JPH06105414A - Driving controller - Google Patents

Abstract

PURPOSE:To reduce the number of entire fuzzy rules by facilitating extraction of a fuzzy rule before starting of driving and refinement of the rule after start ing of the driving. CONSTITUTION:The driving controller comprises an objective value generator 25 for generating an objective value of a control object by a fuzzy rule based on a driving control knowledge to be transited from a control pattern to be decided according to the control object to be controlled to a control pattern, and a stabilizing controller 26 for generating a manipulated variable of the control object by the rule based on an optimum driving control knowledge in a steady state to be decided according dynamic characteristics of the control object.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an operation control device for a vehicle such as a traffic system whose control purpose changes from dynamic characteristics such as acceleration and weight and speed control to position control.

[0002]

2. Description of the Related Art Fuzzy control characterizes expressions such as "high affinity", "high" and "low" in terms of linguistic "high affinity", which are easy for humans to understand, by means of membership functions, etc. Control knowledge can be expressed by a small number of if-then rules. Therefore, it is also applied to a vehicle operation control device used in a traffic system or the like. Further, in order to semi-automatically learn the control knowledge expressed by the fuzzy rule, various fusion methods of neural network and fuzzy control have been proposed. In particular, application to an adaptive controller that stabilizes a target system whose dynamic characteristics fluctuate, and application to an adaptive controller that tunes the feedforward manipulated variable from a target generator are active.

By the way, in a vehicle driving control device used in a traffic system or the like, based on the knowledge of a skilled driver, a simple driving operation is synthesized to realize intelligent control corresponding to many control purposes and complicated dynamic characteristic fluctuations. In order to realize operation control, "a) optimum operation control knowledge in a steady state determined by the dynamic characteristics of the vehicle" that is possessed by the conventional fuzzy control system is not sufficient, and "b)" as a higher concept. Operation control knowledge for transitioning from a control pattern determined by the control purpose of the vehicle to the control pattern is required.

However, conventional fuzzy if-then
When these rules are used to represent all of these driving control knowledge in a brute force manner, a huge number of fuzzy rules are required to represent many complex if conditions and driving operations corresponding to those conditions. . Moreover, since the driving operation knowledge corresponding to each condition is not always clear,
It becomes difficult to extract driving control knowledge from a skilled driver before starting driving. Further, it is difficult to refine the extracted operation control knowledge after the start of operation due to the influence of vague interference between fuzzy rules whose conditions are close to each other.

[0005]

The present invention has been made under such circumstances, and facilitates the extraction of fuzzy rules before the start of driving and the refinement of fuzzy rules after the start of driving, and the number of fuzzy rules as a whole. It is an object of the present invention to provide an operation control device that reduces power consumption.

[0006]

In order to solve such a problem, the operation control device of the first invention is, in an operation control device for performing operation control by fuzzy control, from a control pattern determined by a control object of a control target to a control pattern. The target value generator that generates the target value for the control purpose by the fuzzy rule based on the operation control knowledge and the operation of the control purpose by the fuzzy rule based on the optimum operation control knowledge in the steady state determined by the dynamic characteristics of the controlled object A stabilizing controller section for generating a quantity.

According to a second aspect of the present invention, in the operation control device of the first aspect, the target value generator section has a plurality of target generators each having an adjustable parameter and for inputting a state quantity to be controlled, and these target generators. A target value synthesizer that multiplies the output of the target generator by a weight variable and sums it, an adder that sums the output of this target value synthesizer and the target value correction signal, and A first state evaluator for determining a state evaluation value,
A first fuzzy reasoner for determining a weight variable of the target value synthesizer by fuzzy inference according to a state evaluation value of the first state evaluator, a parameter of the target generator and the first fuzzy inference unit And a first parameter rule learning device that learns the rule in accordance with the state of the control system or the target value correction signal, wherein the stabilizing controller section has adjustable parameters, and the target value generator Control unit that inputs the target value and the state quantity of the controlled object that are output from the control unit, the manipulated variable synthesizer that sums the outputs of these stabilized controllers by weighting variables, and this manipulated variable synthesizer. An adder that sums the output of the controller and the manipulated variable correction signal, a second state evaluator that determines a state evaluation value according to a change in the dynamic characteristics of the controlled object, and a weight variable of the manipulated variable combiner. State evaluation of the second state evaluator A second fuzzy reasoner that is determined by fuzzy reasoning according to the above, and a parameter for the stabilizing controller and a rule for the second fuzzy reasoner that are learned according to the state of the control system or the manipulated variable correction signal. Two
And a parameter rule learning device of.

