JP2997515B2 - Transformer protection system and protection method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a transformer protection system, and more particularly, to sample and obtain terminal current, terminal voltage, etc. of a transformer to be protected, and to learn in advance. The present invention relates to a protection system and a protection method for a transformer, which detects an internal fault of the transformer based on a relation between an operation state of the transformer and terminal current and terminal voltage information and outputs a trip signal of a circuit breaker.

[Conventional technology]

2. Description of the Related Art Conventionally, transformer protection at the time of abnormality such as short-circuiting of a winding of a transformer is performed by detecting a difference current between terminal currents of the transformer. Since the difference current is generated not only by the winding accident but also by the transformer's inrush, the protection system must prevent the malfunction due to the inrush by any method.

The conventional transformer protection system, which does not operate with the transformer inrush current but operates due to an internal accident, uses the terminal current of the transformer and the terminal current and terminal voltage of the transformer as input information. It is roughly divided into two.

A first conventional system that uses a terminal current of a transformer as input information is a so-called second harmonic suppression method that utilizes a property that a second harmonic component content in an inrush current is large. Transformer protection systems that prevent malfunction due to current are most often used. In this system, a difference current is derived from a terminal current of a transformer, a fundamental wave component and a second harmonic component in the difference current are obtained, and a second high-wave component content is determined.

A second conventional system using terminal current and terminal current of a transformer as input information is disclosed in Japanese Patent Application Laid-Open No. Sho 62-89.
No. 424 has proposed a digital method. This is because when the transformer is considered as a multi-terminal network and is expressed by an admittance equation, the transfer admittance at the time of excitation inrush and at the time of an internal fault is almost the same as that at the time of sound, whereas the drive point admittance or is derived by this. Based on the fact that the parallel admittance is greatly different between the time of inrush and the time of an internal accident, the following processing is performed. Since there is no suitable word to express the reciprocal of the inductance, this is referred to as admittance here.

(1) The transmission admittance of the transformer to be protected is obtained from the leakage inductance, and is stored in advance in the protection system as a constant.

(2) The terminal current and terminal voltage of the transformer to be protected are sampled at appropriate time intervals.

(3) A difference current is obtained from each sampled terminal current.

(4) When the difference current exceeds a predetermined detection level, it is an internal accident state or an exciting inrush state. Therefore, a driving point admittance or a parallel admittance is obtained from each current, voltage, and transmission admittance stored in advance, and it is determined whether the state is an internal accident state or an exciting inrush state according to the magnitude of the value.

(5) When the determination of the internal accident state continues for a predetermined number of samples, a command to open the circuit breaker is given to disconnect the transformer from the power supply.

[Problems to be solved by the invention]

However, the conventional technique has the following problems.

The first prior art has a problem that malfunction occurs when the content of the second harmonic component in the inrush current of the magnet decreases. Due to the high performance of transformer core materials and advances in transformer manufacturing technology, the operating flux density of the transformer core has been increased, and the difference from the saturation flux density is small. The content of the second harmonic component tends to decrease, and such malfunctions are likely to occur.

Furthermore, if the terminal voltage of the transformer rises transiently and the core of the transformer saturates on both the positive side and the negative side, the content of the even harmonic component in the difference current is low. However, there is a problem that a malfunction cannot be avoided.

Such a problem stems from the fact that only the fundamental component and the second harmonic component in the difference current have a small amount of information in order to determine the magnetic flux saturation of the transformer core and the internal fault.
Therefore, it is conceivable to use other harmonic components. However, obtaining various harmonic components causes an increase in calculation time. Further, for each harmonic component, it is necessary to know the relationship between various transformer core saturation phenomena and internal accident phenomena, but such a relationship is not clear in the current technology.

In the second prior art, the terminal voltage of the transformer is also required as input information. However, the problem of the first prior art can be solved, and a high-speed and high-reliability transformer protection system can be constructed. Since harmonic components are not used for discriminating between the excitation inrush phenomenon and the internal fault phenomenon, even if the fault current includes low-order harmonic components near the second harmonic in the system condition, the internal fault is correctly detected. There are features that can be detected.

