JP2021015098A - Partial discharge diagnostic device, partial discharge diagnostic method, learning device, learning method, partial discharge diagnostic system, and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a partial discharge diagnosis device, a partial discharge diagnosis method, a learning device, a learning method, a partial discharge diagnosis system and a computer program capable of providing information for determining whether or not the diagnosis result of partial discharge is reliable. To provide. A partial discharge diagnostic apparatus according to an embodiment includes a state diagnosis unit, a reliability information generation unit, and an output unit. The state diagnosis unit is a classifier generated by performing predetermined machine learning using learning data in which information on a partial discharge signal generated from a predetermined electric device and information on the state of the predetermined electric device are associated with each other. Based on the discharge data related to the partial discharge signal generated from the electrical device to be diagnosed, the state of the electrical device to be diagnosed is diagnosed. The credit information generator generates information on the reliability of the diagnosis result based on the state of the electrical device identified by the classifier. The output unit outputs the information on the reliability and the result of the diagnosis to a predetermined output device. [Selection diagram] Fig. 1

Description

Embodiments of the present invention relate to a partial discharge diagnostic device, a partial discharge diagnostic method, a learning device, a learning method, a partial discharge diagnostic system, and a computer program.

In recent years, in a partial discharge diagnostic apparatus, diagnosis using machine learning has been studied in diagnosing the presence or absence of partial discharge in an electric device. In the diagnosis using machine learning, the partial discharge diagnostic device diagnoses the electric device based on the discriminator generated based on the past diagnosis result and the newly acquired data on the discharge. In such a diagnostic method, it is required that the data used for generating the discriminator and the newly acquired data are similar. However, it can be difficult to determine if the data used to generate the classifier and the newly acquired data are similar. For this reason, it may be difficult for the diagnostician to determine whether or not the diagnosis result of the partial discharge is a reliable result.

JP-A-2018-165926

The problem to be solved by the present invention is a partial discharge diagnostic apparatus, a partial discharge diagnostic method, a learning apparatus, a learning method, and a portion capable of providing information for determining whether or not the diagnosis result of partial discharge is reliable. It is to provide a discharge diagnostic system and a computer program.

The partial discharge diagnostic apparatus of the embodiment includes a state diagnosis unit, a credit information generation unit, and an output unit. The state diagnosis unit performs identification generated by performing predetermined machine learning using learning data in which information on a partial discharge signal generated from a predetermined electric device and information on the state of the predetermined electric device are associated with each other. The state of the electrical device to be diagnosed is diagnosed based on the discharge data related to the partial discharge signal generated from the device and the electrical device to be diagnosed. The credit information generation unit generates information regarding the reliability of the result of the diagnosis based on the state of the electric device identified by the classifier. The output unit outputs the information on the reliability and the result of the diagnosis to a predetermined output device.

The functional block diagram which shows the functional structure of the partial discharge diagnostic apparatus 100 of 1st Embodiment. The figure which shows a specific example of the learning data of 1st Embodiment. The figure which shows a specific example of the evaluation data of 1st Embodiment. The figure which shows a specific example of the output of the intermediate layer of the neural network in 1st Embodiment. The figure which shows a specific example of the generated diagnostic result of 1st Embodiment. The flowchart which shows a specific example of the process flow of the generation of the classifier of 1st Embodiment. The flowchart which shows a specific example of the process flow of the generation of the diagnosis result of 1st Embodiment. The functional block diagram which shows the functional structure of the partial discharge diagnostic apparatus 100a of 2nd Embodiment. The figure which shows a specific example of the generated diagnosis result of 2nd Embodiment. The flowchart which shows a specific example of the flow of the process of generating the diagnosis result of the 2nd Embodiment.

Hereinafter, the partial discharge diagnostic apparatus, the partial discharge diagnostic method, the learning apparatus, the learning method, the partial discharge diagnostic system, and the computer program of the embodiment will be described with reference to the drawings.

(First Embodiment)
FIG. 1 is a functional block diagram showing a functional configuration of the partial discharge diagnostic apparatus 100 of the first embodiment. The partial discharge diagnostic device 100 is an information processing device such as a personal computer, a server, a tablet computer, or a smart device. The partial discharge diagnostic device 100 is provided in the vicinity of the box body in which the electrical equipment is housed. The partial discharge diagnostic device 100 diagnoses the state of the electric device based on the partial discharge signal generated from the electric device to be diagnosed. The state of the electric device is represented by a plurality of types such as failure A, failure B, or normal. The partial discharge diagnosis device 100 functions as a device including a communication unit 101, an output unit 102, a learning data storage unit 103, a classifier storage unit 104, and a control unit 105 by executing a partial discharge diagnosis program.

