JP2001083188A - Current detector and current detection method - Google Patents

Abstract

(57) [Problem] To provide a current detector and a current detection method capable of accurately detecting a current flowing through each conductor even when a plurality of conductors are arranged in parallel. And A plurality of measured conductor 10 ₁ to 10 _n, provided corresponding to each of the plurality of measured conductor, a plurality of magneto-electric conversion element for measuring the current flowing to each of the plurality of measured conductor 20 ₁ ２０20 _n and current measurement values obtained from a plurality of magneto-electric conversion elements are input as input patterns, and errors between an output pattern corresponding to the input patterns and a teacher signal which is a current value actually flowing through the plurality of conductors are corrected. And a neural network formed in the processor 30 for correcting current measurement values obtained from the plurality of magnetoelectric conversion elements based on the learning result and outputting the corrected current value as a detected current value.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a current detector for detecting a current flowing in an electric circuit mounted on a device such as an automobile, and more particularly to a current detector for detecting a current using a neural network and a current detecting method. .

[0002]

2. Description of the Related Art Conventionally, for example, in a motor vehicle, a current from a power supply is installed in an electric junction box for distribution, and a current flowing through a conductor housed in the electric junction box is detected by using a magnetoelectric conversion element. Current detectors are known.

In a current detector using this magnetoelectric transducer, a plurality of conductors are housed in an electric junction box.
There has been a problem that an accurate current cannot be detected due to interference of magnetic flux generated by a current flowing through a conductor other than the conductor through which the current to be detected flows.

Therefore, a conductor through which a current to be detected flows (hereinafter referred to as a conductor)
In order to prevent the measured conductor from being affected by the magnetic flux generated by the current flowing through the other conductors,
The arrangement position and direction of the magnetic core, the current path of the conductor and the magnetoelectric conversion element are devised. For example, JP-A-63-63
No. 974 discloses a current detecting device for electric wires.
As shown in, the conductor B which the detected current I ₁ flows are arranged substantially at right angles to the other conductors A, is disposed substantially at right angles to the further magnetic core 1a has the conductor B which the conductor B is penetrated ing.

With such a configuration, the magnetic fields H ₂ and H ₃ generated by the current I ₂ flowing through the other conductor A are canceled in the magnetic core 1a, and the magnetic field H ₁ generated by the current I ₁ flowing through the conductor B Only ₁ passes through the magnetic core 1a. Therefore, the magnetoelectric conversion element 1b placed in the gap (gap) of the magnetic core 1a, since there is no subject to interference from other conductors B, it is possible to accurately detect the current I ₁ flowing through the conductor B.

[0006]

However, in this conventional electric wire current detecting device, when it is desired to detect a current flowing through a plurality of conductors to be measured in an electric connection box, the arrangement of the conductors to be measured is complicated. There was a problem that would be.
Also, if there is an error in mounting the conductor to be measured or the magnetic core,
In other words, there is a problem that the influence of other conductors will occur unless the placement accuracy is improved.

An object of the present invention is to provide a current detector and a current detection method capable of accurately detecting a current flowing through each conductor even when a plurality of conductors are arranged in parallel.

[0008]

According to a first aspect of the present invention, there is provided a current detector, comprising: a plurality of conductors to be measured; and a plurality of conductors to be measured. Correspondingly provided, a plurality of magneto-electric conversion elements for measuring a current flowing through each of the plurality of conductors to be measured, and input a current measurement value obtained from the plurality of magneto-electric conversion elements as an input pattern, Learning to correct an error between the output pattern and a teacher signal that is a current value actually flowing through the plurality of conductors to be measured, based on the learning result, a current measurement value obtained from the plurality of magnetoelectric conversion elements. And a neural network for correcting and outputting the detected current value.

According to the current detector of the first aspect of the present invention, by learning so that the influence between the conductors to be measured can be corrected by using a neural network, the complexity of the conductor to be measured and the magnetoelectric conversion element can be improved. Without the need for simple installation
In addition, since errors in mounting and mounting the conductor to be measured, the magnetoelectric conversion element, and the like can be corrected, the current flowing through each of the conductors to be measured can be accurately detected.

