JPH0454243B2 - - Google Patents

Description

[Detailed description of the invention] [Technical field to which the invention pertains] The present invention identifies process characteristics during closed-loop control of a process, and automatically adjusts sample value control constants of a sample value PID control calculation unit to optimal values based on the results. The present invention relates to a sample value PID control device having the function of

[Prior art and its problems]

In order to optimally control the response of a process to be controlled, a procedure is required to measure the dynamic characteristics of the process. The procedure for measuring the dynamic characteristics is called dynamic characteristic identification. In order to perform this identification, there are cases where identification is performed in an open loop state, separated from the control device that controls the process, and identification is performed in a closed loop state while connected to and controlled by the control device that controls the process.
In order to identify the dynamic characteristics of this process, it is desirable to be able to quickly identify them in a closed loop state from the viewpoint of improving the operating efficiency and economic efficiency of the process.

The closed-loop automatic sample-value PID controller has the function of identifying the dynamic characteristics of the controlled process under closed-loop operation and automatically adjusting the sample-value control constants based on the identification results. Tuning method (No.
The 19th SICE Academic Conference 1201) will be held.

According to this method, while the process to be controlled is under sample value PID control, an identification signal called Persistent Exciting is injected into the control system.
The pulse transfer function of the process is identified by the online parameter identification section, and from the identification results, the sample value control constant of the sample value PID control calculation section is automatically designed and used to automatically adjust the sample value control constant during closed-loop control. It is possible to configure a sample value PID control device with the function of Furthermore, by setting the forgetting coefficient of the online parameter identification unit to a value smaller than 1.0, it is possible to follow the case where the characteristics of the process slowly change. However, in a process with little noise, the covariance matrix of the online parameter identification filter becomes too small, making it difficult for the parameters to quickly follow up if the process characteristics change due to disturbance during identification.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide a sample value PID control device that can quickly follow changes in characteristics of a process being identified and can complete automatic adjustment of control constants in a short time.

[Summary of the invention]

The present invention includes a sample value PID control calculation unit that controls a process to be controlled by sample value, and applies a predetermined identification signal to a control loop controlled by the sample value PID control calculation unit, and based on the sample value PID control calculation unit, A sample value PID control device configured to automatically calculate and auto-tuning control constants of the sample value PID control calculation unit by identifying parameters of a dynamic characteristic model of the process. The parameters of the characteristic model are identified, and the model error of the process is calculated from these parameters and the operation signal and output signal of the process, and the error covariance matrix of the parameters indicating the magnitude of the error of the parameter is determined, and this covariance is a parameter identification means for calculating a forgetting coefficient signal from a matrix and the model error; and a transfer function calculation means for calculating a lower-order parameter of a transfer function in an S (Laplace operator) domain from the parameters calculated by the parameter identification means. , identification completion detection means for determining that the identification of the process characteristic is completed when the forgetting coefficient signal and the low-order parameter each satisfy a predetermined condition; Said sample values from lower order parameters
A sample value PID control device comprising: sample value control constant calculation means for calculating a control constant of a PID control calculation section.

〔Effect of the invention〕

According to the present invention, a permissive and exciting identification signal is added to the output signal of the sample value PID control and injected into the process as an operation signal, and a process output signal obtained at each sample period is obtained while performing closed-loop control. , from the input signal of the process, the parameters of the pulse transfer function of the process can be identified online, and from the results, the best sample value control constants can be automatically adjusted. Moreover, even if the characteristics of the process change due to disturbance during identification, the forgetting coefficient is changed and the covariance matrix is largely reset, so that changes in the characteristics can be quickly followed.

In other words, since process characteristics can be identified and control constants can be automatically adjusted under closed-loop control, there is almost no need for actual adjustment of the control device. Also,
Automatic adjustments can be made even while the process is operating, so
The plant can be operated at a high rate of operation, which is often of great benefit.

Furthermore, when the proportional gain of the controller is fixed to zero, it becomes an open loop state, but since the identification signal is applied at the output of the controller, the dynamic characteristics of the process can be identified not only in the closed loop but also in the open loop.

In addition, since an online parameter identification filter similar to a Kalman filter is used to identify the pulse transfer function of the process, statistical processing is automatically performed and stable process characteristics are obtained without being affected by observation noise. identification and automatic adjustment of control constants.

[Embodiments of the invention]

An embodiment of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram showing the circuit configuration of a sample value PID control device according to the present invention.

