Distribution line accident cause estimation model generation device, distribution line accident cause estimation device, distribution line accident cause estimation model generation program, and distribution line accident cause estimation program

Abstract

To provide a distribution line accident cause inferring model generation device, a distribution line accident cause inference device, a distribution line accident cause inferring model generation program, and a distribution line accident cause inference program for improving the accuracy of inferring a cause of a distribution line accident.SOLUTION: A distribution line accident cause inference device 20 includes an accident waveform data acquisition unit 21 that acquires, as accident waveform data 44, a waveform of measurement information when a distribution line accident has occurred, an accident feature amount data generation unit 22 that generates accident feature amount data 45 by extracting a feature amount from the accident waveform data, and a cause inference unit 26 that inputs the accident feature amount data to a learned model 303 to output an inference result of the cause of the distribution line accident. The learned model is generated by machine learning of learning feature amount data on the basis of a prescribed algorithm.SELECTED DRAWING: Figure 17

Description

[Technical Field]

[0001] The present invention relates to a distribution line accident cause estimation model generation device, a distribution line accident cause estimation device, a distribution line accident cause estimation model generation program, and a distribution line accident cause estimation program. [Background technology]

[0002] When an abnormality occurs in a distribution line, such as when an obstacle such as a bird or animal or flying object comes into contact with the line, or when the line comes into contact with another line, the abnormality may be detected by a relay installed in the substation, causing the circuit breaker of the distribution line to open, resulting in an accident (hereinafter referred to as a "distribution line accident") that causes a power outage on the distribution line.

[0003] BACKGROUND ART When a power distribution line accident occurs, a technique is known for estimating the cause of the accident using waveform data of current and voltage at the time of the accident (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2019-30193 Summary of the Invention [Problem to be solved by the invention]

[0005] Incidentally, when estimating the cause of a power distribution line accident using waveform data of current and voltage at the time of the accident, features are extracted from each waveform, and the extracted features are used as input to compare with models generated by machine learning or statistical methods.

[0006] However, in conventional technologies for estimating the cause of power distribution line accidents using models generated by machine learning or statistical methods, when the current and voltage waveforms and the model waveform are the same waveform but out of phase, the two waveforms are recognized as different waveforms. For this reason, further improvement in the accuracy of estimating the cause of power distribution line accidents using conventional technologies for estimating the cause of power distribution line accidents using models generated by machine learning or statistical methods is required.

[0007] The present invention has been made in view of the above-mentioned problems, and has an object to improve the accuracy of estimating the cause of a power distribution line accident. [Means for solving the problem]

[0008] A power distribution line accident cause estimation device according to a representative embodiment of the present invention includes an accident waveform data acquisition unit that acquires a waveform of measurement information as accident waveform data when a power distribution line accident occurs; an accident feature data generation unit that generates accident feature data by extracting features from the accident waveform data by removing the effect of phase difference; and a cause estimation unit that inputs the accident feature data to a trained model that has been generated by machine learning training feature data based on a predetermined algorithm, and outputs an estimation result of the cause of the power distribution line accident, wherein the training feature data is generated by extracting features from training waveform data that is associated with cause data of the power distribution line accident when the waveform of the measurement information at the time of the power distribution line accident occurred. [Effects of the Invention]

[0009] According to the present invention, it is possible to improve the accuracy of estimating the cause of a power distribution line accident. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a diagram showing a system configuration of a power distribution line accident cause estimating device according to an embodiment of the present invention; [Figure 2] FIG. 1 is a diagram showing an outline of a power distribution line accident. [Figure 3] FIG. 1 is a diagram illustrating an example of a supply disruption accident. [Figure 4] FIG. 10 is a diagram illustrating an example of a successful reclosing fault. [Figure 5] 1 is a diagram showing a functional block configuration of a power distribution line accident cause estimation model generation device according to an embodiment of the present invention; [Figure 6] 1 is a diagram illustrating a hardware configuration of a server device that realizes a power distribution line accident cause estimation model generation device according to the present embodiment. [Figure 7] 10 is a flowchart illustrating an example of a model generation process performed by the power distribution line accident cause estimation model generation device. [Figure 8] FIG. 10 is a diagram illustrating an example of a process for acquiring learning waveform data. [Figure 9] FIG. 1 is a diagram illustrating an example of a cause of a power distribution line accident. [Figure 10] FIG. 10 is a diagram showing an example of the waveform of measurement information when the cause of the power distribution line accident is a cable accident. [Figure 11] FIG. 10 is a diagram showing an example of the waveform of measurement information when the cause of a distribution line accident is a switch insulation breakdown. [Figure 12] 10 is a table showing an example of a target for estimating the cause of a power distribution line accident. [Figure 13] FIG. 10 is a schematic diagram showing an example of causes corresponding to power distribution line accidents in learning waveform data. [Figure 14] 10A and 10B are diagrams showing examples of waveforms of measurement information having different phases but the same waveform. [Figure 15] 10A and 10B are diagrams illustrating an example of learning waveform data and learning feature amount data of a sine wave. [Figure 16] 10A and 10B are diagrams illustrating an example of learning waveform data and learning feature amount data of a square wave. [Figure 17] 1 is a diagram showing a functional block configuration of a power distribution line accident cause estimating device according to an embodiment of the present invention; [Figure 18] 10 is a flowchart illustrating an example of a cause estimation process performed by the power distribution line accident cause estimation device. [Figure 19] FIG. 2 is a diagram illustrating an example of a screen interface of the power distribution line accident cause estimation device. [Figure 20] FIG. 10 is a diagram illustrating an example of the relationship between the reliability and the result of cause estimation by the power distribution line accident cause estimation device. [Figure 21] 10 is a table showing the relationship between the cause of an accident and the estimation result in terms of the F value used to evaluate the result of the cause estimation process performed by the power distribution line accident cause estimation device. [Figure 22] 10 is a table showing evaluation results of the cause estimation process performed by the power distribution line accident cause estimation device. DETAILED DESCRIPTION OF THE INVENTION

[0011] 1. Overview of the embodiment First, a typical embodiment of the invention disclosed in this application will be outlined. In the following description, for example, reference numerals in the drawings corresponding to components of the invention are written in parentheses.

