JP3204731B2 - Distributed controller and control device using this controller - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a decentralized controller used in various control systems including plant control, and a control device using the controller. The present invention relates to a distributed controller for performing autonomous determination and treatment, and a control device using the controller.

[0002]

2. Description of the Related Art Conventionally, for example, in plant operation,
2. Description of the Related Art A control device that controls a process state quantity by a plurality of control loops in an operation process to control an operation terminal to a final target value is employed. By the way, in such a control device, normally, a plant is controlled according to a predetermined plan by a decentralized controller constituting each control loop, and when an abnormality occurs in a certain control loop, the abnormality increases or increases due to the abnormality. An alarm is issued by making a determination by comparing a certain reduced process state quantity with an alarm set value. Therefore, when an alarm is issued, the operator checks various state quantities that are likely to be related to the abnormality, identifies the control loop having the abnormality by comprehensive consideration, and automatically controls the control loop. From the control to the manual control. If the operating terminal is abnormal, the control is switched to the standby unit, backup is performed by another control loop, and measures such as plant trip are performed.

When a sensor is abnormal, the decentralized controller determines the abnormality by using a self-diagnosis method based on an alarm check of the upper and lower limits of the control amount and the rate of change transmitted by the sensor. Here, among the sensor abnormalities, a method of detecting a disconnection of a thermocouple by a burnout circuit is also used.

[0004] In order to prevent a control loop from becoming abnormal due to a sensor abnormality or the like, the distributed controller also employs a system in which a sensor for measuring a control amount is duplicated or tripled. Further, a sensor may be newly added for the purpose of detecting an abnormality in the control loop.

On the other hand, there is also a method of using a detailed mathematical model of a controlled object as a distributed controller and judging an abnormality based on a difference between a state quantity indicated by the model and a state quantity of an actual plant.

There is also a method of judging an abnormality by a specially provided centralized abnormality diagnosis device.

[0007]

As described above, in a control device using a controller using a plurality of distributed controllers, it takes time to detect an abnormality, and it is difficult to identify an abnormality. There is a problem that appropriate measures and countermeasures cannot be taken. In other words, in the method in which the abnormality of the control loop is known from the alarm of the process state quantity, it takes time for a certain process state quantity to reach the alarm value after the occurrence of the abnormality in the control loop. Since it may take time to specify the abnormal control loop, an abnormality in a certain control loop may greatly disturb the situation of the entire plant.

Further, in the method of judging an abnormality by an alarm check of a control amount, there is a problem that it is impossible to judge whether the sensor is normal and the plant state quantity of the control loop is largely changed, and whether the sensor is abnormal. In addition, the method of adding a redundant sensor has a problem that equipment costs and maintenance costs increase.

Further, in the method using the mathematical model, it is impossible to determine an accurate mathematical model in many cases, it is necessary to prepare mathematical models corresponding to various operating conditions, and it is necessary to prepare an accurate sensor model. If the degree is slightly deteriorated, there is a problem that a correct judgment cannot be made.

In the case of diagnosing in a centralized manner, there is a problem in the reliability of the abnormality diagnosing apparatus. If the diagnosis is performed mutually in order to avoid this problem, inconsistency may occur in the diagnosis result. Depending on the centralized method, abnormalities may not be specified.

The present invention has been made in view of the above circumstances, and an autonomous distributed controller capable of quickly and accurately specifying a control loop in which an abnormality has occurred and taking appropriate action for the abnormality. And a control device using the controller.

[0012]

According to a first aspect of the present invention, there is provided a decentralized controller used in a system which is decentralized and controlled by a plurality of control loops, wherein an operation amount is controlled so that a control amount matches a target value. A feedback control mechanism for giving the self-control object and outputting a state quantity in the self-control loop including the control quantity of the self-control object; and a feedback control mechanism for receiving the state quantity of the self-control loop from the feedback control mechanism. A determination element for determining an abnormality of another control loop based on
Enter the judgment result of this judgment element and set the coupling coefficient,
By using the coupling coefficient and the coupling coefficient input from another control loop, the output sides of the plurality of elements having a predetermined threshold function are respectively coupled to the input sides of the other plurality of elements, and are applied to these input sides. The predetermined input signal is input to each element as an internal state signal by adding a feedback signal from the output side according to the coupling coefficient, and the internal state signal is normalized by the threshold function by each of these elements to perform each control. An interconnected neural network for determining the presence or absence of a loop abnormality;
If the self-control loop is determined to be abnormal by the mutual connection type neural network, the controller is an abnormal-state processing element that outputs an abnormality countermeasure signal to the feedback control mechanism in response to the abnormality.

According to a second aspect of the present invention, there is provided a distributed control apparatus in which a plurality of distributed controllers are provided in a system and the system is distributedly controlled by respective control loops, wherein each of the distributed controllers has a target value. A feedback control mechanism for providing an operation amount to the self-control target such that the control amounts match and outputting a state amount in the self-control loop including the control amount of the self-control target; and a self-control loop based on the feedback control mechanism. Receiving a state quantity of the above, a determination element for determining the presence or absence of an abnormality in the other control loop based on the two state quantities, and a determination result of the determination element are input to set a coupling coefficient, and the coupling coefficient and other control The output sides of the plurality of elements having a predetermined threshold function are respectively coupled to the input sides of the other plurality of elements by the coupling coefficient input from the loop, and the predetermined The input signal is applied to each element as an internal state signal by adding a feedback signal from the output side according to the coupling coefficient, and the internal state signal is normalized by the threshold function by each of these elements to cause an abnormality in each control loop. An interconnected neural network that determines the presence or absence of, and an abnormal processing element that outputs an abnormality countermeasure signal to the feedback control mechanism in response to the abnormality when the self-control loop is determined to be abnormal by the interconnected neural network. And, based on the presence or absence of an abnormality in each control loop determined by the mutual connection type neural network of each distributed controller, the presence or absence of abnormality in each control loop is determined by majority decision logic. When it is different from the judgment by the neural network, the judgment result by the majority logic is given priority. A distributed control system in which a majority logic providing a centralized decision signal to the distributed control unit for.