Electrical equipment is composed of equipment such as circuit breakers, disconnectors, current transformers, and transformers (all not shown). Electrical equipment energizes high voltage and large current from the outside via a power cable. The electric device has a function of shutting off the energization in the event of an abnormality. The electric device may be any device as long as it may generate a partial discharge, such as a power transformer, a gas-insulated switch, a generator, a motor, or a reactor. The box body is, for example, a power board such as a switch gear. A grounding electrode is connected to the lower part of the box.

The communication unit 101 is a network interface. The communication unit 101 communicates with an external communication device via a network. The communication unit 101 may communicate by a communication method such as a wireless LAN (Local Area Network), a wired LAN, Bluetooth (registered trademark) or LTE (Long Term Evolution) (registered trademark).

The output unit 102 outputs data to the user of the partial discharge diagnostic device 100 via an output device (not shown) connected to the partial discharge diagnostic device 100. The output device may be configured by using, for example, a device that outputs an image or characters to the screen. For example, the output device can be configured by using a CRT (Cathode Ray Tube), a liquid crystal display, an organic EL (Electro Luminescence) display, or the like. Further, the output device may be configured by using a device that prints (prints) an image or characters on a sheet. For example, the output device can be configured by using an inkjet printer, a laser printer, or the like. Further, the output device may be configured by using a device that converts characters into voice and outputs the characters. In this case, the output device can be configured by using a voice synthesizer and a voice output device (speaker). The output device may be configured by using a light emitting device such as an LED (Light Emitting Diode). The output unit 102 may transmit the determination result to another information processing device via the communication device provided in the partial discharge diagnostic device 100. The user may be any person who may use the partial discharge diagnostic apparatus 100, such as a diagnostician who diagnoses an electric device.

The learning data storage unit 103 is configured by using a storage device such as a magnetic hard disk device or a semiconductor storage device. The learning data storage unit 103 stores one or more learning data. The training data is used to generate the classifier. FIG. 2 is a diagram showing a specific example of the learning data of the first embodiment. The training data has a data number, data, and a state in association with each other. The data number is identification information of the learning data stored in the learning data storage unit 103. The data is data relating to a partial discharge signal generated from a predetermined electrical device. The data may be represented by a waveform of a partial discharge signal, or may be represented by an image such as a φ−q−n pattern. The data may be of any format as long as it is possible to know what kind of discharge signal appears in which phase. In FIG. 2, the data is data represented by a waveform. The state represents the state of a predetermined electric device. The state represents any of the states that the electric device can take, such as failure A, failure B, or normal. The learning data storage unit 103 may store the learning data in advance. Further, when the learning data storage unit 103 does not store the learning data, the learning data storage unit 103 may store the learning data received from the external device.

Returning to FIG. 1, the description of the partial discharge diagnostic apparatus 100 will be continued. The classifier storage unit 104 stores the classifier. The classifier is generated by the classifier generator 106. The classifier identifies the state of the electrical device based on the input evaluation data. Specifically, the classifier discriminates, for example, the state of an electric device into any one of a plurality of states. The classifier outputs the status of the identified electrical device. The plurality of states may be states possessed by the training data used to generate the classifier.

The input evaluation data is generated by the evaluation data generation unit 107. The evaluation data is data relating to a partial discharge signal of an electric device. The evaluation data is data used for evaluation regarding the state of the electrical equipment to be diagnosed. The evaluation data may be represented by a waveform of a partial discharge signal, or may be represented by an image such as a φ−q−n pattern. The evaluation data is data in the same format as the data column of the training data. For example, when the data column of the training data is the waveform of the partial discharge signal, the evaluation data is represented by the waveform of the partial discharge signal. The discriminator storage unit 104 may store the discriminator generated by the discriminator generation unit 106, or may store a discriminator that has been learned in advance. The evaluation data may be data that shows what kind of intensity and what kind of discharge signal appears in which phase.

FIG. 3 is a diagram showing a specific example of the evaluation data of the first embodiment. The evaluation data of FIG. 3 is represented by a waveform of a partial discharge signal. In the evaluation data of FIG. 3, the horizontal axis represents time. In the evaluation data of FIG. 3, the vertical axis represents the signal strength. The evaluation data is not limited to the aspect shown in FIG. For example, the evaluation data may be in any format as long as it has the same format as the data column of the training data used to generate the discriminator.

