JPH02181283A - Signal processing circuit using neural net - Google Patents

Abstract

PURPOSE:To easily select the analog input signals which are decided based on the ranking of their values by inhibiting the supply of the input signal decided as the highest signal level via a neural net in the next decision. CONSTITUTION:The excluding elements 4 are all in conducting state in an initial state and all analog input signals I1 - In are supplied to a neural net 1. In the net 1 only the output to which the analog input signal having the highest level is supplied is set at logic 1. A sense element 2 detects the output of level 1 in the net 1 and supplies the number of the detected output to a control element 3 as the number of the maximum analog input signal. The element 3 stores the number of the maximum input signal into a store area for maximum signal level of a storage/output element 5 and at the same time outputs an excluding command to the element 4 to release the switch corresponding to the maximum analog input number. Thus the element 3 repeats this operation until the ranking of values is decided among the levels of all signals I1 - In.

Description

[Detailed Description of the Invention] [Purpose of the Invention (Industrial Application Field) This invention relates to signal processing using neural network technology, which is used to discriminate between a large number of similar signals and control plants, etc. Regarding circuits.

(Prior technology) Gas turbine combustion temperature control, boiler evaporator tube temperature control, fuel cell cell stack temperature control, fuel cell reheater temperature control, heating furnace temperature control, chemical reaction tank 9 m a III
? 10. When controlling a distribution system such as room temperature control or water pipe network pressure control, a large number of analog input signals of the same kind are taken in, and the average value excluding the maximum value, intermediate value, minimum value, maximum value and minimum value, Using several average values near the intermediate value as the control amount,
A case can be considered in which a controlled object is controlled by operating one operating end.

This control method requires signal processing to find the maximum value, minimum value, intermediate value, etc. of the signal levels (eg, current values, voltage values) of a large number of analog input signals.

This signal processing is usually performed by repeatedly comparing the signal levels of two input signals. For example, when determining the maximum value, minimum value, and average value excluding the maximum and minimum values of the signal levels of N analog input signals, processing as shown in the flowchart of FIG. 13 may be performed. Furthermore, when checking whether the signal level of an analog input signal exceeds the upper limit value or is below the lower limit value, it is necessary to compare the signal level of the input signal with the upper limit value or the lower limit value for each input signal.

(Problems to be Solved by the Invention) In the above-described method, iterative processing is required to obtain the maximum value, minimum value, intermediate value, etc. of the signal level of the input signal, and it takes time. Therefore, if such signal processing is employed, an extra delay is added to the feedback control loop, making it impossible to improve controllability. Especially when high-speed control is required,
The delay caused by this signal processing is unacceptable. In this case, it is also conceivable to select one of the many analog input signals and use it as the control variable. However, if the sensor that outputs the selected analog input signal becomes abnormal or the selected analog input signal fluctuates differently from other analog input signals, the entire distribution system will be controlled based on the selected analog input signal, which will cause problems. There is a risk of disturbing the distribution system.

[Object of the Invention] The present invention utilizes the parallel processing properties of mutually inhibiting neural networks to eliminate or minimize the repetitive operations in the signal processing described above, and to provide a signal processing system that can perform high-speed signal processing. The purpose of the present invention is to provide a processing circuit.

(Means and effects for solving the problem) In order to solve the problem, a signal processing circuit according to the present invention receives a plurality of analog signals, and processes a signal having a maximum signal level among the plurality of analog signals. We decided to equip a neural network that uses mutually inhibiting connections to make decisions.

By adopting such a configuration, the signal having the maximum signal level can be detected, and the order of the signal levels of the signals can be determined by repeating the same operation for other signals excluding that signal.

In addition, the neural
The signal with the minimum signal level can be detected by inputting it to the net.

By adding an element that averages the signal level, it is possible to obtain the average signal level of the signals excluding signals having the maximum signal level or the minimum signal level that may be abnormal signals.

By converting the input signal into a complement signal with respect to a predetermined signal range, multiplying this complement signal and the original signal in an analog manner, and supplying the resultant signal to the neural network, the input signal within the signal range is The signal with the signal level closest to the intermediate value can be detected.

After an input signal having a signal level closest to the intermediate value is detected, the operation of excluding that signal from input to the neural net and detecting the next input signal having a signal level close to the intermediate value is repeated. By this, several input signals near the intermediate value can be discriminated. The average value of the signal levels of these several signals can be determined by the averaging element.

(Example) An example of a neural network used in the present invention will be described below.

Figure 11 shows the neural system realized using analog electronic circuits.
An example of a net is shown. One amplifier in the circuit of Figure 11 (
Amplifier) API to APN correspond to the cell body of one neuron in the living body. A feedback line of an inverted voltage vj of the output voltage Vj of the j-th amplifier is connected to an input line of the i-th amplifier via a resistor (the conductance, which is the reciprocal of the resistor, is Tij). This connection corresponds to inhibitory synaptic connections in living organisms.