[0008]

In the present invention, the fuzzy rules for operation control are changed from "a) optimal operation control knowledge in a steady state determined by the dynamic characteristics of the vehicle" to "b) control pattern determined by the control purpose of the vehicle". For the purpose of controlling the vehicle by the stabilizing controller section based on "a) Optimal operation control knowledge in a steady state determined by the dynamic characteristics of the vehicle". Since the target value generator section is configured to perform the control based on the "operation control knowledge for changing from the determined control pattern to the control pattern", the number of fuzzy rules can be significantly reduced. Further, when the rule base is constructed, only the rules in each steady state need to be extracted from the skilled driver, the prospect of rule acquisition is improved, and oversight of control rules can be prevented. In addition, it enables online learning to refine the rules after they go live.

[0009]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an operation control device according to an embodiment of the present invention. The operation control device shown in the figure is divided into a target value generator unit 25 and a stabilization controller unit 26. The target value generator unit 25 and the stabilization controller unit 26 have the same structure. As a result, the entire operation control device has a hierarchical structure. The signals represented by thick arrow lines in the figure are all vector quantities.

Here, the target value generator 25 performs parameter learning in advance so as to generate an optimum target value according to the control purpose (speed control, fixed position stop control, etc.) of the vehicle 22 to be controlled. With a plurality of target value generators 10, 10, ...

The target value synthesizer 11 adds the outputs r1 ^* , r2 ^* , ... rn ^* of these target value generators 10, 10 ... In proportion to the magnitudes of the respective weight variables Wr1, Wr2 ,. To synthesize.

The adder 12 outputs the output r of the target value synthesizer 11.
After adding the target value correction signal to ^* , the addition result is output to the stabilization controller 26.

The state evaluator A13 determines a state evaluation value based on the state quantity of the vehicle 22 and an evaluation function with the control purpose as an input.

The fuzzy inference unit A14 infers the state evaluation value based on the target value generation rule 56,
A weight Pr Wr1, Wr2, ... Wrn that determines the parameter Pr of the target generator 10 and the ratio of their output amounts obtained from the matching rules is determined.

The parameter rule learning device A15 receives a target value correction signal from the outside so that the target value correction signal decreases, and the parameter Pr of the target value generators 10, 10 ... And the target value of the fuzzy inference unit A14. Generation rule 5
Fix 6.

On the other hand, the stabilization controller unit 26 is used for the vehicle 2
A plurality of stabilization controllers 16 that have been subjected to parameter learning in advance so as to generate an optimum manipulated variable in accordance with the dynamic characteristic fluctuation of 2 (acceleration / deceleration fluctuation, vehicle weight fluctuation, etc.),
Has 16, ...

The stabilization controller 16 inputs the target value which is the output of the adder 12 and the state quantity y of the controlled object, and outputs the manipulated variables u1, u2, ..., Un corresponding to each situation.

The operation amount synthesizer 17 adds and synthesizes the outputs u1, u2, ..., Un of the stabilization controller 16 in proportion to the magnitudes of the respective weight variables Wc1, Wc2 ,.

The adder 18 outputs the output u of the manipulated variable synthesizer 17.
After adding the manipulated variable correction signal to, the signal is output to the vehicle 22 that is the control target.

The same design method can be applied to the stabilization controller 16 by using not only the feedback compensation element but also the feedforward compensation element.

The state evaluator B19 determines a state evaluation value by an evaluation function or the like with the state quantity of the vehicle and the dynamic characteristic variation value as inputs.

The fuzzy inference unit B20 infers the state evaluation value based on the operation amount generation rule 64,
The weighting variables Wc1, Wc2, ..., Wcn that determine the parameter Pc of the stabilization controller 16 and the ratio of their output amounts obtained from the matching rules are determined.

