Fault point locating device, fault point locating system, and fault point locating method

Abstract

To improve efficiency of maintenance and maintainability of a power system by locating a fault point with high accuracy during fault occurrence.SOLUTION: A data acquisition part 130 acquires current information and voltage information at a predetermined point during fault occurrence. A locating result inference part 140 outputs a locating result PFrst obtained by digitizing an existence probability at a fault point in a power system with the current information and the voltage information acquired at the data acquisition part 130, and an estimation fault cause CAest based on the current information and the voltage information as inputs.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to an accident point locating device, an accident point locating system, and an accident point locating method. [Background technology]

[0002] When an accident occurs in a power transmission line of a power system, it is important to find the location where the accident occurred (hereinafter referred to as the "fault point") early in order to quickly restore the accident. As a fault point locating device for this purpose, Japanese Patent Laid-Open Publication No. 2003-114249 (Patent Document 1) describes a technology for locating the fault section by a neural network method based on current information of a power transmission line fault detected by a sensor attached to the overhead ground wire of the overhead transmission line.

[0003] Furthermore, the following non-patent document 1 describes an impedance method, which is one of the methods for locating a power line fault in a power system, in which an impedance is calculated from the measured values ​​of current and voltage at the time of a fault to locate the fault point. In particular, as an example of the impedance method, a digital impedance method that can be incorporated into a digital relay for protecting a power line is described. In the digital impedance method, when the measured value of only one terminal is used, the fault point is located under the condition that the resistance of the fault point is assumed to be a pure resistance and the voltage and current at the fault point are in phase. Then, the fault point can be located by using the impedance calculated under the above assumption and system information including the topology and impedance of the power system (power transmission system and power distribution system). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2003-114249 A [Non-patent literature]

[0005] [Non-Patent Document 1] Shoichi Urano, "Technical Trends in Transmission Line Fault Location in Power Systems," IEEJ Transactions on Power and Energy, Vol. 137, No. 4, pp. 261-264 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the fault point locating device of Patent Document 1 requires the placement of a sensor on an overhead ground wire and a configuration for transmitting measurement information from the sensor, which raises concerns about the high costs of building the system (initial costs) and operating it (running costs).In addition, there are concerns that Patent Document 1 cannot locate the fault point in cases where the fault current does not flow through the overhead ground wire, such as in the case of a short-circuit fault, and that it cannot be applied to underground cables or distribution lines that do not have overhead ground wires.

[0007] On the other hand, in the impedance method of Non-Patent Document 1, while it is possible to obtain measurement information for locating the fault point using existing equipment installed in electrical stations such as power plants and substations, there is a possibility that there are multiple estimated fault points whose electrical distance from the electrical station matches the calculated impedance, and in such cases, there is a concern that it may not be possible to identify the fault point early.

[0008] The present disclosure has been made to solve such problems, and an object of the present disclosure is to improve the efficiency of maintenance and preservation of power systems by locating the point of an accident at low cost and with high accuracy when an accident occurs. [Means for solving the problem]

[0009] In one aspect of the present disclosure, there is provided an accident point locating device for a power system. The accident point locating device includes a data acquisition unit and an inference unit. The data acquisition unit acquires current information and voltage information at a predetermined point when an accident occurs. The inference unit receives the current information and voltage information acquired by the data acquisition unit and an estimated cause of the accident based on the current information and voltage information when an accident occurs, and outputs a location result that quantifies the probability of the existence of an accident point on the power system.

[0010] In another aspect of the present disclosure, a method for locating a fault point in a power system is provided, which acquires current information and voltage information at a predetermined point when a fault occurs, and outputs a location result in which a probability of the existence of a fault point in the power system is quantified using the acquired current information and voltage information and an estimated cause of the fault based on the current information and voltage information as inputs. Effect of the Invention

[0011] According to the present disclosure, by performing accident point location by adding the cause of the accident (estimated cause of the accident) as an input, the location of the accident point can be determined probabilistically from multiple points with the same electrical distance (impedance), thereby enabling the accident point to be located with high accuracy when an accident occurs, thereby making it possible to efficiently maintain and protect the power system at low cost. [Brief description of the drawings]

[0012] [Figure 1] FIG. 2 is a schematic diagram illustrating an accident point locating device according to a comparative example. [Diagram 2] 1 is a schematic diagram illustrating a configuration of an accident point locating system including an accident point locating device according to a first embodiment. [Diagram 3] 3 is a block diagram illustrating an example of a functional configuration of the fault point locating device shown in FIG. 2. FIG. [Figure 4] 3 is a block diagram illustrating an example of a functional configuration of the accident point locating server shown in FIG. 2. [Diagram 5] FIG. 2 is a block diagram illustrating an example of the configuration of a computer system for realizing the functions of an accident point locating device and an accident point locating server. [Figure 6] FIG. 2 is a block diagram illustrating input and output in the learning phase of the accident point inference model in the first embodiment. [Figure 7] FIG. 2 is a block diagram illustrating input and output of a statistical model. [Figure 8] FIG. 13 is a conceptual diagram illustrating an example of an output of a statistical model. [Figure 9] 5 is a flowchart for explaining the process of inputting and storing lineage information shown in FIGS. 3 and 4. [Figure 10] 7 is a flowchart explaining the process of generating and storing learning data and a learned model related to the accident point inference model shown in FIG. [Figure 11] 7 is a flowchart illustrating the process of generating and saving the statistical model shown in FIG. 6. [Figure 12] 4 is a flowchart for explaining the processing contents for locating a fault point when a system fault occurs in the fault point locating system according to the first embodiment. [Figure 13] FIG. 2 is a block diagram for explaining fault point location according to the first embodiment. [Figure 14] FIG. 13 is a conceptual diagram illustrating an example of an output of a fault point location result. [Figure 15] FIG. 4 is a block diagram illustrating a configuration example of an accident point locating device according to a modified example of the first embodiment. [Figure 16] FIG. 11 is a block diagram illustrating input and output in the learning phase of the accident point inference model in the second embodiment. [Figure 17] FIG. 11 is a block diagram illustrating fault point location according to a second embodiment. [Figure 18] 10 is a flowchart for explaining the processing contents for locating a fault point when a system fault occurs in the fault point locating system according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the following, the same or corresponding parts in the drawings are denoted by the same reference characters, and their description will not be repeated in principle.

[0014] Embodiment 1 (Explanation of Comparative Example) FIG. 1 is a schematic diagram illustrating a fault point locating device according to a comparative example.

[0015] 1, a fault point locator 100# according to a comparative example locates a fault point when a fault occurs in a power system 5. In the simplified example of Fig. 1, the power system 5 has a power station 10 including a bus LM and an AC power source 13, power transmission lines LA1 and LA2 connected to the bus LM to which power is supplied from the AC power source 13, a power transmission line LB1 branched off from the power transmission line LA1, and a power transmission line LB2 branched off from the power transmission line LA2. The power station 10 includes a power plant, a substation, and the like.

[0016] In the electric power station 10, a voltage transformer (VT) 15 for measuring voltage is provided on the bus LM, and current transformers (CT) 11, 12 for measuring current are provided on the transmission lines LA1 and LA2, respectively. Measurement values ​​of the voltage transformer 15 and the current transformers 11, 12 are transmitted to a fault point locator 100#.

[0017] The fault point locator 100# is disposed in, for example, an electric power station 10, and is communicatively connected to a central monitoring system 300 via a communication network 50. The electric power station 10 may be operated unmanned, and a fault point locator result PFrst calculated by the fault point locator 100# using current information and voltage information at the time of an accident is transmitted to the central monitoring system 300 via the communication network 50. The voltage information includes at least a part of the phase voltages of A phase, B phase, and C phase, the zero-phase voltage, the line voltages of AB phase, BC phase, and CA phase, and the like. The current information includes at least a part of the phase currents of A phase, B phase, and C phase, the zero-phase current, the line currents of AB phase, BC phase, and CA phase, and the like. These current information and voltage information can be measured and stored by the CT and VT of the electric power station 10, and then transmitted from the CT and VY.

[0018] The central monitoring system 300 corresponds to a manned command center or the like, and can provide the accident point information indicated by the accident point location result PFrst to a maintenance worker from an operator at the command center or the like, or automatically from the central monitoring system 300 or a system linked to the central monitoring system 300. By dispatching a maintenance worker, the actual accident point can be identified and the cause can be investigated, leading to recovery work.

[0019] The fault point locating device 100# of the comparative example is configured by storing an inference model 101M based on the digital impedance method described in Non-Patent Document 1. Therefore, the inference model 101M receives as input current information and voltage information at the time of the fault measured by current transformers (CT) 11, 12 and a potential transformer (VT) 15, and outputs a fault point location result located based on an impedance calculation value from the electric power station 10 to the fault point.

[0020] Specifically, the fault point can be located by indicating the electrical distance on the power transmission path from the substation 10 using the impedance calculation value and system information including the topology and impedance of the power system. The inference model 101M can be configured, for example, by a machine-learned neural network model, as in Patent Document 1.

[0021] However, a point where the impedance from the electric power station 10 is equal exists for each power transmission path. For example, in the example of Fig. 1, when an accident is detected in the system of the power transmission line LA2 from the current information and voltage information measured by the current transformers (CT) 11, 12 and the voltage transformer (VT) 15 in the electric power station 10, PF1 on the power transmission line LA2 and PF2 on the path of the power transmission line LA2-LB2 exist as points having an electrical distance equivalent to the impedance calculated value from the current information and voltage information. However, in the inference model 101M based on the impedance method, it is difficult to identify which of PF1 and PF2 is the accident point.

[0022] Another problem that reduces the accuracy of fault location is that the arc resistance at the fault location changes depending on the cause of the accident. In general, the arc resistance has different values ​​depending on the cause of the accident. For example, the arc resistance value is relatively high in the case of an accident caused by contact with a tree, and is low in the case of an accident caused by contact with metal objects.

[0023] First, since the impedance value calculated from the current information and the voltage information includes the arc resistance value, the existence of the arc resistance value itself may cause an error between the accident point located based on the impedance value and the actual accident point. Furthermore, when the learning model is used to absorb the error of the arc resistance value, there is a concern that the change in the arc resistance value depending on the above-mentioned accident cause will become a learning error and reduce the accuracy of the location result of the accident point.

[0024] In this way, while the accident point locating device 100# of the comparative example is capable of constructing a learning model without requiring a new sensor arrangement as in Patent Document 1, it is understood that there is room for improvement in the accuracy of locating the accident point.

[0025] (Description of the First Embodiment) The accident point locating device according to the first embodiment is configured as described below in order to improve the accuracy of locating the accident point compared to the comparative example.

