed from a reference time. The length of the predetermined period is not particularly limited. For example, it may be less than one day, more than one day, or more than one week.

[0031] Here, the relationship between the data groups such as topographical information, location information, insulation performance information, meteorological information, and zero-phase voltage information acquired by the input data acquisition unit 11 and the power outage occurrence section acquired by the label acquisition unit 12 will be described.

[0032] The possibility of a power outage occurring can be estimated from information such as the weather and topography of area A, the location and insulation performance of insulating members in the power system of area A, and the zero-phase voltage of the power system. For example, it can be estimated that a section where insulating members are present near a coastline is more likely to experience a power outage. Even if the distance between the coastline and the insulating members is the same, the direction of the salt-containing sea breeze changes depending on the wind direction, the path of a typhoon, etc., and the amount of salt adhering to the insulating members changes depending on these weather conditions and the relative positions of the coastline and the insulating members.

[0033] In addition, even if the distance from the coastline to the insulating member and the weather conditions are the same, the possibility of a power outage varies depending on the type of terrain. For example, insulating members of power systems in sections that exist on alluvial fans facing the sea, sections where rivers with wide mouths run through urban areas, and sections where low-lying land continues inland along rivers, tend to be more susceptible to salt adhesion. In addition, if there are buildings, trees, etc. located between the coastline and the power system, even if the power system is located near the coastline, the buildings, trees, etc. block the sea breeze carried from the sea and prevent salt from adhering to the insulating members, so it can be estimated that the section is less likely to experience a power outage.

[0034] For example, it can be estimated that the insulation performance increases with the length of the insulating member, making it less likely that a power outage will occur. For example, by considering the location and insulation performance of the insulating member, topographical information of area A, meteorological information, and other information related to the magnitude and direction of the zero-phase voltage, the section where a power outage will occur can be predicted more accurately. Such causes of power outages and sections where power outages have occurred in the past are learned as teacher data, so that a highly accurate learning model can be constructed.

[0035] The storage unit 13 stores the topographical information and meteorological information of the area A acquired by the input data acquisition unit 11, the positional information and insulation performance information of the insulating members, and the zero-phase voltage information of the power system as a set of mutually linked data groups. The storage unit 13 then stores the input data, which is a set of data groups, and the label, which is a section in area A where a power outage has occurred for the input data, as a set of teacher data, by linking them together. The storage unit 13 also stores the learning model constructed by the learning model construction unit 14.

[0036] The learning model construction unit 14 constructs a learning model for predicting a section where a power outage will occur due to adhesion of salt to an insulating member by performing supervised learning using pairs of input data and labels stored in the memory unit 13 as teacher data. Then, the learning model construction unit 14 transmits the constructed learning model to the power outage occurrence section prediction device 2.

[0037] The learning model construction unit 14 can be realized, for example, by using a Support Vector Machine (hereinafter also referred to as SVM).

[0038] The power outage occurrence section prediction device 2 will be described with reference to Fig. 3. Fig. 3 is a functional block diagram of the power outage occurrence section prediction device 2. The power outage occurrence section prediction device 2 includes a control unit 20, a storage unit 24, a communication unit 25, and a display unit 26.

[0039] The control unit 20 is a part that controls the entire power outage occurrence section prediction device 2, and realizes various functions in this embodiment by appropriately reading and executing various programs from a storage area such as a ROM, RAM, flash memory, or hard disk (HDD). The control unit 20 may be a CPU. The control unit 20 includes a prediction data acquisition unit 21, a power outage occurrence section prediction unit 22, and an output unit 23.

[0040] The prediction data acquisition unit 21 acquires prediction data for the power outage occurrence section prediction unit 22 to predict a section where a power outage will occur. The prediction data acquisition unit 21 has a topographical information acquisition unit 211, a positional information acquisition unit 212, an insulation performance information acquisition unit 213, a meteorological information acquisition unit 214, and a zero-phase voltage information acquisition unit 215.

[0041] The topographical information acquisition unit 211 acquires topographical information of the region A. The topographical information of the region A may be information included in the teacher data used in the machine learning, or may be the latest topographical information at the present time acquired separately from the information included in the teacher data. Note that the present time refers to the time when the user starts work to predict the section in the region A where a power outage will occur.

[0042] The position information acquisition unit 212 acquires position information of each of the multiple insulating members in the area A. The position information may be information included in the teacher data used in the machine learning, or may be the latest position information at the present time acquired separately from the information included in the teacher data. The position information acquisition unit 212 may acquire the position information from, for example, a management system that manages the power grid.

[0043] The insulation performance information acquisition unit 213 acquires information indicating the insulation performance of each of a plurality of insulating members in the region A. The insulation performance information may be information included in the teacher data used in the machine learning, or may be the latest insulation performance information at the present time acquired separately from the information included in the teacher data. The insulation performance information acquisition unit 213 may acquire the information from, for example, a power system management system or the like.

