JP6966245B2 - Support device and support method - Google Patents

Description

The present invention relates to a support device and a support method that support the creation of a model by displaying a mathematical model used in process control.

There is a model control method that predicts the future response of a control variable using a mathematical model such as a transfer function model or a state-space representation model, and calculates an operation variable such that the predicted response approaches a reference trajectory or a target value. In order to implement a control method using such a mathematical model, it is necessary to create (estimate) the model. There are several methods for creating this model, but one of the typical methods is a method called system identification. In system identification, input / output time series data to be modeled is statistically processed, and a model that reproduces the input / output is created by estimating using the least squares method or the like.

In model creation by system identification, in general, an engineer creates multiple models by trial and error, evaluates the created multiple models from the viewpoint of accuracy, etc., and the evaluation result is high and it seems appropriate for the purpose of use. It often goes through a procedure such as selecting the one to be used. Therefore, many models are created through the model creation work. In this way, trial and error is necessary because there are many setting conditions in the model creation work.

Examples of setting conditions include a preprocessing method (noise removal and trend removal) of input / output time series data used to create a model, a model type and order (number of model parameters), and an estimation method. In addition, when the time series data that can be used to create the model exists for a long period of time, the selection of the period is also one of the setting conditions. In this way, the conditions at the time of model creation are composed of a combination of many setting conditions, so that the number of model candidates inevitably increases as trial and error is repeated.

Lennart Ljung, "Matlab & Simulink, System Identification Toolbox Getting Started Guide", The MathWorks, Inc., R2015b, p. 4-15, 2015.

As described above, in the creation of the model used for the model control method, the number of model candidates may increase due to trial and error, and it is a laborious task to select an appropriate model from these.

In process control, transient characteristics are often emphasized when the target value is changed or when a disturbance occurs. In order to improve this transient characteristic, it is important to use a mathematical model with high accuracy of transient response. However, it is not easy to select a model with high transient response accuracy from a large number of model candidates for two reasons described later.

First, in the field of system identification, it is not common to arrange models based on evaluation indexes. For example, in the technique of Non-Patent Document 1, a screen is disclosed in which the step responses of estimated model candidates are displayed side by side in a grid pattern, and an engineer selects from the displayed model candidates. However, on this screen, the models are only arranged in the estimated order. It can be sorted manually, but there is no function to sort according to the evaluation index. Therefore, even if the number of model candidates is small, if the number of candidates is large, it becomes difficult to know which model has a good evaluation result.

Second, evaluation indicators such as RMSE (Root Mean Square Error), which is generally used in the field of system identification, and the precision rate calculated based on RMSE, are the accuracy of the transient response of the model. Is often not reflected. RMSE is calculated from the output error of the actual control target and the model, but this error depends on the input time series data. However, in process control, operation input that greatly fluctuates the process is often avoided. For this reason, RMSE is also often calculated by giving an input such that there is little change / fluctuation and the steady state lasts for a long time. Since the transient response is the response output when there is a change in the input, the RMSE calculated from the time series data with little change / fluctuation does not reflect the transient characteristics so much.

Due to these two issues, it is a laborious task for the engineer to select a model having excellent transient characteristic accuracy from many model candidates, that is, a model having good transient characteristic of the control system.

The present invention has been made to solve the above problems, and many prototype models have been prototyped even in a situation where the model must be evaluated using time-series data in which the steady state occupies most of the data. The purpose is to enable efficient selection of models with excellent transient characteristic accuracy.

The support device according to the present invention stores input / output data composed of a model storage unit that stores a plurality of models to be controlled, an operation input to the control target, and a control output output from the control target by the operation input. The difference between the input / output data storage unit to be stored and the model output and control output of each of the multiple models using the operation input reflects the accuracy of the transient characteristics of the model output, and the evaluation index of each of the multiple models is set. An evaluation calculation unit configured to calculate, an order determination unit configured to determine the display order of a plurality of models according to the size of the evaluation index, and a plurality of models arranged side by side in the order determined by the order determination unit. It is provided with a display control unit configured to be displayed on the display unit.

In the above support device, the evaluation calculation unit passes the model output calculation function unit, which is configured to obtain the model output which is the output of each of the plurality of models using the operation input, and the control output and the model output through a filter. The evaluation index is calculated based on the error between the emphasis control output and the emphasis model output, and the transient response emphasis function unit configured to obtain the emphasis control output and the emphasis model output that emphasize the transient response. It is equipped with an evaluation index calculation function unit.

In the above support device, the evaluation calculation unit includes a model correction function unit configured to generate a plurality of correction models in which the DC gains are corrected to the same value for the plurality of models, and a plurality of correction models using operation inputs. Evaluation index calculation configured to calculate the evaluation index based on the error between the control output and the correction model output and the model output calculation function unit configured to obtain the correction model output, which is each output of. It has a functional unit.

In the above support device, the evaluation calculation unit is a model output calculation function unit configured to obtain a model output which is an output of each of a plurality of models using an operation input, and a transient response from each of the control output and the model output. The transient state section extraction function unit, which is configured to extract the interval and obtain the extraction control output and the extraction model output, is configured to calculate the evaluation index based on the error between the extraction control output and the extraction model output. It is equipped with an evaluation index calculation function unit.