According to a third aspect of the present invention, there is provided a distributed control apparatus in which a plurality of distributed controllers are provided in a system and the system is distributedly controlled by respective control loops. The controller is configured to provide an operation amount to the self-control target such that the control amount matches the target value, and output a state amount in the self-control loop including the control amount of the self-control target, and a feedback control mechanism. A determining element for receiving the state quantity of the own control loop from the feedback control mechanism and determining whether there is an abnormality in the other control loop based on the two state quantities; and an abnormality countermeasure signal to the feedback control mechanism for the abnormality of the own control loop. Which has a predetermined threshold function with each coupling coefficient set based on the determination result of the determination element of each controller. The output side of the element is respectively coupled to the input side of another plurality of elements, and a predetermined input signal applied to these input sides is added to a feedback signal from the output side by the coupling coefficient to generate an internal state signal. To determine whether or not each control loop is abnormal by normalizing the internal state signal with the threshold function by each of these elements, and when it is determined that there is an abnormality, to the abnormality processing element of the corresponding distributed controller. This is a distributed control device provided with an interconnected shared neural network that provides a signal for performing a process at the time of abnormality.

According to a fourth aspect of the present invention, there is provided a distributed control apparatus wherein a plurality of distributed controllers are provided in a system, and the distributed control of the system is performed by respective control loops. The controller is configured to provide an operation amount to the self-control target such that the control amount matches the target value, and output a state amount in the self-control loop including the control amount of the self-control target, and a feedback control mechanism. A determining element for receiving the state quantity of the own control loop from the feedback control mechanism and determining whether there is an abnormality in the other control loop based on the two state quantities; and an abnormality countermeasure signal to the feedback control mechanism for the abnormality of the own control loop. And an abnormality processing element for outputting a predetermined value, and is provided in parallel with each other, and is provided in a predetermined manner by each coupling coefficient set based on the determination result of the determination element of each controller. The output side of the plurality of the plurality of elements having a threshold function coupled to the input side of the other of the plurality of elements,
A predetermined input signal applied to these inputs is added to the feedback signal from the output side by the coupling coefficient and input to each element as an internal state signal. A plurality of interconnected neural networks that determine the presence or absence of an abnormality in each control loop by normalizing in each of the control loops, and a majority decision logic based on the presence or absence of an abnormality in each control loop determined by each of the interconnected neural networks. This is a distributed control device provided with a majority logic decision mechanism that determines whether or not there is an abnormality and, when it is determined that there is an abnormality, gives a signal for causing an abnormal processing element of the corresponding distributed controller to perform an abnormal process.

[0016]

Therefore, according to the present invention, by taking the above measures, the feedback control mechanism sends out the state quantity of the own control loop, and the judgment element determines the control quantity of this self control loop and the state quantity of the other control loop. The mutual coupling type neural network determines the presence or absence of an abnormality in the other control loop based on the above, and the mutual coupling type neural network sets the coupling coefficient based on the determination result to determine the abnormality of the own control loop. Is transmitted, the abnormality processing element sends an abnormality countermeasure signal corresponding to the abnormality to the feedback control mechanism, so that the control loop in which the abnormality has occurred can be quickly and accurately specified, and the abnormality can be appropriately dealt with. Can be treated.

Further, according to the present invention, the majority logic mechanism receives a judgment result of each control loop sent from each of the interconnected neural networks in the plurality of distributed controllers, and
The decision result for each control loop is checked by majority logic, and when the decision result of each distributed controller for its own control loop contains an error, a centralized decision signal is sent to the distributed controller in order to eliminate the decision result. Therefore, it is possible to monitor the determination of each distributed controller with respect to its own control loop, and to improve the reliability of the abnormal-time process.

Still further, according to the present invention, one interconnected neural network is provided for each of a plurality of feedback control mechanisms, a plurality of judgment elements, and a plurality of abnormality processing elements.
Network, and this interconnected neural network is shared by all control loops, so that the distributed controllers of each control loop are simplified.

Further, according to the present invention, a plurality of interconnected neural networks are provided in parallel for a plurality of feedback control mechanisms, a plurality of decision elements and a plurality of abnormality processing elements, and these interconnected neural networks are provided. -Since the decision result by the neural network is confirmed by the majority logic, the interconnected neural network shared by all control loops is multiplexed, and the reliability can be improved.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Before that, an interconnected neural network for determining whether or not each control loop is abnormal will be described.

First, the mutual coupling type neural network couples the output sides of a plurality of elements having a predetermined threshold function to the input sides of other plural elements by a predetermined coupling coefficient. A predetermined input signal applied to the input side is input to each element as an internal state signal by adding a feedback signal from the output side according to the coupling coefficient, and each element converts the internal state signal into the threshold function. , And is constituted by, for example, an analog electronic circuit as shown in FIG.

Referring to FIG. 1, a mutual coupling type neural network includes a constant voltage source (not shown), N input conductances for converting the constant voltage into constant current inputs Y _{1 to N} , respectively. A parallel circuit of N resistors r and capacitors C to which the inputs Y _{1 to N} are respectively inputted, and N amplifiers AP _{1 to N} , for example, each having an input connected to the input conductance side of these parallel circuits; The output side of these amplifiers AP is branched to form a plurality of other amplifiers AP.
And a coupling coefficient T composed of a plurality of feedback conductances for applying a feedback signal to the input side of the input signal.

