JPH0677212B2 - Plant modeling equipment - Google Patents

Description

TECHNICAL FIELD OF THE INVENTION The present invention provides a dynamic characteristic model necessary for designing a control system, pre-evaluation, abnormality diagnosis of a controlled object, etc. from a response waveform of a controlled object of unknown dynamic characteristics. , A plant modeling apparatus for estimating a transfer function.

[Technical background of the invention and its problems]

A model representing the dynamic characteristics of the plant, such as a transfer function, is necessary for designing the control system of the plant, performing numerical tests, and monitoring. Conventionally, there have been proposed a number of devices for estimating a transfer function from response waveform information of a plant whose unknown dynamic characteristics are known, but each has its application limit, and there is no one that can be applied to all plants and environmental conditions. For example,
When the data includes disturbances and noise, the accuracy of estimation is improved by pre-processing the data to remove the disturbances from a certain amount of prior knowledge, but a unified transfer function estimation device that includes data pre-processing procedures Not proposed.
Further, even if some information about the structure of the plant to be estimated is known in advance, no systematic rack estimation device that effectively uses the information has been proposed. Therefore, depending on the situation, humans must select effective means and perform work heuristically and by trial and error. The operator (operator) requires a high degree of specialized knowledge, and a lot of time and labor is required. Had to spend.

[Object of the Invention]

In order to solve the above-mentioned problems, the present invention provides a plurality of means for estimating and modifying a transfer function from a response waveform of a plant, or modifying response waveform data again from the result, and making a human judgment. Estimate the transfer function for all types of plants and situations in a short time with high precision by efficiently and effectively driving those means by inference by a set of rules stored in a database. The object is to provide a plant modeling device.

[Outline of Invention]

The present invention has the following functions as means for estimating a transfer function from a response waveform of a plant to be controlled, thereby realizing an improvement in estimation accuracy and an increase in work efficiency.

(1) If the response waveform data of the plant contains disturbances or abnormal data that may lead to a decrease in estimation accuracy, the transfer function is estimated once from the data, and then the validity of the data is determined from the model. , A function that corrects the data by deleting the data that is inconvenient for estimation and interpolating it with the data that substitutes, and re-estimates the transfer function.

(2) Once the transfer function estimated from the response waveform data is stored, its moment sequence is stored, and the structure, the function of deforming the order, the dead time are extracted, or the partial structure of the transfer function is constrained to a presumed form. The function to perform the transformation of the transfer function.

(3) The operator inputs each command by making the individual functions relating to transfer function estimation, correction, and data correction independent in the minimum unit and associating a command (command) with each one. An interactive model correction function that can drive the function corresponding to it and perform transfer function estimation, reduction of dimension, correction, etc. in a short time based on the operator's judgment.

(4) Evaluating a model that has a database that can store the procedures for driving these functions as rules, and compares a series of work procedures performed by the operator with the response waveform of the estimated transfer function and the original data. Alternatively, a new rule is generated based on the evaluation score of the model goodness of fit evaluated by the operator, and the learning function of storing it in the database and the a priori information on these rules stored in the database and the structure of the estimation target A function that infers, based on these rules, the sequence and conditions under which a series of functions should be driven based on the rules related to knowledge of characteristics, instruct the operator, or automatically drive the functions.

〔The invention's effect〕

 The present invention has the following effects.

(1) When the data related to the response waveform of the plant to be estimated includes disturbance or observation noise, abnormal data is detected and removed by comparing the response waveform of the estimated transfer function with the original data. As a result, the estimation accuracy can be prevented from deteriorating.

(2) After estimating the transfer function of the plant, it can be transformed into a form convenient for designing the control system.

(3) Since individual functions related to model estimation and correction are independent in a minimum unit, the system is rich in modularity, and it is easy to change the system by adding new functions.

(4) Since each function can be operated interactively in a command format, modeling work can be done efficiently in a single time.
It is easy to perform trial-and-error complicated work that requires human judgment.

(5) Based on the inference of the database based on the rules, the operator can be instructed the optimum procedure of modeling or can automatically perform modeling, and the operator does not need to have specialized knowledge about modeling.

(6) A rule-based reasoning mechanism is asymptotically optimal because a new rule is generated and meaningless rules are deleted from the estimation and correction procedures and the evaluation of the resulting model. It becomes a thing.

(7) Modeling including known structural information of the plant can be performed according to the operator's intention or the knowledge and rules stored in the database.

Example of Invention

One embodiment of the present invention is shown in FIGS. 1 (a) and 1 (b).
The plant modeling apparatus p of the present invention includes a series of transfer function estimation functions by the response waveform data acquisition unit 1, the data preprocessing unit 4, the order determination unit 5, and the transfer function estimation unit 6, and the transfer function correction of steps 8 to 11. Function group, response waveform calculation unit 12,
Estimated model evaluation unit 13, estimation / correction function management unit 7 for managing these functions, graphic display terminal 14, console 15 for interfacing with an operator, database 16, processing procedure inference unit 17, processing procedure learning unit
Composed of 18.