A current detector according to a second aspect of the present invention is the invention according to the first aspect, further comprising a memory for storing a learning result, and another neural network different from the neural network. Another neural network inputs a current measurement value obtained from the plurality of magnetoelectric transducers as an input pattern, and outputs an output pattern for the input pattern and a teacher signal which is a current value actually flowing through the plurality of conductors to be measured. Learning, the learning result is stored in the memory, and the neural network calculates a current measurement value obtained from the plurality of magnetoelectric conversion elements based on the learning result stored in the memory. It is characterized in that it is corrected and output as a detected current value.

According to the current detector of the second aspect of the present invention, if a processor having normal performance is used as a neural network and a high-performance processor is used as another neural network, a large amount of data is generally obtained. Learning that requires time can be performed in a short time, and a neural network can be configured at low cost.

According to a third aspect of the present invention, in the current detector according to the first or second aspect, each of the neural network and the other neural network has the same number as the number of the conductors to be measured. An input layer having a plurality of input neurons and inputting current measurement values obtained from the plurality of magnetoelectric transducers as an input pattern to the plurality of input neurons, and a predetermined number of intermediate neurons; An intermediate layer in which each of the intermediate neurons is mutually connected to each of the input neurons; and a plurality of output neurons having the same number as the number of conductors to be measured. An output layer which is mutually connected to each neuron and outputs the output pattern corresponding to the input pattern from each output neuron; An error between the output pattern and the teacher signal is obtained, and based on the error, a load indicating the strength of connection between each of the input neurons and each of the intermediate neurons, and each of the intermediate neurons and the output A learning unit that changes a load indicating the strength of connection with each neuron for use and repeats the process of changing the load until the error becomes equal to or less than a certain value.

According to the current detector of the third aspect of the present invention, the current measurement values obtained from the plurality of magneto-electric conversion elements are input to the plurality of input neurons as an input pattern, and the intermediate neurons are input. Each neuron is interconnected with each input neuron, each output neuron is interconnected with each intermediate neuron, and each output neuron outputs an output pattern corresponding to the input pattern. An error between the error signal and the teacher signal is obtained, the load is changed based on the error, and the process of changing the load is repeated until the error becomes equal to or less than a certain value.

That is, by using a neural network or another neural network, learning is performed so that the influence (the above-mentioned error) between the conductors to be measured can be corrected, so that the complicated arrangement of the conductor to be measured, the magnetoelectric conversion element, and the like can be achieved. Since the mounting and mounting errors of the conductor to be measured, the magnetoelectric conversion element, and the like can be corrected without being required, the current flowing through each of the conductors to be measured can be detected with high accuracy.

According to a fourth aspect of the present invention, there is provided a current detecting method comprising:
The current flowing through each of the plurality of conductors to be measured is measured by a plurality of magnetoelectric transducers provided in one-to-one correspondence with the plurality of conductors to be measured, and the current measurement values obtained from the plurality of magnetoelectric transducers are used as neural signals. Learning is performed so as to correct an error between an output pattern corresponding to the input pattern and a teacher signal, which is a current value actually flowing through the plurality of conductors to be measured, based on the learning result. The current measurement value obtained from the magneto-electric conversion element is corrected and output as a detected current value.

According to a fifth aspect of the present invention, there is provided a current detecting method comprising:
In the invention according to claim 4, another neural network different from the neural network inputs current measurement values obtained from the plurality of magnetoelectric conversion elements as an input pattern, and outputs an output pattern corresponding to the input pattern and the plurality of received signals. Learning to correct an error with a teacher signal which is a current value actually flowing through the measurement conductor, storing the learning result in a memory, and the neural network, based on the learning result stored in the memory, The current measurement value obtained from the magneto-electric conversion element is corrected and output as a detected current value.