[0023] Here, the coupling coefficient when connecting the output voltage V _j of the j-th amplifier AP _j to the input side of the i-th amplifier AP _i and T _ij. This connection corresponds to a synaptic connection in the living body, and the amplifier AP corresponds to one neuron in the living body. When the coupling coefficient T _ij is a negative value,
Although not shown, connected to the input side of the amplifier AP _i an inverted voltage v _j of the amplifier output voltage V _j via the feedback conductance, and inhibitory synaptic connections.

[0024] Now, in the input side of the amplifier connected AP _i by coupling coefficient T _ij, input Y _i as the current ΣT _ij · V _j and the inflow current flowing through the coupling coefficients T _ij as the feedback conductance sum And a capacitor C _i and a resistor r
_Grounded via the parallel circuit of _i . Accordingly, the terminal voltage of the parallel circuit is applied to the i-th amplifier AP _i as an input voltage U _i (internal state signal).

The amplifier AP _i generates an output voltage corresponding to the input voltage U _i in the interval [0,
And outputs the normalized state V _i to 1. This threshold function includes, for example, an input / output static characteristic g as shown in FIG.
Use a sigmoid function with (U _i ).

The characteristics of such a mutual connection type neural network are expressed by the following equations (1) to (4).

[0027]

(Equation 1) Further, such an interconnected neural network can be used when synaptic connections are symmetric (T _ij = T _ji and T _ii =
0), the energy function (Lyapunov function) shown in the following equation (5) decreases. Where (5)
The minimum point of the energy value E in the equation corresponds to the attractor (steady state) of the circuit.

[0028]

(Equation 2) Here, the third term on the right side of Expression (5) is a small value that can usually be ignored.

Next, specific coupling coefficient T _ij and input Y
How to set _i will be described. These coupling coefficients T _ij
The setting of the input Y _i is performed based on a determination result when a determination element (not shown) determines the presence or absence of an abnormality among a plurality of control loops in the system.

Here, the judging element is for judging the presence or absence of an abnormality in another control loop based on the state quantity of its own control loop and the state quantity of another control loop. It is provided for each control mechanism (not shown).

Firstly, among a plurality of control loops, define the state V _i of the i-th control loop as follows.

V _i = 1: Normal, 0: Abnormal.

The determination result _Hji of the control loop j in the determination element of the i-th control loop (hereinafter referred to as control loop i) is determined as follows.

H _ji = −1: when the determining element of the control loop i determines that the control loop j is abnormal, 1: when the determining element of the control loop i determines that the control loop j is normal.

As described above, the state V _i and the judgment result H _ji
Is evaluated by the value of the evaluation function J _{i in} the following equation (6).

[0036]

(Equation 3) Here, depending on the value of the evaluation function J _i obtained by changing the state V _i as follows, to evaluate the value of J _i by substituting the evaluation of altered V _i was again function J _i.

When J _i (H, V)> 0: Set V _i = 1 When J _i (H, V) = 0: V _i = No change When J _i (H, V) <0: Here, V _i = 0. Here, the presence or absence of abnormality in the control loop is determined based on the value of the evaluation function J _i when the state V _i has not changed.

Next, the energy function E 'is defined as in equation (7) (where H _ii = 0).

[0039]

(Equation 4) From this equation (8), it can be seen that the energy -E 'decreases as long as the state V _i is changed by the value of the evaluation function J _i .

Here, the equations (5) and (7) are compared, and the coupling coefficient T _ij and the input Y _i of the mutual connection type neural network are set by the following equations (9) and (10). .

[0041]

(Equation 5) For example, all the initial values of the amplifiers AP are set to “1”, and FIG.
When such a circuit is operated, only the amplifier AP corresponding to the control loop determined to be abnormal as a result of the mutual determination becomes "0", so that the mutual connection type neural network is provided with each control loop. Can be determined whether or not there is an abnormality.

Next, how to determine the determination result _Hji for another control loop among a plurality of control loops in the system in the above description will be specifically described. Here, a boiler steam temperature control system will be described as an example with reference to FIG.

First, in this process, water generated by an evaporation system (not shown) is sprayed with water by a first desuperheater 1 to reduce the temperature. 2 heated. The steam heated by the first heater is again cooled by the second desuperheater 3, and the deheated steam is heated by the second heater 4, and finally becomes main steam and becomes steam. It is configured to be sent to a turbine (not shown).

On the other hand, the boiler temperature control system of this process is constituted by the following four control loops.

The steam temperature at the outlet side of the second heater 4 is detected by the main steam sensor 16 and this detected temperature is compared with the comparator 1.
7, a feedback control mechanism 18 is operated so that the deviation becomes zero as compared with the target value, and a fourth control loop for outputting as a target value of the outlet temperature of the second desuperheater 3 is formed.

The steam temperature at the outlet of the second desuperheater 3 is detected by a temperature sensor 12 and this detected temperature is compared with a comparator 13.
By comparing with the target value obtained from the fourth control loop,
A third control loop that operates the third feedback control mechanism 14 so that the deviation becomes zero, and provides an opening command to the spray valve 15 that controls the amount of water supplied to the second desuperheater 3. Is configured.

The steam temperature at the outlet side of the first heater 2 is detected by a temperature sensor 9, this detected temperature is compared with a target value by a comparator 10, and a feedback control mechanism 11 is operated so that the deviation becomes zero. A second control loop for operating and outputting as a target value of the outlet temperature of the first heater 2 is configured.

The outlet temperature of the first desuperheater 1 is detected by a temperature sensor 5, and this detected temperature is compared by a comparator 6 to a second temperature.
The feedback control mechanism 7 is operated so that the deviation becomes zero, as compared with the target value obtained from the control loop, and the spray valve for controlling the spray amount of water supplied to the first desuperheater 1 is controlled. To give the opening command to the spray valve 8
Of the control loop.

Here, the temperatures detected by the temperature sensors 5, 9, 12, 16 in the first, second, third, and fourth control loops are 340 degrees, 430 degrees, 390 degrees, and 515 degrees, respectively. is there.

