JPH056204A - Controller construction processing system - Google Patents

Abstract

PURPOSE:To easily construct a controller to handle a nonlinear controlled system in a general control rule form concerning the controller construction processing system for constructing the controller used for controlling the controlled system. CONSTITUTION:The system is provided with a learning processor 2 to set a signal conversion function to a teaching signal group, a virtual target managing device 5 to manage experiential knowledge between control states, a manipulation corrected variable calculator 6 to calculate the corrected variable of a control manipulated variable from a control state variable and a virtual target value specified from the management data of the managing device, the control state variable is inputted to the processor 2, and the output is handled as the control manipulated variable and applied to the controlled system. This controller construction processing system constructs the processor 2 as the desired controller while obtaining the teaching signals by correcting the control manipulated variable according to the corrected variable from the calculator at that time, the processor 2 inputs a differential value between the control state variable and a target control state variable, and the managing device manages the experiential knowledge between the control state variables with the differential value from the control state variable to be the control target as a parameter.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a control device construction processing system for constructing a control device used for controlling an object to be controlled, and more particularly to a device for handling the non-linear object easily and generally. The present invention relates to a control device construction processing system that enables construction in various control rule formats.

A method of designing a control device by applying modern control theory because a classical PID correspondence cannot cope with a complicated controlled object such as a 1-input 2-output system of inverted pendulums. Is taken. However, when the linear control theory of the modern control theory is used, the control performance is deteriorated outside the linear region because the controlled object model is constructed by linearizing the equation of motion of the controlled object.

Further, when the nonlinear control theory of the modern control theory is used, it is necessary to accurately describe the equation of motion, so that it is necessary to accurately identify the parameter of the controlled object, which is extremely difficult. It has the drawback of being work. Against this background, a control device having a new configuration using a neural network is being proposed in recent years.

[0004]

2. Description of the Related Art A neural network has a property that, when a teacher signal group is given, the input / output characteristics of those teacher signal groups are provided by learning. Is given,
It has a property of realizing an adaptive data processing function of outputting an output signal like that.

When such a neural network is constructed as a control device, a sufficient number of control data are acquired from the controlled object, and the acquired control data is used as a teacher signal to execute learning. By mapping the control rules for the controlled object on the neural network, a method of constructing as a control device will be adopted.

However, it is often impossible to obtain the control data of the controlled object in reality when the controlled object becomes complicated. So, as a way to deal with this,
Recently, a certain amount of a priori knowledge has been obtained qualitatively, but for a control object that has many unknown parts quantitatively, a teacher signal is obtained by trial, and this is used Proposal of Mapping Control Rules for a Neural Network on a Neural Network (Saito, Kitamura, "Stabilized Learning Control of Inverted Pendulum Using Multilayer Neural Network", Robotics and Mechatronics '90 Proceedings, p.
283-286, 1990).

This new method employs a configuration including a neural network, a virtual target generation unit, and an evaluation unit in order to stop the inverted pendulum on the carriage at the origin.

This newly provided virtual target value generator expresses a priori knowledge that "the farther the position of the carriage is from the origin, the more the virtual target angle of the pendulum is tilted from the vertical direction to the origin side". When the position and speed of the carriage are given, a virtual target value of the pendulum angle and angular velocity for moving the position and speed of the carriage to the origin is generated.

On the other hand, the newly provided evaluation unit evaluates how the difference between the generated virtual target value and the control output (angle / angular velocity of the pendulum) should be after one sampling. Finding the amount of correction of the force to be applied to the dolly,
The force corrected by the correction amount is specified as a teacher signal. Then, the neural network inputs the pendulum angle and each speed at each sampling time, and the position and speed of the carriage, and outputs the force to be applied to the carriage.
The learning of the neural network is performed by the back propagation method based on the teacher signal generated by the evaluation unit.

By adopting this configuration, a certain amount of a priori knowledge is qualitatively obtained, but a teacher signal is obtained by trial for a control target that has many unknown quantitatively. Then, by using this to map the control rules for the controlled object on the neural network, it is possible to construct the neural network as a control device for the controlled object.

Conventionally, there has been devised a control device which represents an intermediate state that can be taken by the system by a virtual value called a virtual target value, and controls by learning an input / output relationship that can realize the virtual target value. However, since only one virtual target value is used for these, there is a problem that the control performance deteriorates when the system state changes in a complicated manner. Therefore, in the present invention, a plurality of virtual target values are set according to the state of the system to improve the control performance of the system.