[0012] [1] A model generation device (30) for estimating the cause of a power distribution line accident, comprising: a learning waveform data acquisition unit (31) that acquires, as learning waveform data (41), a waveform of measurement information at the time of a power distribution line accident, by correlating it with cause data (42) of the power distribution line accident when the waveform of the measurement information occurred; a learning feature data generation unit (32) that generates learning feature data (43) by extracting features by removing the influence of phase difference from the learning waveform data (41); and a trained model generation unit (36) that generates a trained model (303) for causing a computer (1) to function to estimate the cause of the power distribution line accident based on the waveform of measurement information acquired when an input power distribution line accident occurred, by performing machine learning on the learning feature data (43) based on a predetermined algorithm.

[0013] [2] In the power distribution line accident cause estimation model generation device (30) described in [1], the learning feature data (43) is generated by extracting features of the waveform shape of the learning waveform data (41).

[0014] [3] In the power distribution line accident cause estimation model generation device (30) described in [1] or [2], the learning feature data (43) removes the influence of a phase difference from the waveform of the learning waveform data (41) by a Hilbert transform.

[0015] [4] In the model generation device (30) for estimating the cause of a power distribution line accident described in any one of [1] to [3], the cause data (42) of the power distribution line accident associated with the learning feature data (43) is information that can identify the event that caused the accident, the component in which the accident occurred, and the equipment in which the accident occurred.

[0016] [5] A program for generating a model for estimating the cause of a power distribution line accident, which causes a computer (1) to execute the following steps: a step (S101) of acquiring, as learning waveform data (41), a waveform of measurement information at the time of a power distribution line accident, by correlating the waveform of the measurement information with cause data (42) of the power distribution line accident when the waveform of the measurement information occurred; a step (S102) of generating learning feature data (43) by extracting features by removing the influence of phase difference from the learning waveform data (41); and a step (S103) of generating a trained model (303) for causing the computer (1) to function to estimate the cause of the power distribution line accident based on the waveform of measurement information acquired when an input power distribution line accident occurred, by machine learning the learning feature data (43) based on a predetermined algorithm.

[0017] [6] A power distribution line accident cause estimation device (20) comprising: an accident waveform data acquisition unit (21) that acquires a waveform of measurement information as accident waveform data (44) when a power distribution line accident occurs; an accident feature data generation unit (22) that generates accident feature data (45) by extracting features from the accident waveform data (44) by removing the effect of phase difference; and a cause estimation unit (26) that inputs the accident feature data (45) into a trained model (303) that is generated by machine learning the learning feature data (43) based on a predetermined algorithm, and outputs an estimation result of the cause of the power distribution line accident, wherein the learning feature data (43) is generated by extracting features from learning waveform data (41) that is associated with cause data (42) of the power distribution line accident when the waveform of the measurement information at the time of the power distribution line accident occurs.

[0018] [7] In the power distribution line fault cause estimation device (20) described in [6], the fault feature data (45) is generated by extracting features of the waveform shape of the fault waveform data (44).

[0019] [8] In the power distribution line fault cause estimation device (20) described in [6] or [7], the fault feature data (45) is obtained by removing the influence of a phase difference from the waveform of the fault waveform data (44) by a Hilbert transform.

[0020] [9] In the power distribution line accident cause estimation device (20) described in any one of [6] to [8], the causes of the power distribution line accident estimated by the cause estimation unit (26) are the event that caused the accident, the component in which the accident occurred, and the equipment in which the accident occurred.

[0021]

[10] In the power distribution line accident cause estimating device (20) according to any one of [6] to [9], the cause estimating unit (26) outputs a plurality of estimated causes of the power distribution line accident.

[0022]

[11] In the power distribution line accident cause estimation device (20) according to any one of [6] to

[10] , the cause estimation unit (26) outputs the estimated cause of the power distribution line accident together with the reliability of the cause.

[0023]

[12] A program for causing a computer (1) to execute the following steps: a step (S201) of acquiring a waveform of measurement information as accident waveform data (44) when a distribution line accident occurs; a step (S202) of generating accident feature data (45) by extracting features from the accident waveform data (44) by removing the influence of phase difference; and a step (S203) of inputting the accident feature data (45) into a trained model (303) generated by machine learning training feature data (43) based on a predetermined algorithm, and outputting an estimation result of the cause of the distribution line accident, wherein the training feature data (43) is generated by extracting features from training waveform data (41) associated with cause data (42) of the distribution line accident when the waveform of the measurement information at the time of the distribution line accident occurs.

[0024] 2. Specific examples of embodiments Specific examples of embodiments of the present invention will be described below with reference to the drawings. In the following description, components common to each embodiment will be given the same reference numerals, and repeated explanations will be omitted. Furthermore, the drawings are schematic, and the dimensional relationships and ratios of each element may differ from the actual objects.

[0025] FIG. 1 is a diagram showing a system configuration of a distribution line accident cause estimation model generation device 30 and a distribution line accident cause estimation device 20 according to an embodiment of the present invention. As shown in FIG. 1, the distribution line accident cause estimation model generation device 30 and the distribution line accident cause estimation device 20 are realized by a server device 1, which is an example of a computer capable of executing a computer program. The server device 1 is communicably connected via a network NW to a computer 2 that uses the distribution line accident cause estimation device 20 or a client-side computer terminal such as a smartphone 3. The network NW is, for example, a wide area network (WAN) such as the Internet. The network NW can be connected to the server device 1 and the client-side computer terminal via wired or wireless communication.