Note that the sample value PID control device of the present invention is based on a device that is equipped with a sample value PID control calculation unit that controls a process to be controlled using sample values, and that is controlled by the sample value PID control calculation unit. Apply a predetermined identification signal within the loop,
For example, regarding the "sample value PID control device equipped with an auto-tuning function that automatically calculates the control constants of the sample value PID control calculation section by identifying the parameters of the process dynamic characteristic model based on this," Since it is described in detail in JP-A No. 57-39412 (Japanese Patent Publication No. 63-18202), detailed explanation thereof will be omitted in the present specification.

The controlled object is a process 1 such as temperature, flow rate, and pressure in a steel or chemical industrial plant, and the process control amount y(t) is determined by a process output signal by a first sampler 2 that operates at a predetermined sampling period τ. It is returned to the device as y ^* (k). On the other hand, the target value r(t) is determined by a second sampler 3 operating in synchronization with the first sampler, which generates a target value signal r ^* (k), which is determined by the difference between the process output signal y ^* (k) and the target value signal r*(k). Arithmetic unit 4
Then, the deviation signal e ^* (k) (e ^* (k)=r ^* (k)−y ^* (k)) is calculated. Based on this deviation signal e ^* (k), the output signal u _O ^* (k) for controlling process 1 is the sample value PID
It is generated by the control calculation unit 5. A persistently exciting identification signal v ^* (k) is generated by the identification signal generator 6 when identifying the process 1. This identification signal v ^* (k) and output signal u _O ^* (k) are
The input signal U ^* (k) of the process is added by the adder 7.
is calculated, and U ^* (k) is input to the O0-order hold unit 8 and becomes the operation signal u(t) of process 1. The above constitutes a basic sample value PID control device. Here, k is a discrete time, meaning that the actual time t is kτ.

From the process input signal U ^* (k) and the process output signal y ^* (h)(k), the model of process 1 is input to the online identification unit 9 and identified. Furthermore, in preparation for changes in process characteristics during identification, this online identification section 9 detects changes in characteristics based on the model error of the online identification filter (the difference between the predicted value estimated from the model and the actually observed control amount). In order to
Forgetting Factor (hereinafter referred to as forgetting factor) λ(k)
is calculated, and the parameter error covariance matrix (P(k), which indicates the magnitude of parameter error and is hereinafter referred to as covariance matrix) is reset as the forgetting coefficient λ(k) changes due to changes in characteristics. On the other hand, the online identification section 9 outputs the forgetting coefficient λ(k) and the identification parameter θ^(k).
Next, this identification parameter θ^(k) is input, and the transfer function calculation unit 10 calculates the low-order parameter Gi(k) (i=0, 1, 2,
3...) Compute the signal.

This low-order parameter signal Gi(k) is input, and the sample value control constant calculation unit 11 calculates the sample value PID
Constant, Kc (proportional gain), Ti (integral time constant), Td
(differential time constant).

On the other hand, the low-order parameter signal Gi(k) and the forgetting coefficient λ
(k) is input, and the completion detection section 12 calculates and outputs an automatic adjustment completion signal CC(k). Control part 1
3 inputs this automatic adjustment completion signal CC(k) and automatic adjustment start signal ST(k), and outputs an identification signal generation section 6, an online parameter identification section 9, and a transfer function calculation section 1.
0, a control signal CRL that controls each calculation of the sample value control constant calculation section 11 and the completion detection section 12.
(k) is output.

Next, each operation will be explained using the timing chart shown in FIG.

Based on the pulse of the automatic adjustment start signal ST(k),
The control section 13 uses the control signal CRL(k)
Turn on. By turning on this control signal CRL(k), the identification signal generator 6 generates the identification signal u ^* (k).
occurs. Furthermore, the online parameter identification unit 9 inputs the input signal U ^* (k) to the process 1 and the output signal y ^*
(k) is input, and the parameters of the model in Process 1 are passed through an online identification filter, in this example, an RML1 filter that approximates the maximum likelihood method, and the identification begins.
Fig. 3 shows an equivalent model of process 1 focusing on input signal U ^* (k) and output signal y ^* (k). Here, the equivalent parameters of the process are as shown in block 20, the equivalent parameters of noise are as shown in block 21, and the equivalent parameters for the DC component corresponding to the equilibrium point of the process are as shown in block 22. Also, unknown parameters to be identified are ai, bi, ci (where i=1, 2,...
m, j=1, 2,...n), d^ is expressed as a vector θ^, then θ^ ^T = [a^ ₁ , a^ ₂ ,..., a^m, b^ ₁ , b^ ₂ ,... b^n, c^ ₁ , c
^ ₂ ..., c^m, d^]...Equation 1 (T represents transposition), and the algorithm of the RML1 identification filter can be expressed by the following Equations 2 to 5, which are similar to the Kalman filter. can.