The parameter / rule learning device B21 receives a manipulated variable correcting signal from the outside so that the manipulated variable correcting signal decreases so that the parameter Pc of the stabilization controller 16 and the manipulated variable generating rule 64 of the fuzzy inference unit B20. To fix.

As described above, in the operation control device of this embodiment,
Fuzzy rules for driving control are "a) Optimal driving control knowledge in steady state determined by vehicle dynamics."
And “b) operation control knowledge for changing from a control pattern determined by the control purpose of the vehicle to the control pattern” and “a) optimal operation control knowledge in a steady state determined by the dynamic characteristics of the vehicle”. Stabilizing the control based on the controller unit 2
6, the target value generator unit 25 performs control based on "b) operation control knowledge for changing from a control pattern determined by the control purpose of the vehicle to the control pattern". Therefore, "a) Optimal operation control knowledge in a steady state determined by the dynamic characteristics of the vehicle" and "b) Operation control knowledge for transitioning from the control pattern determined by the vehicle control purpose to the control pattern" are all integrated. It is possible to significantly reduce the number of fuzzy rules, compared with the conventional method in which the fuzzy rules are expressed as a combination of. Further, when the rule base is constructed, only the rules in each steady state need to be extracted from the skilled driver, the prospect of rule acquisition is improved, and oversight of control rules can be prevented. In addition, it enables online learning to refine the rules after they go live.

Next, a more specific embodiment of the present invention will be described.

FIG. 2 is a diagram showing the system configuration of this embodiment.

The system shown in the figure is such that the vehicle 30 moves along a guide rail 35. The vehicle 30 includes a propulsion device 34 and a drive device 33 that drives the propulsion device 34, and a speed / speed controller such as a pulse encoder that detects the speed / position of the operation control device 32 and the vehicle 30 according to the present invention.
A position sensor 31 is provided.

FIG. 3 is a diagram showing a hardware configuration of the operation control device 32.

As shown in the figure, the operation control device 32 is composed of a microcomputer 43 in which a microprocessor 40, a memory 41, an interface 42, etc. are connected on a bus 44. Where the interface 42
Is for inputting a speed signal and a position signal from the speed / position sensor 31, and outputting an operation amount signal to the drive device 33.

FIG. 4 is a functional block diagram of the operation control device 32.

Here, when the vehicle 22 linearly travels on a horizontal plane at a constant acceleration αr, the equation of motion for time t, speed v, and position x is Is represented as

[0033]

[Equation 1] However, x0 and v0 are the vehicle 30 at the time t0.
Position and speed. When the relationship between the position and the speed of the vehicle 22 is calculated by these two equations,

[Equation 2] Becomes

Therefore, the speed control target generator 5 shown in FIG.
0 is constituted by the function shown below, and the acceleration target value α _v ^* , the speed target value v _v ^*, and the position target value x _v ^* are output.

[Equation 3] Further, the position control target generator 51 is configured by the functions shown below, and outputs the target values α _x ^* , v _x ^* , x _x ^* .

[0035]

[Equation 4] The target value synthesizer 11 adds the respective outputs of the speed control target generator 50 and the position control target generator 51 constituting the target value generator 10 in proportion to the magnitudes of the weight variables Wv and Wx, respectively, and obtains the target value. Output α ^* , v ^* , and x ^* . That is, the target values α ^* , v ^* , x ^* are

[Equation 5] Becomes

The state evaluator A13 uses the remaining running time T up to the estimated stop time of the vehicle 22 as an evaluation value for evaluating the state of control purpose change from speed control to position control.
Calculate r. If x0 is the target value Xs of the stop position, then Tr is

[Equation 6] Becomes

In the fuzzy inference unit A14, the following target generation rule 56 is used in order to switch the speed control and the position control of the Tr.

[Equation 7] Reasoning according to each target generator 50, 51
Output the goodness of fit a _ri for. However, C _ri (0 ≤ C
_ri ≤) is a function degree coefficient indicating the degree of connection between each rule and the target value generating function of the consequent part. In the description of FIG. 1, the fuzzy inference unit A14, which had been adapted to perform inference about the parameters of the switching and function of the target function is performed wherein only the switching of the objective function for the sake of simplicity, a _r , X ₀ , v _0, etc. are given as constants.