[0026] The server device 1 is communicatively connected to a sensor-equipped switch 6 provided on a distribution line 5 via a network NW. The sensor-equipped switch 6 is supported on a utility pole 52 together with an electric wire 51 constituting the distribution line 5. The sensor-equipped switch 6 can acquire measurement information, for example, of at least one of the zero-phase voltage value and the zero-phase current value of the electric energy flowing through the electric wire 51. In the following description, the measurement information of the electric energy flowing through the electric wire 51 acquired by the sensor-equipped switch 6 is the zero-phase current value. The sensor-equipped switch 6 acquires the measurement information at predetermined intervals or continuously and outputs it to the network NW. The measurement information acquired by the sensor-equipped switch 6 can be acquired by the server device 1 via the network NW. The server device 1 uses the waveform of the acquired measurement information to perform functions as a distribution line accident cause estimation device 20 and a distribution line accident cause estimation model generation device 30, which will be described below.

[0027] In this embodiment, the power distribution line accident cause estimating device 20 and the power distribution line accident cause estimating model generating device 30 are both described as being realized by the server device 1, but the present invention is not limited to this configuration. For example, the power distribution line accident cause estimating device 20 and the power distribution line accident cause estimating model generating device 30 may be realized by separate computers or dedicated devices.

[0028] Fig. 2 is a diagram showing an overview of a distribution line accident. As shown in Fig. 2, a distribution line accident is an accident in which, when an abnormality occurs in a distribution line 5, for example, an obstacle such as a flying object F or a tree T comes into contact with an electric wire 51, the abnormality causes a circuit breaker 7 installed in a substation to operate, resulting in a power outage in the distribution line 5. Causes of a distribution line accident include, in addition to contact with an obstacle, for example, contact between electric wires 51 and failure of equipment such as a transformer 53.

[0029] FIG. 3 is a diagram showing an example of a supply disruption accident. FIG. 4 is a diagram showing an example of a successful reclosing accident. Types of distribution line accidents will be described with reference to FIGS. 3 and 4. In FIGS. 3 and 4, the distribution line 5 is divided into three sections, section 1 to section 3, downstream of substation S. Sensor-integrated switches 6A, 6B, and 6C are provided in each section of the distribution line 5. In addition, in FIGS. 3 and 4, black triangles indicate the on state of the sensor-integrated switch 6, and white triangles indicate the off state of the sensor-integrated switch 6.

[0030] In the supply disruption accident shown in Figure 3, if an accident such as contact with a tree T1 occurs on the distribution line 5, the circuit breaker installed at substation S will trip, causing a power outage in the entire section of the distribution line 5. After that, in order to identify the location AP1 where the cause of the accident occurred in the distribution line 5 (hereinafter referred to as the "fault point"), the sensor-equipped switches 6A to 6C are turned on for each section. By performing this operation, a distribution line accident will occur again in the two sections in Figure 3 where the fault point AP1 is located. After identifying the section where the accident occurred as described above, workers will go to the site and identify the fault point AP1 in the case of a supply disruption accident.

[0031] In the successful reclosing accident shown in Figure 4, similar to the supply disruption accident, an accident such as contact with a tree T2 occurs on the distribution line 5, causing the circuit breaker installed at substation S to trip and resulting in a power outage across the entire section of the distribution line 5. However, the cause of the accident may subsequently be temporarily eliminated at the fault point AP2 of the distribution line, for example, because the tree T2 moves away from the distribution line 5. In this case, even if the sensor-integrated switches 6A to 6C are turned on for each section to identify the fault point AP2, power transmission is successful across all sections, making it impossible to identify the fault point AP2. In other words, in the successful reclosing accident, workers must inspect all sections of the distribution line 5 to identify the fault point AP2. In particular, in the successful reclosing accident, the far-end length L2 of the power outage is longer than the far-end length L1 of the supply disruption accident, which makes it time-consuming to identify the fault point AP2.

[0032] Therefore, the distribution line accident cause estimation device 20 focuses on the fact that the waveforms of the measurement information of the zero-phase sequence voltage value or the zero-phase sequence current value at the time of the accident differ depending on the cause of the accident, and estimates the cause of the distribution line accident by the following processing: The distribution line accident cause estimation device 20 inputs information based on the waveforms of the measurement information at the time of the accident into a trained model generated in advance, and estimates the cause of the distribution line accident as described above.

[0033] The distribution line accident cause estimation model generation device 30 generates a trained model that the distribution line accident cause estimation device 20 uses to estimate the cause of a distribution line accident. The trained model is generated based on the waveform of measurement information acquired when an input distribution line accident occurs, by machine learning training feature data based on a predetermined algorithm. The training feature data is data obtained by extracting features by removing phase differences from training waveform data, which is the waveform of the measurement information at the time of the distribution line accident. The training waveform data is associated with cause data (42) of the distribution line accident when the waveform occurred. The distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 will be described below.

[0034] [Model generation device for estimating the cause of power distribution line accidents] Fig. 5 is a diagram showing a functional block configuration of a power distribution line accident cause estimation model generation device 30 according to this embodiment. Fig. 6 is a diagram showing a hardware configuration of a server device 1 that realizes the power distribution line accident cause estimation model generation device 30 according to this embodiment.

[0035] As shown in FIG. 6, the server device 1 includes, as hardware resources, an arithmetic unit 101, a storage device 102, an input device 103, an I / F (Interface) device 104, an output device 105, and a bus 106, for example.

[0036] The arithmetic device 101 is configured by a processor such as a CPU (Central Processing Unit). The storage device 102 has a storage area for storing programs 1021 for causing the arithmetic device 101 to execute various data processing operations, and data 1022 such as parameters and calculation results used in the data processing by the arithmetic device 101, and is configured by, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), and a flash memory.