According to a sixth aspect of the present invention, there is provided a current detecting method comprising:
In the invention according to claim 4 or 5, each of the neural network and the other neural network is obtained from the plurality of magnetoelectric transducers to a plurality of input neurons of the same number as the number of conductors to be measured. The measured current value is input as an input pattern, a predetermined number of intermediate neurons are mutually coupled with the input neurons, and a plurality of output neurons of the same number as the number of conductors to be measured. Are mutually coupled with each of the intermediate neurons, the output pattern with respect to the input pattern is output from each of the output neurons, an error between the output pattern and the teacher signal is obtained, and based on the error, A load indicating the strength of connection between each of the input neurons and each of the intermediate neurons; Changing the load indicating the strength of the coupling between the neurons for serial output, and repeating the process of changing the 該荷 heavy until the error is below a predetermined value.

[0018]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a current detector and a current detecting method according to an embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a circuit diagram showing a configuration of a current detector according to an embodiment of the present invention. This current detector has n
(N is an integer of 1 or more) conductors to be measured 10 _i (i = 1,
, N), n magnetoelectric conversion elements 20 _i , a processor 30 and a memory 40. Among them, the n conductors to be measured 10 _i and the n magnetoelectric transducers 20 _i are housed in the electric connection box.

The n conductors to be measured 10 _i are conductors through which a current to be detected flows. In this embodiment, a conductor drawn from a power source is branched into n branches in an electric junction box, and each of the conductors is connected to a load. It is configured to be connected.
The connection of the conductor to be measured is not limited to the above-described branched form, but may be any form.

The magnetoelectric conversion element 20 _i is arranged near the conductor to be measured 10 _i . As the magnetoelectric element, for example, a Hall element, a magnetoresistive element, or the like can be used. The magnetoelectric conversion element 20 _i generates a voltage signal corresponding to the strength of the magnetic field generated by the current flowing through the conductor to be measured 10 _i ,
Supplied to the processor 30 as a current measurement value H _i. Accordingly, the processor 30, the n current measurements H _i generated by, respectively, respectively of n magneto-electric conversion element 20 _i is adapted to be supplied.

It should be noted that each of the conductors 10 _{i to} be measured may penetrate the magnetic core, and the magneto-electric conversion element 20 _i may be arranged in a gap between the magnetic cores. In this case, most of the magnetic flux generated by the current flowing through each of the conductors to be measured 10 _i passes through each magnetic core.
The magnetoelectric conversion element 20 _i links with more magnetic flux. As a result, the current detection sensitivity can be increased.

The processor 30 includes n magnetoelectric conversion elements 2
0 _i to process the n current measurements H _i inputted respectively from, respectively, and detecting a current flowing through each measured conductor on the basis of the n current measurements H _i. The processor 30 is constituted by, for example, a central processing unit (hereinafter, referred to as a “CPU”), and a neurocomputer is realized by using its function.

That is, the processor 30 has the configuration shown in FIG.
The neural network shown in FIG. This neural network has an input layer 3 having the same number of input neurons U as the number “n” of conductors to be measured.
1, an intermediate layer 32 having a predetermined number of intermediate neurons, each intermediate neuron being interconnected with each of the input neurons, and an output neuron of the same number as the number "n" of conductors to be measured And the output layer 33 mutually connected to the intermediate neuron, and a back propagation rule (error back propagation learning rule) is applied.

In the input layer 31, n current measurement values _Hi are input as input patterns corresponding to the number of input neurons equal to the number "n" of conductors to be measured. Output layer 3
3, n current detection values O _i are output as output patterns. The number of neurons in the middle layer 32 is 10 to 50
It is about.

From each neuron of the input layer 31 to the intermediate layer 32
And the strength of the connection from each neuron of the intermediate layer 32 to each neuron of the output layer 33 are defined by the weight _Wji . This load _Wji changes adaptively by learning.

FIG. 3 shows the interaction between neurons. Each neuron interacts with all other neurons via synaptic connections. As shown in FIG. 3, when the ith neuron is targeted, all the other neurons j (j = 1, 2, 3,... N,
However, j = i is excluded. ) And the input signal. Therefore, assuming that the load from the neuron j to the neuron i is W _ji (j = 1, 2, 3,... N), the input signal to the neuron i is obtained by the following equation.