However, since the driving point admittance and the parallel admittance are obtained by special arithmetic expressions, special arithmetic software is required for this, and the calculation takes time. Further, in order to determine the presence or absence of an internal accident, it is necessary to grasp in detail the values of a plurality of drive point admittances and parallel admittances in various operating states in advance. Requires knowledge.

The object of the present invention is to eliminate the above-mentioned disadvantages of the prior art,
An object of the present invention is to provide a highly reliable transformer protection system that can effectively and effectively utilize input information and can easily and objectively set an internal accident detection function.

[Means for solving the problem]

The present invention, in order to achieve the above object, to detect the terminal current of the transformer to be protected or the difference current obtained from the terminal current,
In a transformer protection system that determines the presence or absence of an internal fault in the transformer based on the obtained current information and outputs a trip signal of the circuit breaker, the terminal current or the terminal current in a specific operating state of the transformer to be protected is obtained. A plurality of types of samples of the difference current information obtained are learned in advance in correspondence with each operation state, and when the difference current information obtained from the terminal current or the terminal current of the transformer is input, an output corresponding to the learning result is output. The operating state simulating means to be obtained, the input means for inputting the terminal current of the transformer or the difference current information obtained from the terminal current to the operating state simulating means, and the occurrence of an internal fault in the transformer is determined from the output signal of the operating state simulating means And output means for outputting a trip signal of the circuit breaker of the transformer.

Further, in order to achieve the above object, the present invention detects a terminal current of a transformer to be protected or a difference current obtained from the terminal current, and determines whether there is an internal accident of the transformer based on the obtained current information. In a transformer protection system that outputs a trip signal of a circuit breaker, a terminal current in a specific operation state of a transformer to be protected or a plurality of types of samples of difference current information obtained from the terminal current is associated with each operation state. Operating state simulating means for obtaining an output corresponding to the learning result when terminal current of the transformer or difference current information obtained from the terminal current is input, and obtained from the terminal current or terminal current of the transformer. Input means for inputting the difference current information to the operating state simulation means, and one or more samples of terminal current or difference current information obtained from the terminal current in a specific operating state of the transformer The learning means for learning the operating state simulating means and the occurrence of an internal accident of the transformer are determined from the output signal of the operating state simulating means so as to obtain output signals of different modes for each current state. Output means for outputting a trip signal of the circuit breaker of the transformer.

Further, in order to achieve the above object, the present invention, when terminal current and terminal voltage information of the transformer to be protected is input,
In the transformer protection method, which includes an operation state simulation means for obtaining an output corresponding to the learning result and determines whether there is an internal accident in the transformer and outputs a trip signal of the circuit breaker, a specific operation of the transformer to be protected is specified. A plurality of types of samples of the terminal current and terminal voltage information in the state are made to correspond to each operation state, and the operation state simulation means learns, and the difference current obtained from the protection target transformer and the terminal current exceeds a predetermined magnitude. When the operating state determination means learns, input the terminal current and voltage information of the transformer to the learned operation state obtaining means, and obtain an output corresponding to the learning result. The occurrence is determined and a trip signal of the circuit breaker of the transformer is output.

[Action]

In the converter protection system according to the present invention, the following describes how each of the means functions.

In the neural circuit model as a specific example of the operating state simulating means, the learning means for determining a learning sample such as a terminal current or a terminal voltage in a specific operating state of the transformer to be protected as a connection strength in the circuit, the learning means comprises: The connection strength in the circuit is corrected by iterative calculation in the direction in which the deviation between the signal and the output signal decreases. Even when another learning sample is presented to the input layer, the in-circuit connection strength is corrected in the same manner so that the deviation between the corresponding teacher signal and the output signal decreases. Thus, a plurality of learning samples corresponding to a plurality of driving states can learn in the same neural network model in the same direction.

Therefore, as the teacher signal, it is sufficient to simply define the target output pattern so that the signal is output only to a specific portion of the output layer for each operating state such as an excitation rush or an internal accident. Can learn the driving state of the vehicle.

The output means for judging the operating state by using the operating state simulating means in which learning of various operating states has been completed in advance by the above means, detecting an internal accident and outputting a trip signal of the circuit breaker, When the input signal is converted according to the connection strength in the circuit determined at the time of state learning, the operating state is determined based on the part of the output layer that is output as a strong signal and the strength of the output signal, and the internal accident state is determined. Output a trip signal of the circuit breaker.