The support method according to the present invention is composed of a first step of storing a plurality of models of a controlled object in a model storage unit, an operation input to the controlled object, and a control output output from the controlled object by the operation input. The second step of storing the input / output data in the input / output data storage unit and the difference between the model output and the control output of each of the multiple models using the operation input reflect the accuracy of the transient characteristics of the model output. The third step of calculating each evaluation index of each model, the fourth step of determining the display order of a plurality of models according to the size of the evaluation index, and the display unit arranging a plurality of models according to the order determined in the third step. It is provided with a fifth step of displaying on.

In the above support method, the third step is the sixth step of obtaining the model output which is the output of each of the plurality of models using the operation input, and the emphasis that the control output and the model output are filtered to emphasize the transient response. It includes a seventh step of obtaining a control output and an emphasized model output, and an eighth step of calculating an evaluation index based on an error between the emphasized control output and the emphasized model output.

In the above support method, the third step is the ninth step of generating a plurality of correction models in which the DC gains are corrected to the same value for the plurality of models, and the output of each of the plurality of correction models using the operation input. The tenth step of obtaining the correction model output and the eleventh step of calculating the evaluation index based on the error between the control output and the correction model output are provided.

In the above support method, the third step is the sixth step of obtaining the model output which is the output of each of the plurality of models using the operation input, and the extraction control by extracting the transient response section from each of the control output and the model output. It includes a twelfth step of obtaining an output and an extraction model output, and a thirteenth step of calculating an evaluation index based on an error between the extraction control output and the extraction model output.

As described above, according to the present invention, even in a situation where the model must be evaluated using time-series data in which the steady state occupies most of the data, the accuracy of the transient characteristics is higher than that of many prototype models. The excellent effect is that a good model can be selected efficiently.

FIG. 1 is a configuration diagram showing a configuration of a support device according to a first embodiment of the present invention. FIG. 2 is a flowchart for explaining the support method according to the first embodiment of the present invention. FIG. 3 is a configuration diagram showing a configuration of a support device according to a second embodiment of the present invention. FIG. 4 is a flowchart for explaining the support method according to the second embodiment of the present invention. FIG. 5 is a configuration diagram showing a display example by the support device according to the second embodiment. FIG. 6 is a configuration diagram showing a display example according to the conventional technique. FIG. 7 is an explanatory diagram for explaining the state of the operation input and the control output of the stepped waveform. FIG. 8 is an explanatory diagram for explaining the state of the control output of the stepped waveform. FIG. 9 is a characteristic diagram showing the result of confirming the effect in the second embodiment by the spectrum analysis. FIG. 10 is a configuration diagram showing a configuration of a support device according to a third embodiment of the present invention. FIG. 11 is a flowchart for explaining the support method according to the third embodiment of the present invention. FIG. 12 is a configuration diagram showing a configuration of a support device according to a fourth embodiment of the present invention. FIG. 13 is a flowchart for explaining the support method according to the fourth embodiment of the present invention. FIG. 14 is a configuration diagram showing a configuration of a support device according to a fifth embodiment of the present invention. FIG. 15 is an explanatory diagram for explaining a plurality of models stored in the model storage unit 101. FIG. 16 is an explanatory diagram showing a state in which the model set is displayed for each group. FIG. 17 is an explanatory diagram showing a state in which the model set is displayed for each group. FIG. 18 is a configuration diagram showing a hardware configuration of the support device according to the embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described.

[Embodiment 1]
First, the support device according to the first embodiment of the present invention will be described with reference to FIG. This support device includes a model storage unit 101, an input / output data storage unit 102, an evaluation calculation unit 103, an order determination unit 104, a display control unit 105, a display unit 106, and an operation reception unit 107.

The model storage unit 101 stores a plurality of models to be controlled. The model storage unit 101 stores the model in a known form such as a transfer function, a differential equation, and a step response. Each model can be prototyped by means such as system identification.

The input / output data storage unit 102 stores input / output data composed of an operation input to the control target and a control output output from the control target by the operation input. The operation input and control output are time series data. The input / output data is a set of an operation input to the actual control target and a control output output as a result of this operation input. It is preferable that the data has a constant time interval because it is easy to carry out. In addition, since it is necessary to evaluate the transient characteristics, it is important that the time interval of the data is short enough to reproduce the shape of the transient response in sufficient detail. For example, it is difficult to evaluate the transient characteristics with data having only 0 to 5 points until the target value is changed and the control output from the controlled object is set.

The evaluation calculation unit 103 calculates the evaluation index of each of the plurality of models by reflecting the accuracy of the transient characteristics of the model output on the difference between the model output of each of the plurality of models using the operation input and the control output. do.

The order determination unit 104 determines the display order of the plurality of models according to the magnitude of the evaluation index. The display control unit 105 displays a plurality of models side by side on the display unit 106 in the order determined by the order determination unit 104. The operation reception unit 107 receives an operation instruction by an engineer. For example, the model to be controlled is stored in the model storage unit 101 in response to an operation instruction received by the engineer received by the operation reception unit 107.