The control unit 105 controls the operation of each unit of the partial discharge diagnostic apparatus 100. The control unit 105 is executed by a device including a processor such as a CPU (Central Processing Unit) and a RAM (Random Access Memory), for example. The control unit 105 functions as a discriminator generation unit 106, an evaluation data generation unit 107, a state identification unit 108, a credit information generation unit 109, and a diagnosis result generation unit 110 by executing a diagnosis program.

The classifier generation unit 106 generates a classifier based on the learning data stored in the learning data storage unit 103. Specifically, the classifier generation unit 106 generates a classifier by using predetermined machine learning. Predetermined machine learning is a learning algorithm capable of outputting an intermediate layer such as a neural network. The predetermined machine learning may be, for example, R-CNN (Region-based CNN). The predetermined machine learning may be, for example, LSTM (Short-Term Memory). The predetermined machine learning may be, for example, Fast R-CNN. The classifier generator 106 may use any machine learning as long as it can output the intermediate layer. The classifier generation unit 106 records the generated classifier in the classifier storage unit 104.

The evaluation data generation unit 107 generates evaluation data based on the acquired discharge signal. Specifically, the evaluation data generation unit 107 acquires a discharge signal from the sensor. The sensor is provided on the surface of the box containing the electrical equipment to be diagnosed. The sensor measures a physical quantity related to a partial discharge generated from an electric device. The sensor may be, for example, a CT (Current Transformer) sensor. The sensor may be, for example, an AE (Acoustic Emission) sensor. The sensor may be, for example, TEV (Transient Earth Voltage. The sensor may be any sensor as long as it can measure a physical quantity from a partial discharge. The physical quantity may be, for example, potential, current, electromagnetic wave, or the like. Any physical quantity generated by partial discharge such as vibration or sound may be used. The evaluation data generation unit 107 generates evaluation data based on the acquired physical quantity. The evaluation data generation unit 107 generates training data. Data in the same format as the data stored in the storage unit 103 is generated as evaluation data. For example, the evaluation data generation unit 107 may generate a partial discharge waveform as evaluation data. The evaluation data generation unit 107 may generate the evaluation data. , A scatter diagram (Φ−q plot) may be generated as evaluation data. In this case, the evaluation data generation unit 107 generates the phase (Φ) of the power supply voltage applied to the partial discharge signal for a certain period. The horizontal axis and the partial discharge charge amount (q) are the vertical axes. The evaluation data generation unit 107 quantizes the axes of the scatter diagram (Φ−q plot) and expresses the occurrence frequency n corresponding to each axis. The −q−n pattern may be generated as evaluation data. The evaluation data is a specific example of the discharge data. The discharge data is data relating to a partial discharge signal generated from the electrical device to be diagnosed.

The state identification unit 108 identifies the state of the electric device based on the evaluation data and the classifier. Specifically, the state identification unit 108 acquires the classifier from the classifier storage unit 104. The state identification unit 108 inputs the generated evaluation data to the classifier. When the evaluation data is input to the classifier, the classifier outputs the status of the electric device. The state identification unit 108 outputs the state of the electric device output by the classifier to the diagnosis result generation unit 110. The state identification unit 108 is a specific example of the state diagnosis unit. The state diagnosis unit diagnoses the state of the electrical device to be diagnosed based on the classifier and the evaluation data.

The credit information generation unit 109 generates information regarding the reliability of the state of the electric device identified by the state identification unit 108. The credit information generation unit 109 outputs the generated reliability information to the diagnosis result generation unit 110. The credit information generation unit 109 includes a feature amount acquisition unit 191.

The feature amount acquisition unit 191 acquires the feature amount of the evaluation data or the learning data. The feature amount acquisition unit 191 outputs the acquired feature amount to the diagnosis result generation unit 110 as information regarding the reliability of the state of the electric device identified by the state identification unit 108. The feature amount is information obtained by performing a predetermined conversion on the evaluation data or the learning data. The predetermined conversion may be, for example, a process related to dimension reduction such as principal component analysis.