The conversion of the input signal consists of only a simple function operation and a multiply-accumulate operation, and eliminates the need for harmonic analysis and special operations as in the conventional method, so that the processing time is short and sufficient real-time processing is possible. is there.

〔Example〕

One embodiment of a transformer protection system according to the present invention, including a transformer main circuit to be protected, is shown in FIG. 1 and will be described below.

In FIG. 1, 11 and 12 are primary and secondary windings of a transformer to be protected, 21 and 22 are primary and secondary current transformers, and 31 and 32.
Is a primary and secondary voltage transformer, 41 and 42 are primary and secondary circuit breakers, and 1100 to 1500 is a transformer protection system 1000.

The transformer protection system 1000 includes the following units. That is, an input means 1100 for converting input information such as a terminal current and a terminal voltage of the protection target transformer detected by the current and voltage transformers to create operating state simulation data 1100 and learning data 1120; Learning sample storage means 1200 for storing a plurality of phenomena in a specific operating state as learning samples in a predetermined manner based on the learning data 1120, and a plurality of learning sample information stored in the learning sample storage means 1200. Learning management means for learning 1210 to the neural circuit model 1300 as a driving state simulation means described later in a predetermined procedure
1400 and the learning input data 1410 presented from the learning management means 1400 are stored as in-circuit connection strength, and are recalled based on the in-circuit connection strength using the driving state simulation data 1110 input from the input means 1100. Neural circuit model as the driving state simulating means for outputting signal 1310
1300 and the occurrence of an internal accident in the transformer to be protected is determined based on the recall signal 1310 and the trip signal 1510 of the circuit breakers 41 and 42 is determined.
And an output means 1500 for outputting

The protection system of the transformer can be constituted by, for example, a micro computer system having hardware such as an arithmetic device, a storage device, and an input / output device.
The function of each unit described above is constituted by these hardware and a supplied program.

The neural network model 1300 as the driving state simulation means includes, for example, a function of defining an input layer, an intermediate layer, an output layer and a connection network thereof in a memory space, a calculation function at the time of learning and a driving state simulation, and a circuit. And a function of storing and holding a calculation result such as an internal connection strength distribution, and is configured by a program and a memory. The number and arrangement of the input layer, the intermediate layer, and the like are determined from outside as structural parameters.

The neural circuit model 1300 of the present embodiment can store each function constituting the neural circuit model 1300 by storing it as a program and data. If the state once learned is fixed, it may be replaced with circuit elements and configured as hardware. That is, a neural circuit model IC may be used.

Next, a neural circuit model 1300 as a driving state simulation means
The learning management means 1400 will be described.

FIG. 2 shows one unit model 1320 constituting the neural network model 1300.

Here, the input signals x ₁ to Yunitsuto, x _2, the ... x _n, takes the value range (0, 1), synaptic weights _{w 1, w 2, ... w} n are range (-
∞, + ∞). Here, if the input u _i transmitted from the i-th input signal x _i to the unit is u _i = x _i x _i (1), the total input U to the unit is Becomes

The unit output y is Defined by Where Uo is a bias in the equation (3).

In this embodiment, the unit model 1320 described above is the third model.
As shown in the figure, a neural circuit model 1300 is arranged in a layered manner and the output signal from each unit 1320 is used as an input signal to each unit 1320 in the next layer.

Regarding the unit model 1320 and the neural circuit model 1300, see the MIT Press, Niuro Computing Foundations of Research, 198
8 years, pages 318 to 362 (The MIT Press, Neurocomput
ing Foundations of Research, 1988, PP 318-362).

In this paper, as shown in FIG. 3, when a certain input signal pattern 1330 is given to the input layer, the output signal pattern 1340 from the output layer becomes a desired signal pattern, that is, a teacher signal pattern 1350. A learning algorithm (referred to as backpropagation) for correcting the connection strength of the input section to each unit of the intermediate layer and the output layer, that is, the synaptic weight, according to the error between them, is shown. Also in the learning management means 1400 of the present embodiment, the learning algorithm itself uses the back propagation shown in the above-mentioned paper.