[0026] FIG. 2 is a schematic diagram illustrating the configuration of an accident point locating system including the accident point locating device 100 according to the first embodiment.

[0027] 2, the accident point locating system includes an accident point locating device 100 and an accident point locating server 200, which are communicatively connected via a communication network 50. A central monitoring system 300 similar to that shown in FIG. 1 is also connected to the communication network 50. Furthermore, an accident cause estimation device 150 is disposed as an external device of the accident point locating system.

[0028] As in FIG. 1, current information and voltage information measured by the current transformers (CT) 11, 12 and the voltage transformer (VT) 15 are transmitted to the fault location device 100 and the accident cause estimation device 150. For example, the current information and voltage information can be transmitted by connecting the current transformers (CT) 11, 12 and the voltage transformer (VT) 15 to the fault location device 100 and the accident cause estimation device 150 with cables. In the following, the current information and voltage information transmitted from the current transformers (CT) 11, 12 and the voltage transformer (VT) 15 are also referred to as current and voltage (actual values). In this way, it is noted that the application of this embodiment does not require the placement of a special detector (sensor). This allows fault location to be performed without increasing costs as in Patent Document 1.

[0029] When an accident occurs, the accident cause estimation device 150 outputs an estimated accident cause CAest based on the current and voltage (actual values) from the data collection unit 20. The accident cause estimation device 150 can be realized by any known technology, and can be configured, for example, by an information processing device described in JP 2021-19480 A. In this disclosure, the method of generating the estimated accident cause CAest by the accident cause estimation device 150 is arbitrary, and the accident cause estimation device 150 may be obtained by further using information other than the current and voltage (actual values) input to the accident point locating device 100. The accident cause estimation device 150 is input to the accident point locating device 100 via a communication network 50 using the Internet Protocol (IP).

[0030] The accident point locator 100 according to the first embodiment and the accident cause estimator 150 can basically be arranged in a power station 10, similarly to the accident point locator 100# in Fig. 1. Therefore, as shown in Fig. 2, it is possible to input current and voltage (actual values) from the data collector 20 to the accident point locator 100 and the accident cause estimator 150 without passing through a communication network 50 using the Internet Protocol (IP).

[0031] Alternatively, the current and voltage (actual values) may be input from the data collector 20 to the accident point locating device 100 or the accident cause estimating device 150 via a communication network 50 using the Internet Protocol (IP). This improves the degree of freedom in arranging the accident point locating device 100 and the accident cause estimating device 150.

[0032] 2, the fault point locating device 100 receives a model created by the fault point locating server 200 via the communication network 50, and outputs a fault point locating result PFrst that infers the fault point using the model, the current and voltage (actual values) at the time of the fault occurrence from the data collecting unit 20, and the estimated fault cause CAest from the fault cause estimating device 150. In this embodiment, the fault point locating result PFrst is a numerical representation of the probability of the existence of a fault point on the power system 5.

[0033] The fault point location result PFrst from the fault point locator 100 is transmitted to the central monitoring system 300 via the communication network 50, as in Fig. 1. That is, the fault point location result PFrst is provided to a maintenance worker by an operator at a command center where the central monitoring system 300 is located, or automatically from the central monitoring system 300 or a system linked to the central monitoring system 300. By improving the accuracy of the fault point information based on the fault point location result PFrst, it is possible to identify the fault point, investigate the cause, and shorten the restoration time (of a power outage, etc.). This makes it possible to make the maintenance and preservation of the power system 5 more efficient.

[0034] The central monitoring system 300 corresponds to, for example, an EMS (Energy Management System) installed in a central load dispatching center, a core system load dispatching control center, a local load dispatching center, etc., or a SCADA (Supervisory Control And Data Acquisition) system, or a power distribution automation system, etc.

[0035] In this manner, in this embodiment, the accident point locator 100 and the accident point locator server 200 constituting the accident point locator system are connected via the above-mentioned communication network 50, and information can be transmitted and received between the accident point locator 100 and the accident point locator server 200. In addition, the accident point locator 100 and the accident point locator server 200 constituting the accident point locator system are also connected to the accident cause estimator 150 and the central monitoring system 300 via the communication network 50, so that the accident point locator 100, the accident cause estimator 150, the accident point locator server 200, and the central monitoring system 300 can transmit and receive data and information to and from each other via the communication network 50.

[0036] FIG. 3 is a block diagram illustrating an example of a functional configuration of the fault point locating device shown in FIG. As shown in FIG. 3, the fault point locating device 100 includes a communication unit 110, a storage unit 120, a data acquisition unit 130, and a location result inference unit 140.

[0037] The communication unit 110 has a function for communicating with external devices via the communication network 50 .

[0038] The data acquisition unit 130 acquires the current and voltage (actual values) from the current transformers (CTs) 11 and 12 and the voltage transformer (VT) 15 received by the communication unit 110. In this way, current information and voltage information at the current transformers (CTs) 11 and 12 and the voltage transformer (VT) 15 of the electric power station 10, which is an example of a "predetermined point", is acquired.

[0039] The storage unit 120 can write and read data between the communication unit 110, the data acquisition unit 130, and the location result inference unit 140. In addition, the storage unit 120 can transmit and receive data to and from external devices connected to the communication network 50, including the accident point location server 200, via the communication unit 110.

[0040] The storage unit 120 has storage areas for model information 121 , system information 122 , measurement data 124 , accident cause estimation results 126 , and calculation data 128 .

[0041] The model information 121 includes information for constructing a trained model and a statistical model, which will be described later. The system information 122 includes information related to the topology and impedance of the power system 5. The model information 121 and the system information 122 are transmitted from the fault point locating server 200 via the communication network 50, received by the communication unit 110, and then stored in the storage unit 120.

[0042] The measurement data 124 includes the current and voltage (actual values) acquired by the data acquisition unit 130. The current and voltage (actual values) are transmitted by the communication unit 110 to the accident point locating server 200 via the communication network 50, and are also stored on the accident point locating server 200 side.

[0043] The accident cause estimation result 126 includes an estimated accident cause CAest outputted at the time of the occurrence of an accident from the accident cause estimation device 150. The calculation data 128 includes an accident point location result PFrst obtained by the location result inference unit 140.

[0044] When an accident occurs in the power system 5, the location result inference unit 140 acquires from the memory unit 120 the current and voltage (actual values) (measurement data 124), which are the current and voltage information at the time of the accident, the estimated accident cause CAest (accident cause estimation result 126), and the learned model and arc resistance value statistical model (model information 121), as well as the topology and impedance information (system information 122) of the power system 5.

[0045] Then, as will be described in detail later, the location result inference unit 140 inputs the current and voltage (actual values) at the time of the accident occurrence and the estimated accident cause CAest from the accident cause estimation device 150 into the trained model to calculate the accident point location result. The location result inference unit 140 corresponds to one embodiment of the "inference unit".

[0046] Furthermore, as will be described later in detail, the location result inference unit 140 inputs the estimated accident cause CAest from the accident cause estimation device 150 into a statistical model of the arc resistance value, acquires a probabilistic distribution of the arc resistance value, and integrates the calculation result by the learned model with the probabilistic distribution to output the accident point location result PFrst. The accident point location result PFrst output from the location result inference unit 140 is stored in the storage unit 120 as the calculation data 128, and is transmitted by the communication unit 110 to the accident point location server 200 and the central monitoring system 300 via the communication network 50.

[0047] Therefore, the fault point locator 100 is typically installed in an electric power station 10 (such as a power station or a substation) and has the following functions.

[0048] (1) A process is executed to receive the learned model and the statistical model of the arc resistance value from the fault point locating server 200 and store them as model information 121.

[0049] (2) When an accident occurs in the power system 5, current information and voltage information (current and voltage (actual values)) at the time of the accident are acquired from the potential transformers (VT) and current transformers (CT) of the power station 10, stored, and further transmitted to the accident point location server 200.

[0050] (3) When an accident occurs in the power system 5, the estimation result of the cause of the accident (estimated accident cause CAest) is received (acquired) from a separately installed accident cause estimation device 150, and a process of storing it is executed.

[0051] (4) When an accident occurs in the power system 5, the above current information and voltage information (current and voltage (actual values)) and the estimated result of the cause of the accident (estimated accident cause CAest) are used as input data for the learned model and the arc resistance value statistical model to locate the accident point, output the result (fault point location result PFrst), and store it.

[0052] (5) A process is executed to transmit the fault point location result (fault point location result PFrst) to the central monitoring system 300 where the operator is located.

[0053] FIG. 4 is a block diagram illustrating an example of a functional configuration of the accident point locating server shown in FIG. As shown in FIG. 4 , the accident point locating server 200 includes a communication unit 210, a memory unit 220, an input receiving unit 230, a learning data generating unit 240, an arc resistance calculating unit 250, a statistical model generating unit 260, and a learning model generating unit 270.

[0054] The communication unit 210 has a function for communicating with an external device via the communication network 50. When an accident occurs, the communication unit 210 receives the current and voltage (actual values) from the accident point locating device 100 via the communication network 50 at any timing, and also receives the estimated accident cause CAest output at the time of the accident from the accident cause estimating device 150. In this way, when a grid accident occurs, the accident point locating server 200 can acquire the current and voltage (actual values) and the estimated accident cause CAest.

[0055] The storage unit 220 can write and read data between the communication unit 210, the input reception unit 230, the learning data generation unit 240, the arc resistance calculation unit 250, the statistical model generation unit 260, and the learning model generation unit 270. In addition, by passing through the communication unit 210, the storage unit 220 can also transmit and receive data to and from external devices connected to the communication network 50, including the fault point locating device 100.

[0056] The storage unit 220 has storage areas for system information 222 , measurement data 224 , accident-related information 226 , and calculation data 228 .

[0057] The system information 222 includes information related to the topology and impedance of the power system 5, and information indicating the surrounding conditions of each of a plurality of points in the power system 5.

[0058] The measurement data 224 includes various types of current and voltage information. The current and voltage (actual values) received by the communication unit 210 in response to the occurrence of an accident can be stored in the storage unit 220 as the measurement data 224.

[0059] The accident-related information 226 includes the accident point and accident cause (actual results), which are actual values ​​at the time of the accident occurrence, and the arc resistance value (calculated value) calculated by the arc resistance calculation unit 250. Furthermore, the estimated accident cause CAest received by the communication unit 210 and output from the accident cause estimation device 150 at the time of the accident occurrence can also be stored in the storage unit 220 as the accident-related information 226.

[0060] The calculation data 228 includes learning data (teacher data) used for machine learning of the model, and data indicating the learned model and the statistical model.