Next, the operation (support method) of the support device according to the first embodiment will be described with reference to FIG. First, in step S101, a plurality of models prototyped by an engineer by trial and error for use in target model control are stored in the model storage unit 101 (first step). Next, in step S102, the input / output data composed of the operation input to the control target and the control output output from the control target by the operation input is stored in the input / output data storage unit 102 (second step). ..

Next, in step S103, the evaluation calculation unit 103 uses the operation input stored in the input / output data storage unit 102 for each of the plurality of models stored in the model storage unit 101. The evaluation index is calculated by reflecting the accuracy of the transient characteristics of the model output in the difference between the model output and the control output of (3rd step).

When the above-mentioned calculation of the evaluation index is performed for all the models stored in the model storage unit 101 (y in step S104), in step S105, the order determination unit 104 determines a plurality of models depending on the size of the evaluation index. The display order of is determined (4th step). Next, in step S106, the display control unit 105 displays a plurality of models side by side in the order determined by the order determination unit 104 on the display unit 106 (fifth step).

As described above, according to the first embodiment, not only the error of the existing RMSE or the like but also the evaluation index in which the difference is emphasized by reflecting the accuracy of the transient characteristic in the output of the model is used as the evaluation index. .. By displaying the models side by side according to this evaluation index, it is possible to easily identify the model having excellent transient characteristic accuracy. As a result, the work load of selecting a model that improves the transient characteristics of the control system from among a large number of model candidates by engineers can be reduced.

It should be noted that solving only one of the two problems does not solve the problem. For example, displaying a list of models in ascending order of RMSE can be easily realized only by existing technology. However, even if the models are displayed side by side in the order of the indexes that do not reflect the accuracy of the transient characteristics, the burden on the engineer is not reduced. Moreover, even if an index that emphasizes the accuracy of transient characteristics is used, it cannot be said that it is easy to understand which model is superior unless the index value is reflected in the order in which the models are arranged.

[Embodiment 2]
Next, the support device according to the second embodiment of the present invention will be described with reference to FIG. This support device includes a model storage unit 101, an input / output data storage unit 102, an evaluation calculation unit 103, an order determination unit 104, a display control unit 105, a display unit 106, and an operation reception unit 107.

In the second embodiment, the evaluation calculation unit 103 includes a model output calculation function unit 131, a transient response emphasis function unit 132, and an evaluation index calculation function unit 133. Other configurations are the same as those in the first embodiment described above.

The model output calculation function unit 131 obtains the model output, which is the output of each of the plurality of models, using the operation input. The model output calculation function unit 131 inputs the operation input stored in the input / output data storage unit 102 to each model stored in the model storage unit 101, and performs time-series data by a well-known simulation or the like. Get the model output that is. In addition, in order to evaluate equally, it is preferable that the data applied to each model is the same, and it is common.

The transient response emphasis function unit 132 passes the control output and the model output through a filter to obtain the emphasis control output and the emphasis model output in which the transient response is emphasized. The transient response emphasis function unit 132 applies a frequency filter to the control output and the model output of the evaluation data to obtain the data in which the transient response is emphasized. For example, the low frequency range is filtered through a high-pass filter to emphasize the mid-high frequency range, or the mid-frequency range is emphasized through a bandpass filter. In this way, the steady-state response is cut to obtain the emphasis control output and the emphasis model output in which the transient response is emphasized.

Signal processing by a high-pass filter or a band-pass filter can be easily performed by using numerical calculation software such as MATLAB (registered trademark) Signal Processing Toolbox. Here, it is necessary to determine the cutoff frequency (cutoff frequency) of the filter, and there are several methods for this.

For example, the cutoff frequency of the filter can be determined by using the knowledge of the modeling target. For example, when modeling the temperature change of the reactor, when the flow rate of the heating steam supplied to the reactor is increased by a certain amount and the flow rate is maintained, it takes about 30 minutes for the temperature to settle after that. It is a finding that requires. Based on this finding, the cutoff frequency of the filter is determined so as to attenuate components having a long period of more than 30 minutes (that is, low frequency components). That is, the cutoff frequency of the high-pass filter ^{can be calculated as "8.3 x 10 -4} = 1 / (60 x 30) [Hz]".

Further, the cutoff frequency may be determined based on the frequency characteristics of the model. For example, the cutoff frequency of the model (the frequency at which the gain attenuates to half of the DC gain on the Bode diagram) is obtained, and half of this median value is used as the cutoff frequency of the filter. By using the statistic of the frequency characteristics of the model, the cutoff frequency of the filter can be determined so that the difference in the dynamic characteristics between the models remains.

The evaluation index calculation function unit 133 calculates the evaluation index based on the error between the emphasis control output and the emphasis model output. The evaluation index calculation function unit 133 calculates the evaluation index for all the models.

RMSE, which is generally used as an evaluation index, is calculated by the following equation using the control output output by the control target (modeling target) and the model output.