FIG. 4 specifically illustrates the algorithm of the back propagation. This figure focuses on the k-th output signal y _3k of the output layer to make the algorithm easy to understand, and shows a synapse weight correction procedure for matching this with the teacher signal _ytk .

Hereinafter, the algorithm shown in FIG. 4 will be specifically described.

First, the error e _k between the k-th output signal y _3k and the teacher signal y _tk
Is e _k = y _tk −y _3k (4). Assuming that the degree of influence of the error on the operation level U _{3k of the} unit is d _3k , d _3k is represented by d _3k = e _k f′3 _k (U _3k ) (5). Here, f '(U) is as follows.

Accordingly, the correction amount Δw _3k , _2i (N +) of the synapse weights w _3k , _{2i at} the j-th input section in the k-th unit of the output layer
1) is represented by the following equation.

_{Δw 3k, 2i (N + 1} ) = ηd 3k y 2i ... (7) where, N represents a symbol indicating the last time, eta is referred to as a learning constant. Y _2i indicates the j-th output signal of the intermediate layer. However, in order to realize stable convergence, the correction amount obtained by the equation (7) is not used as it is, but is corrected by a method expressed by the following equation, and a new synapse weight w _3k , _2i (N + 1) is calculated by the following equation. Get like.

w _3k , _2i (N + 1) = w _3k , _2i (N) + Δw _3k , _2i (N + 1) + αΔw _3k , _2i (N) (8) where α is referred to as a stabilization constant.

As described above, the method of correcting the synapse weight of the input section of the output layer has been described.

Next, a method of correcting the synapse weight of the input unit of the hidden layer will be described. FIG. 4 shows a method of correcting the synapse weights w _2i and _1i in the i-th input section of the j-th unit in the intermediate layer.

In this case, the degree of influence d _2i of the error on the operation level U _2i of the unit should be determined in consideration of the error of all the unit outputs of the output layer. It is represented by Accordingly, the correction amount Δw _2i , _1i of the synapse weight at the i-th input unit in the j-th unit of the hidden layer
(N + 1) is represented by the following equation.

Δw _2i , _1i (N + 1) = ηd _2i y _1i (10) where N is a symbol indicating the previous time, and η is a learning constant. Also, y _1i indicates the i-th output signal of the input layer. However, as in the case of the output layer, in order to achieve stable convergence, the correction amount obtained by Expression (10) is not used as it is, but is corrected by the method expressed by the following expression, and a new synapse weight w _2i , _1i (N
+1).

w _2i , _1i (N + 1) = w _2i , _1i (N) + Δw _2i , _1i (N + 1) + αΔw _2i , _1i (N) (11) where α is a stabilization constant.

The error e _k can be minimized by repeating the arithmetic processing of the above equations (4) to (11). That is, the output signal pattern from the output layer can be made to match the teacher signal pattern, and as a result, the input signal pattern is stored (learned) as a synapse weight distribution (ie, in-circuit connection strength distribution) in the neural circuit model. Will be.

Also, if another input signal pattern is presented to the input layer, and the teacher signal pattern also presents another pattern corresponding to this, the above algorithm operates and is stored (learned) as a new synapse weight distribution. Become.

By using such an algorithm, it is possible to store a plurality of learning samples in the same neural network model. Also, if the neural circuit learned in this way is used,
Even when an unlearned pattern is input, an output signal pattern corresponding to the most similar one of the learned patterns is obtained from the output layer.

In the above method, information such as current and voltage in a specific operation state of the transformer to be protected is associated with the operation state,
Train the neural network model 1300.

Next, a method of converting input information such as a terminal current and a terminal voltage of the transformer to be protected by the input unit 1100 to generate the operating state simulation data 1110 and the learning data 1120 will be described.

First, several methods will be described for the case where a terminal current of a transformer to be protected is used as input information.

In the first method, the terminal current of the transformer to be protected is sampled at predetermined time intervals and synthesized to obtain a difference current, and a plurality of samples of the difference current, for example, data for one cycle of the system frequency, The operating state simulation data 1110 is used. Since the data of one cycle of the difference current contains not only the second harmonic component but also information of many harmonic components as in the conventional method, by using this data,
It is possible to distinguish the excitation inrush phenomenon and the internal accident phenomenon with higher accuracy.