[0061] When the input receiving unit 230 receives the above-mentioned input of information related to the topology and impedance of the power system 5 or information related to the surrounding conditions, the input receiving unit 230 stores the information in the storage unit 220 as system information 222. Similarly, when the input receiving unit 230 receives input of the accident point and the actual value of the accident cause when a system accident occurs (more precisely, after an accident patrol carried out after the system accident), the input receiving unit 230 stores the input in the storage unit 220 as accident-related information 226.

[0062] The learning data generating unit 240 can configure an instantaneous value analysis model of the power system 5 by acquiring information related to the topology and impedance of the power system 5 from the system information 222 stored in the storage unit 220. The instantaneous value analysis model can be configured using known simulation tools such as, for example, EMTP (Electro-Magnetic Transient Program), PSCAD (registered trademark), XTAP (registered trademark), etc., but is not limited to these examples.

[0063] In addition, on the instantaneous value analysis model, the current and voltage at the installation point (electric station 10 such as a power plant or a substation) of the VT (15) and CT (11, 12) when an accident occurs at any various points (accident points) can be simulated and calculated to generate calculated values ​​of current data and voltage data (hereinafter, also referred to as current and voltage (calculated values)). As a result, the learning data generation unit 240 can acquire a set of the accident point and the corresponding current data and voltage data (calculated values) for each assumed accident condition as learning data to be used for machine learning. The calculated learning data can be stored in the storage unit 220 as calculated data 228.

[0064] The arc resistance calculation unit 250 acquires the accident point and accident cause (actual results) and the current and voltage (actual values) at the time of the accident from the accident-related information 226 stored in the storage unit 220, and further acquires information related to the topology and impedance of the power system 5 from the system information 222 stored in the storage unit 220, thereby calculating an arc resistance value (calculated value) for each accident. The calculated arc resistance value (calculated value) is stored in the storage unit 220 as the accident-related information 226.

[0065] The statistical model generation unit 260 acquires accident causes (actual) (or accident causes (estimated)) and arc resistance values ​​(calculated values) for multiple past accidents from the accident-related information 226 in the storage unit 220. Furthermore, a statistical model of the arc resistance is generated using the acquired data. The generated data representing the statistical model is stored in the storage unit 220 as calculated data 228.

[0066] The learning model generation unit 270 acquires information indicating the topology of the power system 5, information related to impedance, and surrounding conditions of each point of the power system 5 from the system information 222 stored in the storage unit 220. Furthermore, the learning model generation unit 270 acquires a set of the current and voltage (calculated values) obtained by the above-mentioned simulation calculation, which is learning data, from the calculation data 228 stored in the storage unit 220, and the corresponding accident point and accident cause (actual or estimated) from the accident-related information 226 stored in the storage unit 220, and executes machine learning to generate an accident point inference model. The accident point inference model will be described in detail later. Information for indicating the accident point inference model obtained as a learned model as a result of the machine learning is stored in the storage unit 220 as the calculation data 228.

[0067] In addition, the accident point locating server 200 can be installed at any location such as a server base or a business office, and may be constructed by a cloud system, etc., as long as it can communicate with the accident point locating device 100 via the communication network 50. According to the configuration of FIG. 4, the accident point locating server 200 has the following functions.

[0068] (6) Information such as the topology, impedance, and surrounding conditions of each point of the power system 5 is acquired by accepting input (or by referring to a database external to the fault location system via the communication network 50) and stored.

[0069] (7) When a system accident occurs, current information and voltage information (current and voltage (actual values)) measured by a CT (current transformer) or VT (voltage transformer) of the power station 10 are received (acquired) from the fault point locating device 100 as the current and voltage (actual values) at the time of the accident and stored.

[0070] (8) The accident point (actual) and accident cause (actual) confirmed during inspection after the system accident are acquired and stored by accepting input (or by referring to a system accident-related database located outside the accident point location system via the communication network 50).

[0071] (9) Using system information (topology, impedance), the above-mentioned instantaneous value analysis model of the target power system 5 is generated, and the calculated values ​​of current and voltage at the CTs and VTs of the substation 10 under each accident condition obtained by instantaneous value analysis simulating various accident conditions (various accident points and accident aspects) are obtained and stored. The stored calculated values ​​of current and voltage become learning data to be used in the machine learning described later.

[0072] (10) Calculate the arc resistance value by performing electrical calculations from the current and voltage (actual values) measured during the system accident, the accident point (actual values), system information (topology, impedance), etc., and save the results. Note that any calculation method may be used to calculate the arc resistance value, but it is considered most efficient to use the simulation tool for instantaneous value analysis described above.

[0073] (11) A statistical model is generated and saved from the cause of the accident (basically, actual results are used, but estimated results can also be used) and the arc resistance value (calculated value). As the statistical model, for example, a histogram, or a probability density function, probability distribution function, cumulative distribution function, etc. obtained from the histogram, etc. using a parametric method or a non-parametric method can be generated and saved. Note that, as a parametric method, the least squares method, etc. can be used, but the method is not limited thereto. Similarly, as a non-parametric method, a kernel density estimation method, etc. can be used, but the method is not limited thereto.

[0074] (12) A machine learning model (trained model) is generated and saved by machine learning using the learning data generated in (9) above. Specifically, the current and voltage (calculated values) and the possible (possible) causes of the accident based on the surrounding conditions at each point of the power system 5 are set as "features (input data for the machine learning model at the time of inference)," and the accident point (the accident point set (simulated) as the accident condition at the time of instantaneous value analysis, or the section including the accident point) is set as the "answer (i.e., output data for the machine learning model)," and machine learning is executed to generate and save an accident point inference model, which is a trained model.

[0075] The machine learning method is not specified, but for example, a decision tree, a random forest, an ANN (neural network), an SVM (support vector machine), etc. can be used. Machine learning can basically be used as a "regression method", in which case the accident point can be indicated by a combination of the transmission route and distance from the substation 10 on the power system 5. Machine learning can also be used as a "classification method". In this case, the transmission lines and distribution lines in the power system 5 are divided into a plurality of sections, and which of the plurality of sections the accident point is included in is obtained by the learning model, and for example, the probability of the existence of the accident point in each section can be obtained from the learning model.

[0076] (13) A process is performed to transmit the above-mentioned “learned model,” “statistical model,” “system topology,” and “impedance” to the corresponding fault point locator 100.

[0077] FIG. 5 is a diagram showing an example of the configuration of a computer system for realizing the accident point locating device 100 and the accident point locating server 200. As shown in FIG.

[0078] As shown in FIG. 5, computer system 40 can have a typical configuration having a display unit 41, an input unit 42, a network interface (I / F) 43, a memory 44, a CPU (Central Processing Unit) 45, a HDD (Hard Disk Drive) 46, and a bus 47.

[0079] The fault point locator 100 shown in FIG. 3 can be realized by using the computer system 40 as follows.

[0080] Specifically, the function of the data acquisition unit 130 in Fig. 3 is realized by an analog / digital converter (A / D converter) (not shown) in the input unit 42 in Fig. 5. Also, the function of the communication unit 110 in Fig. 3 can be realized by the CPU 45 and network I / F 43 in Fig. 5. Also, the function of the storage unit 120 in Fig. 3 can be realized by using a partial area of ​​the HDD 46 in Fig. 5. Also, the function of the orientation result inference unit 140 in Fig. 3 is realized by the CPU 45 in Fig. 5 executing a program stored in the HDD 46 or memory 44 in Fig. 5.

[0081] Similarly, the accident point locating server 200 shown in Fig. 4 can be realized as follows using the computer system 40. Specifically, the function of the communication unit 210 in Fig. 4 is realized by the CPU 45 and network I / F 43 in Fig. 5. Also, the function of the storage unit 220 can be realized by using a partial area of ​​the HDD 46 in Fig. 5. Furthermore, the function of the input receiving unit 230 in Fig. 4 can be realized by using the input unit 42 and display unit 41 in Fig. 5. By using the display unit 41, it is possible to display the input results to the user.

[0082] In addition, the functions of the learning data generation unit 240, the arc resistance calculation unit 250, the statistical model generation unit 260, and the learning model generation unit 270 in FIG. 4 are realized by the CPU 45 in FIG. 5 executing a program stored in the HDD 46 or the memory 44.

[0083] Next, the accident point inference model and the statistical model will be explained in detail with reference to FIGS.

[0084] FIG. 6 shows the input / output relationship in the learning phase of the accident point inference model. The fault point inference model 510 is configured to receive as input current information INFI and voltage information INFV at the time of an accident in a CT (current transformer) or VT (voltage transformer) of the electric power station 10, and the cause of the accident, and output an estimation result of the fault point. Here, the estimation result of the fault point is represented by a combination of the transmission path and distance from the electric power station 10 on the power system 5, using machine learning based on a regression method. The fault point inference model 510 corresponds to one embodiment of the "first inference model".

[0085] The accident cause can be input to the accident point inference model 510 as a multi-bit accident cause code in which one bit is assigned to each of a plurality of causes (lightning strike, tree contact, bird / animal contact, metal contact, etc.) and each bit is set to "1: possible" or "0: not possible." For example, if the first bit of the accident cause code is defined to correspond to the possibility of "lightning strike," the second bit to the possibility of "tree contact," and the third bit to the possibility of "bird / animal contact," and the codes are listed in order from the right, the accident cause code "011" is input when there is a possibility of lightning strike and tree contact, but no possibility of bird / animal contact.

[0086] In the learning phase of the fault point inference model 510, the simulation results by the instantaneous value analysis model 500 reflecting the system information (topology and impedance of the power system 5) are used as learning data (teaching data) for machine learning. Specifically, in order to generate learning data, various accident conditions (accident points and accident aspects) are set, and the current and voltage (calculated values) at the CT and VT of the electric power station 10 when an accident occurs at an accident point under each accident condition are simulated on the instantaneous value analysis model 500. From the current and voltage (calculated values), current information INFI(S) and voltage information INFV(S) are obtained as simulation values. In this simulation, the arc resistance is not taken into consideration, and the current and voltage (calculated values) at the CT and VT of the electric power station 10 when the arc resistance value = 0 under each accident condition (accident point) are obtained. The cause of the accident under each accident condition can be set as a possible (possible) one based on the surrounding circumstances at the corresponding accident point.

[0087] The accident modes include one-phase earth fault (A phase earth fault, B phase earth fault, C phase earth fault), phase-to-phase short circuit (AB phase-to-phase short circuit, BC phase-to-phase short circuit, CA phase short circuit), two-phase earth fault (AB phase earth fault, BC phase earth fault, CA phase earth fault), three-phase short circuit, three-phase earth fault, etc. Since the above-mentioned current information and voltage rise behavior differ for each accident mode, the simulation values ​​by the instantaneous value analysis model 500, i.e., the current information INFI(S) and the voltage information INFV(S), differ. In addition, it is possible to identify the accident mode of the system accident that has occurred from the measurement values ​​(i.e., current and voltage (actual values)) by the CT and VT of the substation 10 after the system accident has occurred.