The response waveform data acquisition unit 1 takes in response waveform data from the plant 3 to be estimated or the data recorder 2 in which it is recorded, and temporarily stores it in the database 16.

The data pre-processing unit 4 subjects the response waveform data to the following processing, and modifies it so that it is convenient for subsequent estimation.

(1) In the case of continuous signal data such as the plant response waveform 19 as shown in FIG. 2 (a), after passing through a high frequency cutoff filter that cuts off high frequency components of the signal by sampling so as not to cause aliasing. Sampling the filtered response waveform 20 as shown in FIG. 2 (b) and sampling the response waveform as shown in FIG. 2 (c).
21 {y ₁ , y ₂ ... Y _N } is obtained.

(2) When the pre-processing unit is activated again after the transfer function has been estimated once, the response waveform of the estimated transfer function is obtained by the response waveform calculation unit 12 and compared with the data.
Modify the data according to the following rules.

Rule 1; First, compare the estimated response waveform of the model (= transfer function) with the original data, and obtain the following amount.

(However, Y _i is the original data, Is the response waveform sample value of the estimation model. ) If N data is N ₀ For the data within If, then Y _i is removed as abnormal data, Change.

Rule 2; If the formula (1) is satisfied for N ₀ or more data, it is determined that the estimation model is inappropriate, and the conditions are re-estimated and estimated again.

Rule 3; From the estimated model information, if it is determined that there is waste time in the plant by inference by the rule group of the database or the operator's judgment, it corresponds to the waste time from the beginning of the original data Delete the data that you want to re-estimate.

Rule 4; If the plant is determined to be an atypical system from the information of the estimated model or from the a priori information on the structure of the plant stored in the database, sample data Y _i (i = 1, ..., N), (I = 1, ..., N, τ: sampling period) Is re-estimated as data.

(3) When the operator inputs a command regarding data correction from the console, the following basic processing is performed according to the command.

(A) Decimation of data: Data is thinned out every l to make data again.

Y ₁ , Y ₂ ... → Y ₁ , Y _2l , ... (b) Use of data part: A part of response waveform data is taken out and used as data again.

Y ₁ , Y ₂ , ..., Y _N → Y _i , Y _{i + 1} , ..., Y _j (c) Average of a plurality of data groups: Average a plurality of data sets to obtain new data.

The order determining unit 5 determines the order from the information of the sample data by the following procedure.

(1) The impulse response sequence g _i of the plant is calculated from the sample data Y _i .

(2) Obtain the singular value of the Hankel matrix H.

(3) For the threshold value ε, find an integer n such that σ _n >ε> σ _{n + 1} and set it as the order of the transfer function.

The transfer function estimation unit 6 obtains the transfer function by the following procedure.

A matrix consisting of the first n columns of the matrix U of equations (1) and (4),
A matrix consisting of the first n columns of V Then, the matrix A, the vector B, the ^CT , and the scalar D are obtained.

(2) The matrix A is diagonally compared using an appropriate conversion matrix.

(3) Matrix A _c , vector B _c , C _c , scalar D
_{Find c} .

(4) The transfer function G (s) is obtained from the following equation.

G (s) = C _c (sIn−A _c ) ⁻¹ B _c + D _c (8) (In is an n × n identity matrix) In the estimation / correction function management unit 7, the above transfer function estimation function group and the following One transfer function modification function group described in
One function is to activate a function corresponding to a command input by the operator from the console 15 in correspondence with a command (command word). Also, a database of data and estimated models
It has functions such as writing / reading to / from 16 and interface with an operator through the graphic display terminal 14 and the console 15.

The following steps 8 to 13 have the functions related to correction / evaluation of the estimated transfer function, pole / zero point, elimination / change / addition function.
8; Eliminate, change, and add poles and zeros of the estimated transfer function while preserving a finite number from the lower order side of the moment series (9).

_{G (s) = g 0 +} g 1 s + g 2 s 2 + ... ... (9) for example, the following modifications.

Parameter change function 9; Change steady gain time scale of transfer function, dead time, etc.

Transfer function reduction function 10; Change to a transfer function having a denominator / numerator of an arbitrary order while retaining the moment series (9). That is, the new denominator / numerator order (n ′,
m ′) solves the following system of linear equations ai (i = 1,2,
, N '), bi (i = 0,1,2, ..., m') are obtained.

From this, a reduced-dimensional transfer function of the following equation is obtained.

Waste time extraction function 11: Uses the moment sequence expression (21) of the given transfer function G (s) as waste time element; Based on the resulting moment series divided by, (1
The transfer function G '(s) is obtained again by the method of equations (0) and (11). As a result, the dead time 1 can be extracted as in the following equation.

G (s) → e ^ls G ′ (s) Response waveform calculation unit 12; Response waveform of the estimated and corrected transfer function is calculated by the Runge-Kutta-Gill method, and the original data {Y ₁ , Y ₂ , ... Y _N } and displayed on the graphic display terminal 14.

Estimated model evaluation unit 13: Sample data {i} of the response waveform of the estimated transfer function is compared with the original data {Yi}, evaluated according to the following rules, and the result is displayed on the terminal and written in the database.