The second method is to use each terminal current itself instead of the difference current in the first method. Since each terminal current contains the operation information of the transformer more than the difference current, it can be expected that the operation state of the transformer can be determined with higher accuracy. Operating state simulation means 1300
In order to improve the learning efficiency, it is desirable to convert each terminal current into an arbitrary reference number of turns and obtain operating state simulation data 1110.

In creating the operating state simulation data, it is optional to use the current sampling data from which time point. However, for that purpose, it is necessary to make the neural circuit model 1300 as the driving state simulation means learn many cases corresponding to this. Therefore, it is desirable to determine in advance from which point the current sampling data should be used in order to improve the learning efficiency. For example, sampling data from a specific phase of current or voltage is used.

Since a difference current occurs when an internal accident occurs, one specific method is to use sampling data from the time when the difference current exceeds a predetermined magnitude. By doing so, there is no need to learn an operating state in which a difference current is not generated, and there is neither an internal accident nor magnetic saturation of the iron core, so that learning efficiency and operating state simulation accuracy can be improved.

How to assign the operation state simulation data 1110 for the plurality of samples to each unit 1320 of the input phase in the neural circuit model 1300 shown in FIG. 3 is arbitrary, as long as it corresponds to the learning method. Good.

The operating state simulation data 1110 is updated for a plurality of predetermined samples, for example, every cycle. It is desirable that the operating state simulation data 1110 be normalized to a value range from 0 to 1 or a value range from -1 to +1.

Next, several methods will be described for the case where the terminal current and the terminal voltage of the transformer to be protected are used as input information.

In the first method, in the above-described method using the terminal current of the transformer to be protected, sampling data of the terminal voltage is added to the operation state simulation data 1110. Since the amount of information is increased by the addition of the terminal voltage information, the simulation accuracy of the transformer operating state is improved, and there is an advantage that data for one cycle is not necessarily required as in the method using only terminal current information. .

A more effective method of creating the operating state simulation data 1110 will be described below in association with the voltage and current relational expressions of the transformer.

FIG. 5 shows a schematic diagram of a two-winding transformer. In FIG. 5, 1 is a two-winding transformer, 10 is an iron core, and 11 and 12 are primary and secondary windings. v ₁ , v ₂ , and i ₁ , i ₂ indicate the primary and secondary terminal voltages, and the primary and secondary terminal currents, but are converted to arbitrary reference turns for simplicity of the following description. And Generally, the resistance of a power transformer is negligible compared to the inductance.
There is the following relationship:

In the equation (12), L ₁₁ and L ₂₂ are self-inductances of the primary and secondary windings, and L ₁₂ = L ₂₁ is a mutual inductance between the primary and secondary windings. Therefore, the characteristics of the transformer are related to the terminal voltage and the time derivative of the terminal current.

Therefore, the second method using the terminal current and the terminal voltage of the transformer to be protected as input information is to obtain a time differential value of the terminal current, and to convert the current differential value information and the terminal voltage information into the operating state. That is, the simulation data 1110 is used. In the first method, the sampling data for about one cycle is required for the fault judgment. On the other hand, in this method, only a few samples of data for obtaining the current differential value are required. Can be done. Further, in the neural circuit model as the driving state simulating means shown in FIG. 3, since the number of units of the input phase is small, effective learning and determination can be performed.

In equation (13), y ₁₁ and y ₂₂ are driving point admittances, and y ₁₂ = y ₂₁ is a transmission admittance. Although these have dimensions of the reciprocal of the inductance, they are referred to as admittance for convenience here because there is no appropriate expression method. The transformer shown in FIG. 5 is represented by an equivalent circuit shown in FIG. 6, and is related to equation (13). Note that y ₁₀ and y ₂₀ are parallel admittances, There is a relationship. As described above, the characteristics of the transformer are related to the integral value of the terminal voltage and the terminal current.