On the other hand, in the second embodiment, the following evaluation index is calculated by using the error between the emphasis control output and the emphasis model output output by the transient response emphasis function unit 132. Hereinafter, this evaluation index is referred to as a corrected RMSE.

In this evaluation index, of the difference between the two output time series data of the emphasis control output and the emphasis model output, the low frequency component corresponding to the steady response is attenuated, so that the influence on the value of the evaluation index is small. .. On the other hand, of the difference between the two output time series data, the medium frequency component corresponding to the transient response passes through the filter and is reflected in the value of the evaluation index.

In the second embodiment, as described above, the evaluation index is obtained for each model, the order determination unit 104 determines the display order of the plurality of models according to the magnitude of the evaluation index, and the display control unit 105 is determined. A plurality of models are arranged side by side in the order of the above and displayed on the display unit 106.

The evaluation index in the second embodiment is the difference between the control output and the model output after signal processing, and the property that the error is evaluated is the same as that of RMSE. Therefore, if the errors are displayed in ascending order of the evaluation index in the second embodiment, the model with the highest accuracy of the transient characteristics is displayed first.

Next, the operation (support method) of the support device according to the second embodiment will be described with reference to FIG. First, in step S101, a plurality of models prototyped by an engineer by trial and error for use in target model control are stored in the model storage unit 101. Next, in step S102, the input / output data composed of the operation input to the control target and the control output output from the control target by the operation input is stored in the input / output data storage unit 102.

Next, in step S131, the model output calculation function unit 131 uses the operation input to obtain the model output, which is the output of this model, for the model stored in the model storage unit 101 (sixth step).

Next, in step S132, the transient response enhancement function unit 132 passes the control output and the model output through a filter to obtain the emphasis control output and the emphasis model output in which the transient response is emphasized (7th step).

Next, in step S133, the evaluation index calculation function unit 133 calculates the evaluation index based on the error between the emphasis control output and the emphasis model output (8th step).

When the above-mentioned calculation of the evaluation index is performed for all the models stored in the model storage unit 101 (y in step S104), in step S105, the order determination unit 104 determines a plurality of models depending on the size of the evaluation index. Determine the display order of. Next, in step S106, the display control unit 105 causes the display unit 106 to display a plurality of models side by side in the order determined by the order determination unit 104.

There are various display methods for each model, but step response is often emphasized in process control. In this case, for example, as illustrated in FIG. 5, each model is displayed side by side in the order determined together with the step response. The display form is not limited to this, and model information may be displayed by various known techniques.

When representing a model in a step response plot, it is desirable to have the same range on the time axis (horizontal axis) and output axis (vertical axis) so that the differences between model candidates are clear. For example, the upper left corner is the display order 1, and thereafter, the step response is displayed on the screen in the order of the numbers assigned to the step response. Compared with the display based on RMSE described later, in the display according to the second embodiment, the display is performed in descending order of the accuracy of the transient characteristics, so that the engineer can easily design a control system having excellent transient characteristics.

Here, for comparison, FIG. 6 shows a display example when model candidates are displayed based on RMSE, which is a general evaluation index. With respect to the display example shown in FIG. 5 and the display example shown in FIG. 6, attention is paid to the response speed, which is one of the parameters representing the transient characteristics. Looking at the leftmost column in FIG. 6, the models 1, 2, and 6 respond faster than the models 3, 4, and 5. As described above, in the example shown in FIG. 6, models having different transient characteristics are displayed side by side.

On the other hand, in the display example of the second embodiment shown in FIG. 5, the model candidate having a high response speed is displayed first, and the model having a slow response speed is displayed later. In this example, the model with a faster response speed has a transient characteristic closer to the actual control target. According to the second embodiment, it is possible to examine the model having excellent transient characteristic accuracy when viewed from the models arranged first. On the other hand, when they are arranged in the order of RMSE as shown in FIG. 6, it is difficult to judge the superiority or inferiority of the accuracy of the transient characteristics from the display screen, and the efficiency of the work of selecting the model remains low. In addition to RMSE, the evaluation index includes FPE (Final Prediction Error), AIC (Akaike Information Criterion), etc., but this problem can occur regardless of the type of evaluation index.

As described above, it is difficult to evaluate the accuracy of transient characteristics by using general RMSE. On the other hand, according to the second embodiment, the accuracy of the transient characteristic can be evaluated by obtaining the emphasis control output and the emphasis model output by the transient response emphasis function unit 132. This point will be described below.

For example, the operation input is input to the control target in a step-like waveform as shown in FIG. 7 (a), and the control output is output from the control target as shown by the solid line in FIG. 7 (b). Consider the case where the model output indicated by the alternate long and short dash line and the dotted line is output from the two models. Note that FIG. 8A shows an enlarged response portion of the rise step of the control output and the model output. Further, FIG. 8B shows an enlarged response of the stationary portion of the control output and the model output.

In this example, there are a model with a fast response (dotted line) and a model with a slow response (dotted line), but as shown in FIG. 8 (a), the accuracy of the transient response is the model with a fast response (dotted line). Is higher.