A specific example when the state identification unit 108 uses the neural network as the classifier will be described. First, the feature amount acquisition unit 191 inputs evaluation data or learning data into the classifier. Here, the classifier is a classifier stored in the classifier storage unit 104. The classifier is a classifier used to discriminate the state of electrical equipment. FIG. 4 is a diagram showing a specific example of the output of the intermediate layer of the neural network in the first embodiment. In FIG. 4, the classifier outputs an n-dimensional intermediate layer from z1 to zn. The feature amount acquisition unit 191 acquires the output of the n-dimensional intermediate layer with respect to the data input to the classifier. The feature amount acquisition unit 191 reduces the dimension of the acquired output of the n-dimensional intermediate layer. For example, the feature amount acquisition unit 191 may perform dimension reduction by using a dimension reduction method such as principal component analysis. For example, the feature amount acquisition unit 191 may acquire a two-dimensional feature amount composed of a first principal component and a second principal component by using principal component analysis. The feature amount acquisition unit 191 acquires the feature amount for each of the learning data stored in the learning data storage unit 103, the generated evaluation data, and the respective data. The feature amount acquisition unit 191 outputs the acquired feature amount to the diagnosis result generation unit 110. In the present embodiment, the feature amount acquisition unit 191 has acquired the two-dimensional feature amount, but the feature amount acquisition unit 191 is not limited to the two-dimensional feature amount. For example, the feature amount acquisition unit 191 may acquire a one-dimensional feature amount or a three-dimensional or higher-dimensional feature amount. The credit information generation unit 109 outputs the feature amount acquired by the feature amount acquisition unit 191 to the diagnosis result generation unit 110.

The diagnosis result generation unit 110 generates a diagnosis result of the electric device based on the identification result output by the state identification unit 108 and the feature amount output by the credit information generation unit 109. The diagnosis result is information in which the result of identifying the state of the electric device and the feature amount are associated with each other. The diagnosis result generation unit 110 outputs the generated diagnosis result to the output unit 102.

FIG. 5 is a diagram showing a specific example of the generated diagnostic result of the first embodiment. FIG. 5 includes an identification result 210 and a scatter plot 220. The diagnosis result generation unit 110 outputs an image in which the identification result 210 and the scatter diagram 220 are arranged side by side to the output unit 102 as the diagnosis result. The identification result 210 represents the identification result output by the state identification unit 108. According to the identification result 210, the electrical device to be diagnosed is identified in the state 1. Scatter plot 220 shows the feature amount output by the credit information generation unit 109. The diagnosis result generation unit 110 generates a scatter diagram 220 based on the feature amount output by the credit information generation unit 109.

Scatter plot 220 plots features acquired from evaluation data or training data. Both axes in Scatter Plot 220 are, but are not limited to, the first and second principal components in the principal component analysis. Also, the scatter plot is not limited to two dimensions. The scatter plot may be generated according to the number of dimensions of the acquired features. For example, when the feature quantity is three-dimensional, the scatter plot may be generated in three dimensions. Further, when the feature quantity has four or more dimensions, it may be represented by a matrix.

Scatter plot 220 includes a region 221 and a region 222. Region 221 represents the shape of the data plotted in scatter plot 220. According to region 221 the shape of the data is associated with the state. For example, the data associated with state 1 is plotted in a quadrangle in scatter diagram 220. For example, the data associated with state 2 is plotted in a circle in scatter plot 220. For example, the data associated with states 3-5 are plotted in scatter plot 220 with triangles of different orientations. Region 222 represents the color of the plotted data. The area 222 indicates that the evaluation data and the training data have different colors plotted in the scatter diagram 220. The display mode of each data shown in the scatter diagram 220 is not limited to these. The scatter diagram 220 may be drawn in a manner in which the shape and the color can be distinguished according to the content of the data.

The user can determine the credibility of the identification result with respect to the state of the electric device based on the diagnosis result output to the output unit 102. For example, a case where the evaluation data is plotted at the position A (evaluation data A) will be described. The evaluation data A is identified in state 1. The evaluation data A is plotted at the same position as the training data having the state 1. Therefore, the user can determine that the evaluation data A can be trusted for the result identified in the state 1.

The case where the evaluation data is plotted at the position B (evaluation data B) will be described. The evaluation data B is identified in state 1. The evaluation data B is plotted between the learning data having the state 1 and the learning data having the state 3. Therefore, it is considered that the evaluation data B may be identified in the state 3. The user can determine that the evaluation data B cannot be trusted for the result identified in the state 1.

The case where the evaluation data is plotted at the position C (evaluation data C) will be described. The evaluation data C is identified in state 1. The evaluation data C is plotted at the same position as the training data having the state 3. Therefore, it is considered that the evaluation data C was desirable data to be identified in the state 3. The user can determine that the evaluation data C cannot be trusted for the result identified in the state 1.

The case where the evaluation data is plotted at the position D (evaluation data D) will be described. The evaluation data D is identified in state 1. The evaluation data D is plotted at a location distant from the training data. Therefore, it is considered that the evaluation data D is not similar to the training data. The user can determine that the evaluation data D is unreliable for the result identified in the state 1.