[0012]

Although the newly proposed method certainly has an advantage that a control device used for a non-linear controlled object can be easily constructed, the virtual target value generation unit However, when the position and speed of the trolley are given, a virtual target of the pendulum angle and angular velocity for moving this to the origin is generated. The neural network outputs a virtual target value of the pendulum angle at each sampling time.
Since the configuration is such that the absolute value of the control state quantity is input, such as the input of the angular velocity and the position / velocity of the carriage, when the target value of the control state quantity is changed, it is not necessary to re-learn each time. There was a problem that it should not happen.

Although not mentioned above, the evaluation unit must obtain the correction amount of the force to be applied to the trolley according to a complicated evaluation formula that changes the response characteristic of the control target. There is also a problem that the control rule that is not faithful to the response suppression of the above is set and that the control device cannot be constructed in a short time.

It is an object of the present invention to provide a new device construction processing system that enables a control device that handles a non-linear controlled object to be easily constructed in a general control rule format.

Another object of the present invention is to set a plurality of virtual target values according to the state of the system and select them according to the distance to the target value so that the target value can be stably reached. That is.

[0016]

FIG. 1 is a block diagram showing the principle of the present invention. In the figure, 1 is a data processing device, 2 is a learning processing device, 3 is a control target, 4 is a target value setting device, 5 is a virtual target management device, 6 is an operation correction amount calculation device, and 7 is a first differencer. , 8 are second differencers.

The data processing device 1 has a variable signal conversion function, and when a teacher signal group is given, the signal conversion function can be set to one that realizes the input / output characteristics of the teacher signal group. Take. This data processing device 1 is
When the control state amount of the controlled object 3 and its target value are given, the control device outputs a control operation amount for controlling the controlled object 3 to the target control state. I will go.

The data processing device 1 receives one or a plurality of inputs and internal state values to be multiplied with these inputs to obtain a product sum value, and converts the product sum value by a predetermined function. It is composed of a network structure composed of internal connections of basic units for obtaining output values.

Further, the data processing apparatus 1 uses the IF-THEN to determine the qualitative data relationship between the control state quantity and the control operation quantity.
A fuzzy device may be used which describes the qualitative attributes of the control state amount and the control operation amount described in the IF-THEN rule by the membership function, as described in the rule.

The learning processing device 2 learns the signal conversion function of the data processing device 1 so as to realize the input / output characteristics of the teacher signal group when the teacher signal group is given. When the data processing device 1 is configured by the network structure unit, the learning processing device 2 executes a well-known learning algorithm such as a back propagation method.

The control target 3 is a control target controlled by the data processing device 1 constructed as a control device. It is preferable that an actual controlled object is used as the controlled object 3, but the controlled object model may be used instead of the actual controlled object.

The target value setting device 4 sets a target value of the control state quantity representing the desired control state of the controlled object 3. The virtual target management device 5 manages a priori knowledge of the data relationship between the control state quantities, which is obtained to realize the desired control state of the controlled object 3. The virtual goal management device 5 of the present invention manages this a priori knowledge management data using the difference value between the control state quantity and the target value as a parameter.

The operation correction amount calculation device 6 has a control state amount of the controlled object 3 and a virtual target value of the control state amount specified from the management data of the virtual target management device 5 corresponding to the control state amount. From this, the correction amount of the control operation amount for the controlled object 3 required to realize the target value of the control state amount is calculated.

This operation correction amount calculation device 6 is controlled by the controlled object 3
There may be a configuration in which the correction amount of the control operation amount is calculated by multiplying the difference value between the control state amount of the virtual target management device 5 and the virtual target value output by the virtual target management device 5 by the proportional coefficient.

The first difference unit 7 calculates a difference value between the target value of the control state amount set by the target value setting device 4 and the control state amount of the controlled object 3, and the difference value is subjected to data processing. Input to the device 1 and the learning processing device 2. At this time, in order to adjust the dynamic range, the output value of the first difference unit 7 may be multiplied by the proportional coefficient.

The second differencer 8 calculates a difference value between the control operation amount output from the data processing device 1 and the correction amount of the control operation amount output from the operation correction amount calculation device 6, and the difference value. That is, the control operation amount output from the data processing device 1 is corrected by the calculated correction amount and input to the learning processing device 2.

[0027]

In the present invention, when the signal conversion function of the data processing device 1 is set to, for example, the initial state, the control target 3
When the initial value of the control state quantity is output from, the first difference unit 7 calculates the difference value between the target value of the control state quantity set by the target value setting device 4 and the initial value of the control state quantity. And input it to the data processing device 1. In response to this input, the data processing device 1 calculates the control operation amount defined by the initial state signal conversion function and outputs the control operation amount to the control target 3, and the control target 3 receives the output process of the control operation amount. Transition to a control state different from the initial state. Hereinafter, this process is repeated until the control state of the controlled object 3 reaches the specified limit.