[0037] The program 1021 stored in the storage device 102 includes a distribution line accident cause estimation model generation program that causes a computer to function as the distribution line accident cause estimation model generation device 30, and is pre-installed in the storage device 102, for example.

[0038] The data 1022 stored in the storage device 102 includes, for example, various data such as the training waveform data 41, the distribution line accident cause data 42, and the training feature data 43 shown in FIG. 5, the accident waveform data 44 and the accident feature data 45 shown in FIG. 17, and the generated trained model 303.

[0039] The program 1021 and the data 1022 may be distributed via a network, or may be written to a non-transitory computer readable medium such as a CD-ROM or a memory card and distributed.

[0040] The input device 103 is a device that detects input of information from the outside, and examples thereof include a keyboard, a mouse, a pointing device, a button, a touch panel, etc. The I / F device 104 is a device that transmits and receives data to and from the outside, and is composed of, for example, a communication control circuit, an input / output port, an antenna, etc. for wired or wireless communication.

[0041] The output device 105 is a device that outputs information obtained by data processing by the arithmetic device 101. Examples of the output device 105 include external storage devices such as SSDs and HDDs, and display devices such as LCDs (Liquid Crystal Displays) and organic EL (Electro Luminescence) displays. The bus 106 interconnects the arithmetic device 101, the storage device 102, the input device 103, the I / F device 104, and the output device 105, enabling data to be transmitted and received among these devices.

[0042] In the server device 1, the arithmetic device 101 executes calculations in accordance with a program 1021 stored in the storage device 102. The server device 1 executes calculations and controls the storage device 102, the input device 103, the I / F device 104, the output device 105, and the bus 106, thereby realizing each functional block (learning waveform data acquisition unit 31, learning feature data generation unit 32, storage unit 35, and trained model generation unit 36) of the power distribution line accident cause estimation model generation device 30 shown in FIG.

[0043] The power distribution line accident cause estimation model generation device 30 is realized by, for example, the above-mentioned server device 1 or an information processing device (computer) such as a personal computer (PC). The power distribution line accident cause estimation model generation device 30 generates training feature data in accordance with an installed power distribution line accident cause estimation model generation program, and generates a trained model by performing machine learning on the generated training feature data based on a predetermined algorithm.

[0044] As shown in Fig. 5, the distribution line accident cause estimation model generation device 30 has, as functional blocks for generating a trained model, a training waveform data acquisition unit 31, a training feature data generation unit 32, a storage unit 35, and a trained model generation unit 36. These functional blocks are realized by the hardware resources of the server device 1 shown in Fig. 6 constituting the distribution line accident cause estimation model generation device 30 working together with installed software (various programs including a distribution line accident cause estimation model generation program).

[0045] FIG. 7 is a flowchart showing an example of a model generation process performed by the power distribution line accident cause estimation model generation device 30.

[0046] The learning waveform data acquisition unit 31 acquires waveforms used to generate a trained model, that is, learning waveform data (step S101).

[0047] The learning feature data generating unit 32 generates learning feature data from which features are extracted by removing the phase difference from the learning waveform data acquired by the learning waveform data acquiring unit 31 (step S102).

[0048] The trained model generation unit 36 ​​performs machine learning on the training feature data based on a predetermined algorithm to generate a trained model that causes the computer to estimate the cause of the power distribution line accident based on the waveform of the measurement information (accident waveform data) acquired when the input power distribution line accident occurred (step S103).

[0049] Hereinafter, each functional block of the power distribution line accident cause estimation model generation device 30 will be described in detail.

[0050] The learning waveform data acquiring unit 31 acquires waveforms (learning waveform data) of measurement information of zero-phase-sequence voltage values ​​or zero-phase-sequence current values ​​at the time of a distribution line accident, which are acquired from the sensor-integrated switchgear 6 via the network NW. Specifically, the learning waveform data acquiring unit 31 acquires, as learning waveform data, waveform data of the waveforms of measurement information of zero-phase-sequence voltage values ​​or zero-phase-sequence current values ​​at the time of a distribution line accident, which the user has stored in various storage areas such as the storage unit 35, from the various storage areas.

[0051] 8 is a diagram showing an example of a process for acquiring learning waveform data. In FIG. 8, waveform data D1 of the measurement information has, for example, a time of 1.1 seconds, a sampling frequency of 9.6 kHz, and 10,580 plotted points. The learning waveform data acquiring unit 31 divides the acquired learning waveform data D into a predetermined number (for example, 55 pieces) of data D2 each having a predetermined length, for example, a time of 0.02 seconds, and 192 plotted points.

[0052] FIG. 9 is a diagram showing an example of the cause of a distribution line accident. FIG. 10 is a diagram showing an example of the waveform of measurement information when the cause of the distribution line accident is a cable accident. FIG. 11 is a diagram showing an example of the waveform of measurement information when the cause of the distribution line accident is a switch insulation breakdown. As shown in FIGS. 9 to 11, the waveform of measurement information for a distribution line accident differs depending on the cause. Specifically, the frequency of the waveform of measurement information for a cable accident that occurred at fault point AP3 shown in FIG. 9 (see FIG. 10) is higher than the frequency of the waveform of measurement information for switch insulation breakdown that occurred at fault point AP4 (see FIG. 11). Waveform data at the time of the accident can be acquired from the sensor-integrated switch 6 (see FIG. 1) installed on the distribution line 5.

[0053] Fig. 12 is a table showing an example of an estimation target for the cause of a power distribution line accident. Fig. 13 is a schematic diagram showing an example of a cause corresponding to a power distribution line accident in the learning waveform data.