In the conventional technique, an output error is formed for the entire time shown in FIG. 6, and RMSE is calculated from this error. As described above, since large fluctuations / changes are often avoided in process control, most of the calculation targets of RMSE are in a steady state as shown in FIG. 6B. Therefore, in RMSE, the magnitude of the error in this steady state has a strong influence. This means that the accuracy of the model's DC gain and low frequency characteristics dominates the RMSE rather than the transient response. For example, as shown in FIG. 6B, in the steady state, the slow response model (dotted line) has higher DC gain accuracy than the fast response model (dotted chain line), so the error is smaller. .. Since this difference becomes the difference in RMSE almost as it is, the model with a slow response is evaluated higher in RMSE.

In the second embodiment, in order to suppress the influence of the difference in the steady state on the evaluation index and emphasize the difference in the transient state, a frequency filter such as a high-pass filter or a bandpass filter is used to emphasize the transient response. We are looking for output and emphasized model output. The result of confirming this effect by spectrum analysis is shown in FIG.

In FIG. 9A, errors are calculated from two output time series data for both the fast response model (dotted line) and the slow response model (solid line) without using the transient response enhancement function unit 132. The spectral density is calculated and plotted. On the other hand, in FIG. 9B, the emphasis control output and the emphasis model output are obtained by using a high-pass filter having a ^{cutoff frequency of 0.84 × 10 -3 [Hz] in the transient response enhancement function unit 132, and between them.} The error is calculated and the spectral density is plotted. The error calculated from the data in which the transient response is emphasized by the transient response enhancement function unit 132 reduces the power in the frequency range lower than the cutoff frequency. As a result, the error in the steady state is not reflected so much in the evaluation index, and the error in the transient state is emphasized in the evaluation index.

As described above, in the second embodiment, the transient response enhancement function unit obtains the emphasis control output and the emphasis model output in which the transient response is emphasized by passing the control output and the model output through a filter, and between them. The evaluation index is calculated based on the error. By displaying the models side by side according to this evaluation index, it is possible to easily identify the model having excellent transient characteristic accuracy. As a result, according to the second embodiment, the work load of selecting a model in which the transient characteristics of the control system is improved from among a large number of model candidates by the engineer can be reduced.

[Embodiment 3]
Next, the support device according to the third embodiment of the present invention will be described with reference to FIG. This support device includes a model storage unit 101, an input / output data storage unit 102, an evaluation calculation unit 103, an order determination unit 104, a display control unit 105, a display unit 106, and an operation reception unit 107.

In the third embodiment, the evaluation calculation unit 103 includes a model correction function unit 134, a model output calculation function unit 135, and an evaluation index calculation function unit 136. Other configurations are the same as those of the above-described first and second embodiments.

The model correction function unit 134 generates a plurality of correction models in which the DC gain is corrected to the same value for the plurality of models. The model correction function unit 134 changes (corrects) the DC gains of all the models stored in the model storage unit 101 to the same value to obtain a correction model. This change can be made, for example, by fixing the denominator polynomial coefficients of the transfer function model, dividing all the molecular polynomial coefficients by the DC gain of the model, and then multiplying by the same value. The same value for changing the DC gain is preferably the average value of the DC gains of all models. Alternatively, other statistics such as the median, minimum, and maximum DC gains of all models may be adopted. Further, when an approximate DC gain is assumed from the knowledge of the modeling target, it may be the same value.

The model output calculation function unit 135 obtains the correction model output, which is the output of each of the plurality of correction models, using the operation input stored in the input / output data storage unit 102. The model output calculation function unit 135 obtains the correction model output for all the models stored in the model storage unit 101.

The evaluation index calculation function unit 136 calculates the evaluation index based on the error between the control output and the correction model output. The evaluation index calculation function unit 136 calculates RMSE for the error between the control output and each correction model output as shown in the following equation. The formula is as follows.

In the third embodiment, as described above, the evaluation index is obtained for each model, the order determination unit 104 determines the display order of the plurality of models according to the size of the evaluation index, and the display control unit 105 is determined. A plurality of models are arranged side by side in the order of the above and displayed on the display unit 106.

Next, the operation (support method) of the support device according to the third embodiment will be described with reference to FIG. First, in step S101, a plurality of models prototyped by an engineer by trial and error for use in target model control are stored in the model storage unit 101. Next, in step S102, the input / output data composed of the operation input to the control target and the control output output from the control target by the operation input is stored in the input / output data storage unit 102.

Next, in step S134, the model correction function unit 134 generates a plurality of correction models in which the DC gains are corrected to the same value for the plurality of models (9th step).

Next, in step S135, the model output calculation function unit 135 obtains the correction model output, which is the output of each of the plurality of correction models, using the operation input stored in the input / output data storage unit 102 (10th step). ).

Next, in step S136, the evaluation index calculation function unit 136 calculates the evaluation index based on the error between the control output and the correction model output (11th step).

When the above-mentioned calculation of the evaluation index is performed for all the models stored in the model storage unit 101 (y in step S104), in step S105, the order determination unit 104 determines a plurality of models depending on the size of the evaluation index. Determine the display order of. Next, in step S106, the display control unit 105 causes the display unit 106 to display a plurality of models side by side in the order determined by the order determination unit 104.