[0088] As shown by the dotted lines in FIG. 6, the current information INFI(S) and voltage information INFV(S) corresponding to the current and voltage (calculated values) obtained by the simulation are used as teacher data for the current information INFI and voltage information INFV, which are the feature values ​​(input) of the accident point inference model 510. Furthermore, teacher data for the accident cause (accident cause code), which is the feature value (input) of the accident point inference model 510, is created in response to the surrounding circumstances of the accident point of each accident condition. Furthermore, the accident point of each accident condition is used as teacher data for the accident point (calculated value), which is the answer (output) of the feature value (input) of the accident point inference model 510. Machine learning of the accident point inference model 510 is performed using a set of these teacher data created for each accident condition.

[0089] As described above, the behavior of the current information and voltage rise differs for each accident aspect, so the accident point inference model 510 is created for each accident aspect. On the other hand, since the accident aspect that occurred can be identified from the current and voltage (actual values), the accident point inference model 510 created for each accident aspect can be uniquely selected at the time of inference. For this reason, the accident aspect does not need to be included in the feature quantity (input) of the accident point inference model 510.

[0090] When the accident point locating server 200 acquires an accident point inference model 510 (for each accident aspect) as a trained model resulting from the machine learning, data for expressing the trained model is stored in the storage unit 220 as calculated data 228. Furthermore, the data for expressing the trained model (calculated data 228) is transmitted to the accident point locating device 100 via the communication network 50 and stored in the storage unit 120 as model information 121.

[0091] In the fault point locating device 100, the location result inference unit 140 can output the fault point location result by inference using the learned model, i.e., the fault point inference model 510 (learned model) of the occurred fault state when a grid fault occurs. During fault point location (inference phase), the current information INFI and voltage information INFV of the input of the fault point inference model 510 (learned) are set to the current and voltage (actual values) acquired when the grid fault occurs, and the fault cause code is set to a binary code set according to the estimated fault cause CAest by the fault cause estimation device 150. As a result, the fault point inference model 510 (learned) outputs the fault point (calculated value) indicated by the power transmission route and distance from the power station 10.

[0092] By introducing the accident cause code, it becomes possible to make an inference that takes into account the surrounding conditions for each point of the power system 5, that is, each candidate point of the accident point. For example, as in the comparative example of FIG. 1, inference based only on impedance calculates both PF1 in the forest and PF2 on the plain as candidate points, and it is difficult to identify which of the two is the accident point. However, by performing machine learning using the accident cause corresponding to the surrounding conditions for each point on the power system 5, it becomes possible to obtain an accident point location result that takes into account the surrounding conditions. This solves the problem that it is difficult to identify the accident point from multiple candidate points that are equivalent in terms of impedance (electrical distance), as pointed out in the comparative example of FIG. 1, and it is possible to generate an accident point location result. At this time, the accident point location result can be expressed as the candidates of the accident point and the existence probability of the accident point at each candidate.

[0093] Next, the statistical model of the arc resistance value will be explained. Figure 7 shows a block diagram explaining the input and output of the statistical model, and Figure 8 shows an example of the output of the statistical model. As mentioned above, the probability distribution of the arc resistance value mainly depends on the cause of the accident, and the difference due to the difference in the accident point is relatively small.

[0094] Therefore, as shown in FIG. 7, the statistical model 520 of the arc resistance value is configured to input the cause of the accident and output a histogram, a probability density function, a probability distribution function, a cumulative distribution function, or the like, for indicating the probabilistic distribution of the arc resistance value. When the statistical model 520 is generated (learning phase), an arc resistance value (calculated value) is calculated for each accident using the current and voltage (actual values) acquired when a past system accident occurred and information related to the topology and impedance of the power system 5. Specifically, the arc resistance value (calculated value) is calculated by the arc resistance calculation unit 250 of the accident point locating server 200 shown in FIG. 4. Then, the arc resistance value (calculated value) is collected for each cause of the accident and statistical processing is performed to generate the statistical model 520 of the arc resistance value. Specifically, the statistical model 520 is generated by the statistical model generation unit 260 of the accident point locating server 200 shown in FIG. 4.

[0095] For example, as shown in Fig. 8, the statistical model 520 of the arc resistance value is configured so as to output a probabilistic distribution of the arc resistance value Rac for each of n (n: natural number) accident causes. In the example of Fig. 8, for cause 1 (e.g., metal contact), the arc resistance value Rac is distributed in a relatively low region, and for cause n (e.g., tree contact), the arc resistance value Rac is distributed in a relatively high region.

[0096] For example, when the statistical model 520 of the arc resistance value is generated in the accident point locating server 200, data for expressing the statistical model is stored in the storage unit 220 as the calculated data 228. Furthermore, the data for expressing the statistical model (the calculated data 228) is transmitted to the accident point locating device 100 via the communication network 50, and is stored in the storage unit 120 as the model information 121.

[0097] In the fault point locator 100, when a system fault occurs, the fault cause code set according to the fault cause CAest estimated by the fault cause estimator 150 is input to the statistical model 520. This makes it possible to obtain a probabilistic distribution of arc resistance values ​​corresponding to the fault cause estimated by the fault cause estimator 150 as an output of the statistical model 520. Specifically, the probabilistic distribution is acquired by the location result inference unit 140 of the fault point locator 100 shown in FIG.

[0098] As described above, the accident point (calculated value) output from the accident point inference model 510 is obtained from a trained model using the simulation results when the arc resistance value is set to 0, while the current information and voltage information (current and voltage (actual values)) at the time of the accident include the influence of the arc resistance value. Therefore, the accident point location result from the accident point inference model 510 may include an error caused by the arc resistance value.

[0099] Therefore, in the first embodiment, the fault point location result by the fault point inference model 510 (learned model) is combined with the arc resistance value (probabilistic distribution) estimated by the statistical model 520, thereby improving the accuracy of the fault point location result.

[0100] Next, a process flow by the fault point locating device 100 and the fault point locating server 200 will be described with reference to Figs.

[0101] FIG. 9 is a flowchart illustrating the process of inputting and storing lineage information. Referring to FIG. 9, in step (hereinafter simply referred to as "S") 110, the fault point locating server 200 receives input of information related to the topology and impedance of the power system 5, and information related to the surrounding conditions, via the input receiving unit 230 (FIG. 4), and stores the information in the memory unit 220 (FIG. 4) as system information 222.

[0102] Furthermore, in S120, the fault point locating server 200 transmits, via the communication unit 210, information relating to the system topology and impedance from the system information 222 stored in the memory unit 220 in S110 to the fault point locating device 100 via the communication network 50.

[0103] In response to this, in S130, the fault point locating device 100 receives the information related to the system topology and impedance transmitted by the fault point locating server 200 in S120, and stores it in the storage unit 120 (FIG. 3) as system information 122. Through the process of FIG. 9, information related to the system topology and impedance related to the power system 5 is stored in both the fault point locating server 200 and the fault point locating device 100. Furthermore, information related to the surrounding conditions of the power system 5 is stored in the fault point locating server 200. The process of S130 is executed by the communication unit 110 (FIG. 3).

[0104] FIG. 10 is a flowchart explaining the process of generating and storing learning data and a learned model related to the accident point inference model 510.

[0105] 10, in S210, the fault point locating server 200 uses the information related to the topology and impedance of the power system 5 among the system information 222 stored in S110 (FIG. 9) by the learning data generating unit 240 to generate an instantaneous value analysis model 500 (FIG. 6) of the power system 5. Then, in S220, the fault point locating server 200 sets various accident conditions (fault point, accident aspect, accident cause) to be input to the instantaneous value analysis model 500 generated in S210 by the learning data generating unit 240. The accident cause under each accident condition reflects the surrounding circumstances of the accident point. As described above, it is possible to set the accident cause by the accident cause code expressed in binary, but the accident cause may be set by a method different from this example.

[0106] As for the accident state, at least a part of the states included in the above-mentioned one-phase ground fault, phase-to-phase short circuit, two-phase ground fault, three-phase short circuit, three-phase ground fault, etc. is set sequentially for each accident point. Note that these accident conditions may be automatically set (generated) by the learning data generating unit 240, or may be input from outside the accident point locating server 200 by the input receiving unit 230.

[0107] In S230, the accident point location server 200 executes a simulation calculation to calculate current data and voltage data at VT (15) and CT (11, 12) for each accident point under various accident conditions set in S220 on the instantaneous value analysis model 500 generated in S210 by the learning data generation unit 240. The calculated values ​​of current and voltage (simulation results) obtained under each accident condition are combined with the accident conditions (accident point, accident aspect, accident cause) and saved as learning data.

[0108] In S240, the accident point location server 200 executes the machine learning described in FIG. 6 by the learning model generation unit 270 using the learning data obtained in S230. That is, for the accident point inference model 510 for each accident aspect, the current and voltage (calculated values) and the accident cause, which are the simulation results under the accident conditions of the item aspect, are set as "feature values ​​(i.e., input of the accident point inference model 510)", and the accident point under the accident condition (the accident point set (simulated) as the accident condition during the instantaneous value analysis, or the section including the accident point) is set as "answer (i.e., output of the accident point inference model 510)", and a supervised type of machine learning is executed. In S240, a learned model of the accident point inference model 510 for each accident aspect is generated by the machine learning. Data for expressing the learned model is stored in the storage unit 220 as calculated data 228.

[0109] In S250, the accident point locating server 200 transmits data for expressing the trained model stored in the memory unit 220 in S240 to the accident point locating device 100 via the communication network 50 by the communication unit 210.

[0110] In response to this, in S260, the accident point locating device 100 receives the data for expressing the trained model transmitted in S250 by the accident point locating server 200 via the communication unit 110. The received data is stored as model information 121 in the storage unit 120 (FIG. 3).

[0111] 10, the processes of S210 to S230 of the accident point locating server 200 are executed by the learning data generating unit 240 of Fig. 4, the process of S240 is executed by the learning model generating unit 270 of Fig. 4, and the process of S250 is executed by the communication unit 210. Moreover, S260, which is the process of the accident point locating device 100, is executed by the communication unit 110 of Fig. 3.

[0112] FIG. 11 is a flowchart illustrating the process of generating and saving the statistical model 520.

[0113] 11, in S270, the fault point locating server 200 uses the current and voltage (actual values) and the cause of the fault (actual results) recorded in the fault-related information 226 of the storage unit 220 for each of a plurality of past system faults, and information related to the topology and impedance of the power system 5 recorded in the system information 222 of the storage unit 220, to calculate an arc resistance value (calculated value) in each fault by the arc resistance calculation unit 250. The calculated arc resistance value is stored in the storage unit 220 as the fault-related information 226.