FIG. 6 is a flowchart showing a specific example of the flow of processing for generating the classifier according to the first embodiment. The classifier is generated at a predetermined timing. The predetermined timing may be a timing designated in advance, or may be a timing in which a predetermined number of learning data is stored in the learning data storage unit 103. The classifier generation unit 106 acquires the learning data stored in the learning data storage unit 103 (step S101). The classifier generation unit 106 generates a classifier by performing predetermined machine learning on the acquired learning data (step S102). The classifier generation unit 106 records the generated classifier in the classifier storage unit 104 (step S103).

FIG. 7 is a flowchart showing a specific example of the flow of processing for generating the diagnosis result of the first embodiment. The diagnosis result may be generated at the timing when the discharge signal is acquired from the electric device. The generation of the diagnostic result may be performed after a predetermined period of time when the discharge signal is acquired. The generation of the diagnosis result may be performed at a predetermined timing. The diagnosis result may be generated at any timing after the discharge signal is acquired. The evaluation data generation unit 107 generates evaluation data based on the acquired discharge signal (step S201). The evaluation data generation unit 107 generates data in the same format as the data stored in the learning data storage unit 103 as evaluation data. The state identification unit 108 acquires the classifier from the classifier storage unit 104. The state identification unit 108 identifies the state of the electrical device based on the generated evaluation data and the acquired classifier (step S202).

The feature amount acquisition unit 191 acquires the learning data stored in the learning data storage unit 103. The feature amount acquisition unit 191 acquires a classifier from the classifier storage unit 104. The feature amount acquisition unit 191 acquires the feature amount of the evaluation data or the learning data (step S203). For example, the feature amount acquisition unit 191 inputs evaluation data or learning data to the classifier. The feature amount acquisition unit 191 acquires the output of the n-dimensional intermediate layer with respect to the data input to the classifier. The feature amount acquisition unit 191 reduces the output dimension of the n-dimensional intermediate layer based on the principal component analysis. For example, the feature amount acquisition unit 191 acquires a two-dimensional feature amount from the output of the n-dimensional intermediate layer. The feature amount acquisition unit 191 acquires the feature amount for each of the evaluation data and the learning data.

The diagnosis result generation unit 110 generates a scatter plot based on the feature amount (step S204). For example, when the feature amount is two-dimensional, the diagnosis result generation unit 110 generates a scatter diagram in which the first principal component and the second principal component are plotted on both axes. The diagnosis result generation unit 110 may generate a scatter plot according to the number of dimensions of the acquired feature amount.

The diagnosis result generation unit 110 generates a diagnosis result based on the identification result and the scatter plot (step S205). For example, the diagnosis result generation unit 110 may generate an image in which the identification result and the scatter diagram are arranged as the diagnosis result. The diagnosis result generation unit 110 outputs to the output unit 102. The output unit 102 outputs the diagnosis result (step S206).

In the partial discharge diagnostic apparatus 100 configured in this way, the state identification unit 108 identifies the state of the electrical device to be diagnosed based on the classifier and the evaluation data. The classifier is generated by performing predetermined machine learning using the training data. The learning data is data in which information on a signal of partial discharge generated from a predetermined electric device and information on a state of the predetermined electric device are associated with each other. Further, the credit information generation unit 109 generates information on the reliability of the identification result based on the output of the intermediate layer generated by the classifier. In this way, the partial discharge diagnostic apparatus 100 can provide information for determining whether or not the diagnosis result of the partial discharge is reliable.

Further, the feature amount acquisition unit 191 of the credit information generation unit 109 acquires the feature amount by performing a predetermined conversion on the evaluation data and the learning data. The feature amount acquisition unit 191 outputs the acquired feature amount to the diagnosis result generation unit 110. Therefore, the diagnosis result generation unit 110 plots the difference between the state of the identified electric device and the state associated with the learning data and the difference between the state evaluation data and the learning data based on the feature amount. Diagrams can be generated. In particular, when the feature amount acquisition unit 191 acquires the feature amount using the output of the intermediate layer of the neural network, the data converted for identification by the neural network can be used. Therefore, the feature amount acquisition unit 191 can acquire the feature amount more easily.

(Second Embodiment)
Next, the partial discharge diagnostic apparatus 100a according to the second embodiment will be described. The partial discharge diagnostic apparatus 100 in the first embodiment generates a scatter plot as information indicating the reliability of the identification result. The user determines whether the identification result is reliable or not based on the generated scatter plot. For this reason, the judgment criteria have become subjective. In the partial discharge diagnostic apparatus 100a in the second embodiment, the reliability of the identification result is calculated as numerical information. Therefore, the partial discharge diagnostic apparatus 100a can ensure the objectivity of the identification result.