Therefore, a third method using the terminal current and the terminal voltage of the transformer to be protected as the input information is to calculate the integral value of the terminal voltage, and simulate the operating state of the voltage integrated value information and the terminal current information. Data 1110. In the second method, the current differential value is obtained. Generally, the current differential value of the power system is largely affected by high-frequency noise having a sharp rise. On the other hand, this method using the integrated value of the voltage has a feature that it is hardly affected by high frequency noise.

As the input device, in the above-described method using the terminal current and the terminal voltage information of the transformer to be protected, the sampling data when the difference current obtained from the terminal current exceeds a predetermined value is used. Since there is no need to learn an operating state that is neither an internal accident nor a magnetic saturation of the iron core, it is effective in improving learning efficiency and operating state simulation accuracy.

In particular, in the third method, if the integrated value up to the present time starting from the time when the difference current exceeds a predetermined value is used as the integrated value of the terminal voltage, the operating state of the transformer changes. Since only the necessary and sufficient information after use is used, it can be correctly determined in a short time whether the current time is an internal accident state or a magnetic saturation state of the iron core.

Next, the internal accident determination method in the output means 15000 is made to correspond to the setting method of each unit 1320 of the input layer and the output layer of the neural network model 1300 as the driving state simulating means.
The case of a winding transformer will be described as an example.

FIG. 7 shows a schematic diagram of a three-winding transformer, and FIG. 8 shows an equivalent circuit thereof.

In FIG. 7, 1 is a three-winding transformer, 10 is an iron core, and 11,1
Reference numerals 2 and 13 are primary, secondary and tertiary windings. v ₁ , v ₂ , v ₃ and
i ₁ , i ₂ , i ₃ denote the primary, secondary, and tertiary terminal voltages and the primary, secondary, and tertiary terminal currents.
It is assumed that it has been converted to an arbitrary reference number of turns.

FIG. 8 shows an equivalent circuit of the three-winding transformer shown in FIG. In FIG. 8, y ₁₀ , y ₂₀ and y ₃₀ are parallel admittances. Although the value of each parallel admittance changes depending on the internal state of the transformer, as in the case of the two-winding transformer described above, the aspect of these changes depends on the differential value of each terminal current and the terminal voltage or the terminal voltage. It is related to the integral value and the terminal current.

FIG. 9 shows a configuration example of the windings of the three-winding transformer. FIG. 9 is a schematic cross-sectional view of the interior of the transformer, where 10 is an iron core, 11, 12, and 13 are primary, secondary, and tertiary windings, and each winding is concentrically wound around the iron core. ing. FIG. 10 shows an example of how the parallel admittance changes in various operating states of the three-winding transformer having the winding arrangement as shown in FIG.

In Figure 10, (a) represents a characteristic at the time of magnetizing inrush, power-on phase theta, irrespective of the residual magnetic flux density B _R of the core, the value of each parallel Adomitsutansu is substantially constant.

(B), (c), and (d) show the characteristics in the event of a short-circuit fault in the primary winding, the secondary winding, and the tertiary winding, respectively. The horizontal axis represents the number of turns of the short-circuited portion with respect to the total number of turns of each winding. Indicates the ratio. The value of each parallel admittance differs depending on the type of the short-circuited winding and the number of short-circuited windings.

Therefore, for example, when the integral value of the terminal voltage of the transformer and the terminal current are given to the unit of the input phase 1320 in the neural circuit model 1300 as the operating state simulating means, in order to simulate the states of the above four cases, At least four units may be set as output phases, and each may correspond to the above four cases.

Specifically, at the time of learning the neural network model 1300 as the driving state simulation means, the input information at the time of the excitation rush is input signal 13
When 30 is set, the teacher signal 1350 is set so that the output signal 1340 is 1 for the first unit and 0 for all other units. Further, at the time of learning at the time of an accident inside the primary winding, the teacher signal 1350 is set as the output signal 1340 so that the signal of the second unit is 1 and all others are 0. Similarly, at the time of learning at the time of an accident inside the secondary winding or tertiary winding, the teacher signal 1350 is set as the output signal 1340 so that the signal of the third or fourth unit is 1 and all others are 0. .