Next, the effect in the third embodiment will be described. Table 1 shows the results when the conventional correction is not performed, and Table 2 shows the results according to the third embodiment. In order to show the effect of the corrected RMSE according to this example, it is compared with the normal RMSE. Table 1 shows normal (no correction) RMSE. Table 2 shows the results according to the embodiment. In addition, in order to investigate the contribution of the steady state and the transient state, the evaluation period was divided into five segments, and the integration error in each segment was also shown. This segmentation was done to explain the effect.

The input time series data is a step-like data similar to the example described with reference to FIG. 7, and segments 1, 3 and 4 are in a steady state, and 2 and 5 are in a transient state. More specifically, segment 1 has a constant input value of 0 and rises in segment 2, segments 3 and 4 have a constant input value of 1, and segment 5 has a fall of input value.

In Table 1, the error integration values in the steady state (segments 1, 3, and 4) are different for each model, but in Table 2, they are the same for all models. On the other hand, in the results shown in Table 2, there is a difference in the error integration value between the models in the transient states segments 2 and 5. In this way, the corrected RMSE calculated from the correction model serves as an evaluation index in which the difference in the transient state is emphasized and the difference in the steady state is suppressed.

The above-mentioned effect is obtained by making the DC gain of the model the same. If the DC gains are the same, the ratio of input and output in the steady state will be the same. Therefore, the errors in the steady state of the models are the same, and there is no difference between the models. As a result, the evaluation index according to the third embodiment emphasizes the difference in dynamic characteristics.

As described above, in the third embodiment, the model correction function unit generates a plurality of correction models in which the DC gains are corrected to the same value for the plurality of models, and the evaluation index is based on the error between them. To calculate. By displaying the models side by side according to this evaluation index, it is possible to easily identify the model having excellent transient characteristic accuracy. As a result, according to the third embodiment, the work load of selecting a model in which the transient characteristics of the control system is improved from among a large number of model candidates by the engineer can be reduced. Further, according to the third embodiment, it is possible to carry out the modeling even if the modeling target is unknown and the frequency range to be focused on is unknown.

[Embodiment 4]
Next, the support device according to the fourth embodiment of the present invention will be described with reference to FIG. This support device includes a model storage unit 101, an input / output data storage unit 102, an evaluation calculation unit 103, an order determination unit 104, a display control unit 105, a display unit 106, and an operation reception unit 107.

In the fourth embodiment, the evaluation calculation unit 103 includes a model output calculation function unit 131, a transient state section extraction function unit 137, and an evaluation index calculation function unit 138.

The model output calculation function unit 131 obtains the model output, which is the output of each of the plurality of models, using the operation input. The model output calculation function unit 131 inputs the operation input stored in the input / output data storage unit 102 to each model stored in the model storage unit 101, and performs time-series data by a well-known simulation or the like. Get the model output that is. In addition, in order to evaluate equally, it is preferable that the data applied to each model is the same, and it is common. The model output calculation function unit 131 is the same as that of the second embodiment.

The transient state section extraction function unit 137 extracts a transient response section from each of the control output and the model output to obtain the extraction control output and the extraction model output. The transient state section extraction function unit 137 determines a section in the transient state from the operation input and control output of the time series data stored in the input / output data storage unit 102, and determines the transient state section from each of the control output and the model output. Extract only the time series data of.

There are several possible methods for determining the transient state interval, but as an example, a method using time-frequency analysis of the operation input is shown.

[First step] The operation input is divided into sections of a certain length.

[Second step] The power spectral density of the operation input is calculated for each divided section. It can be calculated by MATLAB (registered trademark) Signal Processing Toolbox, spectrogram function, etc.

[Third step] The power spectral density corresponding to a frequency lower than a predetermined frequency is integrated.

[Fourth step] A section in which the integrated power spectral density exceeds a predetermined threshold value is determined as a steady section.

[Fifth step] Data of a section determined to be a steady section is removed from each of the control output and the model output to obtain the extraction control output and the extraction model output.

Since the predetermined frequency described above determines which frequency is in the low frequency range, it can be determined in the same manner as the cutoff frequency in the second embodiment.

The determination of the transition state section is not limited to the method described above. In addition to this, it is also possible to perform time-frequency analysis by short-time Fourier transform or wavelet transform, calculate the power of low-frequency components, and determine the steady-state interval. Further, it is preferable to determine the transient state section using the operation input rather than the control output.

The evaluation index calculation function unit 138 calculates the evaluation index based on the error between the extraction control output and the extraction model output obtained by the transient state interval extraction function unit 137.

In the fourth embodiment, the evaluation index is obtained for each model as described above, the order determination unit 104 determines the display order of the plurality of models according to the size of the evaluation index, and the display control unit 105 is determined. A plurality of models are arranged side by side in the order of the above and displayed on the display unit 106.