[0114] In S280, the accident point locating server 200 generates the statistical model 520 described in Fig. 7 and Fig. 8 from the accident cause and the arc resistance value (calculated value) obtained in S270 by the statistical model generating unit 260. Data for indicating the generated statistical model 520 is stored in the storage unit 220 as the calculated data 228. Note that, for the cause of the accident, basically, the past record is used, but an estimated result may be used.

[0115] As described above, data for indicating a statistical model may be, for example, a histogram, or a probability density function, probability distribution function, cumulative distribution function, etc., obtained from the histogram etc. using a parametric method (for example, the least squares method can be applied, but the method is not limited thereto) or a non-parametric method (for example, a kernel density estimation method can be applied, but the method is not limited thereto).

[0116] In S290, the accident point locating server 200 transmits the data for expressing the statistical model, which was stored in the storage unit 220 in S280, to the accident point locating device 100 via the communication network 50 by the communication unit 210.

[0117] In response to this, in S295, the accident point locating device 100 receives the data for expressing the statistical model transmitted in S290 by the accident point locating server 200 through the communication unit 110. The received data is stored in the storage unit 120 as model information 121.

[0118] 11, the process of S270 of the fault point locating server 200 is executed by the arc resistance calculation unit 250 in Fig. 4, the process of S280 is executed by the statistical model generation unit 260 in Fig. 4, and the process of S290 is executed by the communication unit 210. Also, the process of S295 of the fault point locating device 100 is executed by the communication unit 110 in Fig. 3.

[0119] FIG. 12 is a flowchart for explaining the processing contents for locating the fault point when a system fault occurs in the fault point locating system according to the first embodiment. The processing shown in FIG. 12 is started when the fault point locating device 100 detects the occurrence of an accident in the power system 5. The fault point locating device 100 may detect the occurrence of an accident by acquiring and monitoring the current and voltage (actual values) measured by the CT (11, 12) and VT (15) of the electric power station 10, or may detect the occurrence of an accident by receiving a trip signal from a system protection relay (not shown) or the like separately installed in the electric power station 10. In addition, an operator at a command center or the like where the central monitoring system 300 is present may manually detect the occurrence of an accident via the communication network 50, that is, it is also possible to manually start the fault point locating device 100 from the outside so that the operation of FIG. 12 is performed.

[0120] 12, in S310, the fault point locating device 100 acquires, by the data acquiring unit 130, the current and voltage (actual values) at the time of the occurrence of the accident measured by the CT (11, 12) and the VT (15) of the substation 10. The acquired current and voltage (actual values) at the time of the occurrence of the accident are stored in the storage unit 120 as measurement data 124.

[0121] In S320, the accident point locating device 100 transmits the current and voltage (actual values) at the time of the accident occurrence, stored in S310, to the accident point locating server 200 by the communication unit 110 via the communication network 50. Note that S320 may be executed any time after the accident occurs, and does not necessarily have to be executed at the timing shown in the flowchart of Fig. 12. For example, batch processing may be performed during a time period (e.g., nighttime) when the processing capacity of the accident point locating device 100 and the accident point locating server 200 and the communication condition of the communication network 50 are sufficient.

[0122] In response to this, in S390, the accident point locating server 200 stores the current and voltage (actual values) at the time of the accident transmitted by the accident point locating device 100 in S320 as measurement data 224 in the storage unit 220 via the communication unit 210. Note that S390 does not necessarily have to be executed at the timing shown in the flowchart of Fig. 12, and may be executed after the accident point locating device 100 executes S320.

[0123] Furthermore, in S395, the accident point locating server 200 accepts input of the accident point (record) and accident cause (record) confirmed by the patrol after the system accident through the input accepting unit 230, and stores them in the storage unit 220 as accident related information 226. As described above, the accident point (record) and accident cause (record) may be acquired by referring to a system accident related database (not shown). Also, S395 does not necessarily have to be executed at the timing shown in the flowchart of FIG. 12, and may be executed at any time after the accident point locating server 200 executes S390. In this way, the processing timings of S320, S390, and S395 are not limited to those of the flowchart of FIG. 12.

[0124] Meanwhile, in S330, the accident point locating device 100 receives the accident cause estimation result (estimated accident cause CAest) from the accident cause estimating device 150 (FIG. 2) via the communication unit 110, and stores it in the storage unit 120 as the accident cause estimation result 126. As described above, the accident cause estimating device 150 is provided separately from the accident point locating device 100 according to this embodiment, and when an accident occurs, it operates in parallel with the accident point locating device 100 in response to input of current information and voltage information (current and voltage (actual values)) to output the estimated accident cause CAest.

[0125] In S340, the location result inference unit 140 of the accident point locating device 100 inputs the current and voltage (actual values) at the time of the accident occurrence acquired in S310 and the inference result of the accident cause (estimated accident cause CAest) acquired in S330 to the accident point inference model (trained model) 510 stored as the model information 121. As described above, at this time, the accident point inference model 510 corresponding to the accident aspect identified from the current and voltage (actual values) is selected. In addition, the inference result of the accident cause can be input to the accident point inference model 510 as a binary code set according to the estimated accident cause CAest.

[0126] As a result, the accident point (calculated value) is acquired as the output of the accident point inference model (learned model) 510. The acquired accident point (calculated value) is stored in the memory unit 120 as calculation data 128. As described above, the accident point (calculated value), which is the output value of the accident point inference model 510, can be expressed as accident point candidates indicating specific points or sections on the power system 5, and the existence probability of the accident point at each candidate. It is also possible to determine one accident point (calculated value) assuming the existence probability to be 100(%).

[0127] In S350, the fault point locating device 100 inputs the estimation result of the cause of the accident (estimated fault cause CAest) acquired in S330 to the statistical model 520 stored as the model information 121 by the location result inference unit 140. As a result, a probabilistic distribution of the arc resistance value corresponding to the cause of the accident (estimated result) is acquired as an output of the statistical model 520.

[0128] In S360, the fault point locating device 100 converts the probabilistic distribution of the arc resistance value acquired in S350 into a distance by the location result inference unit 140, taking into consideration the impedance of the actual power transmission line, etc., and calculates the probabilistic distribution of the distance. Specifically, the probabilistic distribution of the arc resistance value acquired in S350 is converted into a probabilistic distribution of distance, taking into consideration the topology and impedance of the system (power transmission line, power distribution line, etc.) existing before and after the fault point (calculated value) as the starting point. At this time, the impedance of the power transmission line, power distribution line, etc. starting from the fault point (calculated value) is specified from the information on the topology and impedance of the power system 5 (system information 122 of the storage unit 120), so that the arc resistance value can be converted into the distance on the path from the fault point (calculated value). The acquired probabilistic distribution of the distance is stored in the storage unit 120 as calculation data 128.

[0129] In S370, the accident point locating device 100 uses the accident point (calculated value) acquired in S340 and the probabilistic distribution of the distance obtained in S360 to calculate the probability distribution of the location of the accident point as the accident point location result PFrst by the location result inference unit 140. The accident point location result PFrst calculated in S370 is stored in the storage unit 120 as the calculation data 128.

[0130] 13 shows a block diagram for explaining fault point location according to the embodiment 1. The function of the block diagram in FIG. 13 is realized by the processes of S340 to S370 in FIG.

[0131] 13, the trained fault point inference model 510 receives current information INFI and voltage information INFV, which are the current and voltage (actual values) at the time of the accident, and the cause of the accident (estimated fault cause CAest) by the process of S340, and outputs the fault point PF0. The fault point PF0 corresponds to the above-mentioned fault point (calculated value). When the fault point PF0 indicates a point on the power system 5, it is specified by the transmission route and distance from the substation 10. Furthermore, the fault point PF0 may be specified as one section including the fault point among a plurality of sections into which the transmission lines and distribution lines in the power system 5 are divided in advance. For example, the fault point PF0 corresponds to a point inferred by the fault point inference model 510 as a point on the power system 5 where the probability of the existence of the fault point is maximum.

[0132] Alternatively, the accident point inference model 510 can output the accident point PF0 so as to include a plurality of points on different routes from the electric power station 10. In this case, for the plurality of points, a calculated value of the existence probability of the accident point is inferred in accordance with a combination of the estimated result of the accident cause (estimated accident cause CAest) and the surrounding conditions of each of the plurality of points. Specifically, the existence probability of the accident point is distributed among the plurality of points included in the accident point PF0 under the assumption that the sum of the existence probabilities of the accident point at each of the plurality of points is 1.0. For example, in the example of FIG. 1, when the estimated result of the accident cause (estimated accident cause CAest) indicates "contact with a tree", both of the points PF1 and PF2 may be output from the accident point inference model 510 as the accident point PF0, with the existence probability of the accident point assigned. In this example, an inference result is obtained in which the existence probability of the accident point at the point PF1 (e.g., 80(%)) is higher than the existence probability at the point PF2 (e.g., 20(%)).

[0133] The statistical model 520 of the arc resistance value receives the accident cause (estimated accident cause CAest) as an input by the process of S350, and outputs the probabilistic distribution Rac(0) of the arc resistance value corresponding to the accident cause (estimated result). The probabilistic distribution Rac(0) can be represented by a histogram, a probability density function, a probability distribution function, a cumulative distribution function, or the like of the arc resistance value Rac, as described in FIG.

[0134] The integrated processing unit 530 receives the accident point PF0 from the accident point inference model 510 and the probabilistic distribution Rac(0) of the arc resistance value from the statistical model 520. The integrated processing unit 530 converts the probabilistic distribution Rac(0) of the arc resistance value into a distance from the accident point PF0 on the power transmission route including the accident point PF0, using the topology and impedance of the surrounding power transmission lines, power distribution lines, etc. starting from the accident point PF0. As a result, in the process of S360, the probabilistic distribution Rac(0) of the arc resistance value is converted into a probabilistic distribution of the distance between the accident point PF0 and the actual accident point, which corresponds to the arc resistance value.

[0135] Furthermore, the integration processing unit 530 outputs the accident point location result PFrst by the process of S370, using the accident point PF0 and the above-mentioned probabilistic distribution of the distance between the accident point PF0 and the actual accident point.

[0136] FIG. 14 is a conceptual diagram for explaining an example of output of the fault point location result PFrst. As shown in FIG. 14, the fault point location result PFrst is shown, for example, as a probability distribution of the location of the fault point at each point on the power transmission path from the substation 10 identified to indicate the fault point PF0.