The current ΣT1j-Vj (-1 ΣT1j-vj) flowing through the resistor and the current 11 flowing from the outside are added together and grounded through the capacitor CI and the resistor rl. As a result, the i-th amplifier AP
An input voltage Uj is applied to I.

As the output function of the amplifier AP, the input/output static characteristics g
(Ui) is a sigmoid function as shown in FIG.
Output i.

The characteristics of the neural network shown in FIG. 11 can be expressed by equations (1) to (4).

It is known to do.

... (1) ... (2) Vi Gt)--Vi (t)=g [Ui
(t)]... (3) When the synaptic connections are symmetrical (Tij=Tji), the neural net shown in FIG. 11 operates in the direction in which the energy function (Ryabunov function) shown in equation 5 decreases. Then, the minimum point of the energy value E in equation (5) is 7 of the circuit.
1 herakta (steady state). Therefore, the mutually inhibiting coupling (
If we set T jj > Q, Tij-0 for j-, the first term on the right side of equation (5) means that the amplifier with output V-"1"
If the other amplifier outputs are "O", the minimum value is O. Next, the second term (-Σ11 ・Vi) is 1 when the output of the amplifier to which the gummy current with the maximum current value is input is "1".
1.1j, becomes the minimum. , and the third term is usually a minute chain, so it can be ignored. Furthermore, if the suppression coupling conductance chains Tij (i'#j) are made equal and the first and second terms are considered together, then the amplifier output to which the maximum analog current is input is 1", and the other amplifier outputs are " O.

When , the energy function E becomes a minimum and the circuit is in a steady state. That is, this neural net has the ability to instantaneously determine the largest one among N analog inputs, if the time constant 01·R1 of the circuit is made sufficiently small. Therefore, by combining this neural net with other circuit elements, it is possible to perform various signal processing on a large number of analog input signals. Specifically, for example, the following signal processing can be performed at high speed.

(1) The analog input signal having the maximum signal level can be detected from a large number of analog input signals. moreover,
By detecting the signal with the highest signal level among the other analog input signals other than that analog input signal and repeating this operation, it is possible to determine the order of the signal levels of a large number of analog input signals ( It is not necessary to determine all the rankings, just some of them).

(2) The analog input signal having the minimum signal level can be detected by inverting the polarity of the analog input signal or converting it to its complement for a certain range and feeding it to the neural net.

(3) It is possible to obtain the average value of the signal levels of the analog input signals, excluding the signal with the maximum signal level and the signal with the minimum signal level, which may be abnormal signals.

(4) Check the upper limit alarm only for the analog input signal that has the maximum signal level, and check the lower limit alarm only for the analog input signal that has the minimum signal level, thereby making alarm judgments for many similar analog input signals. thing.

(5) Converting the analog input signal to a complementary value signal for a predetermined signal range. That is, if the analog input value is X lunge lower limit value is 8 lunge upper limit value is L, the complement value is (S
+L)-X. Then, this complement signal and the original signal X are multiplied in an analog manner.

Y-X [(S+L) -X [X-(S+L)/2] 2 + [(S+L)/2] 2...(6) This signal Y is , is maximum when the original analog input signal has the middle level of the range. Therefore, if the analog input signal converted to signal Y is input to the mutually inhibiting neural network and the analog input signal with the maximum level is detected, the signal closest to the middle value of the range among the original analog input signals will be detected. You can detect things that have a high level.

As a result, if the upper and lower limits of the range are determined so that a certain set value becomes the middle value of the range, it is possible to instantly determine the input signal having the signal level closest to the set value.

Furthermore, by setting the upper and lower limits of the range to the maximum and minimum levels of a large number of analog input signals, it is possible to detect an analog input signal having a signal level closest to the intermediate value between the maximum and minimum levels.

(6) After the analog input signal with the signal level closest to the aforementioned intermediate value is detected, remove that signal from the input to the neural net and select the next analog input signal with the signal level closest to the intermediate value. By repeating the sensing operation, several analog input signals near the intermediate value can be discriminated and the average value of their signal levels can be determined.

A first embodiment of a signal processing circuit that enables the above-mentioned signal processing will be described with reference to FIG.

The circuit of FIG. 1 comprises a neural net 1 of mutually inhibiting connections shown in FIG. Neural net 1 is N(
(N is an integer of 2 or more) input terminals and output terminals. A sense element (circuit, block) 2 is connected to the output end (amplifier output) of the neural net 1. Sense element 2 determines which amplifier output of neural net 1 is “1”
Determine whether it is. Analog input signals (input currents) 11 to IN from temperature sensors, pressure sensors, etc. are supplied to the input end of the exclusion element 4. A control element 3 is connected to the sense element 2 and the exclusion element 4.

The control element 3 receives the detection signal from the sense element 2, outputs an exclusion command to the exclusion element 4, and controls the exclusion element 4 so that the analog input signal identified by the sense element 2 is no longer supplied to the neural net 1. Control.

The control element 3 is connected to a storage/output element 5 that stores and outputs the signal numbers of the analog input signals 11 to IN in descending order of signal level.