θ^(k)=θ^(k-1)+p(k-1)・(k)/λ(k)
+ ^T (k)・p(k−1)(k)・ε(k)……2nd formula Here, the covariance matrix p(k) is shown by the following formula (n
+2m+1)×(n+2m+1) matrix.

p(k)=[p(k-1)-p(k-1)(k) ^T (k)p(
k-1)/λ(k)+ ^T (k)・p(k-1)・(k)]/λ
(k)...Equation 3 In addition, ε(k) is the model error expressed by the following equation.

ε(k)=y(k)−(k) ^T・θ^(k−1)……4th formula Furthermore, (k) is given by the following formula (n+2m+1)
It is a vector.

(k) ^T = [−y ^* (k−1), ..., −y ^* (k−m), U ^*
(k-1), ..., U ^* (k-n), ε(k-1), ...
..., ε(k-m), 1] ...Equation 5 In addition, λ(k) is a forgetting coefficient, which is calculated as follows.

λ(k)=μ(k)+√μ(k) ² +4(k) ^T P(k-1)(k)
/2 ...Equation 6 Here, μ(k)=1-(k) ^T P(k-1)(k)-ε ² (k)/J ^* ...Equation 7 That is, λ(k) is a varying forgetting coefficient driven by the model error ε(k), and all changes in the characteristics of process 1 appear in the model error ε(k), so if there is a change in the characteristics of process 1 during identification, By calculating the sixth and seventh equations, λ(k) becomes smaller than 1.0. As a result, the covariance matrix P(k) is reset, and the online parameter identification section 9 can follow changes in characteristics. On the other hand, when the parameter θ^(k) converges, the model error ε(k) becomes smaller, and the forgetting coefficient λ(k) becomes 1.0.
approach.

Note that the initial value P(o) of the covariance matrix P(k) is set as follows.

P(o)=β·I...Equation 8 Where, β is a large positive number and I is a unit matrix.

In addition, the initial value θ^(o) of the unknown parameter θ^(k) is
If the general parameters of process 1 or the previous parameters are known, set their values. If you don't know at all, set it as follows.

θ^(o)=O ...Equation 9 Furthermore, regarding the parameter J ^* set in Equation 7, if the variance value of model error ε(k) is measured as σ ² 〓, approximately the following Set it as follows.

J ^* = 60 to 140σ ² 〓 . . . Equation 10 Since the internal operation of the online parameter identification section 9 has been explained above, the operation will be explained again with respect to the timing chart of FIG. 2.

The transfer function calculation unit 10 calculates the transfer function G(s) in the S (Laplace operator) region of the process from the identified parameter θ^(k): G(k,s)=1.0/(G ₀ (k)+G ₁ ( k)S+G ₂ (k)S ² +G ₃
(k)S ³ +…) ...Low-order parameter Gi(k) of equation 11 (i=0, 1, 2,...)
Calculate.

These low-order parameters of the S-domain transfer function represent the characteristics of the low to middle frequency band that are important in tuning the control system, and are used to detect process characteristic changes or sample value PID.
It can be used to calculate control constants for the control calculation section.

During the identification, when a characteristic change occurs as shown in Figure 2, the forgetting coefficient λ(k), which starts to approach 1.0, decreases.
When the value becomes smaller than 1.0, the online parameter identification unit 9 immediately starts tracking the parameter.
As it follows, λ(k) approaches 1.0 and
The low-order parameter Gi(k) of the transfer function in the S domain (i=
0, 1, 2,...) also become stable. Therefore, the completion detection unit 12 inputs the forgetting coefficient signal λ(k) and the low-order parameter signal Gi(k), and monitors the completion of identification from the next calculation.

λ(k)≧0.97 ...Equation 12 |Gi(k)−Gi(k−1)/Gi(k−1)|≦αi(i=0
, 1) ...Equation 13 (however, αi = a small positive number of about 0.01) Furthermore, when the conditions of Equations 12 and 13 are satisfied multiple times, five times in this example, completion is detected. The unit 12 generates a completion pulse of an automatic adjustment completion signal CC(k). When the completion pulse of this automatic adjustment completion signal CC(k) is input to the control section 13, the control signal CRL(k) to each calculation section is reset to an off state. Due to the fall of this CRL(k) signal,
The sample value control constant calculation unit 11 generates a low-order parameter signal Gi(k) of the transfer function (i=0, 1, . . . ),
Based on the sample period τ, the sample value control constant
Kc, Ti, and T _D are calculated and transferred to the sample value PID control calculation section 5. At the same time, by turning off this control signal CRL(k), the identification signal generating section 6,
The calculation operations of the online parameter identification unit 9, transfer function calculation unit 10, and completion detection unit 12 are stopped, and process 1 is controlled using the newly automatically set sample value control constant, and a series of control This completes the system's automatic adjustment procedure.