On the other hand, if an accident occurs in a system outside the transformer, a large current may pass through the transformer, causing a considerable difference current due to errors in the current transformer. is there. A unit corresponding to such an operation state can be additionally provided in the output layer of the neural circuit model 1300 and learned, which is effective in further improving the determination system.

By learning as described above, the output means 1500 determines which unit output signal of the output layer of the neural circuit model 1300 is close to 1 and which unit output signal is close to 0. It is possible to know whether or not an internal accident has occurred, and to know the location of the internal accident. When the occurrence of an internal accident is detected, the trip signal of the circuit breakers 41 and 42 is output.

Also, if the output signal pattern is stored in an appropriate form, the situation at the time of an accident can be correctly grasped.
Effective post-accident handling can be performed.

Conversely, if only an internal fault is detected, it is not necessary to provide a unit for each fault winding in the output layer of the neural circuit 1300. Reducing the number of units in the output layer has the effects of simplifying the configuration of the neural network model 1300, increasing learning efficiency, and shortening the determination time during driving.

When determining with a small number of sample data, such as when using the differential value of the terminal current and the terminal voltage, the tripping signal of the circuit breaker is not output immediately after one internal accident determination, but to prevent malfunction. It is desirable to output when the internal accident determination is continued a plurality of times.

In addition, outputs 1510 through 1510 can be combined with the outputs of other suitable relay elements to increase the reliability of the circuit breaker operation. For example, the trip signal of the circuit breaker is determined based on the AND condition of the difference current and the output of the passing current suppression relay element.

The specific configuration of each means in the present invention has been described above.

Next, in a three-winding transformer, an example in which an integral value of a terminal voltage and a terminal current value are used as input information will be described.
FIG. 11 shows an example of the operation flow of the entire transformer protection system according to the present invention, and the contents will be described for each step.

First, in step S _1, the terminal at time t voltage v
₁ (t), v ₂ (t), v ₃ (t) and terminal currents i ₁ (t), i ₂
(T), i ₃ (t) are sampled. It is assumed that each voltage and current is converted into an arbitrary reference number of turns.

In step S2, the difference current Σi (t) is obtained from equation (15).

In _{Σi (t) = i 1 (} t) + i 2 (t) + i 3 (t) ... (15) step S _3, the absolute value is judged whether large or below a predetermined detection level K ₁ of .sigma.i (t) . If it is smaller than K _1, the process proceeds to step S _11. If K ₁ or a transformer inrush or internal accident, the process proceeds to step S _4.

In step S4, it determines whether the absolute value of the difference current .SIGMA.i (t) is the time of first sampling large and has fallen below the detection level K _1. K ₁ continues from the previous sampling
If larger, the process proceeds to step S _6. If it is the first time,
Advances to step S _5.

In step S _5, the initial setting in the following manner. First t ₀
= T, and i ₁ (t ₀ ), i ₂ (t ₀ ), i ₃ (t ₀ ) are stored. Next, φ ₁ (t ₀ ), φ ₂ (t ₀ ), and φ ₃ (t ₀₎ are cleared as integration areas of the voltages v ₁ (t), v ₂ (t), and v ₃ (t).
After initial setup, the process proceeds to step S _12.

In step S _6, the current increment _{A 1 (t), A 2} (t), A 3 (t)
Is obtained from equation (16).

In step S _7, the voltage integral _{φ 1 (t), φ 2} (t), φ
₃ (t) is obtained from equation (17) by applying, for example, the trapezoidal rule.

In step S _8, and input to the learned neural circuit model the current increment, a voltage integral value, the output _{F 1 (t), F 2} (t),
F ₃ (t) and F ₄ (t) are obtained. Each output corresponds to, for example, an excitation rush and an internal fault of the primary winding, the secondary winding, and the tertiary winding, respectively.

In step _{S 9, S 10, F (} t), F 2 (t), F 3 (t), F 4
The possibility of occurrence of an internal accident is determined based on the value of (t).

And excisional judgment flag at step S ₁₂ when there is a possibility of an internal accident, and resetting the decision flag in step S ₁₁ when there is no possibility, both proceed to step S _13.

Step judges the continuity of the determination flag in S _13, S _14.