During this processing, when the virtual target management device 5 receives the control state quantity from the controlled object 3, it specifies the virtual target value of the control state quantity according to the management data. According to the processing of the virtual target management device 5, for example, the control target 3 is 1
In the case of the example of the control system of the input 2 output system, a virtual target value of one control state quantity output from the controlled object 3 for the other control state quantity is specified.

When the virtual target management device 5 specifies the virtual target value of the control state quantity, the operation correction amount calculation device 6 uses this virtual target value to realize the target value of the control state quantity. The correction amount of the control operation amount for the controlled object 3 necessary for that is calculated. According to this correction amount calculation process,
In order to realize the target control state amount set by the target value setting device 4, it is decided how the control operation amount output by the data processing device 1 at the time of the processing should be corrected. Become.

In this way, when the teacher signal group consisting of the difference value of the control state quantity input to the data processing device 1 and the more preferable control operation quantity at the time of inputting the difference value is obtained, the learning processing is performed. The device 2 executes the learning process of the signal conversion function of the data processing device 1 and sets the signal conversion function to one suitable for realizing a more targeted control state.

Then, according to the signal conversion function of the newly set data processing device 1, the same process as described above is repeated so that the next teacher signal group is generated. The data processing device 1 is constructed as a control device by setting the signal conversion function of the data processing device 1 so as to realize a target control state.

As described above, in the present invention, a certain amount of a priori knowledge is obtained qualitatively, but the teaching signal is tentatively applied to the controlled object 3 in which there are many unknown parts quantitatively. Then, by using this, the control rule for the controlled object 3 is mapped onto the signal conversion function of the data processing apparatus 1, so that the data processing apparatus 1 is constructed as the control apparatus for the controlled object 3. When the target value of the control state quantity is changed, the construction processing of the data processing device 1 is executed according to the difference value from the target value of the control state quantity. Even if there is, you will not have to start learning again.

Even if the target position changes, for example, the input to the controlled object for inversion does not change with respect to the difference value between the target position and the current position. That is, the input to the controlled object changes only by the difference value, and even if the target position changes, if the difference value does not change, the input to the same controlled object may be output. Therefore, if the input to the controlled object related to the difference value is learned, the control is only performed according to the difference value, and it is not necessary to relearn.

According to the present invention, even if the target value of the control state quantity is changed, the relationship between the difference value between the current value and the target value and the corresponding input value to the controlled object is learned in advance. The state quantity can be controlled to a desired value, that is, the controlled object can be brought into a desired control state at an arbitrary target position. Further, since the target value setting device 4 is provided in the present invention, it becomes possible to move the inverted position at a certain speed while the pendulum is inverted.

Further, according to the present invention, a plurality of virtual target curves showing the dependency relation of the variable of the controlled object are provided, and one of the plurality of control units for each virtual target curve is selected according to the control state quantity of the controlled object.

Further, according to the present invention, a plurality of virtual target curves showing the dependency of the variable to be controlled are provided, and one of the variables to be controlled
Control was performed by selecting one of the virtual target curves of the variable according to the area of the part.

[0037]

EXAMPLES The present invention will be described in detail below with reference to examples. FIG. 2 illustrates one embodiment of the present invention. 10 in the figure
Is a neural network, which functions as a control device, 11 is a learning processing device, which executes the learning process of the neural network 10, 12 is an inverted pendulum model, and 1 input 2 to be controlled An output system control system, 13 is a target value setting device,
What sets a target value of the control embodiment amount of the inverted pendulum model 12, 14 is a virtual target calculation device, which calculates a virtual target of the control state amount of the inverted pendulum model 12, 15 is a torque correction amount calculation device Which calculates the correction amount of the torque output from the neural network 10,
Is a first delay device for delaying the control state quantity output from the inverted pendulum model 12 by one sampling time and giving it to the virtual target calculation device 14, and 17 is a second delay device for the inverted pendulum model. The control state quantity output from 12 is delayed by one sampling time, and 18 is a first difference device, which is a target value of the control state quantity set by the target value setting device 13 and the output of the second delay device 17. The difference value from the control state quantity to be calculated is calculated, and the difference value is calculated.
And a second differencer 19 for calculating the difference value between the torque output from the neural network 10 and the torque correction amount calculated by the torque correction amount calculation device 15. The difference value is input to the learning processing device 11. Here, (n) in the figure is
It represents the sampling time.