Next, the operation (support method) of the support device according to the fourth embodiment will be described with reference to FIG. First, in step S101, a plurality of models prototyped by an engineer by trial and error for use in target model control are stored in the model storage unit 101. Next, in step S102, the input / output data composed of the operation input to the control target and the control output output from the control target by the operation input is stored in the input / output data storage unit 102.

Next, in step S131, the model output calculation function unit 131 uses the operation input to obtain the model output, which is the output of this model, for the model stored in the model storage unit 101 (sixth step).

Next, in step S137, the transient state section extraction function unit 137 extracts the transient response section from each of the control output and the model output to obtain the extraction control output and the extraction model output (12th step).

Next, in step S138, the evaluation index calculation function unit 138 calculates the evaluation index based on the error between the extraction control output and the extraction model output obtained by the transient state interval extraction function unit 137 (13th step). ).

When the above-mentioned calculation of the evaluation index is performed for all the models stored in the model storage unit 101 (y in step S104), in step S105, the order determination unit 104 determines a plurality of models depending on the size of the evaluation index. Determine the display order of. Next, in step S106, the display control unit 105 causes the display unit 106 to display a plurality of models side by side in the order determined by the order determination unit 104.

As described above, in the fourth embodiment, the transient state section extraction function unit extracts the transient response section from each of the control output and the model output to obtain the extraction control output and the extraction model output, and between them. The evaluation index is calculated based on the error. By arranging and listing the models according to this evaluation index, it is possible to easily identify a model having excellent transient characteristic accuracy. As a result, according to the fourth embodiment, the work load of selecting a model in which the transient characteristics of the control system is improved from among a large number of model candidates by the engineer can be reduced. Further, the fourth embodiment is also characterized in that it can be carried out even if the modeling target is unknown and the frequency range to be focused on is unknown, as in the third embodiment.

Since the purpose of the present invention is to display the transient characteristics with an emphasis on the accuracy, it is preferable to display them in the order of the obtained evaluation indexes. Other display forms are also conceivable. For example, if the evaluation index obtained by the present invention is displayed based on the weighted sum or the sum of ranks of general RMSE, a balanced display of both evaluations becomes possible.

[Embodiment 5]
Next, the support device according to the fifth embodiment of the present invention will be described with reference to FIG. This support device includes a model storage unit 101, an input / output data storage unit 102, an evaluation calculation unit 103, an order determination unit 104, a display control unit 105, a display unit 106, and an operation reception unit 107. These are the same as the above-described embodiments. Further, the support device according to the fifth embodiment includes a feature amount calculation unit, a classification unit, and a classification display control unit.

The model storage unit 101 stores a plurality of models to be controlled. The model storage unit 101 stores the model in a known form such as a transfer function, a differential equation, and a step response. The input / output data storage unit 102 stores input / output data composed of an operation input to the control target and a control output output from the control target by the operation input.

The evaluation calculation unit 103 calculates the evaluation index of each of the plurality of models by reflecting the accuracy of the transient characteristics of the model output in the difference between the model output of each of the plurality of models using the operation input and the control output. do. The order determination unit 104 determines the display order of the plurality of models according to the magnitude of the evaluation index. The display control unit 105 displays a plurality of models side by side on the display unit 106 in the order determined by the order determination unit 104. The operation reception unit 107 receives an operation instruction by an engineer.

The feature amount calculation unit 108 calculates the feature amount corresponding to the set evaluation item for each of the plurality of models stored in the model storage unit 101. The feature quantity is a numerical value that may be used as a judgment material for an engineer to select a similar model. The evaluation item may be, for example, a DC gain. The evaluation items include an overshoot amount, an undershoot amount, a rise time, a settling time, a waste time, a cutoff frequency, a resonance frequency, and the like. The evaluation items of the feature quantity may be appropriately set when selecting the optimum model for the target model control.

The classification unit 109 classifies a plurality of models stored in the model storage unit 101 into a plurality of groups according to the feature amount calculated by the feature amount calculation unit 108, for example, by a clustering method or the like. In the classified group, models with similar steady-state characteristics are collected. The classification display control unit 110 causes the display unit 106 to display a plurality of groups for each group classified by the classification unit 109.

According to the fifth embodiment, the comparison of the steady-state characteristics is performed by the feature amount calculation unit 108, the classification unit 109, and the classification display control unit 110, and the comparison of the transient characteristics is performed by the evaluation calculation unit 103, the order determination unit 104, and the display. The roles can be divided by the control unit 105. The DC gain can be calculated from a mathematical model.

For example, a case where a plurality of models shown in FIG. 15 are obtained and stored in the model storage unit 101 will be described as an example.

The classification display control unit 110 displays the model set for each group divided by the classification unit 109. For example, as shown in FIG. 16, information such as a representative value (average value, etc.) of the DC gain of the models included in the group, a representative value of RMSE, the number of models included in the group, and information that can understand the characteristics of the group, and A representative diagram plotting the average value of the step response of the models included in this group is displayed for each group. Also, provide a means for selecting a group. The order in which the groups are arranged is preferably in ascending order of RMSE or in descending order of the number of models included in the group.

The representative diagram of the model described above is not limited to the average value of the step response, and may be any model as long as it illustrates the characteristics of the model. Further, if the number of models included in the group is not very large, a method of plotting the step response of the models included in the group on one graph is also possible.