The feedback control mechanisms of the first, second, third and fourth control loops are provided via corresponding first, second, third and fourth determination elements, which will be described in detail later. The first to fourth interconnected neural networks and the first to fourth abnormal-time processing elements are connected, and when an abnormality is determined, these abnormal-time processing elements are provided to the corresponding feedback control mechanism. , An abnormality countermeasure signal for taking a predetermined action is transmitted to minimize the adverse effect on the process due to the abnormality.

Next, a description will be given of a case where a judgment result _Hji is obtained by judging the presence or absence of an abnormality in another control loop among a plurality of control loops.

First, the first determination element determines, based on the first control loop, whether there is an abnormality in the second control loop. Specifically, a first predetermined temperature (340 degrees) which is a control amount of the first control loop is compared with a second predetermined temperature (430 degrees) which is a control amount of the second control loop. The steam that has exited the pre-cooler should be heated by the pre-heater and become high temperature, so it is determined to be normal and H ₂₁ = 1. Conversely, when the first control loop is determined from the second control loop by the second determination element, it is also determined to be normal, and H ₁₂ = 1.

Similarly, since the mutual judgment between the first control loop and the fourth control loop is both normal, H ₄₁ = H ₁₄ =
Let it be 1.

Next, the second judging element judges the third control loop based on the second control loop. Here, while the third-stage spray valve opening degree, which is the operation amount of the third control loop, is set to zero (fully closed) so as not to lower the third predetermined temperature, the control amount of the second control loop is used. The third predetermined temperature (390 degrees), which is the control amount of the third control loop, is lower than a certain second predetermined temperature (430 degrees).

For this reason, the second determining element determines that the third control loop for the second control loop is abnormal,
₃₂ = −1. Conversely, if the third control element determines the second control loop based on the third control loop,
If the spray amount is zero and the control amount of the second control loop is high, it is strange that the control amount is abnormal and it is determined that H ₂₃ = −1.
And

Next, the third judgment element judges the control loop 4 based on the third control loop 3. Here, since the control amount of the fourth control loop is higher than the control amount of the third control loop, it is determined to be normal, and H ₄₁ = 1. Conversely, when the third control loop is determined based on the fourth control loop by the fourth determination element, the second-stage spray valve opening degree, which is the operation amount of the third control loop, is zero, Since the control amount is 390 degrees and has not reached the target value of 400 degrees, which is also the operation amount of the fourth control loop, it is determined that there is an abnormality, and H ₃₄ = −1.

Here, the respective judgment results _Hji are arranged and shown in FIG. As shown in the figure, the determination result H for another control loop
_{Since ji} is obtained, the coupling coefficient T _ij and the input Y _i of the mutual coupling type neural network are set by the above-mentioned equation (9).

Next, the operation of the mutual coupling type neural network for determining whether or not each control loop is abnormal when the coupling coefficient is set as described above will be described.
Since this operation is equivalently represented by the evaluation function J _i (H, V), the operation will be described using the expression (6) indicating J _i . Here, when the determination result _Hji of the boiler steam temperature control system described above is substituted into the expression (6), the following expressions (11) to (14) are obtained.

[0059] _{J 1 = 2V 2 + 2V 4} -2 ... (11) J 2 = 2V 1 -2V 3 -1 ... (12) J 3 = -2V 2 -1 ... (13) J 4 = 2V 1 -1 ... (14) First, each control loop is brought into a normal state (V ₁ = V ₂ = V
₃ = V ₄ = 1), each neuron element (amplifier A
The operation is performed with the initial value of P _{1 to 4} ) set to “1”.

1) J ₁ = 2 + 2-2 = 2> 0: V ₁ = 1 2) J ₂ = 2-2-1 = −1 <0: V ₂ is changed from “1” to “0”.

[0061] _{3) J 3 = -1 <0} : the V ₃ change from "1" to "0".

[0062] _{4) J 4 = 2-1 = 1} > 0: V 4 = 1 5) J 1 = 2-2 = 1> 0: V 1 = 1 6) J 2 = 2-1 = 1> 0: the V ₂ from "0" change to "1".

7) J ₃ = −2-1 = −3 <0: V ₃ = 0 8) J ₄ = 2-1 = 1> 0: V ₄ = 1 9) J ₁ = 2 + 2-2 = 2> _{0: V 1 = 1 10)} J 2 = 2-1 = 1> 0: V 2 = 1 11) J 3 = -2-1 = -3 <0: V 3 = 0 12) J 4 = 2-1 _{= 1> 0: V 4 =} 1 is judged whether the abnormality of the control loop with the value of the evaluation function J _i in this way when the state V _i becomes in a state that does not vary. Here, since V1 = V2 = V3 = 1, V 3 = 0, the third control loop is determined to be abnormal. That is, among the outputs of the amplifiers AP _{1 to} AP ₄ , the amplifier AP ₃
Is "0" only.

Therefore, the mutual connection type neural network can determine the presence or absence of an abnormality in each of the plurality of control loops by the above operation.

When this judgment is made by the mutual connection type neural network relating to the third control loop, the self-control loop is judged to be abnormal, and an alarm is issued by the abnormal-time processing element, and the spray is performed. The mode is switched from the automatic control to the manual control as needed by warning the valve leakage and the temperature sensor abnormality. Further, since such a determination is made also in the interconnected neural network of the fourth control loop, It turns out that the minor control loop of the cascade control is abnormal. For the fourth control loop, if the operation is continued in the automatic control mode in a state where the operation end does not move, the integrator of the feedback control mechanism is saturated, and the excessive operation amount, that is, the target value of the third control loop is excessive. And the third-stage control valve continues to close even if the third control loop returns to normal. As a result, the main steam temperature is greatly disturbed. For this reason, the abnormality processing element of the fourth control loop takes measures such as switching its own control loop to the manual control mode.