FIG. 3 shows a controlled object model in this embodiment.
The inverted pendulum model 12 assumed as described above is illustrated. In this Figure 3
As shown, the inverted pendulum model 12 is
Base link L connected to shaft (Z axis)₁Other end C
And link L₂Is the base link L₁Is connected as a rotation axis
It is configured by Base link L₁And Rin
Black L ₂Rotation angle of θ₁, Θ₂And the mass
Each m₁, M₂And the length is l₁, L₂age,
If the gravitational acceleration is g and the motor torque is T,
The equation of motion of the inverted pendulum model 12 of
Become.

In this embodiment, the torque T generated by the motor is applied to the inverted pendulum model 12 according to this equation of motion.
Is controlled to feed back the states of the respective links L ₁ and L ₂ to invert the pendulum, and further to stop the base link L ₁ at an appropriate target position.

As described above, the inverted pendulum model 12 has

[0041]

[Equation 1]

Since there are four of them, and one of them is the torque T of the motor as the control operation amount,
In the differencer 18 of, the corresponding target value set by the target value setting device 13 is

[0043]

[Equation 2]

If

[0045]

[Equation 3]

The difference value is calculated and input to the neural network 10. In the following, for convenience of description, when expressing the differential value of the angular velocity, (d / dt) may be added before the angle.

From now on, the neural network 10
Needs to have four units as an input device and one unit for outputting the torque T as an output layer. Therefore, in this embodiment, as shown in FIG. An input layer, an intermediate layer having a one-stage structure composed of eight basic units 21, and an output layer composed of one basic unit 21. A neural network 10 having a hierarchical network configuration in which weight values are respectively set for the internal connection between the basic units 21 of the intermediate layer and the basic unit 21 of the output layer is prepared.

The input unit 20 of the input layer distributes the input signal value as it is and outputs it to the basic unit 21 of the intermediate layer, and the basic units 21 of the intermediate layer and the output layer each internally couple a plurality of inputs. A multiplication processing unit that multiplies the weight values of, a cumulative processing unit that adds all the multiplication results, and a threshold processing unit that performs non-linear threshold processing on the cumulative value and outputs one final output. The learning processing device 11 is
In order to realize the input / output characteristics of the teacher signal group, the learning process of the weight values of these internal couplings will be executed.

The virtual target calculator 14 is an inverted pendulum model.
12 obtained for "base link L₁Rotation angle of
θ₁Link L when is away from the target position₂Rotation of
Each θ ₂The base link by tilting the
L₁Rotation angle of₁Approaches the target position ”
Link L based on control knowledge₂Rotation angle of₂Virtual
Target value θ_{d 2}And its angular velocity (d / dt) θ₂Virtual eyes of
Standard value (d / dt) θ_{d 2}And are calculated and output.

That is, the virtual target calculation device 14 is provided with the target value θ _{t 1} of the rotation angle θ ₁ of the base link L ₁ from the target value setting device 13 and the first delay device 16 from the inverted pendulum model. through it, the rotation angle theta ₁ of the base link L ₁ and the angular velocity (d / dt) θ ₁ is given, as a virtual target value theta _{d 2} of the rotation angle theta ₂ of the link L _2, the angular velocity (d /
dt) θ ₂ virtual target value (d / dt) θ _{d 2}

[0051]

[Equation 4]

The calculation is performed according to the calculation formula Where θ _{2 max} is the link L
₂ which is the maximum tilt angle of the rotation angle θ _2. Further, the virtual target calculation device 14 outputs the target value given from the target value setting device 13 as it is for the rotation angle θ ₁ of the base link L ₁ and its angular velocity (d / dt) θ _1. .
It should be noted that although the sigmoid function is used in this equation,
It is also possible to use a relational expression in which ₁ and θ _{d 2} are proportional. As can be seen from this equation, when the rotation angle θ ₁ of the base link L ₁ reaches the target position θ _{t 1} , the virtual target value θ _{d 2} of the rotation angle θ ₂ of the link L ₂ becomes 0, so the pendulum You will be upside down.

The torque correction amount calculation device 15 uses the link L ₂
The torque T applied to the base link L ₁ is corrected so that the rotation angle θ ₂ of the virtual link becomes the virtual target value θ _{d 2} .
That is, the torque correction amount calculation device 15 receives the virtual target value θ _{d 2} of the rotation angle θ ₂ of the link L ₂ and the virtual target of the angular velocity (d / dt) θ ₂ from the virtual target calculation device 14. value as _{(d / dt) θ d 2} , the rotation angle theta ₁ of the base link L ₁ angular velocity (d / dt) θ ₁ of the virtual target value _{(d / dt) θ d 1} ( in this case, actual the target value (d / dt) matching theta _{t 1)} and is provided, the inverted pendulum model 12, the rotation angle of the link L ₂ theta ₂ and its angular velocity (d / dt) θ _2, the base link L ₁
Given the angular velocity (d / dt) θ ₁ of the rotation angle θ ₁ of
Torque correction amount ΔT (n)

[0054]

[Equation 5]

Processing is performed so as to calculate and output according to the calculation formula Where the third term in this equation is
It acts as a damping term and exerts the effect of rapidly stopping in the vicinity of the target position of the rotation angle θ ₁ .