[0054] As shown in FIG. 12, the learning waveform data is associated with cause data of a distribution line accident. The cause data of a distribution line accident is information that enables the cause of the distribution line accident to be identified. The cause data of a distribution line accident defines a plurality of factors corresponding to, for example, the following three types. In the learning waveform data, the types of cause data of a distribution line accident are the accident occurrence event F1, the component where the accident occurred F2, and the equipment where the accident occurred F3. As shown in the example of a distribution line accident in FIG. 13, the cause data of a distribution line accident identifies, for example, that the occurrence event is a broken wire, the component where the accident occurred is a tree T, and the equipment where the accident occurred is a high-voltage line of the distribution line 5.

[0055] The learning waveform data acquiring unit 31 acquires, as learning waveform data, waveforms of measurement information acquired during a power distribution line accident such as those shown in Figures 10 and 11 in association with data on the cause of the power distribution line accident when the waveform shown in Figure 12 occurred. Each piece of learning waveform data is associated with the cause of the accident and stored in, for example, the storage unit 35.

[0056] FIG. 14 is a diagram showing an example of waveforms of measurement information that have the same waveform but different phases. In FIG. 14, the waveform WA shown by the solid line and the waveform WB shown by the dashed line have the same waveform but different phases. In the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30, if no processing of waveform data is performed, the waveform WA and the waveform WB are usually recognized as different waveforms. Waveforms with different phases, such as the waveform WA and the waveform WB, are recognized as different waveforms even if they have the same wave shape. The problem of waveforms with different phases being recognized as different waveforms even if they have the same wave shape can occur between learning waveform data, between multiple fault waveform data, and between learning waveform data and fault waveform data.

[0057] The training feature data generation unit 32 extracts features from the training waveform data without being affected by the phase or time axis. For example, the training feature data generation unit 32 generates training feature data from which phase differences have been removed and features have been extracted by drawing a phase plane for the waveform of the training waveform data. Specifically, the training feature data generation unit 32 generates training feature data using information based on the original waveform on the horizontal axis (time) and information obtained by performing a Hilbert transform, i.e., an inverse Fourier transform with the phase of the original waveform information shifted by 90°, on the vertical axis (voltage value or current value). The waveform of the measurement information after the Hilbert transform is the waveform of measurement information obtained by performing an inverse Fourier transform with the phase of the original waveform data shifted by 90°.

[0058] Fig. 15 is a diagram showing an example of sine wave learning waveform data and learning feature data. Fig. 16 is a diagram showing an example of square wave learning waveform data and learning feature data. In Fig. 15, sine waves W1 and W2 of the learning waveform data have different phases (inverted) but have the same waveform. In Fig. 16, square waves W3 and W4 of the learning waveform data have different phases (inverted) but have the same waveform.

[0059] 15, the learning feature data generation unit 32 extracts phase planes P1 and P2 as learning feature data by performing a Hilbert transform on the acquired learning waveform data (sine wave W1 and sine wave W2). That is, by performing feature extraction by the Hilbert transform, the learning feature data generation unit 32 obtains data from which the influence of phase shifts has been removed. For this reason, the phase planes P1 and P2 can be recognized as the same waveform in the power distribution line accident cause estimation device 20 and the power distribution line accident cause estimation model generation device 30.

[0060] Furthermore, the training feature data generation unit 32 is not limited to cases where the acquired training waveform data is a sine wave, and can also extract training feature data for other waveforms, for example, square waves W3 and W4 as shown in Fig. 16. That is, the training feature data generation unit 32 can extract phase planes P3 and P4 as training feature data by performing a Hilbert transform on the square waves W3 and W4.

[0061] Note that the training feature data generation unit 32 is not limited to generating training feature data from training waveform data using the Hilbert transform described above, and any other method may be used as long as it can generate training feature data from which phase shifts have been removed. For example, the training feature data generation unit 32 may extract features using frequency analysis based on the Wigner distribution or TDA (Topological Data Analysis, persistent homology).

[0062] The trained model generation unit 36 ​​performs machine learning on the training feature data based on a predetermined algorithm. The trained model generation unit 36 ​​can use, for example, a classification decision tree, discriminant analysis, nearest neighbor, or a support vector machine (SVM) as the predetermined algorithm used for machine learning. The trained model generation unit 36 ​​performs machine learning to generate a trained model 303 that causes a computer to function to estimate the cause of a power distribution line accident based on the waveform of input measurement information acquired when the power distribution line accident occurs. The trained model 303 is stored in, for example, the storage unit 35 and used by the power distribution line accident cause estimation device 20 to estimate the cause of the power distribution line accident. Furthermore, when the power distribution line accident cause estimation model generation device 30 and the power distribution line accident cause estimation device 20 are realized by different hardware (computers), the trained model 303 may be available to the power distribution line accident cause estimation device 20 via network HW.

[0063] [Distribution line accident cause estimation device] 17 is a diagram showing a functional block configuration of a power distribution line accident cause estimating device 20 according to this embodiment. The power distribution line accident cause estimating device 20 is realized by the server device 1 shown in FIG. 6, similar to the above-described power distribution line accident cause estimating model generating device 30.

[0064] In the server device 1, the program 1021 stored in the storage device 102 includes a power distribution line accident cause estimation program that causes a computer to function as the power distribution line accident cause estimation device 20, and is installed in the storage device 102 in advance, for example.

[0065] In the server device 1, the arithmetic device 101 executes calculations in accordance with a program 1021 stored in the storage device 102. The server device 1 executes calculations and controls the storage device 102, the input device 103, the I / F device 104, the output device 105, and the bus 106, thereby realizing each functional block (fault waveform data acquisition unit 21, fault feature data generation unit 22, storage unit 25, and cause estimation unit 26) of the power distribution line fault cause estimation device 20 shown in Fig. 17 .