Next, an autonomous distributed controller according to the present invention and a control device using the controller will be described with reference to the drawings.

FIG. 1 is a diagram showing the configuration of an autonomous distributed controller according to the present invention. First, the autonomous distributed controller 21 is used for each of a plurality of control loops in the system, and a feedback control mechanism 22 for controlling a control target in the control loop, and a state of the own control loop received from the feedback control mechanism. A judgment element 23 for judging the presence or absence of an abnormality in another control loop based on the amount; an interconnected neural network 24 for judging the presence or absence of an abnormality in each control loop based on the result of the judgment; New
An abnormality processing element 25 for transmitting an abnormality countermeasure signal to the feedback control mechanism 22 based on the determination of the ral network 24 for its own control loop.

Here, in particular, the judgment element 23, the mutual connection type neural network 24 and the abnormality processing element 25
Determines the presence or absence of an abnormality in the own control loop based on the state quantity of the own control loop received from the feedback control mechanism 22 and the state quantity of the other control loop received from another control loop. Mutual abnormality determination processing mechanism 26
Is configured.

The feedback control mechanism 22 gives an operation amount to the self-control target so that the control amount matches the target value, and sends out a state amount in the self-control loop including the control amount of the self-control target. Has the ability to

The determination element 23 receives the state quantity of the self-control loop and receives the state quantity from another control loop, and determines whether there is an abnormality in the other control loop based on the two state quantities. in, for example, the state quantity of the control loop i with respect to determining the state of the control loop j, and "1" judgment result H _ji if successful, the H _ji if abnormal "-
1 ".

The Mutually Connected Neural Network 2
4 couples the output sides of the plurality of elements having a predetermined threshold function to the input sides of the other plurality of elements by a coupling coefficient based on the determination result of the determination element 23, and adds the input sides to these input sides. The predetermined input signal is added to each element as an internal state signal by adding a feedback signal from the output side by the coupling coefficient, and each element normalizes the internal state signal by the threshold function. This is for determining the presence / absence of an abnormality in each control loop. For example, upon receiving the determination results _Hji from the respective determination elements 23, each coupling coefficient is obtained by equation (9) and internally set. (Amplifiers AP _{1 to N} ) are each set to “1” to operate and settle in a certain steady state to determine whether or not each control loop is abnormal.

When the mutual connection type neural network 24 determines that the self-control loop is abnormal, the abnormal time processing element 25 sends a control amount, a target value, and a control value to the feedback control mechanism 22 in response to the abnormality. An abnormal countermeasure signal indicating a control parameter value or a manual switching command is transmitted. For example, when the self-control loop is determined to be abnormal, the mutual connection type neural network 24 transmits a self-abnormal signal, and the self-abnormal signal is transmitted. When the signal is received, the state quantity is taken in from the determination element 23, the type of abnormality is specified, and an abnormality countermeasure signal predetermined for each type of abnormality is transmitted to the feedback control mechanism 22.

Next, the operation of such an autonomous distributed controller will be described.

First, a plurality of control loops operate in the system, and an autonomous distributed controller is provided for each of these control loops.

First, while controlling the control loop, the feedback control mechanism 22 takes in the state quantity in its own control loop and sends out this state quantity. When receiving the state quantity of the self-control loop and receiving the state quantity of the other control loop from another autonomous distributed controller, the determination element refers to these two state quantities to determine the abnormality of the other control loop. Is determined. For example, the determination here as abnormal is
For example, the following is the case.

(A) In a fluid system in which fluids join on the way, the flow rate on the upstream side is smaller than the flow rate on the downstream side.

(B) In the fluid system, the flow rate does not increase even though the differential pressure between the upstream and the downstream has increased.

(C) The case where the liquid level is lowered even though the inflow amount into the tank is larger than the outflow amount.

(D) A case in which the temperature is expected to decrease due to heat radiation, but the temperature is high.

(E) A case where the fluctuation amplitude at the point B is increasing, although it should be attenuated, relative to the fluctuation amplitude of the control amount at the point A.

In such a case or the like, the judgment element 23 sets the judgment result _Hji to “−1” indicating an abnormality to indicate the mutual coupling type news.
It is sent to the LAL network 24.

This mutual connection type neural network 2
4 receives the judgment result from each judgment element 23 provided in each control loop, and calculates the coupling coefficient T _{ji according to the} equation (9).
And prompted Y _i Function setting. Thereafter, all the amplifiers AP _{1 to N} corresponding to the respective control loops are initialized to “1” and operated to settle down to a certain steady state.

That is, in the judgment result of the judgment element 23, a proper result cannot be obtained when the self-control loop is abnormal in order to judge the presence or absence of an abnormality in the other control loop based on the self-control loop. Utilizing the energy minimizing function of the mutual connection type neural network 24 as shown in the equations (5) and (7), each judgment result based on each of a plurality of control loops is integrated, and each control is performed. The presence or absence of a loop abnormality is determined.

[0084] Here, when a steady state is only amplifier AP _i is indicates "0", the control loop i Interconnected New of - Rarunettowa - clause, abnormality determination to self abnormal signal self control loop i Is sent. The abnormal time processing element 25
When the self-abnormality signal is received, the state quantity is taken in from the determination element, the type of the abnormality is specified, and the feedback control mechanism 22 takes an abnormality countermeasure so that a countermeasure defined in the control loop is taken from a failsafe viewpoint. Send a signal. Here, the measures defined in the control loop are, for example, the following.

(F) Notification of occurrence of abnormality to the operator.

(G) Forcible switching from automatic control to manual control.

(H) Output a predetermined value on the safe side as the manipulated variable.

(I) Switching of control parameter values to values robust to changes in the characteristics of the control object.

(N) Use of target value as estimated value of control amount.

Feedback control mechanism 2 of control loop i
2 receives the abnormality countermeasure signal indicating such a countermeasure,
An abnormality countermeasure is taken based on the content of the abnormality countermeasure signal.