The second subtractor 19 receives the torque correction amount output from the torque correction amount calculation device 15 output from the torque correction amount calculation device 15, and receives the torque correction amount from the neural network 1
The difference value T ′ (n) T ′ (n) = T (n) −ΔT (n) between the torque T (n) output by 0 and the torque correction amount ΔT (n) calculated by the torque correction amount calculation device 15. ) Is calculated, and this difference value is calculated by the neural network 10
The learning processing device 11 is informed that the output torque is a preferable value.

The learning processing device 11 inputs the input value to the neural network 10 thus obtained and a more preferable neural network 1 at that time.
The output of the torque value of 0 is used as a teacher signal, and the learning of the weight value of the internal coupling of the neural network 10 is executed at high speed in accordance with the improved backpropagation method proposed by the applicant as a virtual impedance control method. It will be processed as it does.

To describe the learning method, first, the base link L ₁ (arm) is 30 °, and the link L ₂ (pendulum) is 0.
Set to ° and try. As a result, 20
Suppose that sampling values have been obtained. This is stored in the memory, the torque value of the teacher signal is automatically generated using this sampling value, the weight value of the neural network is updated, and learning is performed.

Next, the arm is set to 30 ° and the pendulum is set to 0 °, and the second trial is performed. In the meantime, based on the weight values determined by the previous 20 sampling values, for example, 40 sampling values are taken and stored as a new sampling value in the memory. A torque value is automatically generated, the weight value of the neural network is updated, and learning is performed.

In this way, by moving the arm from 0 ° to 90 ° and using the weight value finally stored by the neural network, an arbitrary initial position and target position (provided that the initial position and the target position are And a difference between 0 ° and 90 °) is given, it can be inverted at the target position.

A better network can be constructed by moving the entire region of the arm from 0 ° to 90 °, but it is not necessary to move the entire region from experience. Further, even if the difference between the initial position and the target position is a difference between 0 ° and 90 ° or more, control is possible to some extent.

Even if the target position changes, for example, the control torque for inversion does not change with respect to the difference value between the target position and the current position. That is, the control torque changes only according to the difference value, and if the difference value does not change even if the target position changes, the same control torque may be output. Therefore, according to the present invention, if the control torque related to the difference value is learned, the control is thereafter performed only in accordance with the difference value, and it is not necessary to relearn even if the target position changes.

Next, the effectiveness of the embodiment of the present invention configured as above will be described according to the simulation results. The simulation mass m ₁ of the base link L ₁ of the inverted pendulum model 12 "1", the mass m ₂ of the link L ₂ "0.25", the length l ₁ of the base link L ₁ "0.
2 assumed ", the link L ₂ of the length l _2" 0.5 ", the coefficient K ₁ for use in a torque modification value computing device 15, K _2, K ₃ values respectively" 1 "," 1 "" 0.1 " I went.

Then, (d / dt) E ₁ of the difference values output from the first difference unit 18 is multiplied by a coefficient of "0.1" and input to the neural network 10. . The initial value of the rotation angle theta ₁ is 30 °, the initial value of the rotation angle theta ₂ is 0 °, the target value theta _{t 1} of the rotation angle theta ₁ is 0
°, the target value theta _{t 2} is 0 ° rotation angle theta _2, the maximum inclination angle theta _2max of the rotation angle theta ₂ is set to 20 °, the initial value of the weight values of the internal bonds of the neural network 10, ± 0. It was set according to a random value of 01.

The simulation is tried as follows.
It That is, the state of the inverted pendulum model 12 [θ_i(n),
(D / dt) θ_i(n)] (i = 1, 2) is the sampling time
Sample every 0.01 seconds. And at this time
Calculates the torque output of the neural network 10 of
This torque output is used as the base link L of the inverted pendulum model 12.
₁Of the inverted pendulum model 12
State (θ_i(n + 1), (d / dt) θ_i(n + 1)] is simulated
To continue.

At this time, a teacher signal is obtained by calculating a more preferable output torque of the neural network 10 according to the torque correction amount calculation device 15. This process is repeated for 500 steps, that is, for 5 seconds. At this time, if the inverted pendulum is tilted by 45 ° or more, the trial is terminated there. From this, a maximum of 500 teacher signals will be obtained.