For example, while counting up a counter, the sense element 2 compares the signal level of the output signal of the neural net 1 specified by the count value with a predetermined reference value using a comparator.

The sense element 2 supplies the count value when the output signal of the neural net 1 is determined to be larger than the reference signal to the control circuit 3 as the number of the maximum analog input signal.

The exclusion element 4 includes, for example, N switch circuits (such as relays) that are turned on and off by a control signal from the control circuit 3. In this case, one end of each switch circuit receives a corresponding analog input signal, the other end is connected to the corresponding input of the neural net 1, and the control end is connected to the control circuit 3.

The control element 3 is composed of, for example, a microprocessor, a memory storing a control program, and peripheral circuits. Each output port of the microprocessor is connected to the control terminal of the corresponding switch circuit, and the input port is connected to the output terminal of the sense element 2.

The storage/output unit 1g5 is composed of, for example, a dual port memory connected to a microprocessor in the control circuit 3 via a bus. The analog input signal numbers (1 to 1) are stored in the dual port memory in descending order of signal level.
N storage areas are prepared for storing N). Dual port memory is connected to a programmable controller, for example, and its stored contents are used to control the process.

Next, the operation of the signal processing circuit shown in FIG. 1 will be explained.

In the initial state, all exclusion elements 4 are conductive and all analog input signals II-IN are supplied to the neural net 1. In the neural net 1, only the output of the amplifier to which the analog input signal of the maximum signal level (current value in this case) is supplied becomes logic "1" (above the reference level). For example, when the analog input signal 11 has the maximum current value, only the amplifier API (Fig. 11) is "1".
Output # level signal and connect other amplifiers AP2 to APN
(Figure 11) outputs a "0" level signal.

The sense element 2 detects the "1" level output of the neural net 1 and supplies the number of the output to the control circuit 3 as the number of the maximum analog input signal. In the above example, since the current value of the analog input signal (1) is the maximum, the sense element 2 outputs data "1" to the control circuit 3. The control element 3 stores the number of the maximum input signal from the sense element 2 (1 in the above example) in the storage area of the storage and output element 5 for the maximum signal level. Further, the control element 3 outputs an exclusion command to the exclusion element 4 to release the switch corresponding to the maximum analog input number. The exclusion element 4 receives the exclusion command and activates a corresponding switch circuit so that the analog input signal having the maximum current value (It in the above example) is not supplied to the neural net 1. Analog input signals I2-IN are therefore provided to the neural net 1. In this case, it is desirable to ground the input terminal of the corresponding amplifier of the neural net 1.

After a certain period of time has elapsed, the neural net 1 enters a new steady state, and the output of the amplifier to which the analog input signal with the second largest current value is supplied becomes "1".

Sense element 2 detects that number. The control element 3 stores the detected number in the second area of the storage/output element 5. The control element 3 sends an exclusion command to the exclusion element 4. The exclusion element 4 responds to the exclusion command and stops supplying the analog input signal of the second detected number to the neural net 1. Control element 3 has all analog input signals 1
This repetitive operation is managed until the order of signal levels from 1 to IN is determined.

In this manner, by using the circuit of the first embodiment, the magnitude order of the signal levels of N analog input signals can be determined by repeating N-1 times. Thereby, it is possible to easily select an analog input signal determined in association with a magnitude order and use it as a control amount, etc.

Note that the same effect as using the exclusion element 4 can also be obtained by forcibly clamping the output of the corresponding amplifier of the neural net 1 to "0". That is, it is possible to substantially stop supplying the analog input signal to the neural net 1.

Note that it is not necessary to determine the magnitude order of all analog input signals.

A second embodiment will be described with reference to FIG. In addition, the second
In the figure, the same parts as in FIG. 1 are given the same reference numerals.

In the circuit shown in Figure 2, the analog input signals ■1 to IN are the first
is directly supplied to the neural net 1 and also supplied to the polarity inverting element 6. The polarity inverting element 6 is composed of, for example, an operational amplifier with an amplification factor of G--1, and inverts the polarity of the signal level of the input signal and outputs it. Polarity reversal element 6
acts to reverse the magnitude order of the signal levels of the input analog signal.

The output signal of the first neural net 1 is fed to the first sense element 2. The output of the polarity reversal element 6 is fed to a second neural net 11 which operates similarly to the first neural net 1.

The output of the second neural net 11 is fed to a second sense element 12 which operates similarly to the first sense element 2.

Among the amplifier outputs of the first neural net 1, the amplifier output to which the analog input signal of the maximum signal level is supplied is in a steady state of "1". This is sensed by the first sense element 2 and the number of the analog input signal having the highest signal level is determined instantaneously.

At the same time, the analog input signals 11 to IN are connected to the polarity inverting element 6.
The polarity is reversed. Of the amplifier outputs of the second neural net 11, only the amplifier to which the inverted signal that angles the maximum signal level is input is in a steady state of "1". This is sensed by the second sense element 12, and the number of the analog input signal having the minimum signal level among the analog input signals before inversion is instantly determined.

