How to evaluate predictive data, design systems, and select control system design methods

Abstract

The predicted data of the control object that performs the work is evaluated. [Solution] The method for evaluating predicted data of the present invention involves giving an external target to a controlled object, collecting the actual input uo(t) and output yo(t) of the controlled object in step S10 to create input-output data, generating predicted data regarding the input and output of the unknown controlled object based on the input-output data in step S20, calculating the variance of the predicted data, and determining the degree of linearity or nonlinearity of the predicted data based on the variance in step S30.

Description

[Technical Field]

[0001] The present invention relates to prediction of a controlled object. [Background technology]

[0002] A workplace management support system (Patent Publication No. 2020-149452) is known that, in a workplace where workers and machines are present, obtains environmental factor information from environmental sensors installed in the workplace and internal physical condition information from human sensors installed on the workers, associates and stores the internal physical condition information and environmental factor information for each worker, compares this stored information with databased information, determines the internal physical condition of each worker, determines the content of an alert, and issues an alert to the worker. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-149452 Summary of the Invention [Problem to be solved by the invention]

[0004] However, if different or unknown tasks could be treated as control targets and predictions for those tasks could be made based on those control targets, it would be possible to obtain a consistent outlook for those tasks and further improve the efficiency of the entire work. The technology described in Patent Document 1 determines only the ongoing status of known tasks, and does not make predictions for different or unknown tasks.

[0005] The inventor has conducted extensive research into treating different or unknown tasks in a workplace as control objects. Construction machinery such as hydraulic excavators used in workplaces, as well as other work machines that perform some kind of work, work to achieve a certain work goal, and can therefore be considered control objects with a work goal r, a corresponding work output y, and a corresponding input u. Furthermore, because the operators of these work machines operate the work machines to achieve a certain work goal, they can also be considered control objects with a control goal r, a corresponding control output y, and a corresponding input u. In other words, anything that has a target r and an output y, such as a work machine, an operator, or even a worker performing manual work without a machine, can be considered a control object G with an input u and an output y. If the input u and output y of the unknown control object G can be predicted, an optimal work plan can be created, shortening work time and improving work efficiency through labor, energy, and manpower savings.

[0006] Regarding the improvement of the control performance of an unknown control target G, for example, for an input u and an output y that are repeatedly added and subtracted during the running of a known PID control target, initial actual output data is collected, and the unknown input JPEG2026042689000002.jpg87 and output It is possible to predict JPEG2026042689000003.jpg107 (hereinafter also referred to as generation of predicted input / output data or generation of predicted data).

[0007] Specifically, for example, attempts have been made to express the coefficient of a controller in a PID controlled object, such as the inverse of the coefficient C*r of a feedforward controller, using the number of data N. For example, when the controlled object is a pilot, the inverse of C*r may be linear in some cases (e.g., the pilot's operating time), or nonlinear in other cases (e.g., the time required for a skilled pilot to complete a task). Linear values ​​are reliable when generating prediction data, but nonlinear values ​​are questionable when generating prediction data.

[0008] The inventors have focused on the reciprocal of the coefficients of the FIR controller in the controlled object and have concluded that highly accurate prediction data can be obtained by determining whether the reciprocal of the coefficients of the FIR controller is closer to a linear system or a nonlinear system. In other words, they have completed a technology for visualizing the degree of linearity or nonlinearity of prediction data for an unknown controlled object (e.g., a pilot). [Means for solving the problem]

[0009] To this end, the method for evaluating predicted data according to the present invention provides an external target to a controlled object, collects actual inputs and outputs of the controlled object to create input-output data, generates predicted data regarding the inputs and outputs of the unknown controlled object based on the input-output data, calculates the variance of the predicted data, and determines the degree of linearity or nonlinearity of the above-mentioned predicted data based on the variance.

[0010] According to the present invention, the degree of nonlinearity can be calculated using only one parameter, the variance of the predicted input. By easily grasping the characteristics of the controlled object (the strength of nonlinearity), it becomes easier to detect changes in the control system and to reconstruct (improve) the control system.

[0011] A high degree of linearity indicates that the data is not random, and thus can be evaluated as having high accuracy. Conversely, a high degree of nonlinearity indicates that the data is completely random or close to random, and thus can be evaluated as having low accuracy. Therefore, in one aspect of the present invention, it is determined whether the variance is less than a predetermined threshold, and the accuracy of the predicted data is determined based on this determination result. According to this aspect, it is possible to distinguish whether the predicted input-output data is linear or nonlinear using a threshold specified by the user, and the prediction accuracy of the predicted input-output data can be evaluated depending on whether the system is a linear system or a nonlinear system.