[0137] In the example of Fig. 14, the probability of the existence of an accident point at a point PFs different from the point corresponding to the accident point PF0 is maximized by reflecting the probabilistic distribution of the distance corresponding to the arc resistance value. That is, a point on the power system 5 where the probability of the existence of an accident point is maximized is corrected according to the estimated arc resistance value from the output of the accident point inference model 510. In this way, by combining the estimation of the arc resistance value from the accident cause (estimated accident cause CAest), the location accuracy can be improved compared to the accident point location result by the accident point inference model 510 alone.

[0138] For example, when one accident point PF0 is output from the accident point inference model 510, the probability of the existence of an accident point at each point starting from the accident point PF0 becomes equal to the probabilistic distribution of the distance corresponding to the arc resistance value.

[0139] On the other hand, when multiple accident points PF0 are output from the accident point inference model 510 with the probability of existence of the accident points added, the probability of existence of an accident point at each point starting from each accident point PF0 can be obtained by multiplying the probabilistic distribution (distance) of the arc resistance value by the probability of existence of an accident point at the accident point PF0 to obtain the accident point location result PFrst.

[0140] In addition, when the fault point PF0 specifies one section including the fault point among a plurality of sections divided in advance in the power system 5, the fault point location result PFrst can be shown as the existence probability of the fault point in each of the above-mentioned plurality of sections. In this case, the section including the fault point PF0 indicates a point inferred by the fault point inference model 510 as a point where the existence probability of the fault point is maximum on the power system 5, but by reflecting the probabilistic distribution of the distance corresponding to the arc resistance value, the existence probability of the fault point may be maximum in a section other than the section including the fault point PF0.

[0141] In this way, in order to take into account the effect that the arc resistance value varies depending on the cause of the accident, the accident point location result PFrst is generated based on the accident point PF0 from the accident point inference model 510 and corrected by the arc resistance value obtained by the statistical model 520.

[0142] 12 again, in S380, the accident point locating device 100 transmits the probability distribution of the location of the accident point acquired in S370 (the accident point locating result PFrst) by the communication unit 110 to the central monitoring system 300 via the communication network 50. The central monitoring system 300 corresponds to one embodiment of the "manned facility".

[0143] As a result, in the central monitoring system 300, by providing the maintenance worker with fault point information indicating the probability distribution of the location of the fault point, which is indicated by the fault point location result PFrst, the maintenance worker can identify the fault point and investigate the cause by patrol, and can speed up the subsequent recovery work. Note that the fault point information may be displayed visually on the topology of the power system 5 as an estimated result of the fault point.

[0144] As described above, according to the first embodiment, by performing fault point location with the fault cause (estimated fault cause CAest) added as an input, it is possible to obtain a location result (fault point location result PFrst) in which the existence probability of the fault point is quantified by combining extraction of a plurality of points with the same electrical distance (impedance) from the substation 10 to the fault point with inference reflecting the relationship between the fault cause (estimated fault cause CAest) and the surrounding conditions of each point. According to the quantified existence probability, it is possible to narrow down the fault point location result so as to indicate a single point where the existence of the fault point is the greatest.

[0145] As a result, in the comparative example described in Fig. 1, it is not possible to obtain a numerically quantified probability of existence of the accident point for each of the points PF1 and PF2, whereas in this embodiment, the accuracy of locating the accident point is improved. For example, it is possible to realize identification of one of the points PF1 and PF2 in Fig. 1 (i.e., location in which the probability of existence is quantified to 100(%)). Alternatively, it is also possible to obtain a location result (accident point location result PFrst) including multiple accident points by quantifying the occurrence probability of the accident point for both the points PF1 and PF2 in Fig. 1 according to a combination of the estimated result of the accident cause (estimated accident cause CAest) and the surrounding circumstances.

[0146] Specifically, in the learning phase of the accident point inference model 510, machine learning is performed using training data that includes the accident point assumed by simulation and the accident cause that reflects the surrounding circumstances of the accident point, thereby creating a trained model capable of inference based on the above-mentioned combination.

[0147] Furthermore, by locating the fault point in a manner that reflects the distance on the power system 5 corresponding to the arc resistance value that changes depending on the cause of the accident, based on the fault point (calculated value) calculated by the fault point inference model 510, the location accuracy can be improved compared to fault point location using only the fault point inference model 510, as explained in the example of Figure 14.

[0148] In addition, by using the probabilistic distribution of the arc resistance value by the statistical model 520 to obtain the probability of the existence of the accident point as the accident point location result, it is possible to improve the efficiency of post-accident patrol by security workers and shorten the patrol time. For example, if it is confirmed that the accident point did not exist at the point or section with the maximum existence probability, the next point (section) to be patrolled to search for the accident point or the direction to be patrolled can be determined according to the probability distribution. This makes it possible to improve the efficiency of the search work for the accident point.

[0149] Such an improvement in the accuracy of fault point location can reduce manpower and costs for fault search, and can improve power quality by shortening the fault recovery time (power outage time). In addition, the maintenance and preservation of the power system 5 can be made more efficient.

[0150] A variation of the first embodiment. In the first embodiment, it is assumed that the current and voltage measured by the potential transformer (VT) and current transformer (CT) are directly used for the current information INFI and voltage information INFV, which are the feature quantities input to the fault point inference model 510. In contrast to this, a modified example will be described in which processed data obtained by calculation from the current and voltage are also included in the current information INFI and voltage information INFV.

[0151] FIG. 15 is a block diagram illustrating an example of the configuration of an accident point locating device 101 according to a modification of the first embodiment.

[0152] As shown in FIG. 15, the accident point locating device 101 according to the variation of the first embodiment differs from the accident point locating device 100 according to the first embodiment (FIG. 3) in that it further includes a processed data generating unit 160.

[0153] The processed data generating unit 160 calculates processed data using the current and voltage (actual values) from the current transformer (CT) and the voltage transformer (VT) received by the communication unit 110. The processed data includes, for example, the total distortion factor, the third harmonic content, the crest factor, the maximum value, the amount of change in a certain period, the results of frequency component analysis, the impedance value, the rate of change of current before and after the accident, the rate of change of voltage before and after the accident, etc. The processed data (actual values) calculated by the processed data generating unit 160 is stored in the memory unit 220 as calculated data 128.

[0154] When an accident occurs, the processed data (actual value) calculated by the processed data generation unit 160 is stored in the memory unit 220 and transmitted by the communication unit 110 to the accident point locating server 201 related to the modified example of embodiment 1 via the communication network 50.

[0155] In the accident point locating server 201, similarly to the first embodiment, the current information INFI and voltage information INFV before and after the accident including the time of the accident received by the communication unit 210 are stored in the storage unit 220 as measurement data 224. That is, in the modification of the first embodiment, the measurement data 224 stored in the storage unit 220 includes the processed data (actual values) calculated by the processed data generating unit 160 in addition to the current and voltage (actual values) similar to the first embodiment.

[0156] Furthermore, in the fault point locating server 201 according to the modification of the first embodiment, the learning data generating unit 240 is configured to have a calculation function similar to that of the processed data generating unit 160. Then, the learning data generating unit 240 adds processed data (calculated values) acquired by a calculation function similar to that of the processed data generating unit 160 from the currents and voltages (calculated values) at the CTs and VTs of the electric power station 10 obtained as the output of the instantaneous value analysis model 500 (FIG. 6) to the learning data.

[0157] As a result, in the modification of the first embodiment, for each simulated accident, a set of the accident point, the corresponding current data and voltage data (calculated values) before and after the accident, and the processed data (calculated values) is treated as learning data to be used for machine learning, and is stored in the storage unit 220 as calculated data 228. Note that it is not necessary to use all of the corresponding current data and voltage data (calculated values) before and after the accident, and the processed data (calculated values) as learning data, and at least one of these quantities may be used as a feature amount, or multiple quantities may be used as feature amounts.

[0158] In the learning phase of the accident point inference model 510, machine learning is performed using the learning data (teacher data) in such a manner that the items of the current information INFI(S) and the voltage information INFV(S) which are the learning data are increased by adding the processed data (calculated value). As a result, the learning model generation unit 270 can generate the learned accident point inference model 510 in which the input feature values ​​are the current and voltage (actual values) similar to those in the first embodiment and the processed data (actual values) of the added items. As described above, it is not necessary to use all of the current data and voltage data (calculated values) before and after the corresponding accident and the processed data (calculated values) as the learning data, and at least one of these may be used as the feature value, or multiple values ​​may be used as the feature values. The data for expressing the learned model is stored in the storage unit 220 as the calculated data 228, and is also transmitted to the accident point locating device 101 via the communication network 50 and stored in the storage unit 120 as the model information 121.

[0159] In the accident point locating device 101 relating to the modified example of embodiment 1, when an accident occurs, current information INFI and voltage information INFV including the current and voltage (actual values) and processed data (actual values) before and after the accident, and the estimated accident cause CAest by the accident cause estimation device 150 are input to the trained accident point inference model 510.

[0160] Then, the fault point inference model 510 outputs the fault point PF0 similar to that in the first embodiment as an estimation result of the fault point, indicating a point or a section on the power system 5. Other processes of fault point location are similar to those in the first embodiment, and therefore detailed description will not be repeated.

[0161] By adding data processing values ​​based on current and voltage to the current information and voltage information that are input values ​​to the fault point inference model 510, it is expected that the accuracy of locating the fault point will be improved.

[0162] In addition, in the first embodiment, an example has been described in which the learning data used in the learning phase of the accident point inference model 510 is composed of simulation values ​​obtained by using the instantaneous value analysis model 500. However, instead of using only the simulation values ​​as the learning data, the learning data may include, in addition to the simulation values, the current and voltage (actual values) before and after the accident when the accident occurs, which are stored as the measurement data 224, or the current and voltage (actual values) and the processed data (actual values).

[0163] In this way, by creating a trained accident point inference model 510 through machine learning using both simulation values ​​and actual values, it is expected that the accuracy of locating the accident point will be improved.

[0164] Alternatively, it is possible to correct the system information 222 by using at least a part of the current and voltage (actual values) before and after the accident and the processed data (actual values). In particular, since impedance information tends to contain many errors, it is expected that the correction will have a large effect. As a result, it is expected that the accuracy of locating the accident point will be improved by improving the estimation accuracy of the accident point inference model 510 through the improvement of the simulation accuracy by the instantaneous value analysis model 500.

[0165] Embodiment 2 Next, a modified example of the accident point inference model will be described. In each of the following embodiments, the operation and function of the learning data generating unit 240 and the learning model generating unit 270 (accident point locating server 200) in the learning phase, and the operation and function of the location result inference unit 140 (accident point locating device 100) during accident point location (inference phase) are appropriately modified according to the contents of the modified accident point inference model.

[0166] FIG. 16 is a block diagram illustrating input and output in the learning phase of the accident point inference model in the second embodiment.