When the teacher signal group is obtained in this way,
The learning processing device 11 learns the weight of the internal connection of the neural network 10 according to the improved backpropagation method. This learning processing is terminated when the number of times of learning of the back propagation method is 100, and the weight value obtained at that time is set as a new weight value in the internal connection of the neural network 10. Then, by repeating the above-described trial according to the neural network 10 having the newly set weight value, the construction of the neural network 10 for controlling the pendulum inversion of the inverted pendulum model 12 is simulated. Become.

6 to 11 show an example of the simulation data obtained in this simulation process. The simulation data of FIG.
After 0 times, the initial value of the rotation angle θ ₁ of the inverted pendulum model 12 is 90 °, the initial value of the rotation angle θ ₂ is 10 °, and the rotation angle θ ₁
Of the target value θ _{t 1} of 0 ° and the target value θ _{t 2} of the rotation angle θ ₂ of 0
It shows the response when operated as °. Also, the simulation data of FIG. 7 is in the same initial state,
Target value of the rotation angle θ ₁ θ _{t 1} to -30 °, the rotation angle theta ₁ when operating target value theta _{t 2} of the rotation angle theta ₂ as 0 °,
The response of θ ₂ is shown. FIG. 8 shows the torque curve in the simulation of FIG. 6, and FIG. 9 shows the response of the rotation angle θ ₂ and the virtual target value θ _{d 2} in the simulation of FIG.

In any of the simulations shown in FIGS. 6 and 7, the target control state is controlled after about 5 seconds. As described above, according to the present invention, the construction process as the control device of the neural network 10 is executed according to the difference value between the control state quantity and the target value.
Can control the inverted pendulum model 12 to a desired control state even in a target state different from that used in the construction process.

The simulation data of FIGS. 10 and 11 shows that the maximum tilt angle θ _2max of the rotation angle θ ₂ used in the calculation process of the torque correction amount calculation device 15 when the simulation of FIG. 10 °, 30 °, 40
This is set by setting the angle to 0 and performing trials 10 times, and the responses of the rotation angles θ ₁ and θ ₂ are obtained for each of them. From this simulation data, the maximum tilt angle θ _2max of the rotation angle θ ₂
It was revealed that the response of the inverted pendulum model 12 basically does not change even if is changed.

Although the illustrated embodiment has been described, the present invention is not limited to this. For example, in the embodiment, the one in which the control device is constructed by using the neural network is disclosed, but the present invention is not limited to this, and can be applied as it is to any data processing device capable of adjusting the signal conversion function according to the teacher signal. Of.

In the embodiment, the control object is the inverted pendulum, but the present invention is not limited to this and can be applied to all the description objects as they are.
Further, in the embodiment, the control device is constructed not by using the actual controlled object but by using the controlled object model, but the present invention is not limited to this, and the actual controlled object itself is used. If so, an accurate system identification including Coulomb friction and the like can be performed, so that a more appropriate control device can be constructed.

Next, as another embodiment of the present invention, a method of stabilizing the inverted pendulum will be described. The empirical rule used in this embodiment is based on the fact that "when a person moves a certain position while standing a stick like a broom, it moves while tilting the stick in the desired direction". When this is applied to the present embodiment, "when the arm moves to the target position while raising the pendulum, the arm moves while tilting the pendulum toward the target position".

FIG. 12 shows this by a function. θ ₁ is the arm position, and θd ₂ is a virtual target value for the arm position θ ₁ . 12, the origin represents the target position of the arm theta _1, when the arm theta ₁ is at the positive side than the target position, on the contrary inclined pendulum theta ₂ to the negative (target position side), the arm theta ₁ is a target position It represents that the pendulum θ ₂ is tilted to the positive side (target position side) when it is on the more negative side.

FIG. 12 shows two such functions.
is there. The difference between the two is the pendulum θ ₂The size of the tilt of
It's a difference. By reducing the slope of the virtual target value
Therefore, the inverted pendulum can be made hard to fall down. There
Allows you to quickly reach the target position by increasing the inclination.
Can be returned. Therefore, using two types of virtual target values
By doing so, it is hard to fall down from the target position,
It is possible to quickly follow the target position near the target position.
Noh.

In the following function corresponding to the virtual target curve, different functions can be formed by changing the parameter a and the maximum swing angle θ _{2 max} of the pendulum according to the type of the virtual target curve. And, for example, 2
The point (here, the origin) on the two curves (1) and (2) corresponding to one function can be set as the virtual target value.