By picking up the analog input signal judged to have the maximum signal level or the minimum signal level by the sense element 2.12 and comparing it with a predetermined reference level using a comparator or the like, the upper and lower limits can be easily and quickly checked.

Furthermore, a signal having the maximum or minimum signal level among a plurality of similar signals is more likely to be an abnormal signal than other signals. In this embodiment, an analog input signal having the maximum signal level and an analog input signal having the minimum signal level are discriminated, and this is useful when it is desired to use the average level value of other analog input signals as a control amount.

Similar actions and effects can be achieved by using an element that converts the input analog signal into a signal with a signal level corresponding to the complement of a certain predetermined signal range (inverts the polarity and adds bias) instead of the polarity inverting element 6. can get.

The control element 3 and storage/output element 5 of FIG. 1 may be added to the circuit of FIG. 2.

A third embodiment will be described with reference to FIG. In FIG. 3, the same parts as in FIGS. 1 and 2 are given the same reference numerals.

The circuit shown in FIG. 3 includes a switching element 7 that simultaneously switches and outputs the original analog input signals II to IN and the inverted signal from the polarity inverting element 6. The switching element 7 is composed of, for example, a plurality of changeover switches that operate in conjunction with each other. The output of the switching element 7 is fed to the neural net 1 of the mutual inhibition combination.

The output of the neural net 4 is fed to a sense element 2 which determines the "1" level. A control element 13 is connected to the control terminal of the switching element 7 and the output terminal of the sense element 2, and a storage/output element 15 is connected to the control element 13.

The analog input signals 11 to IN are conducted directly to the switching element 7 and via the polarity reversing element 6 to the switching element 7.

In the initial state, the switching element 7 receives the analog input signals 11 to I
Select N and output to neural net 1. Neural net 1 operates in this input state, and only the amplifier output to which the analog input signal with the maximum signal level is supplied is 1.
” is detected by the sense element 2.

The control element 13 stores the number of the analog input signal with the maximum signal level detected by the sense element 2 in the storage location for the maximum level of the storage/output element 15.

Control element 13 then sends a switching command to switching element 7 . The switching element 7 selects and directs the polarity reversal signal from the polarity reversal element 6 to the neural net 1 in response to a switching command. In this input state, the neural net 1 operates again, and only the amplifier output to which the signal having the maximum signal level (the minimum signal level of the original analog input signal) among the polarity inverted signals is input is the logic “ It becomes a steady state of 1#.

The sense element 2 senses the number of this output. Control element 1
3 causes the sensed number to be stored in the storage location for the minimum level of the storage/output element 15. The storage output element 15 outputs the number of the signal having the maximum signal level and the number of the signal having the minimum signal level.

In the third embodiment, the time required for processing is approximately twice that of the second embodiment. However, only one neural net is sufficient. Note that the polarity inversion element 6 may be an element that converts into a complement signal as in the case of the second embodiment. Also,
The excluded element 3 shown in FIG. 1 and the like may be added to the circuit shown in FIG. 3 to enable repeated operations.

A fourth embodiment will be described with reference to FIG.

This circuit includes a determination element 8 that determines which analog input signal has the maximum signal level and an analog input signal which has the minimum signal level, an averaging element 9 that averages the signal levels of a plurality of supplied signals, and an analog input signal It also includes an exclusion element 14 that prevents signals of designated numbers from being supplied to the averaging element 9. Judgment element 8 is shown in Figures 2 and 3.
This circuit has substantially the same configuration as the circuit for determining the maximum signal level and minimum signal level shown in the figure. Averaging factor 9
For example, an adder circuit with N inputs and the output of the adder circuit are
For example, it is composed of an amplifier that multiplies by 1/(N-2).

The exclusion element 14 is composed of, for example, a plurality of switches and a control circuit, and when the switch is turned on, the input analog signal passes through, and when the switch is turned off, the corresponding input end of the averaging element 9 is opened.

The operation of the fourth embodiment will be explained. The determination element 8 outputs the number of the signal having the maximum signal level or the minimum signal level among the input analog signals If-IN. The exclusion element 14 stops supplying the number of analog input signals specified by the decision element 8 to the averaging element 9 . Therefore, the averaging element 9 outputs the average signal level of the analog input signal excluding the signal of the number designated by the determining element 8.

For example, if the decision element 8 determines that I1 has the maximum signal level and I2 has the minimum signal level, the exclusion element 14 supplies the analog input signals 13 to IN to the averaging element 9;
The averaging element 9 outputs a signal having the average signal level of the analog input signals I3-IN.

In the above description, the average value is calculated excluding the maximum level and minimum level, but for example, the average value may be calculated by excluding only the analog input signal having the maximum level or the minimum level.

By adopting the configuration shown in FIG. 4, it is possible to use the average level of analog input signals excluding input signals that may cause sensor abnormality as the control amount.