[0012] The present invention is not limited to a method for evaluating predicted data. The present invention may also be a system for evaluating predicted data, including an initial database unit that provides a target from the outside to a controlled object, collects actual inputs and outputs of the controlled object, and creates input-output data, a predicted data generation unit that generates predicted data regarding the inputs and outputs of an unknown controlled object based on the input-output data, a nonlinearity determination unit that calculates the variance of the predicted data and determines whether the variance is less than a predetermined threshold, i.e., determines the degree of linearity or nonlinearity of the predicted data based on the variance, and an accuracy determination unit that determines the accuracy of the predicted data based on the determination result.

[0013] The design system according to the present invention also includes a linearity determination unit that assigns an external target to a controlled object, collects actual inputs and outputs of the controlled object, and creates input-output data. Based on the input-output data, the system generates predicted data regarding the unknown controlled object's inputs and outputs. The system calculates the variance of the predicted data, determines whether the variance is less than a predetermined threshold, and determines whether the predicted data is a linear or nonlinear system based on the determination result. The system also includes a design method selection unit that selects a control system design method for the controlled object based on the linear or nonlinear system determination result. According to this invention, by distinguishing whether the predicted input-output data is a linear system or a nonlinear system, the prediction accuracy of the predicted input-output data can be evaluated. If the prediction accuracy is high, the predicted input-output data can be utilized in, for example, a "database-driven control method." This system can also handle time-series changes in the characteristics of the controlled object. Note that determining whether the predicted data is a linear or nonlinear system should be understood to mean determining whether the predicted data is close to a linear system or a nonlinear system. Selecting a control system design method for the controlled object involves various determination methods, such as preparing a group of various control system design methods in advance and selecting an appropriate method from the group.

[0014] In one aspect of the present invention, the design system is a design system for a database-driven control method, and the design method selection unit determines whether or not to store prediction data in the database for an unknown controlled object based on the judgment result of the linearity judgment unit. According to this aspect, prediction data with a high degree of nonlinearity is not stored in the database, thereby increasing the accuracy of the database.

[0015] In one aspect of the present invention, the linearity determination unit calculates the degree of nonlinearity of the predicted data, and the design method selection unit determines the control parameters of the unknown controlled object based on the calculated degree of nonlinearity. According to this aspect, system identification of the controlled object is improved.

[0016] Furthermore, the method for selecting a control system design method according to the present invention involves giving an external target to the controlled object, collecting actual inputs and outputs of the controlled object, creating input-output data, generating predicted data regarding the inputs and outputs of the unknown controlled object based on the input-output data, calculating system parameters of the unknown controlled object from the predicted data, determining whether the predicted data is a linear system or a nonlinear system based on the variance of the calculated system parameters or the distance from the calculated system parameters to the system parameters of the controlled object, and selecting a control system design method for the unknown controlled object depending on the determination result of whether it is a linear system or a nonlinear system.

[0017] According to this invention, by distinguishing whether the predicted input-output data is a linear system or a nonlinear system, the prediction accuracy of the predicted input-output data can be evaluated, and if the prediction accuracy is high, it is possible to utilize the predicted input-output data in, for example, a "database-driven control method." It is also possible to deal with cases where the characteristics of the controlled object change over time. [Effects of the Invention]