As described above, the autonomous distributed controller according to the present embodiment autonomously determines the abnormality of the self-control loop in consideration of information from other control loops, and determines an appropriate predetermined abnormality. It can take a precautionary measure and perform a fail-safe function.

That is, when an abnormality is found, the operator immediately issues an alarm for facilitating appropriate processing, automatically sends an appropriate operation amount as an operation amount of the abnormality control loop, A decentralized controller can be obtained that sends a reasonable estimate of the quantity to others.

In addition, it is possible to save labor by automating abnormal time processing and greatly contribute to the stability and safe operation of the plant. Further, it is possible to prevent a decrease in plant efficiency, the occurrence of defective products and the breakage of the plant. As a result, a distributed controller that contributes to environmental conservation can be obtained.

Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 2 is a diagram showing the configuration of the autonomous distributed control device. This distributed control device includes the above-mentioned autonomous distributed controllers ₂₁₁ to _N for each of a plurality of, for example, N control loops in the system, and each of these N autonomous distributed controllers is one. It is configured to be connected to the majority logic 31.

Here, the majority logic 31 determines whether or not each control loop determined by each interconnected neural network 24 ₁ -N in all the autonomous distributed controllers 21 ₁ -N has an abnormality. Based on the majority logic, it is finally determined whether or not each control loop is abnormal. If the final determination differs from the determination by the mutual connection type neural network 24, the final determination result is prioritized. Thus, the centralized determination signal is transmitted to the autonomous distributed controller 21.

Next, the operation of such an autonomous distributed control device will be described. First, N autonomous distributed controllers 21
_{1 to N} each determine the presence or absence of an abnormality in the N control loops based on the steady state of the N artificial neuron elements in the interconnected neural network 24.

Here, each autonomous distributed controller 21 _{1 -N}
Sends the determination results for each of the N control loops to the majority logic mechanism 31 as autonomous distributed determination signals. The majority decision logic mechanism 31 finally decides the judgment results on the N control loops received from the autonomous distributed controllers ₂₁₁ to N by majority logic for each control loop.

More specifically, for example, there are four control loops in the system (when N = 4), and the four autonomous distributed controllers ₂₁₁ to ₄ respectively control each control loop as follows. It is assumed that the result of the determination is sent to the majority logic 31.

Distributed controller: 1st, 2nd, 3rd, 4th, 1st control loop: 1, 1, 1, 1, 2nd control loop: 1, 0, 1, 1, 3rd Control loop: 0, 0, 0, 0, fourth control loop: 1, 1, 1, 1, 1 (“1” indicates normal, “0” indicates abnormal) Here, the majority logic 31 is a majority logic. As a result, the third control loop having a large number of "0" is finally determined to be abnormal, and the result is transmitted to each of the autonomous distributed controllers ₂₁₁ to ₄ as a concentration determination signal. Each autonomous distributed controller 21
_{In 1, 2, and 4} , since this concentration determination signal determines that the self-control loop is normal, even if a self-abnormal signal is generated,
This self-abnormality signal is invalidated to prohibit the processing at the time of the abnormality, and an alarm is issued to the effect that the malfunction has occurred.
In autonomous distributed control unit 21 _3, for concentration check signal it is determined to be abnormal self control loop, take action against abnormal.

That is, an erroneous determination result that the second control loop is abnormal due to, for example, a malfunction of the second autonomous distributed controller causes a second mutual connection type neural network 24 _2. be transmitted from the abnormality processing elements 25 _2, because the entire system is identified as the control loop of the final abnormalities from a number of determination results relative to the other normal control loop a third control loop, the as invalid self abnormality signal transmitted to the abnormality processing element 25 ₂ by malfunction can correct abnormal take measures.

As described above, according to the second embodiment,
Since the decision results of each control loop by the plurality of autonomous distributed controllers 21 _{1 to N} are checked for rationality by majority logic, and finally the presence or absence of abnormality in each control loop is determined.
Even when a self-abnormality signal is generated due to a malfunction of the judgment element 23 of the self-control loop or the mutual connection type neural network 24, it is possible to prevent an erroneous countermeasure from being taken erroneously. Can be realized.

Further, since the malfunctioning distributed controller 21 can be specified by the majority logic, the majority logic 31 can be used as a monitor of the mutual abnormality judging mechanism 26 in each autonomous distributed controller 21.

Next, a third embodiment of the present invention will be described with reference to the drawings. FIG. 3 is a diagram showing the configuration of the autonomous distributed control device. This autonomous distributed control device comprises a plurality of distributed controllers 32 _{1 to} 32 _{1 which} are configured by removing the interconnected neural network 24 from the autonomous distributed controller 21 of FIG. 1 provided in each control loop. _N, and an interconnected shared neural network 33 connected to each of the distributed controllers 32 ₁ to 32 _N.

That is, the mutual connection type neural network 24 which each autonomous distributed controller 21 individually has.
Is a distributed control device having a configuration sharing the same.

Next, the operation of such an autonomous distributed control device will be described. First, the mutual connection type shared neural network 33 is connected to the distributed controllers 31 _{1 to N of the} respective control loops.
Each receives a determination result from obtaining the respective coupling coefficients T _ji and input Y _i by (9). Thereafter, performs internal setting seeking final coupling coefficient T _ji and input Y _i from these coupling coefficients T _ji and input Y _i by majority logic or average value.

When the internal setting is completed as described above, the mutual connection type shared neural network 33 can determine whether or not each control loop is abnormal by creating a steady state.

As described above, according to the third embodiment,
Since the entire system can be autonomously distributed and controlled even with one interconnected neural network 33, the whole apparatus can be made relatively simple, small, and inexpensive.