A fifth embodiment will be described with reference to FIG. In FIG. 5, the same parts as in FIGS. 1 to 4 are given the same reference numerals.

In this circuit, input analog signals 1-IN are supplied to a multiplication element IO and a complementation element 1B. Complementing element 16 converts the input signal into a complement signal having complementary value levels for the upper and lower limits of the specified signal range. Multiplying element 1o multiplies each complement signal by the corresponding original analog input signal in analog fashion.

The output signal of the multiplication element 10 is the neural signal of the mutual inhibition connection.
The output of neural net 1 is fed to sense element 2 which determines which amplifier output is at 1''.

The complementing element 16 inverts the polarity of the analog input signal, adds (range upper limit value) + (range lower limit value) to the signal as a bias, and converts it into a complement signal.

This conversion is performed using, for example, an operational amplifier.

As previously mentioned, such a configuration maximizes the product of the analog input signal and its complement signal having a signal level that most closely approximates the midpoint of the signal range.

Therefore, in this configuration, the number output from the sense element 2 is the number of the analog input signal having the signal level closest to the intermediate value of the signal range among the original analog input signals.

If the upper and lower limits of the signal range are determined so that the desired value is the middle value of the range, it is also possible to detect an analog input signal that has a signal level closest to the desired value.

A sixth embodiment will be described with reference to FIG. In FIG. 6, the same parts as in FIGS. 1 to 5 are given the same reference numerals.

The circuit of FIG. 6 is characterized by adding a determination element 8 to the circuit of FIG. 5, and that the complementing element 16 of FIG. It is replaced by a complementing element 18 that uses the minimum signal level as the lower range limit.

The operation of the circuit shown in FIG. 6 will be explained. The determining element 8 determines the number of the analog input signal having the minimum signal level and the number of the analog input signal having the minimum signal level among the input analog signals 11 to IN.

The determining element 8 supplies the determined number to the complementing element 18 . Complementing element 18 uses the maximum signal level as an upper limit and the minimum signal level as a lower limit based on the number received. Therefore, the analog input signal 11~IN
are converted to complement signals having complementary value levels of the signal range from the maximum signal level to the minimum signal level of these signals. The intermediate value of the signal range is analog input signal 11~
This is the intermediate value between the maximum signal level and the minimum signal level of IN.

Therefore, in this circuit, the sense element 2 detects the number of the signal whose signal level is closest to the intermediate value between the maximum signal level and the minimum signal level of the analog input signal at that time.

The complement element 16 includes, for example, a switch circuit that selects the analog input signal of the number specified by the determination circuit 8 as a reference signal, and an operational amplifier that takes the complement of the input signal based on the signal selected by the switch circuit. be done.

Note that the determining element 8 itself may pick up the analog input signal having the maximum signal level and the analog input signal having the minimum signal level and supply them to the complementing element 16. In this case, the complementing element may have the configuration shown in FIG.

A seventh embodiment will be described with reference to FIG. In FIG. 7, the same parts as in FIGS. 1 to 6 are given the same reference numerals.

In the circuit of FIG. 7, analog input signals 11-IN are supplied to complementing element 26, multiplication element 10 and switching element 17. Complementing element 26 inverts the polarity of the analog input signal;
The maximum and minimum signal levels of the analog input signal are added as biases to convert it into a complement signal. Multiplication element 1
0 analogically multiplies the signal from complementing element 26 and the original analog input signal. Analog input signal 11~IN
, the output signal of the multiplication element IO and the output signal of the complementation element 26 are fed to the switching element 17 . The switching element 17 is 3
One of the two input signals is selected and output. The output of the switching element 17 is fed to the neural net 1 of the mutual inhibition combination. The output of neural net 1 is sense element 2
is supplied to A control element 23 is connected to the complement element 2B, the switching element I7, and the sense element 2.

The control element 23 receives the input signal number from the sense element 2 and outputs it to the storage/output element 25 and the complementing element 26, and also issues a switching command to the complementing element 2B and the switching element 17.

The storage/output element 25 stores and outputs the numbers of analog input signals having the maximum signal level, minimum signal level, and intermediate signal level among the analog input signal generators 1 to IN.

Next, the operation of the circuit shown in FIG. 7 will be explained with reference to the flowchart shown in FIG.

The control element 23 outputs a switching command to the switching element 17, and the original analog input signal 11~IN is output to the neural network 1.
(Step Sl).

In this input state, the neural net 1 operates, and after a predetermined time has elapsed (step S2), the amplifier output to which the analog input signal of the maximum signal level is supplied becomes a steady state of "1". This is detected by sense element 2. That is, the number of the analog input signal having the maximum signal level is determined. The control element 23 receives this number from the sense element 2 and stores it in a predetermined location of the storage output element 25 (step S3).

Next, the control element 23 sends a command to the complementing element 26 to set it in a polarity inversion mode in which the polarity of the analog input signal is inverted (step S4). Further, the control element 23 is the switching element 1
7, the switching element 7 is switched, and the output signal of the complementing element 26 is supplied to the neural net 1 (step S5).