[0044] The weather information acquisition unit 214 acquires weather information for region A from a predetermined time in the past to the present time separately from the weather information included in the teacher data used in machine learning. The period from a predetermined time in the past to the present time is not particularly limited. For example, it may be less than one week, more than one week, or more than one month. The weather information acquisition unit 214 may acquire, for example, values ​​announced by a weather observation station or values ​​obtained from measuring equipment that measures air pressure, wind speed, wind direction, temperature, precipitation, amount of sunlight, etc., as weather information.

[0045] The zero-sequence voltage information acquiring unit 215 acquires zero-sequence voltage information of the power grid in the area A from a past time point to a present time point separately from the zero-sequence voltage information included in the teacher data used in the machine learning. The zero-sequence voltage information acquiring unit 215 may acquire, for example, a measurement value of the zero-sequence voltage constantly measured by a power grid management system or the like as the zero-sequence voltage information.

[0046] The power outage occurrence section prediction unit 22 predicts the section in which a power outage will occur due to the adhesion of salt to insulating components of the power system in area A, based on the prediction data acquired by the prediction data acquisition unit 21 and the learning model.

[0047] The output unit 23 outputs the power outage occurrence section in area A predicted by the power outage occurrence section prediction unit 22 to the display unit 26.

[0048] The storage unit 24 stores the learning model acquired from the machine learning device 1. The storage unit 24 may also store, as a set of input data, the topographical information, the positional information, the insulation performance information, the meteorological information, and the zero-phase voltage information acquired by the prediction data acquisition unit 21, and store, as a label, a section in which a power outage is predicted to occur for the input data.

[0049] The communication unit 25 transmits and receives data to and from the machine learning device 1. For example, the power outage occurrence section prediction system 100 can transmit a learning model constructed by the learning model construction unit 14 from the machine learning device 1 to the power outage occurrence section prediction device 2 via the communication unit 25. Conversely, the power outage occurrence section prediction system 100 can transmit a group of prediction data acquired by the prediction data acquisition unit 21 and a newly occurring power outage occurrence section predicted by the power outage occurrence section prediction unit 22 from the power outage occurrence section prediction device 2 to the machine learning device 1 via the communication unit 25.

[0050] The display unit 26 is a monitor that displays the predicted results of the sections in which power outages will occur in the area A, which are output from the output unit 23 of the control unit 20.

[0051] Next, a description will be given of the operation during machine learning in the power outage occurrence section prediction system 100 according to this embodiment. Fig. 5 is a flowchart showing the operation of the machine learning device 1 during machine learning.

[0052] In step S11, the input data acquisition unit 11 of the machine learning device 1 acquires as input data a set of data including topographical information of area A, position information and characteristic information of each of multiple insulating components of the power system in area A, weather information for area A from before the reference time to the reference time, and zero-phase voltage information of the power system in area A from before the reference time to the reference time.

[0053] In step S12, the label acquisition unit 12 of the machine learning device 1 acquires, as a label, a section in area A where a power outage has occurred after a predetermined period has elapsed from the reference time.

[0054] In step S13, the learning model construction unit 14 of the machine learning device 1 accepts a pair of input data and a label as training data.

[0055] In step S14, the learning model construction unit 14 of the machine learning device 1 executes machine learning using this teacher data.

[0056] In step S15, the learning model construction unit 14 determines whether to end the machine learning or to repeat the machine learning. If the learning model construction unit 14 determines to repeat the machine learning (No in step S15), the process returns to step S11. Then, the machine learning device 1 repeats the same operation. On the other hand, if the learning model construction unit 14 determines to end the machine learning (Yes in step S15), the process proceeds to step S16. Note that the condition for ending the machine learning can be determined arbitrarily. For example, the machine learning may be ended when the machine learning has been repeated a predetermined number of times.

[0057] In step S16, the machine learning device 1 transmits the learning model constructed by the machine learning up to that point to the power outage occurrence section prediction device 2 via a network or the like.

[0058] Furthermore, the storage unit 13 of the machine learning device 1 stores this learning model. As a result, when the learning model is requested by the power outage occurrence section prediction device 2, it is possible to transmit the learning model to the power outage occurrence section prediction device 2. Furthermore, when new teacher data is acquired, further machine learning can be performed on the learning model.

[0059] Next, an example of a process of predicting a section in which a power outage will occur in area A by the power outage occurrence section prediction device 2 will be described with reference to Fig. 6. Fig. 6 is a flowchart showing the process flow of prediction of a section in which a power outage will occur in area A by the power outage occurrence section prediction device 2.

[0060] In step S21, the prediction data acquisition unit 21 acquires prediction data including topographical information of area A, position information and insulation performance information of each of a plurality of insulating components in area A, meteorological information of area A, and zero-phase voltage information.