[0066] The program 1021 stored in the storage device 102 includes a power distribution line accident cause estimation program that causes a computer to function as the power distribution line accident cause estimation device 20, and is pre-installed in the storage device 102, for example.

[0067] The storage device 102 of the server device 1 stores various data such as, for example, learning waveform data 41, distribution line accident cause data 42, and learning feature data 43 (see FIG. 5) used in the distribution line accident cause estimation model generation device 30, as well as fault waveform data 44 and fault feature data 45 (see FIG. 17) used in the distribution line accident cause estimation device 20 described later, and data 1022 such as the generated trained model 303.

[0068] The power distribution line accident cause estimation device 20 is realized by, for example, an information processing device (computer) such as a personal computer (PC) in addition to the above-mentioned server device 1. The power distribution line accident cause estimation device 20 generates accident feature data according to an installed power distribution line accident cause estimation program, and inputs the generated accident feature data to a trained model 303 generated by the power distribution line accident cause estimation model generation device 30, thereby outputting an estimation result of the cause of the power distribution line accident.

[0069] 17, the power distribution line accident cause estimation device 20 has, as functional blocks for generating an estimation result of the cause of a power distribution line accident, a fault waveform data acquisition unit 21, a fault feature data generation unit 22, a storage unit 25, and a cause estimation unit 26. These functional blocks are realized by the hardware resources of the server device 1 shown in FIG. 6 constituting the information processing device as the power distribution line accident cause estimation device 20 working together with installed software (various programs including a power distribution line accident cause estimation program).

[0070] FIG. 18 is a flowchart showing an example of a cause estimation process performed by the power distribution line accident cause estimation device 20.

[0071] The fault waveform data acquiring unit 21 acquires, as fault waveform data, a waveform of measurement information that occurs when a distribution line fault occurs (step S201).

[0072] The accident feature amount data generating unit 22 generates accident feature amount data by extracting feature amounts from the accident waveform data acquired by the accident waveform data acquiring unit 21 (step S202).

[0073] The cause estimation unit 26 inputs the accident feature data into a trained model generated by machine learning the learning feature data generated by the distribution line accident cause estimation model generation device 30 based on a predetermined algorithm, and outputs an estimation result of the cause of the distribution line accident (step S203).

[0074] Each functional block of the power distribution line accident cause estimation device 20 will be described in detail below.

[0075] The fault waveform data acquisition unit 21 acquires, as fault waveform data, the waveform of measurement information that occurs during a distribution line accident, acquired from the sensor-integrated switchgear 6 via the network NW. The fault waveform data is input to the trained model 303 in the distribution line accident cause estimation device 20 and is used to estimate the cause of the distribution line accident.

[0076] FIG. 19 is a diagram showing an example of a screen interface 1051 of the power distribution line accident cause estimation device 20. The screen interface 1051 shown in FIG. 19 is displayed on an output device 105 provided in a computer that realizes the power distribution line accident cause estimation device 20. The screen interface 1051 is provided with a fault waveform data selection unit 1052, an estimation result display unit 1053, and the like. In the fault waveform data selection unit 1052, a storage area and a file name of the fault waveform data that is the target of the distribution line accident cause estimation process are designated by the user. Specifically, the fault waveform data acquisition unit 21 acquires, from various storage areas, for example, fault waveform data that has been stored in various storage areas such as the storage unit 25 and that has been acquired at the time of the power distribution line accident.

[0077] The accident waveform data acquiring unit 21, like the learning waveform data acquiring unit 31 described above, divides the waveform data of the acquired measurement information into a predetermined number of waveform data pieces each having a predetermined length (see FIG. 8).

[0078] The accident feature data generation unit 22 extracts features from the accident waveform data without being affected by the phase or time axis, similar to the above-described learning feature data generation unit 32. The accident feature data generation unit 22 performs the same processing on the accident waveform data as the learning feature data generation unit 32 performed when generating the learning feature data, so that the accident feature data 45 can be compared with the trained model 303 in the cause estimation unit 26 described below. For example, the accident feature data generation unit 22 generates accident feature data from which features have been extracted by removing phase differences by drawing a phase plane for the accident waveform data, similar to the learning feature data (see FIGS. 15 and 16).

[0079] The fault feature data generating unit 22 performs a Hilbert transform on the acquired fault waveform data to generate a phase plane from which features are extracted, i.e., fault feature data. The fault feature data generating unit 22 extracts features from the phase plane using the Hilbert transform, resulting in data from which the effects of phase shifts have been removed. Therefore, fault feature data generated from fault waveform data with different phases can be recognized as the same waveform in the power distribution line fault cause estimation device 20.

[0080] Note that the accident feature data generation unit 22 is not limited to generating accident feature data from accident waveform data by the Hilbert transform as in the learning feature data generation unit 32, and may use other methods as long as they can generate accident feature data from which phase shifts have been removed. That is, the accident feature data generation unit 22 may extract features by, for example, frequency analysis using a Wigner distribution or TDA (Topological Data Analysis, persistent homology).

[0081] The cause estimation unit 26 estimates the cause of a power distribution line accident by inputting the accident feature data generated by the accident feature data generation unit 22 into a trained model stored in the storage unit 25. The cause estimation unit 26 compares the trained model, which is generated from the training feature data and associated with the cause data of the power distribution line accident, with the waveform of the accident feature data, and identifies the waveform data of the trained model that is most likely to be similar to the waveform of the accident feature data.