In this input state, the neural net 1 operates, and after a certain period of time (step S6), the amplifier output to which the analog input signal with the lowest signal level among the original analog input signals is supplied becomes "1", and the sense Detected in element 2. The control element 25 stores the number of the analog input signal detected by the sense element 2 at a predetermined location of the storage/output element 25 (step S7).

The control element 23 sends the numbers of the maximum and minimum analog input signals detected in the above operation to the complementing element 26, and at the same time switches the complementing element 26 to the complementing mode (step S8
). In the complementation mode, the complementing element 26 inverts the polarity of the input analog signal, and further converts the analog input signal ■1 to
The maximum signal level and minimum signal level of IN are added in an analog manner. Complementing element 26 thereby converts the analog input signal into a signal having a complement value level ranging from its maximum signal level to its minimum signal level. moreover,
The control element 23 sends a switching command to the input switching element 17 so that the output of the multiplication element 10 is supplied to the neural network 1 (step S9). In this input state, neural net 1 operates, and after a certain period of time (
Step 510), the number of analog input signals having a signal level closest to the intermediate value between the maximum and minimum signal levels of the original analog input signal is detected in the sense element 2. The control element 23 causes the number detected by the sense element 2 to be stored at a predetermined location in the storage/output element 25.

The storage/output element 25 outputs the number of the signal having the maximum signal level, the number of the signal having the minimum signal level, and the number of the signal having the intermediate signal level.

According to this embodiment, one neural ψ net can calculate the number of analog input signals that have the maximum signal level, the number of analog input signals that have the minimum signal level, and the number of analog input signals that have the signal level of the intermediate value. can be determined.

Next, an eighth embodiment will be described with reference to FIG. 9th
In the figure, the same parts as in FIGS. 1 to 7 are given the same reference numerals.

In the circuit of FIG. 9, analog input signals 11-IN are provided to complementing element 36 and multiplication element 10. In the circuit of FIG.

Complementing element 36 converts the input signal into a complemented signal and provides it to multiplication element 10. The multiplication element 10 analogically multiplies the complement signal and the corresponding original analog input signal and outputs the result. The output of multiplication element 24 is provided to exclusion element 24. The exclusion element 24 excludes the control element 3 from the output signal of the multiplication element 24.
The signal of the number commanded from 3 is not supplied to the neural net 1.

The output signal of the exclusion element 24 is fed to the neural net 1. The output of neural net 1 is fed to sense element 2. The output of sense element 2 is fed to control element 33.

The analog input signal If-IN is also supplied to the input element 34. The input element 34 allows only the signal of the number specified by the control element 33 to pass. The output signal of input element 34 is fed to averaging element 9. Averaging element 33 averages the signal levels of the supplied signals.

The operation of the circuit shown in FIG. 9 will be explained.

In the initial state, the exclusion element 24 operates in a conductive state that leads all input signals to the neural net 1, and the input element 34 operates in an open state that turns off all input signals. In this state, the complementing element 3B outputs a complement signal of the input signal for a predetermined range. Multiplier circuit 10 outputs the product of the input signal and the corresponding complement signal. In neural net 1, only the amplifier output corresponding to the input analog signal having the signal level closest to the intermediate value of the signal range becomes "1",
Sense element 2 detects this. The control element 33 receives this number, outputs a command to the exclusion element 24 to exclude the input signal of that number, and outputs a command to the input element 34 to input the input signal of that number. In this state, neural net 1 operates again, and sense element 2 changes to the intermediate value by 2.
Detecting the number of analog input signals having the closest signal level. The control element 33 receives this information, issues a command to the exclusion element 24 to exclude the signal of that number, and outputs a command to the input element 34 to input the input signal of that number. By repeating this operation several times, the numbers of a plurality of signals having signal levels close to the intermediate value of the signal range are determined. The averaging element 9 then outputs the average value of the signal levels of the analog input signals having signal levels near the intermediate value specified by the number.

In this embodiment, the intermediate value is the intermediate value of the range in which the complement converting element 36 converts the input signal into the complement value of the specified range, and the input signal is changed from the minimum signal level to the maximum value of the input signal. If an element is used to convert to a complementary value of the signal range up to the signal level, it is the intermediate value between the minimum signal level and the maximum signal level of the input signal.

Next, an embodiment in which a plant is actually controlled using the circuit shown in the above embodiment will be described with reference to FIG. 1O.

This embodiment shows an example of controlling the temperature of a gas turbine. A plurality of burners 41 that burn fuel are attached to the gas turbine. Combustion fuel is supplied to the burner 41 via a common fuel pipe 43. The total amount of fuel supplied to the burner 41 is adjusted by the degree of opening and closing of the valve 42.