[0028]

(Equation 1) Also, back propagation rule (error back propagation learning rule)
In, the output signal O _i (output pattern) of the neuron i is obtained using the weight and the threshold value and using a quasi-linear element model as shown in FIG. In this quasi-linear element model, the following function is used.

[0029]

(Equation 2) This function is a graph as shown inside the neuron in FIG.

In this case, the input value I _j of the continuous information from each neuron _j , the weight W _ji and the threshold value h _{i of the} neuron _i
Etc., the sum I _i of the input to the neuron i is obtained by the following equation.

[0031]

(Equation 3) With this I _i as a variable x, the calculated function value becomes the output O _i from the neuron i. That is,

(Equation 4) Becomes

The neural network further includes a learning unit 34. The learning unit 34 compares the output pattern consisting of n current detection values O _{i with the} teacher signal consisting of n signals to determine an error.

As the teacher signal, the conductor 10 to be measured is actually
_{The result} obtained by accurately measuring the current measurement value obtained from the magnetoelectric conversion element 20i when a current flows through _i by another means is used. Then, the load W is determined based on the obtained error.
_ji is updated in the order of the load W _ji between the output layer 33 and the intermediate layer 32 and the load W _ji between the intermediate layer 32 and the input layer 31.

The learning section 34 repeats the above-mentioned updating operation until the difference between the output pattern composed of n current detection values O _i and the teacher signal composed of n signals becomes equal to or smaller than a predetermined value. Such a process is a back propagation rule (error back propagation learning rule). Then, when the value falls below a certain value, the weight _Wji of the synaptic connection at that time is stored in the memory 40. Note that the memory 40 also stores data used when the processor 30 executes processing.

As described above, the neural network formed in the processor 30 is configured such that when n current measurement values _Hi are input from the n magnetoelectric transducers 20 _i , the weight of the synaptic connection stored in the memory 40 is obtained. _Wji and threshold h
By taking out _i and performing a predetermined calculation, n current detection values O _i are calculated. As a result, the n current detection values O _i are used as outputs of the current detector.

Next, the configuration and operation of the current detector according to this embodiment, that is, the current detection method will be described more specifically.

FIG. 5 shows an example of a learning pattern when learning is performed by the current detector. In FIG. 5, for simplicity, an example in which n = “3”, that is, the measured conductor 1
0 corresponds to _1, 10 respectively ₂ and 10 _3, examples of which are illustrated in the case of magneto-electric conversion element 20 ₁ formed of Hall elements 20 ₂ and 20 ₃ are provided.

This learning pattern includes an input pattern and a teacher signal. The input pattern includes current measurement values H ₁ , H _2, and H ₃ , which are current conversion values of the output of the Hall element, and is supplied to the input layer 31. The teacher signal is composed of current values (true values) T ₁ , T ₂ and T ₃ which actually flow through the conductors to be measured 10 ₁ , 10 ₂ and 10 ₃ , respectively.
It is supplied to the learning unit 34.

As for the Hall element, 2.50 V is the origin (0 A)
Has a dynamic range in both positive and negative directions according to the direction of the current. For example, 2.5 mV /
For a gain of A, the dynamic range is ± 100
It becomes 0A.

The input layer 3 of the actual neural network
The signal of the input pattern and the teacher signal input to 1 are scaled so that the entire dynamic range falls within the range of “−1” to “+1”. For example, if the dynamic range is ± 100 A, 10 A is “0.1”, −2
5A is given as "-0.25".

Next, the manufacturing process of this current detector is shown in FIG.
This will be described with reference to the flowchart shown in FIG.

First, the conductor to be measured 1 constituting the current detector
0 _i , the magnetoelectric conversion element 20 _i , and the magnetic core are mounted and attached (step S10). At this time, there may be some errors in mounting and mounting these components.