[0167] 16, the fault point inference model 515 according to the second embodiment is configured to receive current information INFI and voltage information INFV as inputs and output an inference result of the fault point. That is, unlike the fault point inference model 510 according to the first embodiment, the fault point inference model 515 excludes the cause of the accident from the inputs.

[0168] In the learning phase of the fault point inference model 515, the instantaneous value analysis model 500 reflecting system information (topology and impedance of the power system 5) executes a simulation similar to that of the first embodiment, and outputs the current and voltage (calculated values) at the CT and VT of the substation 10 when a system fault occurs under each of various fault conditions (fault point and fault aspect) set for generating learning data. As a result, current information INFI(S) and voltage information INFV(S) are acquired as simulation values ​​from the current and voltage (calculated values).

[0169] As shown by the dotted line in Fig. 16, in the learning phase, the learning data generating unit 240 of the accident point locating server 200 shown in Fig. 4 generates teacher data in which the current information INFI(S) and the voltage information INFV(S), which are simulation values, are input (feature amount) to the accident point inference model 515, and the accident point during the simulation is output (answer) from the accident point inference model 515. By machine learning using the generated learning data (teaching data), the learning model generating unit 270 of the accident point locating server 200 shown in Fig. 4 generates a learned accident point inference model 510 (learning model generating unit 270). In the second embodiment, the process contents at S230 and S240 in Fig. 10 are appropriately changed according to the contents of Fig. 16.

[0170] In the second embodiment as well, the fault point inference model 510 is created for each accident aspect by using the current information INFI(S) and voltage information INFV(S) obtained for each accident aspect for each fault point.

[0171] When a system fault occurs, the fault point inference model 515, if selected according to the nature of the fault that has occurred, can output a calculated value of the fault point by inference using the trained model. When locating the fault point (inference phase), the current information INFI and voltage information INFV input to the fault point inference model 510 (trained) are set to values ​​(actual values) acquired at the time of the fault occurrence. As in the first embodiment, the fault point inference model 515 (trained) outputs the fault point (calculated value) indicated by a combination of the power transmission route and distance from the substation 10. The fault point inference model 515 corresponds to one example of the "second inference model".

[0172] However, the fault point inference model 515 according to the second embodiment performs inference based only on the electrical distance (impedance) from the substation 10, similar to the inference model 101M according to the comparative example. Therefore, if there are a plurality of points on the power system 5 that have the same electrical distance (impedance) from the substation 10, a plurality of fault points (calculated values) are also calculated. For example, the fault point (calculated value) by the fault point inference model 515 is calculated so as to include both the points PF1 and PF2 assumed in the comparative example of FIG. 1. Unlike the fault point inference model 510 in the first embodiment, the fault point (calculated value) calculated by the fault point inference model 515 does not include information indicating the existence probability of the fault point. That is, if there are a plurality of fault points (calculated values) calculated by the fault point inference model 515, the existence probability of the fault point is the same among the plurality of fault points at this stage.

[0173] FIG. 17 is a block diagram for explaining calculation of an accident point in the accident point location according to the second embodiment.

[0174] As shown in FIG. 17, in the second embodiment, a combination of a trained accident point inference model 515 and a rule processing unit 540 realizes functionality equivalent to that of the accident point inference model 510 in the first embodiment.

[0175] 17, the trained fault point inference model 515 receives current information INFI and voltage information INFV indicating the current and voltage (actual values) at the time of the fault occurrence, and outputs one or more fault points PFA(1) to PFA(N) (N: natural number). The fault points PFA(1) to PFA(N) correspond to the fault points (calculated values) described above. Each of the fault points PFA(1) to PFA(N) can be specified as one of a plurality of sections into which the power transmission route and distance from the substation 10, or the power transmission lines and power distribution lines in the power system 5 are previously divided, similar to the fault point PF0 in the first embodiment.

[0176] The rule processing unit 540 receives the accident points PFA(1) to PFA(N) output from the accident point inference model 515, the system information 122 (surrounding conditions of each point of the power system 5), and the accident cause (estimated accident cause CAest). As in the first embodiment, the accident cause can be an accident cause code set according to the estimated accident cause CAest. The system information 122 (surrounding conditions of each point of the power system 5) can be used to acquire the surrounding conditions of each of the accident points PFA(1) to PFA(N).

[0177] When there are a plurality of accident points PFA(1) to PFA(N) (i.e., when N≧2), the rule processing unit 540 can be configured to select one accident point according to a rule processing (e.g., a combination of selection logics, or a machine learning processing provided separately from the generation of the accident point inference model 515) that is preset for a combination of the surrounding conditions of each of the accident points PFA(1) to PFA(N) and the cause of the accident. As an example of the selection logic, when the cause of the accident is "contact with a tree", an accident point whose surrounding conditions are "flat ground" can be deselected. Based on such a predetermined selection logic, the rule processing can be arbitrarily configured.

[0178] Also, the accident point location results of a plurality of points may be given the probability of existence of the accident point and ranked in consideration of the surrounding conditions. In this case, the rule in the rule processing unit 540 is determined not to select one from the plurality of accident points PFA(1) to PFA(N) as described above, but to calculate the probability of existence of each of the plurality of accident points PFA(1) to PFA(N). In particular, in the rule processing unit 540, under the condition that the sum of the probability of existence of the accident point at each of the plurality of accident points PFA(1) to PFA(N) is 1.0, a distribution rule of the probability of existence of the accident point can be determined according to the combination of the surrounding conditions of each of the plurality of accident points PFA(1) to PFA(N) and the accident cause (estimated accident cause CAest), instead of a rule of selection / non-selection as an accident point. In this case, the distribution rule according to the above combination can be arbitrarily configured.

[0179] As a result, the rule processing unit 540 can output the fault point PF0 as information for indicating a point or a section on the power system 5 where the fault point existence probability is maximum, similarly to the fault point inference model 510 according to the first embodiment. Alternatively, it is also possible to output from the rule processing unit 540 each of the fault points PFA(1) to PFA(N) to which the fault point existence probability has been assigned. Therefore, the fault point (calculated value) output from the rule processing unit 540 can be used in the same way as the fault point PF0 in the configuration of FIG. 13.

[0180] FIG. 18 is a flowchart for explaining the process for locating a fault point when a system fault occurs in the fault point locating system according to the second embodiment.

[0181] 18, in the second embodiment, the fault point locating device 100 executes S310 and S320 similar to those in FIG. 12. As a result, the current and voltage (actual values) measured by the CT (11, 12) and VT (15) of the substation 10 at the time of the occurrence of the accident are acquired as the current information INFI and the voltage information INFV at the time of the accident. The acquired current and voltage (actual values) are stored in the storage unit 120 as the measurement data 124, and are also transmitted to the fault point locating server 200 via the communication network 50. In the fault point locating server 200, the current and voltage (actual values) at the time of the occurrence of the accident are stored by S390 and S395 similar to those in FIG. 12.

[0182] Furthermore, when the accident point locating device receives and stores the result of the estimation of the accident cause (estimated accident cause CAest) from the accident cause estimation device 150 (FIG. 1) in S330 similar to that in FIG. 12, in S410, the accident point locating device inputs the current and voltage (actual values) at the time of the accident occurrence acquired in S310 to the trained accident point inference model 515. As a result, the accident points PFA(1) to PFA(N) in FIG. 17 are output as accident points (calculated values) from the accident point inference model 515 (N≧1).

[0183] The accident point locating device 100 judges whether the accident points (calculated values) acquired in S410 are multiple or not in S420, and when N≧2 (multiple) (YES judgment in S420), selects one accident point (calculated value) in S430 by a predetermined rule processing based on a combination of the accident cause (estimated accident cause CAest) and the surrounding circumstances of each of the multiple accident points (calculated values). On the other hand, when N=1 (single) (NO judgment in S420), the accident point (calculated value) acquired in S410 is used as it is as the final one accident point (calculated value). That is, the processing in S420 and S430 is realized by the function of the rule processing unit 540 in FIG. 17.

[0184] 12 is obtained, so that in the second embodiment, after the process of S430 or after the NO judgment in S420, the process of S350 and thereafter in FIG. 12 is similarly executed to obtain the fault point location result PFrst similar to that in the first embodiment. That is, by correcting the fault point (calculated value) obtained by the fault point inference model 515 with the arc resistance value by the statistical model 520, the fault point can be located with high accuracy.

[0185] In this way, in the accident location according to the second embodiment, even if the cause of the accident is not directly incorporated into the training data of the machine learning when generating the accident point inference model 515, the rule processing unit 540 can be used to realize the accident point location with the accident cause (estimated accident cause CAest) added as an input, thereby providing the same effect as that of the first embodiment.

[0186] It is also possible to apply a modification of the first embodiment to the second embodiment. That is, processed values ​​of current and voltage may be added to the current information and voltage information used in the learning phase and inference phase of the fault point inference model 515, and the current and voltage (actual values) at the time of the occurrence of the fault may be added to the learning data used in the learning phase of the fault point inference model 515. It is also possible to correct the system information 222 reflected in the instantaneous value analysis model 500 by using the current and voltage (actual values) and / or the processed data (actual values).

[0187] In the first and second embodiments, the current information INFI and the voltage information INFV used in the learning phase and the inference phase of the fault point inference model 510, 515 are mainly the current and voltage (calculated value) and the current and voltage (actual value) at the time of the accident, but it is also possible to use the current and voltage (calculated value) and the current and voltage (actual value) before and / or after the accident in addition to the time of the accident. For example, the current change rate before and after the accident, the voltage change rate before and after the accident, etc., described in the modified example of the first embodiment, can be input values ​​in the learning phase and the inference phase of the fault point inference model 510, 515. In this way, the fault point location accuracy can be improved by performing learning and inference using the current information INFI and the voltage information INFV that comprehensively reflect the current behavior and the voltage behavior before and after the accident.

[0188] The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present disclosure is defined by the claims, not the above description, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0189] 5 Power system, 10 Electrical station, 11, 12 Current transformer, 15 Potential transformer, 20 Data collection unit, 40 Computer system, 47 Bus, 50 Communication network, 100, 101 Fault point locating device, 101M Inference model, 110, 210 Communication unit, 120, 220 Memory unit, 121 Model information, 122, 222 System information, 124, 224 Measurement data, 126 Accident cause estimation result, 128, 129, 228 Calculated data, 130 Data acquisition unit, 140 Location result inference unit, 150 Accident cause estimation device, 160 Processed data generation unit, 200, 201 Fault point locating server, 226 Accident related information, 230 Input reception unit, 240 Learning data generation unit, 250 Arc resistance calculation unit, 260 Statistical model generation unit, 270 Learning model generation unit, 300 central monitoring system, 500 instantaneous value analysis model, 510, 515 fault point inference model, 520 statistical model, 530 integration processing unit, 540 rule processing unit, CAest estimated fault cause, INFI current information, INFV voltage information, LA1, LA2, LB1, LB2 transmission lines, LM bus, PF0, PFA(1) to PFA(N) fault points, PFrst fault point location result.