FIG. 8 is a functional block diagram showing the functional configuration of the partial discharge diagnostic apparatus 100a of the second embodiment. The partial discharge diagnostic apparatus 100a in the second embodiment is different from the first embodiment in that the control unit 105a is provided instead of the control unit 105, but the other configurations are the same. Hereinafter, the points different from the first embodiment will be described.

The control unit 105a controls the operation of each unit of the partial discharge diagnostic apparatus 100a. The control unit 105a is executed by a device including a processor such as a CPU and a RAM. The control unit 105a functions as a classifier generation unit 106, an evaluation data generation unit 107, a state identification unit 108, a credit information generation unit 109a, and a diagnosis result generation unit 110a by executing a partial discharge diagnosis program.

The credit information generation unit 109a generates information regarding the reliability of the state of the electric device identified by the state identification unit 108. The credit information generation unit 109a outputs the generated reliability information to the diagnosis result generation unit 110a. The credit information generation unit 109a includes a feature amount acquisition unit 191, a specificity calculation unit 192, a certainty degree calculation unit 193, and a credit degree calculation unit 194.

The specificity calculation unit 192 calculates the specificity of the evaluation data. The specificity is information indicating how similar the evaluation data and the predetermined learning data group are. The greater the distance between the evaluation data and the predetermined learning data group, the smaller the similarity between the evaluation data and the training data. The predetermined learning data group is learning data having the same state as the identification result for the evaluation data among the learning data stored in the learning data storage unit 103. The specificity is calculated based on the evaluation data and a predetermined learning data group. The specificity may be quantified information. Specifically, the specificity calculation unit 192 calculates the specificity by inputting the learning data and the evaluation data into a predetermined algorithm. The predetermined algorithm is an algorithm for detecting outliers such as LOF (Local Outlier Factor). The information input to the predetermined algorithm is the features (for example, the first principal component and the second principal component) of the training data and the evaluation data. In this way, the specificity calculation unit 192 calculates how similar the evaluation data is from the learning data. Specificity is a specific example of similar index information. The similarity index information is information indicating how similar the evaluation data is to the learning data.

The certainty calculation unit 193 calculates the certainty of the identification result with respect to the evaluation data. The degree of certainty is information indicating the certainty of the identification result with respect to the evaluation data. Confidence is expressed by how far the evaluation data is from the identification boundaries of the state. The farther the evaluation data is from the identification boundary of the state and the closer to the set of data having the state, the greater the certainty of the identification result for the evaluation data. The state identification boundary represents the boundary of the region in which the training data having any of the states is plotted. The state identification boundary may be represented by, for example, a coordinate group. The coordinate group constitutes the boundary line of the state. The degree of certainty is calculated based on the evaluation data and predetermined learning data. The certainty may be quantified information. Specifically, the certainty calculation unit 193 calculates the certainty by inputting the learning data and the evaluation data into a predetermined algorithm. The predetermined algorithm is, for example, a pattern recognition algorithm such as SVM (Support Vector Machine). The information input to the predetermined algorithm is the features (for example, the first principal component and the second principal component) of the training data and the evaluation data. In this way, the certainty calculation unit 193 calculates the distance indicating how far the evaluation data is from the identification boundary of the state as the certainty. For example, with respect to the evaluation data located near the boundary between the two types of states, such as the evaluation data B in FIG. 5, the classifier may make a mistake in identifying the states. However, by calculating the certainty, the user can notice the error of the discriminator. Conviction is a specific example of distance index information. The distance index information is information indicating the distance between the diagnosis result and the state associated with the learning data.

The credit rating calculation unit 194 calculates the credit rating of the identification result with respect to the evaluation data. The credit rating is information that quantifies how much the identification result for the evaluation data can be trusted. Credit is calculated based on specificity and certainty. The credit rating may be quantified information. Specifically, the credit rating calculation unit 194 calculates the credit rating by multiplying the specificity and the certainty. In this way, the credit rating calculation unit 194 calculates the evaluation based on the specificity and the certainty as the credit rating. The credit information generation unit 109a outputs the calculated specificity, certainty, and credit rating to the diagnosis result generation unit 110.

The diagnosis result generation unit 110a generates a diagnosis result of the electric device based on the identification result, the feature amount, the specificity, the certainty, and the reliability. The diagnosis result is information in which the state, feature amount, specificity, certainty, and credibility of the electric device are associated with each other. The diagnosis result generation unit 110a outputs the generated diagnosis result to the output unit 102.