A plurality of temperature sensors SE are arranged within the gas turbine,
The output signal of each sensor is supplied to a signal processing circuit 44. The signal processing circuit 44 has, for example, the configuration shown in FIG. 4 or FIG. , outputs a signal indicating the average value of signal levels of a plurality of signals near the intermediate value. This signal indicates the average temperature inside the gas turbine. Control circuit 45
performs a commonly known control operation (for example, a PID operation) in response to a signal indicating the average value from the signal processing circuit 44, adjusts the opening/closing degree of the valve 42, and controls the temperature inside the gas turbine. .

In this embodiment, the temperature inside the gas turbine is not constant, but varies depending on the position, resulting in a distributed system.

For this reason, a plurality of temperature sensors SE are arranged within the gas turbine to determine the average value of the combustion temperature.

On the other hand, the temperature inside the gas turbine is extremely high, and there is a high probability that the sensor will fail. Therefore, in this embodiment, the average temperature is determined using only signals with a high probability of being normal signals among the output signals of the sensor SE. Therefore, control reliability is high, and since the signal processing circuit 44 is a circuit using a neural network as shown in FIGS. 1 to 9, high-speed operation is possible, and online control is possible. be.

In the embodiment described above, the input signals were current signals 11 to IN, but the present invention can be similarly applied to cases where the input signals are voltage signals or the like. Further, in the embodiment described above, the neural net is constructed from an analog circuit, but an analog input signal may be converted into a digital signal by an A/D converter, and subsequent processing may be performed by a digital arithmetic circuit. In this case, the function of the neural network is implemented using the differential equation of equation (1) converted into a difference equation.

The amplifier characteristics of the neural net are not limited to those shown in FIG. For example, the maximum saturation value of the amplifier output may not be "1".

In this case, a sense element that detects the saturation value is used as the sense element. Further, instead of the sigmoid function as shown in the figure, an amplifier having characteristics of a monotonically increasing function without a maximum saturation value may be used.

In this case, the sense element used is one that senses when the amplifier output reaches a predetermined value.

[Effects of the Invention] This invention utilizes the parallel processing ability of neural networks to calculate the maximum signal level,
It is possible to obtain a circuit that can almost instantaneously select a signal having a minimum signal level or an intermediate signal level, and furthermore, can quickly determine the order of magnitude. As a result, it is possible to obtain a highly reliable average value excluding signals with a high possibility of input abnormality or a highly reliable average value using several analog input signals near the intermediate value without delay in detection. By using the obtained average value as the control amount, the distribution system can be controlled favorably and with high reliability.

Also, instead of checking one by one whether a large number of analog input signals exceed the upper limit value or below the lower limit value, only the signal with the maximum signal level is checked as the upper limit, and only the signal with the minimum signal level is checked as the lower limit. By doing so, alarm processing can be easily performed.

[Brief explanation of the drawing]

Figure 1 is a block diagram of an embodiment that determines the order of signal levels of input analog signals, and Figure 2 is a diagram of an embodiment that simultaneously detects an input signal with the maximum signal level and an input signal with the minimum signal level. Block diagram: Fig. 3 is a block diagram of an embodiment in which one neural net detects an input signal with the maximum signal level and an input signal with the minimum signal level, and Fig. 4 shows the input signal with the maximum signal level. Or, a block diagram of an embodiment that detects the average value of the signal level of the input signal having the minimum signal level, or the signal excluding both. Figure 5 shows the input signal having the signal level closest to the intermediate value of the specified range. 6 is a block diagram of an embodiment that detects an input signal having a signal level closest to the intermediate value between the maximum signal level and the minimum signal level, and FIG. 7 is a block diagram of an embodiment that detects the maximum signal level. A block diagram of an embodiment for detecting an input signal having a minimum signal level and a signal level closest to an intermediate value between the minimum signal level and the maximum signal level. FIG. 8 shows the operation of the control element in the embodiment of FIG. 7. Flowchart, Figure 9 is a block diagram of an embodiment for calculating the average signal level of several analog input signals having signal levels near the intermediate value, Figure 10 is gas turbine control by a signal processing circuit using a neural network. System configuration diagram, Figure 11 is a configuration diagram of a mutually inhibiting neural net, Figure 12 is a diagram of the input/output characteristics of the neural network amplifier, Figure 13 is the maximum signal level, minimum signal level, maximum and minimum FIG. 2 is a flowchart for calculating the average value of signal levels excluding . 1.11... Neural net, 2.12... Sense element, 3.13.23.33... Control element, 4.
24... Exclusion element, 5.15.25... Storage/output element, 6.12... Polarity reversal element, 7.17... Switching element, 8.18... Judgment element, 9. ...Averaging element, 16.18.2B, 36...Complementing element, 34...
・Input element, 4I...Burner, 42...Valve, 43.
...Fuel vibe, 44...Signal processing circuit, 45...
control circuit. Applicant's Representative Patent Attorney Takehiko Suzue Analog Input Signal Analog Input Signal No. 1 Fig. 1zlN Fig. 2 Analog Input Signal Analog Input Signal Fig. Analog Input Signal Fig. Analog Input Signal Fig. 10 Fig. 11 Fig. V+

Claims (9)