Claims

1. An accident point calibration device for a power system, comprising: a data acquisition unit for acquiring current information and voltage information at a predetermined point at the time of an accident; an inference unit that takes as input the current information and voltage information obtained by the data acquisition unit at the time of the accident and an estimated accident cause based on the current information and voltage information, and outputs a calibration result that quantifies the probability of the existence of an accident point on the power system.

2. A storage unit that stores a machine-learned first inference model that sets the accident cause reflecting the surrounding conditions of each point of the power system, and outputs an accident point calculation value indicating one or more points that are inferred to have a high probability of being the accident point on the power system when the current information and voltage information at the predetermined point and the accident cause are input, wherein the inference unit outputs the calibration result using the accident point calculation value when the current information and voltage information obtained by the data acquisition unit at the time of the accident and an estimated accident cause based on the current information and voltage information are input to the first inference model. The accident point calibration device according to claim 1.

3. The first inference model is Machine-learned with the current information and voltage information at the predetermined point at the time of the simulated accident at the set accident assumption point calculated by a simulator that reflects information on the topology and impedance of the power system as the teacher data of the current information and voltage information at the time of the accident, the accident cause corresponding to the surrounding conditions of the accident assumption point as the teacher data of the accident cause, and the accident assumption point as the teacher data of the accident point calculation value. The accident point calibration device according to claim 2.

4. When the first inference model outputs the accident point calculation value including the plurality of points, the first inference model is configured to output the accident point calculation value accompanied by the calculated value of the probability of the existence of the accident point at each of the plurality of points. The accident point calibration device according to claim 2.

5. The first inference model outputs the accident point calculation value so as to indicate the one point where the probability of the existence of the accident point is the maximum. The accident point calibration device according to claim 2.

6. The memory unit further stores a statistical model configured to output the probability distribution corresponding to the input accident cause from the probability distribution of the arc resistance value at the accident point, which is predetermined for each accident cause using the performance values at the time of past accident occurrences in response to the input of the accident cause. At the time of the accident occurrence, the inference unit inputs the estimated accident cause into the statistical model. The inference unit The accident point calibration device according to claim 2, further comprising an integration processing unit that integrates the accident point calculated value output from the first inference model and the probability distribution of the arc resistance value output from the statistical model, and outputs the distribution of the existence probability of the accident point on the power grid as the calibration result.

7. Further comprising a machine learning-trained second inference model that outputs accident point calculated values corresponding to one or more points with the same electrical distance from the predetermined point, using the current information and voltage information at the predetermined point as inputs, and a memory unit that stores second grid information indicating the surrounding conditions of each point on the power grid. At the time of the accident occurrence, the inference unit inputs the current information and voltage information obtained by the data acquisition unit into the second inference model to obtain the accident point calculated value. The inference unit When the accident point calculated value includes a plurality of points, by applying a predetermined rule to the combination of the surrounding conditions of each of the plurality of points indicated by the second grid information and the estimated accident cause, one point is selected from the plurality of points, or the accident point calculation value is corrected so as to distribute the existence probability of the accident point among the plurality of points, including a rule processing unit. The inference unit outputs the calibration result using the accident point calculated value output from the rule processing unit. The accident point calibration device according to claim 1.

8. The second inference model The accident point calibration device according to claim 7, wherein the current information and voltage information at the predetermined point at the time of the simulated accident occurrence at the set accident assumption point, calculated by a simulator reflecting information on the topology and impedance of the power grid, are used as teacher data for the current information and voltage information at the time of the accident occurrence, and the accident assumption point is used as teacher data for the accident point calculated value for machine learning.

9. The memory unit further stores a statistical model configured to output the probability distribution corresponding to the input accident cause from the probability distribution of the arc resistance value at the accident point, which is predetermined for each accident cause using the performance values at the time of past accident occurrences in response to the input of the accident cause. At the time of the accident, the inference unit inputs the estimated accident cause into the statistical model to obtain the probability distribution of the arc resistance value. The inference unit The accident point calibration device according to claim 7, comprising an integration processing unit that integrates the corrected accident point calculated value from the rule processing unit and the probability distribution of the arc resistance value obtained from the statistical model, and outputs the probability distribution of the existence of the accident point on the power system as the calibration result.

10. The memory unit further stores first system information related to the topology and impedance of the power system. The integration processing unit according to claim 6 or 9, uses the first system information to convert the probability distribution of the arc resistance value into the probability distribution of the distance between the one point or the plurality of points included in the accident point calculated value and the accident point, and calculates the probability distribution of the existence of the accident point using the accident point calculated value and the probability distribution of the distance.

11. The inference unit according to claim 1, generates the calibration result so as to correct the accident point calculated value calculated using the current information and voltage information acquired at the time of the accident and the estimated accident cause, by an amount corresponding to the arc resistance value estimated corresponding to the estimated accident cause.

12. The accident point calibration device according to any one of claims 1 to 9 and 11, wherein the current information and voltage information include the voltage value and current value at the predetermined point, and the processed data of the voltage value and the current value.

13. The accident point calibration device according to claim 3 or 8, wherein the teacher data of the current information and voltage information further includes the current information and voltage information obtained by the data acquisition unit at the time of the accident.

14. The accident point calibration device according to any one of claims 1 to 9 and 11, wherein the current information and voltage information include the current information and voltage information at the time of the accident, and the current information and voltage information before and after the accident.

15. The accident point calibration device according to claim 2, comprising an accident point calibration server communicatively connected to the accident point calibration device via a communication network. The accident point calibration server, using the current information and voltage information at a predetermined point at the time of a simulated accident at a set accident scenario point calculated by a simulator reflecting information on the topology and impedance of the power system as teacher data for the current information and voltage information at the time of the accident, using the accident cause corresponding to the surrounding situation of the accident scenario point as teacher data for the accident cause, and using the accident scenario point as teacher data for the accident point calculation value, machine-learns the first inference model. The storage unit stores the machine-learned first inference model transmitted from the accident point calibration server, and is an accident point calibration system.

16. The accident point calibration device according to claim 7, and an accident point calibration server communicatively connected to the accident point calibration device via a communication network, wherein the accident point calibration server, using the current information and voltage information at a predetermined point at the time of a simulated accident at a set accident scenario point calculated by a simulator reflecting information on the topology and impedance of the power system as teacher data for the current information and voltage information at the time of the accident, and using the accident scenario point as teacher data for the accident point calculation value, machine-learns the second inference model. The storage unit stores the machine-learned second inference model transmitted from the accident point calibration server, and is an accident point calibration system.

17. The accident point calibration device according to claim 6 or 9, and an accident point calibration server communicatively connected to the accident point calibration device via a communication network, wherein the accident point calibration server generates a statistical model configured to output a probability distribution corresponding to the input accident cause from the probability distribution of the arc resistance value at the predetermined accident point for each accident cause using the actual values at the time of past accidents, and the storage unit stores the statistical model transmitted from the accident point calibration server, and is an accident point calibration system.

18. The accident point calibration device is arranged at an electrical station including a substation, and the calibration result by the accident point calibration device is transmitted to a manned facility via the communication network. The accident point calibration system according to claim 15 or 16.

19. A method for calibrating an accident point in a power system, comprising: acquiring current information and voltage information at a predetermined point at the time of an accident. An accident point calibration method that outputs a calibration result obtained by quantifying the probability of the existence of an accident point on the power grid, using the acquired current information and voltage information and the estimated accident cause based on the current information and voltage information as inputs.

20. Save a first inference model that has been machine-learned and sets the accident cause reflecting the surrounding conditions of each point on the power grid, and outputs an accident point calculation value indicating one or more points that are inferred to have a high probability of being the accident point on the power grid, using the current information and voltage information at the predetermined point and the accident cause as inputs. The accident point calibration method according to claim 19, wherein when the current information and voltage information acquired at the time of the accident and the estimated accident cause based on the current information and voltage information are input to the saved first inference model, the calibration result is output based on the accident point calculation value.

21. Further save a statistical model configured to output the probability distribution corresponding to the input accident cause from the probability distribution of the arc resistance value at the accident point predetermined for each accident cause using the actual value at the time of past accidents. At the time of the accident, input the estimated accident cause into the saved statistical model. The accident point calibration method according to claim 20, wherein the accident point calculation value output from the first inference model and the probability distribution of the arc resistance value output from the statistical model are integrated to output the probability distribution of the existence of the accident point on the power grid as the calibration result.

22. Save a second inference model that has been machine-learned and outputs an accident point calculation value corresponding to one or more points having the same electrical distance from the predetermined point, using the current information and voltage information at the predetermined point as inputs. Save second grid information indicating the surrounding conditions of each point on the power grid. Input the current information and voltage information acquired at the time of the accident into the second inference model to obtain the accident point calculation value. When the accident point calculation value includes a plurality of points, one point is selected from the plurality of points, or the probability of the existence of the accident point is distributed among the plurality of points, by applying a predetermined rule to the combination of the surrounding conditions of each of the plurality of points indicated by the second grid information and the estimated accident cause, and the accident point calculation value is corrected, and the corrected accident point calculation value is used to output the calibration result. The accident point calibration method according to claim 19.

23. According to the input of the accident cause, further store a statistical model configured to output the probability distribution corresponding to the input accident cause from the probability distribution of the arc resistance value at the accident point, which is predetermined for each accident cause using the actual values at the time of past accident occurrences. At the time of the accident occurrence, input the estimated accident cause into the statistical model to obtain the probability distribution of the arc resistance value. The accident point calibration method according to claim 22, wherein the corrected accident point calculated value and the probability distribution of the arc resistance value obtained from the statistical model are integrated to output the probability distribution of the existence of the accident point on the power system as the calibration result.

24. The calibration result is generated and output so as to correct the accident point calculated value calculated using the current information and voltage information obtained at the time of the accident occurrence and the estimated accident cause by a correction corresponding to the arc resistance value estimated corresponding to the estimated accident cause. The accident point calibration method according to claim 19.

25. When the first inference model outputs the accident point calculated value including the plurality of points, the accident point calculated value is output together with the calculated values of the existence probabilities of the accident points at each of the plurality of points. The accident point calibration method according to claim 20.

26. The first inference model outputs the accident point calculated value so as to indicate the one point where the existence probability of the accident point is the maximum. The accident point calibration method according to claim 20.

Images