Next, a fourth embodiment of the present invention will be described with reference to the drawings. FIG. 4 is a diagram showing the configuration of the autonomous distributed control device according to the present invention. This distributed control device is an interconnected shared neural network shown in FIG.
Interconnect 33, a plurality of interconnected type
A mutual decision majority decision mechanism 34 for finally judging the presence or absence of abnormality in each control loop by majority logic based on the determination of the ral networks 24 _{1 to} 24 _M is provided.

That is, the configuration is such that the mutual determination majority decision mechanism 34 is connected to each of the distributed controllers 32 _{1 to N} provided in a plurality of control loops in the system. Here, mutual determination majority mechanism 34, front to be connected in parallel a plurality a cross-linked New - Rarunettowa - and click 24 _{1 ~M,} subsequent to these mutually New - Rarunettowa - from the determination results of the click 24 _{1 ~M} Finally, a majority logic decision mechanism 35 for judging the presence or absence of an abnormality in each control loop is provided.

The number of these interconnected neural networks 24 ₁ -M is independent of the number of control loops, but internally the number of elements is made equal to the number of control loops. Setting equal coupling coefficients T among the ral networks 241 to _M ;

Next, the operation of such an autonomous distributed control device will be described. First, a plurality of interconnected neural networks 24 ₁ -M connected in parallel receive the respective judgment elements from the distributed controllers 32 ₁ -N of the respective control loops, and make a final decision similarly to the third embodiment. performs internal setting seeking coupling coefficient T _ji and input Y _i.

When the internal setting is completed as described above, each of the interconnected neural networks 24 _{1 to} 24 _M determines the presence or absence of an abnormality in each control loop by forming a steady state, and determines the determination result by majority logic. It is sent to the decision mechanism 35. The majority logic decision mechanism 35 finally determines the presence / absence of abnormality in each control loop by majority logic based on the results of the determination by the plurality of interconnected neural networks 24 _{1 to} 24M, and performs control indicating the abnormality. Sends a self-abnormal signal to the autonomous distributed controller of the loop.

The autonomous distributed controller that receives the self-abnormal signal specifies the type of the abnormality and performs an appropriate process for the abnormality.

As described above, according to the fourth embodiment,
Since the mutual connection type neural network 24 is multiplexed, it is possible to cope with a malfunction of the mutual connection type neural network 24 and to provide a control device in which the reliability of the entire system is improved.

In the present invention, the threshold function of each artificial neuron element has been described as a sigmoid function.
A threshold function having a static characteristic g (U _i ) other than the sigmoid function can be similarly implemented. Also, the input Y _i has been described as a current corresponding to this value, but the present invention can be similarly implemented as the conductance of a resistor connected to a constant voltage source. Furthermore, it has been described that the interconnected neural network 24 is constituted by an analog electronic circuit, but (1)
The same operation can be performed by changing the equation to a difference equation and using a digital arithmetic circuit.

Furthermore, it has been described that the judgment element 23 can always judge whether it is abnormal or normal. However, if it is impossible to judge whether it is normal or abnormal, _Hji = 0, or if it is surely abnormal, it is set to "-1". H _ji is determined by fuzzy estimation using a membership function that is “1” when the state is normal.
Can be similarly implemented.

[0117] Further, although the decision element 23 has been described to deliver a determination result to the interconnecting neural network, by substituting the value of the determination element 23 is determination result H _ji in (9), New - Rarunettowa The same effect can be obtained by determining the coupling coefficient T _ij and the input Y _i of the negative link and transmitting them to the mutual connection type neural network.

Further, it has been described that the mutual connection type neural network transmits a self-abnormality signal to the abnormal time processing element when the self-control loop is determined to be abnormal. Even when the determination is made, other control loop abnormal signals can be transmitted to the abnormal time processing element and external equipment to perform a predetermined fail-safe action, and the same can be implemented.

The mutual connection type neural network transmits an autonomous dispersion determination signal indicating the presence or absence of abnormality of each control loop held in each artificial neuron element to an external device. Can be similarly implemented.

Further, the processing element at the time of abnormality has been described to invalidate the self-abnormal signal due to malfunction upon receiving the concentration determination signal. Even if the transmitted self or other control loop abnormality signal is accepted, rejected, or suppressed, the abnormality processing control can be performed as described above.

Although the feedback control mechanism has been described as controlling one control amount, the feedback control mechanism may execute other input / output control for simultaneously controlling a plurality of related control amounts simultaneously. It can be implemented similarly. Further, the feedback control mechanism has been described to control the control amount and transmit the state amount. However, the same can be implemented by controlling the control amount and transmitting only the control amount. In addition, the present invention can be implemented with various modifications without departing from the scope of the invention.

[0122]

As described above, according to the present invention, the feedback control mechanism sends out the state quantity of the self-control loop, and the judgment element is based on the control quantity of this self-control loop and the state quantity of the other control loop. The presence / absence of an abnormality in another control loop is determined, and the mutual coupling type neural network sets a coupling coefficient based on the determination result, determines an abnormality in the own control loop, and sends a self-abnormal signal when an abnormality occurs. Thus, the abnormality processing element sends an abnormality countermeasure signal corresponding to the abnormality to the feedback control mechanism, so that the control loop in which the abnormality has occurred can be quickly and accurately specified, and appropriate measures are taken for the abnormality. And a control device using the controller.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of an embodiment of an autonomous distributed controller according to the present invention.

FIG. 2 shows a second example using an autonomous distributed controller according to the present invention.
The figure which shows the structure of the distributed control apparatus which is an Example of FIG.

FIG. 3 shows a third example using the autonomous distributed controller according to the present invention.
The figure which shows the structure of the distributed control apparatus which is an Example of FIG.

FIG. 4 shows a fourth example using the autonomous distributed controller according to the present invention.
The figure which shows the structure of the distributed control apparatus which is an Example of FIG.

FIG. 5 is a diagram showing a configuration of a mutual connection type neural network using an analog electronic circuit.