[Claims] (1) A neural net using mutual inhibition coupling that receives a plurality of analog input signals and repeatedly determines the signal having the maximum signal level among the plurality of input signals; and a neural net connected to the neural net. , an input element that supplies the input signal to the neural net and does not substantially supply the input signal determined by the neural net to have a maximum signal level to the neural net in the next determination; connected to a neural net and the input element;
A signal processing circuit using a neural net for determining the magnitude order of signal levels of a plurality of analog input signals, characterized by comprising: the neural net and a control element for controlling repetitive operations by the input element. (2) a first neural net using mutual inhibition coupling that receives a plurality of analog input signals and determines a signal having a maximum signal level among the plurality of input signals; and a signal of the plurality of input signals. an inverting element that inverts the magnitude relationship of levels; a second neural net that is connected to the inverting element and uses a mutual inhibition connection that determines a signal having the maximum signal level among the output signals of the inverting element; A signal processing circuit using a neural network that determines a signal with a maximum signal level and a signal with a minimum signal level among a plurality of analog input signals. (3) an inverting element that receives a plurality of analog input signals and inverts the magnitude relationship of the signal levels of the plurality of input signals; and a mutual inhibition coupling that determines the signal having the maximum signal level among the supplied analog signals. The neural
a net, connected to the inversion element and the neural net;
a switching element that switches between the plurality of input signals and the output signal of the inversion element and selectively supplies the signal to the neural net; A signal processing circuit that uses a neural net to determine which signal has the lowest signal level. (4) an inverting element that receives a plurality of analog input signals and inverts the magnitude relationship of the signal levels of the plurality of input signals; and a mutual inhibition coupling that determines the signal having the maximum signal level among the supplied analog signals. The neural
a net, connected to the inversion element and the neural net;
a switching element that switches between the plurality of input signals and the output signal of the inverting element and selectively supplies the same to the neural net; a switching element receiving the plurality of input signals and connected to the neural net; a neural net for calculating an average value, comprising: an element for calculating an average value of the plurality of input signals excluding at least one of a signal having a maximum signal level and a signal having a minimum signal level; Signal processing circuit used. (5) a complement/multiplying element that receives a plurality of analog input signals, converts the plurality of input signals into complement signals related to a specified signal range, and outputs a signal multiplied by each original analog input signal; Neural algorithm using mutual inhibition coupling to determine the signal with the maximum signal level among the analog signals
What is claimed is: 1. A signal processing circuit using a neural net that determines a signal having a signal level closest to an intermediate value of a specified signal range among a plurality of analog input signals, comprising: a net; (6) a determination element that receives a plurality of analog input signals and detects a signal having a maximum signal level and a signal having a minimum signal level among the plurality of input signals; a complement signal receiving a signal, determining a complement signal of each of the plurality of input signals with respect to a signal range between the maximum signal level and the minimum signal level, and outputting a signal obtained by multiplying each complement signal by the corresponding input signal; A neural algorithm using a multiplication element and a mutual inhibition combination that determines the signal with the highest signal level among the supplied analog signals.
1. A signal processing circuit using a neural net for determining a signal having a signal level closest to an intermediate value between a maximum signal level and a minimum signal level of an analog input signal, comprising: a net; (7) an inversion/complementation element that receives a plurality of analog input signals and converts the plurality of input signals into a polarity inversion signal and a complement signal; an inversion/complementation element that receives the plurality of input signals and is connected to the complement/multiplication element; a multiplication element that multiplies the complement signal by each of the corresponding analog input signals; a switching element that sequentially selects and outputs the output signals of the multiplication elements; and a mutually inhibiting neural network that is connected to the switching element and detects a signal having a maximum signal level among the output signals of the switching element. , an element connected to the neural net and detecting the output of the neural net; an element connected to the inversion/complementation element, the switching element, the detection element, the inversion/complementation element, the switching element; a control element for controlling operation of the detection element; and a circuit for determining a signal having a signal level closest to an intermediate value between a maximum signal level and a minimum signal level of a plurality of analog input signals. (8) A signal processing circuit using a neural network that receives a plurality of analog input signals and determines a signal having a signal level closest to an intermediate value between a maximum signal level and a minimum signal level of the plurality of input signals; an averaging element that calculates an average value of signal levels of the analog signals that have been input, and an input that receives the plurality of input signals and inputs the input signal determined to have a signal level closest to the intermediate value to the averaging element. an exclusion element connected to the signal processing circuit to prevent the input signal determined to have a signal level closest to an intermediate value from being supplied to the signal processing circuit; the signal processing circuit; and the average. an element that is connected to the input element and the exclusion element and controls the repetitive operation of the signal processing circuit, the averaging element, the input element and the exclusion element. A signal processing circuit using a neural network that discriminates several analog input signals having signal levels near an intermediate value and calculates an average value of the signal levels of the several input signals. (9) A signal processing circuit comprising a neural net using mutual inhibition coupling that receives a plurality of analog signals and determines the signal having the maximum signal level among the plurality of analog signals.