Further, when any group is selected from the display states in FIG. 16, the model candidates included in the selected group are individually displayed on the display unit 106 as shown in FIG. This display order is determined according to, for example, the evaluation index in the third embodiment. The evaluation calculation unit 103 in the third embodiment calculates the evaluation index of each model included in the designated group, and the order determination unit 104 determines the evaluation index based on the calculated display order in the group of each model. , The display control unit 105 displays the information of each model in the display unit 106 according to this order. This evaluation index is obtained by generating a correction model in which the DC gains of the models included in each group are the same value, and calculating the correction RMSE with this correction model. When the feature amount calculation unit 108 adopts the DC gain as the feature amount, it is preferable that the same value is the set average of the DC gains of the models included in each group.

By doing so, the classification unit 109 can divide the model into groups based on the steady-state characteristics and evaluate the transient characteristics in the order of accuracy within each group, so that efficient model selection becomes possible.

According to the fifth embodiment, it is most effective when a model having poor accuracy of steady-state characteristics and DC gain is mixed, or when there is a large variation in steady-state characteristics and DC gain between models. According to the third embodiment described above, in the case where the DC gains are roughly matched in all the models and the models with good transient characteristics and the models with poor transient characteristics are mixed, the difference in transient characteristics is emphasized. It is effective. However, since the difference in transient characteristics is emphasized by aligning the DC gains to the same value, the accuracy of the DC gains cannot be compared. Therefore, in the case described above, if only the third embodiment is applied, there is a risk that a model in which the accuracy of the steady-state characteristics is low and the accuracy of the transient characteristics such as the settling time and the rise time is well matched is selected. be.

According to the fifth embodiment, classification is performed by features representing steady-state characteristics, a model set is displayed for each classified group, and the accuracy of transient characteristics is compared within the selected group. Therefore, steady-state characteristics and transient characteristics are compared. It will be easier to select a model with excellent accuracy in both. As a result, it is possible to reduce the risk of erroneously selecting a model with low accuracy of steady-state characteristics.

As shown in FIG. 18, the support device is a computer device including a CPU (Central Processing Unit) 201, a main storage device 202, an external storage device 203, a network connection device 204, and the like. Each of the above-mentioned functions is realized by operating the CPU by the program deployed in the main storage device. The network connection device 204 connects to the network 205. Further, each function may be distributed to a plurality of computer devices.

As described above, according to the present invention, the difference between the model output and the control output of each of the plurality of models using the operation input reflects the accuracy of the transient characteristics of the model output, and each of the plurality of models. Since the evaluation index of is calculated, the accuracy of transient characteristics is superior to that of many prototype models even in situations where the model must be evaluated using time-series data, which is dominated by steady state. Models can be selected efficiently.

The present invention is not limited to the embodiments described above, and many modifications and combinations can be carried out by a person having ordinary knowledge in the art within the technical idea of the present invention. That is clear.

101 ... Model storage unit, 102 ... Input / output data storage unit, 103 ... Evaluation calculation unit, 104 ... Order determination unit, 105 ... Display control unit, 106 ... Display unit, 107 ... Operation reception unit.

Claims (2)

A model storage unit that stores multiple models to be controlled,
An input / output data storage unit that stores input / output data composed of an operation input to the control target and a control output output from the control target by the operation input.
The evaluation index of each of the plurality of models is calculated by reflecting the accuracy of the transient characteristics of the model output in the difference between the model output of each of the plurality of models using the operation input and the control output. The configured evaluation calculation unit and
An order determining unit configured to determine the display order of the plurality of models according to the magnitude of the evaluation index, and
It is provided with a display control unit configured to display the plurality of models side by side on the display unit in the order determined by the order determination unit .
The evaluation calculation unit
A model correction function unit configured to generate a plurality of correction models in which the DC gain is corrected to the same value for the plurality of models.
A model output calculation function unit configured to obtain a correction model output, which is an output of each of the plurality of correction models, using the operation input.
With the evaluation index calculation function unit configured to calculate the evaluation index based on the error between the control output and the correction model output.
Support device according to claim Rukoto equipped with. The first step of storing multiple models to be controlled in the model storage unit,
A second step of storing input / output data composed of an operation input to the control target and a control output output from the control target by the operation input in the input / output data storage unit.
The third is to calculate the evaluation index of each of the plurality of models by reflecting the accuracy of the transient characteristics of the model output in the difference between the model output of each of the plurality of models using the operation input and the control output. Steps and
The fourth step of determining the display order of the plurality of models according to the magnitude of the evaluation index, and
It is provided with a fifth step of arranging the plurality of models side by side in the order determined in the third step and displaying them on the display unit.
The third step is
The ninth step of generating a plurality of correction models in which the DC gains are corrected to the same value for the plurality of models, and
The tenth step of obtaining the correction model output which is the output of each of the plurality of correction models using the operation input, and
The eleventh step of calculating the evaluation index based on the error between the control output and the correction model output.
Support wherein the Rukoto equipped with.

Images