Rule 1; Look at the fitness of the estimated model to the original response waveform data with the following parameters.

Rule 2; When the original response waveform data has a peak, the peak value (maximum limit) and peak time of the first peak, the second peak, etc. are compared with the estimated model response waveform.

Rule 3; If the original response waveform data has a vibration component, compare the vibration period, damping ratio, etc. with the estimated model response waveform.

Rule 4; When the original response waveform data has an inverse response, its magnitude is compared with the estimated model response waveform.

Rule 5; Compare the steady-state gain settling time of the original response waveform data with those of the estimation model.

Graphic display terminal 14 and console 15
Interface with the operator.

The database 16 records plant response waveform data estimation models, estimation and correction procedures, model evaluation degrees, etc., every moment. Further, rules regarding estimation / correction are stored rules regarding the structure of the plant to be estimated, the physical laws, and the like.

The processing procedure inference unit 17 infers what kind of procedure is optimum for the processing requested by the operator based on the rule group stored in the database 16, and drives each function or instructs the operator according to the result. Give instructions.

In the processing procedure learning unit 12, from the modeling procedure by the operation of the operator and the evaluation of the model obtained as a result,
Generate rules for estimating / correcting the model and newly store them in the database 16. Further, the value of the rule used for inference is judged from the evaluation degree of the model obtained as a result, and the rule whose value index is smaller than a certain threshold is deleted from the database.

Fig. 3 shows the response waveform of the model as an estimation result when the step response waveform from the recirculation pump speed set value of the boiling water reactor (BWR) to the core neutral flux is used as data, and it The response waveform of the modified model is shown with the original data. Once estimated transfer function By applying various modifications to the model (Fig. 3 (a)), a more accurate transfer function can be obtained. It can be seen that the model (FIG. 3 (b)) was obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of a plant modeling apparatus as an embodiment of the present invention, FIG. 2 is a curve diagram showing a sampling state of a plant response waveform, and FIG. 3 is a BWR recirculation. It is a curve figure which shows each response waveform of the model estimated based on the step response data from a ring pump speed set value to a core neutron flux, and the model which added correction. 1 ... Response waveform data acquisition unit, 2 ... Data recorder, 3 ... Plant to be estimated, 4 ... Data preprocessing unit, 5 ... Order determination unit, 6 ... Transfer function estimation unit, 7 ... Estimation / correction function management unit, 8 ... Pole / zero elimination / change / addition function, 9 ... Parameter change function, 10 ... Transfer function reduction function, 11 ... Waste time extraction function, 12 ... Response waveform Calculation unit, 13 ... Estimated model evaluation unit, 14 ... Graphic display terminal, 15 ... Console, 16 ... Database, 17 ... Processing procedure inference unit, 18 ... Processing procedure learning unit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-57-23117 (JP, A) JP-A-56-60914 (JP, A) JP-A-59-69813 (JP, A) JP-A-58- 208812 (JP, A) JP 58-172714 (JP, A) JP 57-146314 (JP, A) JP 51-94075 (JP, A)

Claims (4)

[Claims] 1. A data acquisition unit that takes in response waveform data for a known operation signal of a plant to be controlled, a data correction unit that performs sampling, filtering, partial deletion, and interpolation processing on the data, and this data correction. Based on the sample data of the response waveform corrected by the model estimation unit, the model estimation unit having the order determination function and the transfer function estimation function, the model correction unit that corrects the transfer function estimated by this model estimation unit, and this model A storage device for recording / reproducing the transfer function modified by the modifying unit, a display device for displaying the transfer function and its response waveform, a storage device for recording / reproducing the estimation result, and a console terminal for fetching work instructions. In a plant modeling system equipped with, the individual functions for estimating and modifying the transfer function are independent of each other in the smallest unit. Te has, in correspondence with Inochiminato word to each function, plant, characterized in that to execute immediately activate the function corresponding to the command entered in work orders
Modeling equipment. 2. When correcting the estimated transfer function, a function of adding, changing or deleting a structural parameter of a part or the whole of the transfer function while keeping a finite number of moment sequences from the low order of the transfer function. Similarly, the function of reducing the transfer function while saving a finite number of moment sequences from the low order, and the function of extracting the waste time from the transfer function while also saving a finite number of moment sequences from the low order. , The steady-state gain of the transfer function, time scale, waste time,
The plant modeling apparatus according to claim 1, further comprising a function of finely adjusting the coefficient. 3. The transfer function is estimated once from the plant response waveform data, and the original data is compared with that of the response waveform to delete abnormal data and correct the data by interpolation. The plant modeling apparatus according to claim 1, further comprising a function of re-estimating the transfer function. 4. A processing procedure learning for generating, as a rule, information about a use order of individual functions involved in transfer function estimation, correction, and data correction, and a use condition and storing the rule in a database or deleting an unnecessary rule from the database. From the functions and their rules stored in the database, known information about the plant to be estimated, and rules regarding general rules for estimating the transfer function, the transfer function is estimated, modified, and data modified. The plant modeling apparatus according to claim 1, which has a processing procedure inference function of estimating a transfer function by inferring an optimum procedure.