Next, a current is actually applied to the conductor to be measured 10 _i , and a voltage value output from the Hall element as the magnetoelectric conversion element 20 _i is measured (step S 11). here,
The pattern of the current actually passed through the conductor under test 10 _i is used as a teacher signal shown in FIG. 3, and is obtained by converting a voltage value obtained by measuring an output from each Hall element into a current value. It is used as the input pattern shown in FIG.

Next, a learning process is performed (step S1).
2). That is, the input pattern a current value obtained based on the output of the Hall element, learning is performed the pattern of current actually flowing in the measured conductor 10 _i as a teacher signal.

The details of the learning process will be described later. Next, the load W _ji and threshold h _i obtained by the learning is stored in the memory 40 (step S13). That is, the load W _ji of synaptic connections between neurons between neurons and the intermediate layer 32 of the input layer 31 is calculated, the load W of the synaptic connections between neurons during an intermediate layer 32 of the neuron and the output layer 33 _ji Is calculated and stored in the memory 40.

After the above processing is completed, the current detector is shipped (step S14). This current detector is housed in, for example, an electric junction box and mounted on an automobile. And when the operation starts, as described above,
Processor 30, the memory 4 in response to an input pattern consisting of the current measurement value H _i input from the magnetoelectric transducer 20 _i
Takes out a load W _ji and threshold h _i of synaptic connections stored in 0, calculates a current detection value O _i of each measured conductor 10 _i by performing a prescribed operation. The current detection value O _i calculated in this manner is supplied to, for example, a current breaker and used for current interruption control.

Next, the details of the learning process performed in step S12 of FIG. 6 will be described with reference to the flowchart shown in FIG. This learning process is performed based on the input pattern and the teacher signal shown in FIG. First, load W
_ji and threshold h _i is initialized (step S2
0).

[0048] Specifically, the load W _ji and threshold h _i of the memory 40 the initial values are set in an area that is stored. The initial value of the load W _ji and threshold h _i used in the initialization process, can be used, for example random numbers.

Next, the input pattern is input to the input layer 31 of the neural network (step S21). Thereby, each signal constituting the input pattern is propagated through the neural network and output from the output layer 33 as an output pattern.

Next, an error is calculated (step S22). That is, the learning unit 34 compares the output pattern composed of n current detection values O _i from the output layer 33 with the teacher signal composed of n signals to obtain a difference, and determines the difference as an error. I do.

Next, it is checked whether or not the error obtained in step S22 is equal to or smaller than a predetermined value (step S2).
3). Here, when it is determined that the error is not less than the certain value, the load W _ij and the threshold value _hj are changed based on the error (step S24). That is, while changing the strength of connection between each neuron of the input layer 31 and each neuron of the intermediate layer 32 and the strength of connection between each neuron of the intermediate layer 32 and each neuron of the output layer 33, the threshold value is changed. Change. Thereafter, the sequence returns to step S21.

The processes in steps S21 to S24 are repeatedly executed until it is determined in step S23 that the errors of all input patterns have become equal to or smaller than a predetermined value. Further, the processing of steps S21 to S24 is performed for all patterns (13 patterns in FIG. 5). And
When it is determined in step S23 that the errors of all the patterns have become equal to or smaller than a certain value, the learning process ends.

The learning process described above is performed for the conductor 10 to be measured.
_i , the magnetoelectric conversion element 20 _i , the magnetic core, etc. are all attached. Thus, each of the n current measurements H _i which is input to the processor 30, in which the influence of the neighboring conductors are reflected in the other.

This means that even when a plurality of conductors to be measured are provided side by side, learning is performed to eliminate the influence of neighboring conductors in the learning process, so that the current flowing through the conductor to be measured is reduced. This means that a correctly detected current detection value has been obtained.

In the learning process, the above-described process is repeatedly performed for many patterns, so that a large amount of time is generally required. However, cars usually have
In order to reduce the cost, a processor (for example, a microcomputer) with low performance is often mounted.

Therefore, the learning process is performed by another high-performance processor, and the load _Wij and the threshold value obtained by the learning process are stored in the memory 40 mounted on the vehicle. The neural network may be realized using a processor, and may be configured to detect a current.