[0082] FIG. 20 is a diagram showing an example of the relationship between the cause estimation result and reliability by the power distribution line accident cause estimation device 20. Cause estimation result display screens 1054A and 1054B, on which the cause estimation result and reliability shown in FIG. 20 are displayed, are output to, for example, the estimation result display unit 1053 in the screen interface 1051 shown in FIG. 19 . The cause estimation unit 26 outputs the cause data of the power distribution line accident associated with the waveform data of the identified most likely trained model as the cause estimation result displayed on the cause estimation result display screens 1054A and 1054B to the output device 105 of the server device 1, or to the estimation result display unit 1053 of an output device (not shown) of a client-side computer terminal using the power distribution line accident cause estimation device 20 via the network NW. The cause estimation unit 26 may also output the reliability (%) of the cause together with the cause estimation result to the cause estimation result display screens 1054A and 1054B. Furthermore, the cause estimation unit 26 can also output a plurality of cause estimation results and the reliability of the causes, as shown on a cause estimation result display screen 1054B.

[0083] [Evaluation of cause estimation results] The following describes the evaluation of the results of the cause estimation process performed by the power distribution line accident cause estimation device 20. The results of the cause estimation process were evaluated by comparing the accuracy of the cause estimation results for the differences between the feature extraction methods performed by the learning feature data generation unit 32 and the accident feature data generation unit 22, and the machine learning method of the trained model used in the cause estimation unit 26.

[0084] Fig. 21 is a table showing the relationship between the accident cause and the estimation result in the F value used to evaluate the result of the cause estimation process by the power distribution line accident cause estimation device 20. The cause estimation result was evaluated using an index called the F value for the relationship between the accident causes 1 and 2 obtained by the estimation result and the true accident causes 1 and 2 shown in the table of Fig. 21. The F value can be calculated by formula (3) where the recall is formula (1) and the precision is formula (2).

[0085]

number

[0086]

number

[0087]

number

[0088] Fig. 22 is a table showing evaluation results of the cause estimation processing by the power distribution line accident cause estimation device 20. As shown in Fig. 22, the evaluation of the results of the cause estimation processing targeted the phase plane (Hilbert transform), frequency analysis (Wigner distribution), and TDA (persistent homology) as feature extraction methods performed by the learning feature data generation unit 32 and the accident feature data generation unit 22. In addition, the evaluation of the results of the cause estimation processing targeted the classification decision tree, discriminant analysis, naive Bayes, nearest neighbor, and SVM as machine learning methods of the trained model used in the cause estimation unit 26.

[0089] According to FIG. 22, it can be seen that the highest F-value can be obtained when the power distribution line accident cause estimation device 20 uses the phase plane (Hilbert transform) as the feature extraction method and the nearest neighbor as the machine learning method for the trained model.

[0090] [Effects of the embodiment] The distribution line accident cause estimation device 20 described above generates data (accident feature data) in which feature values ​​are extracted by removing phase differences using, for example, a Hilbert transform based on waveforms (accident waveform data) of measurement information of zero-phase voltage values ​​or zero-phase current values ​​output during an acquired distribution line accident. The distribution line accident cause estimation device 20 also outputs an estimation result of the cause of the distribution line accident by inputting the generated accident feature data into a trained model generated by the distribution line accident cause estimation model generation device 30. Here, the trained model is a waveform of measurement information during a distribution line accident, and is generated from training waveform data associated with data on the cause of the distribution line accident when the waveform occurred. Specifically, the trained model is generated by machine learning data (training feature data) in which feature values ​​are extracted from the training waveform data by removing phase differences using, for example, a Hilbert transform, based on a predetermined algorithm (e.g., nearest neighbor).

[0091] The distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 described above can compare accident waveform data with a trained model even if the waveform data has a different phase. In other words, the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 can estimate the cause of an accident even if the accident waveform data has a different phase from the trained model but the same waveform. Therefore, the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 can simply estimate the cause of a distribution line accident.

[0092] Furthermore, the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 described above estimate the cause of the distribution line accident to be estimated for a plurality of types, such as the accident occurrence phenomenon F1, the component F2 where the accident occurred, and the facility F3 where the accident occurred. By outputting such estimation results, the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 can estimate the cause of the distribution line accident in more detail, thereby enabling the discovery of the accident point and the speedy removal of the accident cause based on the estimated cause of the accident.

[0093] Moreover, the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 described above can output a plurality of presumable causes of distribution line accidents. Furthermore, the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 described above can output the cause of a distribution line accident along with a numerical value of the reliability of this cause. In other words, the distribution line accident cause estimation device 20 and the distribution line accident cause estimation model generation device 30 can speed up the discovery of the accident point and the removal of the accident cause by referring to the reliability of the presumed accident cause. [Explanation of symbols]

[0094] 1...server device, 2...computer, 3...smartphone, 5...power distribution line, 6, 6A, 6B, 6C...sensor-integrated switch, 7...circuit breaker, 20...power distribution line accident cause estimation device, 21...fault waveform data acquisition unit, 22...fault feature data generation unit, 25...storage unit, 26...cause estimation unit, 30...power distribution line accident cause estimation model generation device, 31...learning waveform data acquisition unit, 32...learning feature data generation unit, 35...storage unit, 36...model generation unit, 41...learning waveform data, 42...cause data, 43...learning Trained feature data, 44...fault waveform data, 45...fault feature data, 51...electric wire, 52...electric pole, 53...transformer, 101...arithmetic unit, 102...storage device, 103...input device, 104...I / F (Interface) device, 105...output device, 106...bus, 303...trained model, 1021...program, 1022...data, 1051...screen interface, 1052...fault waveform data selection unit, 1053...estimation result display unit, 1054A, 1054B...cause estimation result display screen

Claims

1. a learning waveform data acquisition unit that acquires, as learning waveform data, a waveform of measurement information at the time of a distribution line accident, in association with data on the cause of the distribution line accident at the time the waveform of the measurement information occurred; a learning feature data generation unit that generates learning feature data by extracting features by removing the influence of phase differences from the learning waveform data; a trained model generation unit that performs machine learning on the training feature data based on a predetermined algorithm to generate a trained model for causing a computer to function to estimate the cause of an input power distribution line accident based on the waveform of measurement information acquired when the power distribution line accident occurs; Equipped with the training feature data is generated by drawing a phase plane of the waveform shape of the training waveform data and extracting feature amounts; A model generator for estimating the causes of power distribution line accidents.