FIG. 6 is a diagram showing characteristics of a threshold function of the artificial neuron element.

FIG. 7 is a diagram showing a system configuration of a boiler steam temperature control system.

FIG. 8 is a diagram showing a mutual determination result and a determination state in the system of FIG. 7;

[Explanation of symbols]

22 feedback control mechanism 23 judgment element 24
... Mutual connection type neural network, 25... Abnormal processing element, 31... Majority decision logic mechanism, 33.

Continued on the front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G05B 23/02 G06F 15/16 G06F 15/18 G06G 7/60 G05B 13/02

Claims (4)

(57) [Claims] 1. A decentralized controller used in a system that is decentralized by a plurality of control loops, wherein an operation amount is given to a subject to be controlled such that the control amount matches a target value, and control of the subject to be controlled is performed. A feedback control mechanism that outputs a state quantity in the self-control loop including the quantity, and a determination element that, when receiving the state quantity of the self-control loop from the feedback control mechanism, determines an abnormality of another control loop based on the two state quantities. And a coupling coefficient is set by inputting the determination result of the determination element, and the output sides of a plurality of elements having a predetermined threshold function are determined by the coupling coefficient and the coupling coefficient input from another control loop. Are respectively coupled to the input sides of a plurality of elements, and a predetermined input signal applied to these input sides is added to a feedback signal from the output side by the coupling coefficient to generate an internal state signal for each element. The mutual connection type neural network which inputs and normalizes the internal state signal by the threshold function by each of these elements to determine whether or not each control loop is abnormal, and the self control loop is abnormal by this mutual connection type neural network. And an abnormality processing element for outputting an abnormality countermeasure signal to the feedback control mechanism in response to the abnormality. 2. A distributed control apparatus in which a plurality of distributed controllers are provided in a system and the system is distributedly controlled by respective control loops, wherein each of the distributed controllers adjusts a control amount to a target value. Give the operation amount to the self-control target,
A feedback control mechanism for outputting a state quantity in the self-control loop including the control quantity of the self-control target; and a control loop based on the two state quantities upon receiving the state quantity of the self-control loop from the feedback control mechanism. A determination element for determining the presence / absence of abnormality and a determination result of the determination element are input to set a coupling coefficient, and a predetermined threshold function is determined by the coupling coefficient and the coupling coefficient input from another control loop. The output sides of the plurality of elements are respectively coupled to the input sides of the other plurality of elements, and a predetermined input signal applied to these input sides is added to a feedback signal from the output side by the coupling coefficient to generate an internal state signal. As an input to each element, and the internal state signal is normalized by the threshold function using these elements to determine whether each control loop is abnormal. An abnormality processing element that outputs an abnormality countermeasure signal to the feedback control mechanism when the self-control loop is determined to be abnormal by the mutual connection type neural network, and for each of the distributed controllers. The majority logic is used to determine the presence or absence of abnormality in each control loop based on the presence or absence of abnormality in each control loop determined by the mutual connection type neural network.If the determination result is different from the determination by the mutual connection type neural network, the majority logic is used. A distributed control device characterized by comprising a majority logic mechanism for giving a centralized determination signal to a corresponding distributed controller so as to give priority to a determination result obtained by the control method. 3. A distributed control device in which a plurality of distributed controllers are provided in a system and the system is distributedly controlled by respective control loops, wherein each of the distributed controllers is configured such that each of the distributed controllers has a target value. A feedback control mechanism for providing an operation amount to the self-control target such that the control amounts match and outputting a state amount in the self-control loop including the control amount of the self-control target; and a self-control loop based on the feedback control mechanism. A determination element for determining the presence or absence of an abnormality in another control loop based on the two state quantities upon receiving the state quantity, and an abnormality processing element for outputting an abnormality countermeasure signal to the feedback control mechanism in response to an abnormality in the own control loop. And the output side of a plurality of elements having a predetermined threshold function by each coupling coefficient set based on the determination result of the determination element of each controller, Each of the elements is coupled to the input side of the element, and a predetermined input signal applied to the input side is added to a feedback signal from the output side by the coupling coefficient and input to each element as an internal state signal. The internal state signal is normalized by the threshold function to determine the presence / absence of an abnormality in each control loop, and when it is determined that there is an abnormality, a signal for causing the abnormality processing element of the corresponding distributed controller to perform an abnormality processing is provided. A distributed control device, characterized in that a shared neural network is provided. 4. A distributed control apparatus in which a plurality of distributed controllers are provided in a system and the system is distributedly controlled by respective control loops, wherein each of the distributed controllers is configured such that each of the distributed controllers has a target value. A feedback control mechanism for providing an operation amount to the self-control target such that the control amounts match and outputting a state amount in the self-control loop including the control amount of the self-control target; and a self-control loop based on the feedback control mechanism. A determination element for determining the presence or absence of an abnormality in another control loop based on the two state quantities upon receiving the state quantity, and an abnormality processing element for outputting an abnormality countermeasure signal to the feedback control mechanism in response to an abnormality in the own control loop. And a plurality of multiple units connected in parallel and having a predetermined threshold function with each coupling coefficient set based on the determination result of the determination element of each controller. The output sides of the number of elements are respectively coupled to the input sides of a plurality of other elements, and a predetermined input signal applied to these input sides is added to a feedback signal from the output side by the coupling coefficient to form an internal state signal. Input to each element,
A plurality of interconnected neural networks that determine whether or not each control loop is abnormal by normalizing the internal state signal with the threshold function by each of these elements; and a control loop determined by each of the interconnected neural networks. Majority logic that determines the presence or absence of an abnormality in each control loop based on majority logic based on the presence or absence of an abnormality, and gives a signal to cause the processing element of the corresponding distributed controller to perform abnormal processing when an abnormality is determined. A decentralized control device comprising a decision mechanism.