From the above inference result, the weight variable synthesizer 57 calculates the weight variable according to the following equation.

[Equation 8] The stabilization controller unit 26 first subtracts the state quantity of the vehicle 22 from the target value generated by the target value generator unit 25 to calculate the speed error Δv and the position error Δx. That is, the velocity error Δv and the position error Δx are

[Equation 9] Is.

Stabilizing controller i (i = 1, 2, 3) 16
Is composed of a proportional- _plus- integral controller, and the proportional gain and integral gain of each are G _vPi , G _xPi , G _vIi , and G
_xIi .

The target value of the acceleration is multiplied by the feedforward gains G _vfi and G _xfi to add the manipulated _variable .

Therefore, the manipulated variable α ^ is calculated from the weight variable Wci for each controller.

[Equation 10] Becomes

Then, in the adder 18, the manipulated variable correcting signal Δu _t is added as a teacher signal for learning and α
After setting to _u , it is output to the vehicle 22 as an operation amount.

The state evaluator B19 calculates the acceleration error Δα from the target velocity value and the actual velocity.

In the fuzzy inference unit B20, the acceleration error Δα is first calculated based on the antecedent part membership function 62.
To a membership value.

The associative memory B63 has a target value generator section 25.
It has the same structure as the associative memory A55 in
The following operation amount generation rules 64 with the membership value as input

[Equation 11] Is evaluated and the goodness of fit a _ci for each controller gain is output. However, Cci (0 ≦ Cci ≦ 1) is a degree-of-association coefficient that indicates the degree of connection between each rule and the control gain of the consequent part. When the value approaches 1, the corresponding rule determines the control gain. The rate of contribution is high. The weight variable parameter synthesizer 65 calculates a weight variable according to the following formula.

[0046]

[Equation 12] The parameter rule learning device B21 is used for initial gain acquisition or optimum gain adjustment after operation. When acquiring the initial gain, first, a gain having a relatively small value that can be stably controlled regardless of any acceleration error is obtained. Next, for each of the dynamic characteristic states represented by the label of the membership function of the antecedent part of the gain reasoner (acceleration error is 0, positive and negative states), learning of the optimum gain is performed based on the learning rule of Windrow-Hoff. To do. For G _VP1 which is one of the proportional gain, for example, a current regulator, a manipulated variable correction signal Delta] u _t skilled driver gives the manipulated variable output of a current regulator as a teacher signal, the correction amount of the gain .DELTA.G _VP1
Is corrected as follows.

[0047]

[Equation 13] However, the subscript t represents the discretized time.

At the same time, the control rule is refined by strengthening the degree of association Ccv1 representing the connection between the corresponding rule and the controller gain of the consequent part. Also, when the need for gain adjustment after operation arises, the optimum gain adjustment can be performed by the same procedure as above.

Flowcharts for executing the above control are shown in FIGS.

These control calculations are executed at sufficiently short time intervals so as to follow changes in the speed and position of the vehicle.

FIG. 5 shows the overall control flow.

After initializing at P2, the control loop is entered. The speed and position signals are read in P3, and the target values of the speed and position are calculated in P4 based on them. The manipulated variable is calculated in P5 from these target values, and the manipulated variable is output to the drive device in P6. In P7, it is determined whether parameter rule learning is to be performed. If so, parameter rule learning is performed in P8. Finally, in P9, it is determined whether the operation control is to be ended. If the control is to be continued, the process returns to P3 and the control is repeated.

FIG. 6 shows the flow of target value calculation.

At P21, a fuzzy membership value is calculated from the speed and position, and at P22, the degree of association is determined by fuzzy inference. In P23, a weight variable is calculated from the degree of association.
P24 calculates target values for speed control and position control, and P25
Then, the target value is calculated by interpolating these target values with the weight variable. Return to the main routine at P26. FIG. 7 shows the flow of operation amount calculation.

The acceleration error is calculated in P31, and the fuzzy membership value is calculated in P32 from the value. P3
In 3, the degree of conformity of each controller is determined by fuzzy inference, and P
At 34, the weight variable of the controller is calculated. The error calculation of the speed and the position is performed in P35, and the operation amount is determined in P36 from these. The manipulated variable correction value is added at P37. Return to the mesin routine at P38.