[0057] Further, in the initialization process, rather than performing initialization by normal random number, already prepared typical load W _ji and threshold h _i already been learned, the learning process,
It is also possible to adopt a configuration in which only errors due to mounting and mounting are re-learned and only fine adjustment is performed.

As described above, according to the current detector and the current detection method according to the embodiment, a complicated calculation based on the distance between the conductors to be measured is not required, and the conductor to be measured, the magnetoelectric conversion element, the magnetic core, etc. Can be corrected in a form that includes all mounting errors and the like.

Further, complicated arrangement of the conductor to be measured, the magnetoelectric conversion element, the magnetic core and the like is unnecessary. Further, if re-learning is performed at the time of a periodic inspection or the like after shipment, an error due to aging can be corrected.

In the above-described embodiment, the neural network is configured by the processing of the processor.
It may be configured by hardware.

[0061]

According to the current detector of the first aspect of the present invention, learning is performed so that the influence between the conductors to be measured can be corrected by using a neural network, so that the conductors to be measured and the magnetoelectric conversion elements and the like can be corrected. Since complicated mounting is not required, and mounting and mounting errors of the conductor to be measured and the magnetoelectric conversion element can be corrected, the current flowing through each of the conductors to be measured can be accurately detected.

According to the current detector of the second aspect of the present invention, if a processor having normal performance is used as a neural network and a high performance processor is used as another neural network,
Generally, learning that requires a large amount of time can be performed in a short time, and a neural network can be configured at low cost.

According to the current detector of the third aspect of the present invention, a current measurement value obtained from a plurality of magnetoelectric transducers is input to a plurality of input neurons as an input pattern, and the intermediate neuron is input. Are connected to each of the input neurons, each of the plurality of output neurons is connected to each of the intermediate neurons, and the output pattern corresponding to the input pattern is output from each of the output neurons. An error between the pattern and the teacher signal is obtained, the load is changed based on the error, and the process of changing the load is repeated until the error becomes equal to or less than a certain value. That is, by using a neural network or another neural network, learning is performed so that the influence (the above-mentioned error) between the conductors to be measured can be corrected, so that complicated arrangement of the conductor to be measured and the magnetoelectric conversion element is not required. In addition, since errors in mounting and mounting the conductor to be measured, the magnetoelectric conversion element, and the like can be corrected, the current flowing through each of the conductors to be measured can be accurately detected.

According to the current detecting method of the present invention, the same effect as the effect of the current detector of the present invention can be obtained. According to the current detecting method of the fifth aspect, the same effect as the effect of the current detector of the second aspect can be obtained. According to the current detecting method of the invention described in claim 6, the same effect as the effect of the current detector of the invention described in claim 3 can be obtained.

[Brief description of the drawings]

FIG. 1 is a circuit diagram showing a configuration of a current detector according to an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration of a neural network formed in the processor shown in FIG.

FIG. 3 is a diagram showing interactions between neurons.

FIG. 4 is a diagram showing a quasi-linear element model used in the back propagation rule.

FIG. 5 is a diagram showing an example of a learning pattern when learning is performed by the current detector according to the embodiment of the present invention.

FIG. 6 is a flowchart showing a manufacturing process of the current detector according to the embodiment of the present invention.

FIG. 7 is a flowchart showing details of a learning process shown in FIG. 6;

FIG. 8 is a diagram for explaining a conventional current detector.