[0077]

[Equation 6]

FIG. 13 shows the configuration of another embodiment of the present invention shown in FIG. In FIG. 13, corresponding parts are assigned the same numbers as in FIG. The difference between FIG. 13 and FIG. 1 is that the virtual target management device 5 of FIG. 13 prepares a plurality of curves (functions) for a plurality of virtual target values, and the control unit 10 of FIG. It is composed of a plurality of control units 10 including a net 1 and a learning device 2, and each control unit 10 is provided corresponding to a plurality of curves (functions) of the virtual target management device 5. One control unit learns about a curve (function) corresponding to one virtual target value, and different control units learn about functions corresponding to different virtual target values. That is, in this embodiment, the virtual target value is set in advance.
In this method, the setting is made according to the state of the system, and the control section is changed every time the state of the system changes at the time of learning, and learning is performed at any time.

It is assumed that the control unit 10 is trained so as to output such an output as to approach a given virtual target value according to the current state of the controlled object. At this time, it is possible to make the inverted pendulum difficult to fall down by making the inclination of the virtual target value small. Alternatively, by taking a large inclination, it is possible to quickly return to the target position.

Therefore, 2 represented by the curves (1) and (2)
By using different kinds of virtual target values, even if the arm angle θ ₁ changes according to the curve (1) far from the target value, the change of the pendulum virtual angle θd ₂ is reduced, so that the pendulum falls down. It is possible to quickly follow the target near the target value. It suffices to learn about one virtual target value and then learn about different virtual target values.

At the time of executing the inverted control, the selection circuit 19 uses the current position of the arm coming out of the inverted pendulum to select the control section learned by the curve (function) having a small inclination when the distance is far from the target value. When the value approaches the target value, the control unit learned by the curve (function) having a large slope is selected.

FIG. 14 shows still another embodiment of the present invention. The difference from FIG. 13 is that one control unit 10 is provided,
A plurality of curves (functions) in the virtual target management device 5 are combined to learn one curve (function) indicated by a thick line.

At the time of execution, the curve (function) is calculated by the control unit 1.
Since 0 has been learned, the selection circuit 19 is unnecessary unlike the embodiment of FIG.

[0084]

As described above, according to the present invention,
Qualitatively, a certain amount of a priori knowledge has been obtained, but for a control target that has many unknown parts quantitatively, a teacher signal is obtained by trial, and this is used for the control target. By mapping all the control rules on the signal conversion function of the data processing device, there is a case where the data processing device is constructed as the control device of the control target, and there is a difference from the target value of the control state quantity. Since the construction processing of the data processing device is executed according to the value, it becomes unnecessary to redo the learning even when the target value of the control state amount is changed. As a result, it becomes possible to easily construct a control device that handles a non-linear control target and to construct it in a general control rule format.

The present invention can further change toward a desired target value based on a priori knowledge by following a characteristic synthesized from a plurality of virtual target curves. Further, in the present invention, the control operation amount correction amount calculation device for generating the teacher signal provided for this realization corrects the control operation amount without changing the response characteristic of the control target by following the linear calculation formula. Since the configuration for calculating the quantity is adopted, it becomes possible to set a control rule that is faithful to the response characteristic of the controlled object, and to construct the control device in a short time.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is an example of the present invention.

FIG. 3 is an explanatory diagram of an inverted pendulum used in an example.

FIG. 4 is an explanatory diagram of a motion equation of an inverted pendulum used in an example.

FIG. 5 is a configuration diagram of a neural network.

FIG. 6 is an explanatory diagram of simulation data.

FIG. 7 is an explanatory diagram of simulation data.

FIG. 8 is an explanatory diagram of simulation data.

FIG. 9 is an explanatory diagram of simulation data.

FIG. 10 is an explanatory diagram of simulation data.

FIG. 11 is an explanatory diagram of simulation data.

FIG. 12 is a diagram illustrating another embodiment of the present invention.

FIG. 13 is a configuration diagram of another embodiment of the present invention.

FIG. 14 is a configuration diagram showing a modification of the embodiment shown in FIG.

[Explanation of symbols]

1 Data processing device 2 Learning processor 3 controlled objects 4 Target value setting device 5 Virtual target management device 6 Operation correction amount calculation device 7 First differencer 8 Second differencer

[Procedure amendment]

[Submission date] January 30, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 5

[Correction method] Change

[Correction content]

Claims (12)