Next, the control system is simulated when the vehicle stops at a fixed position, and the stop position error is examined under the following three conditions.

(1) Only speed control is performed, and controller gain switching due to acceleration / deceleration error is also not performed. (2) The speed control is switched to the position control, but the controller gain is not switched.

(3) Switching from speed control to position control and switching of controller gain. The vehicle travels at a speed of approximately 55
It coasts at Km / h, but after entering the deceleration section, deceleration starts at the acceleration / deceleration target value α _r = -2.11 Km / h, and is controlled to stop at the 350 [m] position from the simulation start point. It

Regarding the parameters of the vehicle, the running resistance TL = 2.3 + 4.8X10 ^-3 v + 7.46X10 ^-4 [kgf / t] and the delay time Td = 2.0 [sec] (however, when decelerating from coasting) were used. Acceleration / deceleration error was given by changing the vehicle rated weight used in the model. As the value of the stabilizing controller gain, the values obtained by manually tuning each when the acceleration / deceleration error is + 30%, 0%, and -30% as shown in FIG. 8 are used.

FIG. 9 shows a stop position error when the vehicle acceleration / deceleration error changes from + 30% to -30% with respect to the target acceleration / deceleration.

In the case of the condition (1), if the absolute value of the acceleration / deceleration error exceeds about 15%, a large stop position error will occur.

In the case of the condition (2), the acceleration / deceleration error is from -15%.
In the range of + 30%, the absolute value of the stop position error is controlled within about 0.1 [m]. However, in the region where the acceleration / deceleration error is -15% or less, the stop position error sharply increases and the result is the same as when only the speed control is performed. This is because the gain of the controller is adjusted according to when there is no acceleration / deceleration error, and therefore, when the acceleration / deceleration error has a large negative value, the manipulated variable output from the controller becomes insufficient.

In the case of the condition (3), even if the acceleration / deceleration error changes from + 30% to -30%, the absolute value of the stop position error is controlled within about 0.1 [m].

FIG. 10 shows changes in speed, speed error, position, position error, and acceleration / deceleration operation amount with respect to the time axis when the acceleration / deceleration error is + 30%.

The control system tries to decelerate at the acceleration / deceleration target value of the vehicle, but under this condition, a force larger than the expected braking force is generated, so that the speed error greatly increases in the positive direction. On the other hand, the control system recognizes that the acceleration / deceleration error is in the positive direction and immediately switches the speed controller to one with a small gain to reduce the controller output. Also,
It can be seen that when the stop position approaches, the operation amount output by the position controller gradually increases compared to the operation amount output by the speed controller, and the speed control is smoothly switched to the position control.

FIG. 11 shows changes in the respective amounts when the acceleration / deceleration error is -30%.

Under this condition, a speed force greatly increases in the negative direction because a force smaller than the expected braking force is generated. On the other hand, the control system recognizes that the acceleration / deceleration error is in the negative direction and immediately switches the speed controller to one with a large gain to increase the controller output. Also, switching from speed control to position control has been done successfully. As described above, the control system has a wide range of application in order to adjust the control amount by appropriately switching the control target value and controller gain according to changes in the control purpose and changes in the vehicle dynamic characteristics. Has become.

[0068]

As described above, according to the present invention, the number of fuzzy rules can be significantly reduced as compared with the conventional one. Further, when the rule base is constructed, only the rules in each steady state need to be extracted from the skilled driver, the prospect of rule acquisition is improved, and oversight of control rules can be prevented. In addition, it enables online learning to refine the rules after they go live.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an operation control device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a system configuration of a more specific embodiment of the present invention.

FIG. 3 is a diagram showing a hardware configuration of the operation control device of FIG.

FIG. 4 is a functional block diagram of the operation control device of FIG.

FIG. 5 is a flowchart showing a control flow of the entire operation control device.

FIG. 6 is a flowchart showing a flow of target value calculation of the operation control device.

FIG. 7 is a flowchart showing a flow of operation amount calculation of the operation control device.

FIG. 8 is a table showing gain values of the stabilization controller.

FIG. 9 is a graph of stop position error due to acceleration / deceleration error.

FIG. 10 is a waveform diagram of a simulation when the acceleration / deceleration error is + 30%.