And accident determining in step S ₁₆ when there is continuity, when there is no continuity, integrity in S ₁₅ determines. When the accident determination at step S _16, Toritsupu signal of the circuit breaker, or outputs a Toritsupu allowable signal.

〔The invention's effect〕

As described above, according to the present invention, since the input information can be utilized effectively and maximally and the internal accident detection function can be set easily and objectively, a highly reliable transformer protection system can be constructed. There are features.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an overall configuration of a transformer protection system according to one embodiment of the present invention. FIG. 2 is an explanatory diagram showing one unit model constituting a neural circuit model as operating state simulation means. FIG. 3 is an explanatory diagram showing a basic configuration of a neural circuit model and a basic procedure of a learning algorithm. FIG. 4 is an explanatory diagram showing detailed means of a learning algorithm. FIG. 5 is a schematic diagram showing a two-winding transformer. 6 is an equivalent circuit diagram of a two-winding transformer, FIG. 7 is a schematic diagram showing a three-winding transformer, FIG. 8 is an equivalent circuit diagram of a three-winding transformer, and FIG. FIG. 10 is a diagram showing an example of parallel admittance characteristics of a three-winding transformer, and FIG. 11 is an embodiment of an operation method in the transformer protection system of the present invention. It is a flowchart which shows. 1 Transformer 41, 42 Circuit breaker 1000 Transformer protection system 1100 Input means 1200 Learning sample storage means 1300 Operating state simulation means 1400 Learning Management means, 1500 ... Output means.

Continuation of the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H02H 3/00 H02H 3/28 H02H ⁷ /04-7/045 H02J 3/00 G05B 13/02-13/04

Claims (3)

(57) [Claims] 1. A terminal current of a transformer to be protected or a difference current obtained from the terminal current is detected, the presence or absence of an internal fault in the transformer is determined based on the obtained current information, and a trip signal of a circuit breaker is output. In a transformer protection system, a plurality of types of samples of terminal current or difference current information obtained from terminal current in a specific operation state of a protected transformer are learned in advance in correspondence with each operation state, and the transformer is When the terminal current or the difference current information obtained from the terminal current is input, the operating state simulating means for obtaining an output corresponding to the learning result, and the difference current information obtained from the terminal current or the terminal current of the transformer, From the input means input to the operating state simulating means and the output signal of the operating state simulating means, it is determined whether an internal accident has occurred in the transformer and an output for outputting a trip signal of the circuit breaker of the transformer. Transformer protection system characterized in that it is constituted by a unit. 2. A terminal current of a transformer to be protected or a difference current obtained from the terminal current is detected, and the presence or absence of an internal fault in the transformer is determined based on the obtained current information to output a trip signal of a circuit breaker. In a transformer protection system, a plurality of types of samples of terminal current or difference current information obtained from terminal current in a specific operation state of a protected transformer are learned in advance in correspondence with each operation state, and the transformer is When the terminal current or the difference current information obtained from the terminal current is input, the operating state simulating means for obtaining an output corresponding to the learning result, and the difference current information obtained from the terminal current or the terminal current of the transformer, Input means for inputting to the operating state simulating means, and one or two or more samples of terminal current or difference current information obtained from the terminal current in a specific operating state of the transformer are sequentially given. Learning means for learning the operating state simulating means, and determining the occurrence of an internal accident of the transformer from the output signal of the operating state simulating means, so as to obtain an output signal of a different mode for each current state, Output means for outputting a trip signal of a circuit breaker of the transformer. 3. An operation state simulating means for obtaining an output corresponding to the learning result when terminal current and terminal voltage information of the transformer to be protected is input, and judges whether there is an internal accident of the transformer. In the method for protecting a transformer that outputs a trip signal of a circuit breaker, a plurality of types of samples of terminal current and terminal voltage information in a specific operation state of the transformer to be protected are associated with each operation state to simulate the operation state. When the difference current obtained from the transformer to be protected and the terminal current exceeds a predetermined magnitude, the learned operating state determination means inputs the terminal current and voltage information of the transformer. Obtaining an output corresponding to the learning result, determining the occurrence of an internal accident in the transformer from the output signal of the operating state determining means, and outputting a trip signal of the circuit breaker of the transformer. Transformer protection methods to symptoms.