[Claims] 1. A variable signal conversion function is provided, and when a teacher signal group is given, the signal conversion function can be recognized as one that realizes the input / output characteristics of the teacher signal group. A data processing device (1), a virtual target management device (5) for managing a priori knowledge of a data relationship between control state quantities obtained to realize a desired control state of a controlled object, and a control operation When a quantity is given, the control state quantity of the control target or its control target model and the virtual control state quantity specified from the management data of the virtual target management device (5) corresponding to the control state quantity An operation correction amount calculation device (6) for calculating a correction amount of the control operation amount required to realize a desired control state from the target value, and the data processing device (1) has a control state. As you enter the amount, Is applied to the controlled object or its controlled object model as a control operation amount, and the control operation amount is corrected according to the correction amount from the operation correction amount calculation device (6) at that time. Thus, a teacher signal is obtained, and the signal conversion function is set according to the obtained teacher signal, whereby the data processing device (1) is constructed as a control device for realizing a desired control state. In the control device construction system, the data processing device (1) calculates a difference value between a control state quantity output from a control target or a control target model and a control state quantity serving as a control target, or a value corresponding to the difference value. While adopting a configuration for inputting, the virtual target management device (5)
A control device construction processing system, which is configured to manage a priori knowledge of a data relationship between control state quantities with a difference value from a control state quantity as a control target as a parameter. 2. The control device construction processing system according to claim 1, wherein the operation correction amount calculation device (6) outputs the control state amount of the control target or its control target model and the virtual target management device (5). The control device construction processing system is configured to calculate a correction amount of the control operation amount by multiplying a difference value with respect to the virtual target value by a proportional coefficient. 3. The control device construction processing system according to claim 1, wherein the data processing device (1) receives one or more inputs and an internal state value to be multiplied with respect to the inputs. A control device construction processing system comprising: a network structure unit configured by internal connection of basic units that obtain a product sum value and convert the product sum value by a predetermined function to obtain an output value. 4. The control device construction processing system according to claim 1 or 2, wherein the data processing device (1) uses an IF-T to establish a qualitative data relationship between the control state quantity and the control operation quantity.
A control device construction processing system comprising a fuzzy device which is described by a HEN rule and which describes the qualitative attributes of the state quantity and control operation quantity described by the IF-THEN rule by a membership function. 5. A virtual target management means (5) for calculating a virtual target value for a second variable according to a predetermined virtual target curve from a difference between a current control state quantity related to the first variable and a target control state quantity. ) And the first target output value of the virtual target value and the current controlled object regarding the obtained second variable.
And a second variable, the operation correction amount calculation means (6) for calculating the correction amount of the input signal of the controlled object, the current input signal given to the controlled object, and the corrected amount of the input signal of the controlled object. A first calculation means (7) for forming a new teacher signal from the second calculation means, and a second calculation means for calculating a difference value between the current values of the first and second variables to be controlled and the target values of the first and second variables. The calculation means (8) and the difference signal are input to use the teacher signal to determine the relationship between the first variable and the second variable by the data dependency relationship defined by the virtual target management means (5). Learning so that the data dependence relationship set by the virtual target management means (5) is maintained even when the controlled object reaches the target value given by the target value setting means. Learning is performed using the teacher signal, and at the time of execution, the first performance is performed based on the learning result. A control having a learning function of inputting a response result to the difference value from the means (7) and giving it to the controlled object so that the controlled object achieves a desired purpose at the given target value. A control device construction processing system comprising means (1, 2, 10). 6. The control device construction processing system according to claim 5, wherein the control means comprises a neural network. 7. The control device construction processing system according to claim 6, wherein the neural network learns by a backpropagation algorithm. 8. The control device construction processing system according to claim 5, wherein a plurality of the virtual target curves are prepared, a plurality of control units learned corresponding to each of the plurality of lines are prepared, and a current control target is selected. And a means for selecting one of the plurality of control units according to the output of the control device construction processing system. 9. The control device construction processing system according to claim 5, wherein a plurality of the virtual target curves are prepared, and a value of the second variable with respect to the first variable is selected from a plurality of variables on the virtual target curve. And a control device construction processing system. 10. A virtual target management unit for expressing a part or all of control variables for controlling a controlled object to a given target value by a virtual target curve based on experience, and a virtual target management section on the virtual target curve. Operation control amount that calculates the correction amount of the input signal of the control target using a plurality of control units that set the input to the control target that realizes the target value and the virtual target value and the current output value of the control target Calculating means, calculating means for forming a new control target input signal from the current input signal given to the control target and the correction amount of the control target input signal, and the virtual target management unit in the plurality of control units. A control device comprising: means for learning an input / output relationship between an input relating to a current control state quantity of a control target capable of realizing the calculated virtual target value and an output given to the input of the control target. 11. The control device according to claim 10,
A control device, wherein a plurality of control units are made to learn the input / output relations that can realize a plurality of virtual control target curves, and the plurality of control units are selected according to a control state quantity of a control target. 12. The control device according to claim 10,
A control device characterized in that a virtual target curve is changed according to a control state quantity of a control target, and a control unit is made to learn the input / output relationship that can realize the virtual target value.