FIG. 9 is a diagram showing a specific example of the generated diagnostic result of the second embodiment. FIG. 9 includes identification result 210, evaluation data specificity 211, identification certainty 212, result credit 213 and scatter plot 220. The diagnosis result generation unit 110a outputs an image in which the identification result 210, the evaluation data specificity 211, the identification certainty degree 212, the result credit rating 213, and the scatter diagram 220 are arranged as the diagnosis result to the output unit 102. Since the identification result 210 and the scatter plot 220 have been described above, the description thereof will be omitted.

The evaluation data specificity 211 represents the specificity calculated by the specificity calculation unit 192. According to the evaluation data specificity 211, the specificity of the evaluation data is 0.8. The identification certainty degree 212 represents the specificity calculated by the certainty degree calculation unit 193. According to the discrimination confidence 212, the confidence of the discrimination result with respect to the evaluation data is 0.8. The result credit rating 213 represents the credit rating calculated by the credit rating calculation unit 194. According to the result credit rating 213, the credit rating of the identification result for the evaluation data is 0.64.

The diagnosis result generation unit 110a does not have to generate an image including all of the identification result 210, the evaluation data specificity 211, the identification certainty degree 212, the result credit rating 213, and the scatter diagram 220 as the diagnosis result. For example, the diagnosis result generation unit 110a may generate information including any one or more of the identification result 210, the evaluation data specificity 211, the identification certainty degree 212, the result credit rating 213, and the scatter plot 220 as the diagnosis result. Good. The diagnosis result generation unit 110a provides information including any one or more of the identification result 210, the evaluation data specificity 211, the identification certainty degree 212, the result creditworthiness 213, and the scatter plot 220 by means other than the image such as characters and symbols. It may be generated by.

FIG. 10 is a flowchart showing a specific example of the flow of processing for generating the diagnosis result of the second embodiment. The diagnosis result may be generated at the timing when the discharge signal is acquired from the electric device, after a predetermined period when the discharge signal is acquired, or at a predetermined timing. Good. The diagnosis result may be generated at any timing after the discharge signal is acquired. Since the processes in steps S301 to S303 are the same as the processes in steps S201 to S203 in FIG. 7, the description thereof will be omitted. Since the processes in steps S307 to S309 are the same as the processes in steps S204 to S206 in FIG. 7, the description thereof will be omitted.

The specificity calculation unit 192 calculates the specificity of the evaluation data (step S305). Specifically, the specificity calculation unit 192 calculates the specificity by inputting the learning data and the evaluation data into an algorithm for detecting outliers such as LOF. The certainty calculation unit 193 calculates the certainty of the identification result with respect to the evaluation data. Specifically, the certainty calculation unit 193 calculates the certainty by inputting the learning data and the evaluation data into a pattern recognition algorithm such as SVM. The credit rating calculation unit 194 calculates the credit rating of the identification result with respect to the evaluation data (step S306). Specifically, the credit rating calculation unit 194 calculates the credit rating by multiplying the specificity and the certainty.

In the credit information generation unit 109a of the partial discharge diagnostic apparatus 100a configured in this way, the specificity calculation unit 192 calculates the specificity of the evaluation data. The certainty calculation unit 193 calculates the certainty. The credit rating unit 194 calculates the credit rating. The diagnosis result generation unit 110a displays the specificity, certainty, and credibility. In this way, the user can more objectively judge whether or not the identification result of the state of the electric device is reliable based on the specificity, the certainty, and the reliability.

<Modification example>
The partial discharge diagnostic apparatus 100 is required to acquire a discharge signal at the installation location of the electric device. However, other functions do not have to be performed at the installation site of the electrical equipment. For example, the user arranges the evaluation data generation unit 107 at the installation location of the electric device. The evaluation data generation unit 107 may transmit the evaluation data to the partial discharge diagnostic apparatus 100 through a predetermined communication network such as the Internet. In this case, the partial discharge diagnostic device 100 can diagnose the electric device at a remote place away from the installation place of the electric device.

In addition, the user may arrange the evaluation data generation unit 107 and the output unit 102 at the installation location of the electric device. In this case, the evaluation data generation unit 107 may transmit the evaluation data to the partial discharge diagnostic apparatus 100 through a predetermined communication network such as the Internet. The partial discharge diagnostic device 100 diagnoses an electric device at a remote location away from the installation location of the electric device. The partial discharge diagnostic apparatus 100 outputs the diagnostic result to the output unit 102 via a predetermined communication network. With such a configuration, it becomes possible to carry out a load-intensive calculation process by an information processing device or the like installed in a remote place and having a higher processing capacity.

The partial discharge diagnostic apparatus 100 may be implemented by using a plurality of information processing apparatus connected so as to be communicable via a network. In this case, each functional unit included in the partial discharge diagnostic apparatus 100 may be distributed and mounted in a plurality of information processing apparatus. For example, the classifier generation unit 106 and the state identification unit 108 may be mounted on different information processing devices.