[0061] In step S22, the power outage occurrence section prediction unit 22 predicts the section in which a power outage will occur based on a set of data including the topographical information, location information, insulation performance information, weather information, and zero-phase voltage information acquired in step S21 and the learning model transmitted from the machine learning device 1.

[0062] In step S23, the output unit 23 outputs the power outage occurrence section predicted in step S22 to the display unit 26. As a result, the power outage occurrence section in area A predicted by the power outage occurrence section prediction unit 22 is displayed on the display unit 26.

[0063] According to the machine learning device 1 or the power outage occurrence section prediction device 2 of the power outage occurrence section prediction system 100 according to the present embodiment described above, the following effects are achieved.

[0064] The machine learning device 1 of this embodiment includes an input data acquisition unit 11 that acquires as input data a group of data including location information indicating the respective positions of a plurality of insulating members that electrically insulate the electric wires of the power system from the ground in area A, topographical information of area A, weather information for area A from a time earlier than a reference time to the reference time, and zero-phase voltage information of the power system from a time earlier than the reference time to the reference time, a label acquisition unit 12 that acquires as labels an area in area A where a power outage has occurred due to the adhesion of salt to the insulating members after a predetermined period of time has elapsed from the reference time, and a learning model construction unit 14 that constructs a learning model for predicting an area where a power outage will occur by performing supervised learning using a pair of the input data and the label as teacher data.

[0065] This allows the system to reflect the zero-phase voltage that changes due to the occurrence of a micro-earth fault, along with the location information of the insulating components of the power system within area A, the topographical information of area A, and meteorological information, making it possible to construct a learning model for efficiently predicting with greater accuracy the locations of power outages caused by the adhesion of salt to insulating components.

[0066] Furthermore, in the machine learning device 1 according to this embodiment, the input data acquisition unit 11 further acquires, as input data, insulation performance information of each of the multiple insulating members.

[0067] This makes it possible to construct a learning model with higher prediction accuracy that can predict sections where power outages will occur by taking into account the insulating performance of each insulating component against salt adhesion.

[0068] In addition, the power outage occurrence section prediction device 2 in this embodiment is a power outage occurrence section prediction device 2 that uses a learning model constructed by the machine learning device 1, and includes a prediction data acquisition unit 21 that acquires location information, topographical information, weather information for area A from a predetermined time in the past to the present time, and zero-phase voltage information of the power system in area A from a predetermined time in the past to the present time as a prediction data group, and a power outage occurrence section prediction unit 22 that predicts a section where a new power outage will occur after a predetermined period of time has elapsed from the present time based on the prediction data group and the learning model.

[0069] As a result, by inputting a group of prediction data for a prediction target area into the learning model of this embodiment, it is possible to predict the section where a power outage will occur with higher accuracy. Therefore, by using the information on the section where a power outage will occur predicted by the power outage occurrence section prediction device 2, it is possible to more accurately identify insulating members that need to be cleaned in the section where a power outage will occur, and it is possible to more efficiently prevent power outages and perform recovery work.

[0070] Although the embodiment of the present invention has been described above, the present invention is not limited to the above embodiment and can be modified as appropriate.

[0071] In the above embodiment, the machine learning device 1 performs supervised learning by acquiring topographical information of area A, position information and insulation performance information of each insulating component of the power system in area A, meteorological information of area A, and zero-phase voltage information as input data; however, the device may be configured to perform supervised learning without acquiring insulation performance information as input data. [Explanation of symbols]

[0072] 1 Machine learning device 11 Input data acquisition section 12 Label Acquisition Section 14 Learning Model Construction Department Region A

Claims

1. an input data acquisition unit that acquires as input data a data group including position information indicating the positions of each of a plurality of insulating members that electrically insulate electric wires of a power system from the ground within a specified region, topographical information of the specified region, meteorological information of the specified region from a time point earlier than a reference time to the reference time, and zero-phase voltage information of the power system from a time point earlier than the reference time to the reference time; a label acquisition unit that acquires, as a label, a section in the specified region where a power outage has occurred due to adhesion of salt to the insulating member after a specified period has elapsed from the reference time; A machine learning device comprising: a learning model construction unit that constructs a learning model for predicting the area in which the power outage will occur by performing supervised learning using a pair of the input data and the label as training data.

2. The machine learning device according to claim 1 , wherein the input data acquisition unit further acquires, as the input data, insulation performance information of each of the plurality of insulating members.

3. A power outage occurrence section prediction device using the learning model constructed by the machine learning device according to claim 1 or 2, a prediction data acquisition unit that acquires, as a group of prediction data, the location information, the topographical information, meteorological information for the specified area from a specified time point in the past to a current time, and zero-phase voltage information of the power system from the specified time point in the past to a current time; A power outage occurrence section prediction device comprising: a power outage occurrence section prediction unit that predicts a section in which the power outage will occur newly after a predetermined period of time has elapsed from the present time based on the group of prediction data and the learning model.

Images