[Explanation of symbols]

10 ₁ to 10 _n measured conductor 20 ₁ to 20 _n electric transducer 30 processor 31 input layer 32 intermediate layer 33 output layer 34 learning unit 40 memory

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Yoshinori Ikuta 1500 Onjuku, Susono City, Shizuoka Prefecture Inside Yazaki Sogyo Co., Ltd. (72) Inventor Takashi Gohara 1500 Onjuku 1500 in Susono City, Shizuoka Prefecture Yazaki Sogyo Co., Ltd. F-term (reference) 2G025 AA17 AB01 AB02 AC05 2G035 AA04 AA12 AA27 AB02 AC02 AD28 AD66

Claims (6)

[Claims] A plurality of conductors to be measured, a plurality of magnetoelectric conversion elements provided in one-to-one correspondence with the plurality of conductors to be measured, and measuring a current flowing through each of the plurality of conductors to be measured; A current measurement value obtained from the magneto-electric conversion element is input as an input pattern, and learning is performed so as to correct an error between an output pattern corresponding to the input pattern and a teacher signal which is a current value actually flowing through the plurality of conductors to be measured. A neural network that corrects measured current values obtained from the plurality of magnetoelectric conversion elements based on the learning result and outputs the corrected measured current values as detected current values. 2. The apparatus according to claim 1, further comprising: a memory for storing the learning result; and another neural network different from the neural network, wherein the other neural network receives a current measurement value obtained from the plurality of magnetoelectric transducers. Input as a pattern, learning to correct an error between an output pattern corresponding to the input pattern and a teacher signal which is a current value actually flowing through the plurality of conductors to be measured, storing the learning result in the memory, 2. The current detection device according to claim 1, wherein the neural network corrects a current measurement value obtained from the plurality of magnetoelectric transducers based on a learning result stored in the memory and outputs the corrected current value as a detected current value. vessel. 3. Each of the neural network and the other neural network has a plurality of input neurons of the same number as the number of conductors to be measured, and the plurality of input neurons and the plurality of magnetoelectric conversion elements. An input layer for inputting the current measurement value obtained from the input pattern as an input pattern; an intermediate layer having a predetermined number of intermediate neurons, wherein each of the intermediate neurons is mutually coupled with each of the input neurons; A plurality of output neurons of the same number as the number of conductors to be measured are provided, each of the output neurons is mutually coupled with each of the intermediate neurons, and the output pattern for each of the input patterns is output from each of the output neurons. An output layer for outputting, an error between the output pattern from the output layer and the teacher signal is obtained, and based on the error, A process for changing a load indicating the strength of connection between a neuron and each of the intermediate neurons and a load indicating the strength of connection between each of the intermediate neurons and each of the output neurons, and changing the load 3. The current detector according to claim 1, further comprising: a learning unit that repeats the above until the error becomes equal to or less than a predetermined value. 4. A current flowing through each of the plurality of conductors to be measured is measured by a plurality of magnetoelectric transducers provided in one-to-one correspondence with the plurality of conductors to be measured, and the currents obtained from the plurality of magnetoelectric transducers are obtained. The measured current value is input to the neural network as an input pattern, and learning is performed so as to correct an error between an output pattern corresponding to the input pattern and a teacher signal which is a current value actually flowing through the plurality of conductors to be measured. A current measurement value obtained from the plurality of magneto-electric conversion elements based on the current value and outputs the detected current value as a detected current value. 5. A neural network different from the neural network inputs current measurement values obtained from the plurality of magnetoelectric transducers as an input pattern,
Learning to correct an error between an output pattern corresponding to the input pattern and a teacher signal which is a current value actually flowing through the plurality of conductors to be measured, storing the learning result in a memory; 5. The current detection method according to claim 4, wherein the measured current values obtained from the plurality of magneto-electric conversion elements are corrected based on the learning result stored in the storage device and output as a detected current value. 6. The neural network and the other neural network each include a plurality of input neurons as many as the number of conductors to be measured, and input current measurement values obtained from the plurality of magnetoelectric transducers to an input pattern. And a predetermined number of intermediate neurons are mutually coupled with the input neurons, and a plurality of output neurons of the same number as the number of conductors to be measured are each of the intermediate neurons. And outputting the output pattern for the input pattern from each of the output neurons, obtaining an error between the output pattern and the teacher signal, and, based on the error, A load indicating the strength of connection with each of the intermediate neurons, and each of the intermediate neurons and each of the output neurons; The load indicating the strength of the coupling is changed, according to claim 4 or claim 5 current detection method according the error processing of changing the 該荷 heavy is and repeating until a predetermined value or less.