FIG. 11 is a waveform diagram of a simulation when the acceleration / deceleration error is -30%.

[Explanation of Codes] 10 ... Target value generator, 11 ... Target value synthesizer, 12 ... Adder, 13 ... State evaluator A, 13 ... Fuzzy reasoner A, 1
5 ... Parameter / rule learning device A, 16 ... Stabilization controller, 17 ... Manipulation amount synthesizer, 18 ... Adder, 19 ... State evaluator B, 20 ... Fuzzy reasoner B, 21 ... Parameter
Rule learning device B. 25 ... Target value generator part, 26 ... Stabilization controller part.

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] September 25, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] 0037

[Correction method] Change

[Correction content]

In the fuzzy inference unit A14, the following target generation rule 56 is used in order to switch the speed control and the position control of the Tr.

[Equation 7] Reasoning according to each target generator 50, 51
And outputs a goodness of fit a _ri for. However, C _ri (0
≦ C _ri ≦ 1 ) is a degree-of- association coefficient indicating the degree of connection between each rule and the target value generating function of the consequent part.
Note that in the description shown in FIG. 1, the fuzzy inference unit A14
Was designed to switch the objective function and infer about the parameters of the function, but here, for simplicity, only the switching of the objective function is performed, and parameters such as α _r , x ₀ , v ₀ are given as constants. There is.

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0040

[Correction method] Change

[Correction content]

Further, for the acceleration target value of added <br/> to the operation amount by multiplying the feed forward gain _{G vfi,} the _{G XFI.}

[Procedure 3]

[Document name to be amended] Statement

[Name of item to be corrected] 0045

[Correction method] Change

[Correction content]

The associative memory B63 has a target value generator section 25.
It has the same structure as the associative memory A55 in
The following operation amount generation rules 64 with the membership value as input

[Equation 11] Are evaluated, and the goodness of fit _aci for each controller gain is output. However, C _ci (0 ≦ C _ci ≦ 1)
Is a degree-of-association coefficient that indicates the degree of connection between each rule and the control gain of the consequent part. When the value approaches 1, the corresponding rule contributes to the control gain determination at a higher rate. The weight variable parameter synthesizer 65 calculates a weight variable according to the following formula.

[Procedure amendment 4]

[Document name to be corrected] Drawing

[Name of item to be corrected] Figure 5

[Correction method] Change

[Correction content]

[Figure 5]

Claims (2)

[Claims] 1. A target value for generating a target value for a control purpose by a fuzzy rule based on a driving control knowledge that transitions from a control pattern determined by the control purpose of a control target to a control pattern in a driving control device that performs a fuzzy control operation control. An operation control device comprising: a generator section; and a stabilizing controller section that generates a manipulated variable for control purposes by a fuzzy rule based on optimal operation control knowledge in a steady state determined by the dynamic characteristics of the controlled object. . 2. The operation control device according to claim 1, wherein the target value generator section has a plurality of target generators each of which has an adjustable parameter and which inputs a state quantity of a control target, and these target generators. A target value synthesizer that multiplies the output by a weight variable and sums it, an adder that sums the output of this target value synthesizer and the target value correction signal, and a state evaluation value according to the change of the control objective of the controlled object. A first fuzzy reasoner that determines a weighting variable of the target value synthesizer by fuzzy reasoning according to the status evaluation value of the first condition evaluator; and the target generation. And a first parameter rule learning device that learns the parameters of the first fuzzy inference device according to the state of the control system or the target value correction signal, and the stabilizing controller unit is adjustable. With different parameters A plurality of stabilizing controllers for inputting the target value output from the target value generator and the state quantity of the controlled object, and a manipulated variable synthesizer for multiplying the outputs of these stabilizing controllers by a weight variable and summing them. An adder that sums the output of the manipulated variable combiner and the manipulated variable correction signal; a second state evaluator that determines a state evaluation value according to a change in the dynamic characteristics of the controlled object; A fuzzy reasoner that determines the weighting variable of the second fuzzy reasoner according to the state evaluation value of the second state evaluator by fuzzy inference, and controls the parameters of the stabilizing controller and the rules of the second fuzzy reasoner A second parameter rule learning device that learns according to a system state or the manipulated variable correction signal.