In each of the above embodiments, each functional unit included in the control unit 105 is a software functional unit, but it may be a hardware functional unit such as an LSI.

According to at least one embodiment described above, having the evaluation data generation unit, the state identification unit, and the credit information generation unit can provide reliability for the diagnosis result of partial discharge.

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, as well as in the scope of the invention described in the claims and the equivalent scope thereof.

100, 100a ... Partial discharge diagnostic device, 101 ... Communication unit, 102 ... Output unit, 103 ... Learning data storage unit, 104 ... Discriminator storage unit, 105, 105a ... Control unit, 106 ... Discriminator generation unit, 107 ... Evaluation Data generation unit, 108 ... State identification unit, 109, 109a ... Credit information generation unit, 110, 110a ... Diagnosis result generation unit, 191 ... Feature quantity acquisition unit, 192 ... Specificity calculation unit, 193 ... Confidence calculation unit, 194 … Credit calculation department

Claims (14)

A classifier and a diagnostic target generated by performing predetermined machine learning using learning data in which information on a partial discharge signal generated from a predetermined electric device and information on the state of the predetermined electric device are associated with each other. A state diagnosis unit that diagnoses the state of the electric device to be diagnosed based on the discharge data related to the partial discharge signal generated from the electric device.
A credit information generator that generates information on the reliability of the diagnosis result based on the state of the electrical device identified by the classifier.
An output unit that outputs information on the reliability and the result of the diagnosis to a predetermined output device, and
A partial discharge diagnostic device. A diagnostic result generation unit that generates a scatter plot plotting the discharge data and the learning data as information on the reliability based on the identification result is further provided.
The partial discharge diagnostic apparatus according to claim 1. The diagnosis result generation unit plots the learning data and the discharge data in different modes.
The partial discharge diagnostic apparatus according to claim 2. The diagnosis result generation unit plots the learning data and the discharge data in a different manner for each of the diagnosed states.
The partial discharge diagnostic apparatus according to claim 2 or 3. The credit information generation unit determines similar index information indicating how similar the discharge data is to the learning data based on the discharge data and the learning data.
The partial discharge diagnostic apparatus according to any one of claims 1 to 4. The credit information generation unit determines distance index information indicating the distance between the result of the diagnosis and the state associated with the learning data based on the discharge data and the learning data.
The partial discharge diagnostic apparatus according to claim 5. The credit information generation unit calculates information quantifying the information on the reliability based on the similar index information and the distance index information.
The partial discharge diagnostic apparatus according to claim 6. The classifier is generated by machine learning using a neural network.
The partial discharge diagnostic apparatus according to any one of claims 1 to 7. The credit information generator generates information on the reliability of the identification result based on the output of the intermediate layer of the neural network.
The partial discharge diagnostic apparatus according to claim 8. A partial discharge diagnostic device is generated by performing predetermined machine learning using learning data in which information on a partial discharge signal generated from a predetermined electric device and information on the state of the predetermined electric device are associated with each other. A state diagnosis step for diagnosing the state of the electrical device to be diagnosed based on the discriminator and the discharge data related to the partial discharge signal generated from the electric device to be diagnosed.
A credit information generation step in which the partial discharge diagnostic apparatus generates information regarding the reliability of the result of the diagnosis based on the state of the electrical device identified by the classifier.
An output step in which the partial discharge diagnostic device outputs information on the reliability and the result of the diagnosis to a predetermined output device.
A method for diagnosing partial discharge. A classifier generator that generates a classifier by performing predetermined machine learning using learning data that associates information on a partial discharge signal generated from a predetermined electrical device with information on the state of the predetermined electrical device. With,
Learning device. The learning device generates a discriminator by performing predetermined machine learning using learning data in which information on a partial discharge signal generated from a predetermined electric device and information on a state of the predetermined electric device are associated with each other. Has a classifier generation step,
Learning method. A classifier and a diagnostic target generated by performing predetermined machine learning using learning data in which information on a partial discharge signal generated from a predetermined electric device and information on the state of the predetermined electric device are associated with each other. A state diagnosis unit that diagnoses the state of the electric device to be diagnosed based on the discharge data related to the partial discharge signal generated from the electric device.
A credit information generator that generates information on the reliability of the diagnosis result based on the state of the electrical device identified by the classifier.
An output unit that outputs information on the reliability and the result of the diagnosis to a predetermined output device, and
A partial discharge diagnostic system. A computer program for operating a computer as the partial discharge diagnostic apparatus according to any one of claims 1 to 9.

Images