As explained above, the forgetting coefficient λ(k) is not fixed at a constant value, but is controlled by the model error ε(k), so it is quickly responsive to changes in the characteristics of the process being identified. It follows quickly and can automatically complete automatic adjustment in an extremely short time.

[Other embodiments of the invention]

In the embodiment of the present invention described above, the approximate maximum likelihood method was used to identify the pulse transfer function of the process, but any identification method that can obtain unbiased matching estimates may be the extended least squares method or the auxiliary variable method. , model reference form parameter identification method, etc. can be used.

Further, in the present invention, a persistently exciting signal is particularly used as the identification signal. This persistently exciting signal is not limited to the M-sequence signal used in the above embodiment, but can be a pseudo-Random Binary Signal if it is a signal that belongs to a stationary random process.
A normal random number signal, a white noise signal, etc. may also be used. For these signals, see, for example, I. Gustavsson et al.'s Identification of processing in closed
loop−Identifiability and Accuracy Aspects−
The magazine Automatics published in 1997 as
Defined on page 65. Furthermore, instead of directly using the M-sequence signal, an output signal obtained by passing the generated M-sequence signal through a digital filter may be used.

In particular, by using the M-sequence signal as the identification signal, the M-sequence signal takes binary (+1, -1) or ternary (+1, 0, -1) values, making the identification process easier. can be made to work,
Can be safely identified. Furthermore, since the M-sequence signal includes many frequency components, the matching time can be shortened. Furthermore, since the M-sequence signal is relatively easy to generate and can be obtained with a simple logic configuration, the device can be miniaturized.

According to the present invention, even if the characteristics of the process change due to disturbance during identification, the forgetting coefficient is changed and the covariance matrix is largely reset, so that the change in characteristics can be quickly followed.

In other words, since process characteristics can be identified and control constants can be automatically adjusted under closed-loop control, there is almost no need for actual adjustment of the control device. Also,
Automatic adjustments can be made even while the process is operating, so
The plant can be operated with high operating efficiency.

Furthermore, if the proportional gain of the controller (sample value PID control calculation unit) is fixed to zero, it will be in an open loop state, but since the identification signal is added to the output point of the controller, the process can be operated not only in a closed loop but also in an open loop. Characteristics can be identified.

In addition, since an online parameter identification filter similar to a Kalman filter is used to identify the pulse transfer function of the process, statistical processing is automatically performed and stable process characteristics are achieved without being affected by observation noise. identification and child adjustment of control constants.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the present invention, FIG. 2 is a timing chart explaining this operation, and FIG. 3 is a block diagram showing a process model. 1... Process, 2, 3... Sampler, 5...
Sample value PID control calculation section, 6...Identification signal generation section, 7...Addition section, 8...O-order hold, 9...
Online parameter identification unit, 10...Transfer function calculation unit, 11...Sample value control constant calculation unit, 1
2... Completion detection section, 13... Control section.

Claims (1)

[Claims] 1. A sample value PID control calculation unit that controls a process to be controlled using sample values, and a predetermined identification signal is applied to a control loop controlled by the sample value PID control calculation unit, and In the sampled value PID control device, the sampled value PID control device is configured to automatically calculate and auto-tuning a control constant of the sampled value PID control calculation unit by identifying parameters of the dynamic characteristic model of the process based on the above. Identifying parameters of a dynamic characteristic model of the process, calculating model errors of the process from these parameters and operating signals and output signals of the process, and determining an error covariance matrix of the parameters indicating the magnitude of the error of the parameters; a parameter identification means for calculating a forgetting coefficient signal from the covariance matrix and the model error; and a transfer function for calculating a lower-order parameter of a transfer function in the S (Laplace operator) domain from the parameters calculated by the parameter identification means. calculation means; identification completion detection means for determining that the identification of the process characteristic is completed when the forgetting coefficient signal and the low-order parameter each satisfy a predetermined condition; and the identification completion detection means determines that the identification is complete. The sample value from the low-order parameter when
A sample value PID control device comprising: sample value control constant calculation means for calculating a control constant of a PID control calculation section.