2. A learning waveform data acquisition unit that associates a waveform of measurement information at the time of a distribution line accident with data on the cause of the distribution line accident when the waveform of the measurement information occurred and acquires it as learning waveform data; a learning feature data generation unit that generates learning feature data by extracting features by removing the influence of phase differences from the learning waveform data; a trained model generation unit that performs machine learning on the training feature data based on a predetermined algorithm to generate a trained model for causing a computer to function to estimate the cause of an input power distribution line accident based on the waveform of measurement information acquired when the power distribution line accident occurs; Equipped with The learning feature data is obtained by removing the influence of a phase difference from the waveform of the learning waveform data by a Hilbert transform. A model generator for estimating the causes of power distribution line accidents.

3. The data on the cause of the power distribution line accident associated with the learning feature data is information that can identify the event that caused the accident, the component in which the accident occurred, and the facility in which the accident occurred.

3. The power distribution line accident cause estimation model generation device according to claim 1 or 2.

4. a step of associating a waveform of measurement information at the time of a distribution line accident with data on the cause of the distribution line accident when the waveform of the measurement information occurred, and acquiring the waveform data for learning; generating training feature data by extracting features by removing the influence of the phase difference from the training waveform data; generating a trained model for causing a computer to function to estimate the cause of a power distribution line accident based on the waveform of measurement information acquired when the input power distribution line accident occurs, by performing machine learning on the training feature data based on a predetermined algorithm; on the computer, the training feature data is generated by drawing a phase plane of the waveform shape of the training waveform data and extracting feature amounts; A model generation program for estimating the causes of power distribution line accidents.

5. A step of correlating a waveform of measurement information at the time of a distribution line accident with data on the cause of the distribution line accident when the waveform of the measurement information occurred and acquiring it as learning waveform data; generating training feature data by extracting features by removing the influence of the phase difference from the training waveform data; generating a trained model for causing a computer to function to estimate the cause of a power distribution line accident based on the waveform of measurement information acquired when the input power distribution line accident occurs, by performing machine learning on the training feature data based on a predetermined algorithm; on the computer, The learning feature data is obtained by removing the influence of a phase difference from the waveform of the learning waveform data by a Hilbert transform. A model generation program for estimating the causes of power distribution line accidents.

6. a fault waveform data acquisition unit that acquires a waveform of measurement information as fault waveform data when a distribution line fault occurs; an accident feature data generation unit that generates accident feature data by extracting features by removing the influence of a phase difference from the accident waveform data; a cause estimation unit that inputs the fault feature data into a trained model generated by machine learning the training feature data based on a predetermined algorithm, and outputs an estimation result of the cause of the power distribution line fault; Equipped with The learning feature data is generated by extracting features by plotting a phase plane from the waveform of measurement information at the time of a power distribution line accident, based on the shape of the waveform of the learning waveform data associated with cause data of the power distribution line accident when the waveform of the measurement information at the time of the power distribution line accident occurs. Distribution line accident cause estimation device.

7. A fault waveform data acquisition unit that acquires a waveform of measurement information as fault waveform data when a distribution line fault occurs; an accident feature data generation unit that generates accident feature data by extracting features by removing the influence of a phase difference from the accident waveform data; a cause estimation unit that inputs the fault feature data into a trained model generated by machine learning the training feature data based on a predetermined algorithm, and outputs an estimation result of the cause of the power distribution line fault; Equipped with The accident feature amount data is obtained by removing an influence of a phase difference from the waveform of the accident waveform data by a Hilbert transform. Distribution line accident cause estimation device.

8. The causes of the power distribution line accident estimated by the cause estimation unit include the event that caused the accident, the component in which the accident occurred, and the equipment in which the accident occurred. The power distribution line accident cause estimation device according to claim 6 or 7.

9. the cause estimation unit outputs a plurality of estimated causes of the power distribution line accident; The power distribution line accident cause estimating device according to any one of claims 6 to 8.

10. the cause estimation unit outputs the estimated cause of the power distribution line accident together with the reliability of the cause. The power distribution line accident cause estimation device according to any one of claims 6 to 9.

11. acquiring a waveform of measurement information as fault waveform data when a power distribution line fault occurs; generating accident feature data by extracting feature amounts by removing the influence of a phase difference from the accident waveform data; inputting the fault feature data into a trained model generated by machine learning the training feature data based on a predetermined algorithm, and outputting an estimation result of the cause of the power distribution line fault; It is a program that causes a computer to execute The learning feature data is generated by extracting features by plotting a phase plane from the waveform of measurement information at the time of a power distribution line accident, based on the shape of the waveform of the learning waveform data associated with cause data of the power distribution line accident when the waveform of the measurement information at the time of the power distribution line accident occurs. A program for estimating the causes of power distribution line accidents.

12. A step of acquiring a waveform of measurement information as fault waveform data when a power distribution line fault occurs; generating accident feature data by extracting feature amounts by removing the influence of a phase difference from the accident waveform data; inputting the fault feature data into a trained model generated by machine learning the training feature data based on a predetermined algorithm, and outputting an estimation result of the cause of the power distribution line fault; It is a program that causes a computer to execute the learning feature data is generated by extracting features from learning waveform data that associates a waveform of measurement information at the time of a power distribution line accident with cause data of the power distribution line accident at the time the waveform of the measurement information at the time of the power distribution line accident occurred, The learning feature data is obtained by removing the influence of a phase difference from the waveform of the accident waveform data by a Hilbert transform. A program for estimating the causes of power distribution line accidents.

Images