[0018] Thus, according to the present invention, based on the input / output data of a machine or person (unknown controlled object) working in a workplace, it is possible to evaluate the accuracy of predicted data for another machine or person (unknown controlled object) that is about to perform or is currently performing a similar task. Evaluating the accuracy of predicted data reduces the risk of control in an inappropriate state, and is expected to improve control performance. Furthermore, if the degree of linearity of the predicted data increases, it becomes possible to take appropriate measures, such as replacing the hardware components of the unknown controlled object or improving the parameters of the unknown controlled object. Alternatively, it becomes possible to select (determine) a control system design method appropriate for the unknown controlled object. This contributes to improving work and work efficiency. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a block diagram showing a controlled object according to an embodiment of the present invention. [Figure 2] FIG. 2 is a graph showing actual input / output data (database) according to the first embodiment of the present invention. [Figure 3A] 10 is a flowchart showing a procedure for calculating predicted data and its variance from input / output data in the first embodiment and determining nonlinearity of the predicted data. [Figure 3B] FIG. 3 is a graph showing the procedure for determining the degree of nonlinearity from the actual input / output data (database) on the Nth day in FIG. 2. [Figure 4] FIG. 10 is a graph showing actual input / output data and predicted data (database) according to a second embodiment of the present invention. [Figure 5A] 10 is a flowchart showing a procedure for calculating predicted data and its variance from input / output data in a second embodiment, and determining control parameters for the predicted data. [Figure 5B] FIG. 5B is an explanatory diagram illustrating the steps in FIG. 5A. [Figure 6] FIG. 5B is an explanatory diagram illustrating the steps in FIG. 5A. [Figure 7]10A and 10B are a time chart and a block diagram showing input and output data (database) according to a third embodiment of the present invention. [Figure 8A] 10 is a flowchart showing a procedure for calculating predicted data and its variance from input / output data in a third embodiment and utilizing database-driven control. [Figure 8B] FIG. 8B is an explanatory diagram illustrating the steps in FIG. 8A. [Figure 9] FIG. 8B is an explanatory diagram illustrating the steps in FIG. 8A. [Figure 10] FIG. 8B is an explanatory diagram illustrating the steps in FIG. 8A. [Figure 11] FIG. 10 is a graph illustrating an example of initial data and predicted data of a control target, and system parameters calculated from both data. [Figure 12] FIG. 10 is a graph showing initial data, predicted data, and a graph for determining the strength of nonlinearity of the predicted data. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing a controlled object according to one embodiment of the present invention. The controlled object in FIG. 1 can be applied to various tasks. A motor system is used as the first example, and the specific contents of the block diagram are as follows. The block diagram is stored in a computer that constitutes the evaluation system of the present invention. First, a reference model corresponding to an unknown controlled object (motor system) is prepared. The reference model is determined appropriately in consideration of the characteristics of the controlled object, or is appropriately selected from a group of models prepared in advance. FIG. 2 is a graph showing actual input / output data (database) according to the first embodiment. This database is stored in a computer constituting the evaluation system of the present invention. For example, every day, an unknown controlled object (motor system) measures the actual motor input voltage u(t) (hereinafter also referred to simply as input) and the actual rotation speed y(t) (hereinafter also referred to simply as output) output by the motor, with the time from the start time to the end time being used as a variable t, and data is collected. Next, using this actual input / output data, a feedforward controller and a feedback controller for realizing the desired output characteristics are designed, and the reference model and the controlled object are connected via the designed controller. A response prediction method called ERIT (Estimated Response Iterative Tuning), which can also predict inputs generated by JPEG2026042689000005.jpg87 (hereinafter referred to as predictive input, JPEG2026042689000006.jpg914) and the motor's output rotation speed JPEG2026042689000007.jpg118 (hereinafter referred to as predicted output, The resulting image (also called JPEG2026042689000008.jpg1014) is predicted, and the degree of linearity or nonlinearity of the relationship between the predicted input and the predicted output is calculated.

[0021] As a concrete example, the input and output data u and y on the first day are used to predict the input data. Generate JPEG2026042689000009.jpg87 and Distribution of predicted input data using JPEG2026042689000010.jpg87 Similarly, the predicted input data u and y are calculated using the data u and y on the second day. Variance of JPEG2026042689000012.jpg87 JPEG2026042689000013.jpg99 is calculated. Similarly, the predicted input data is calculated using the daily data u and y up to the Nth day. Variance of JPEG2026042689000014.jpg87 JPEG2026042689000015.jpg99 is calculated. FIG. 3B shows in detail the procedure for calculating the degree of nonlinearity for the Nth day as a representative example.

[0022] If there is no change in the characteristics of the motor system, the variance σ on the first and second days will be 2 On the other hand, if the nonlinearity of the motor system is large, the variance σ on the first day 2 The variance of the second day is σ 2 By utilizing this feature, if the speed of change is particularly strong, it is determined that the nonlinearity will become stronger in the future, and the manager of the motor system is notified immediately.

[0023] Possible reasons for the increase in nonlinearity include deterioration of the motor system over time due to daily operation, and changes in the motor's characteristics due to failures, etc. Generally, on the first day of product delivery in Figure 2, linearity is strong, so the variance is 0 as shown in the bottom of Figure 2, and the degree of nonlinearity is low. Next, on the second day in Figure 2, linearity is also strong, so the variance is 0.05 as shown in the bottom of Figure 2, and although the degree of nonlinearity is low, it is higher than on the first day. In other words, the speed of change can be said to be nearly zero. If the speed of change is zero or nearly zero, or is weak, the motor system manager will not be notified.

[0024] dispersion The magnitude of JPEG2026042689000016.jpg99 can be determined using a threshold value α, as shown in the lower part of Figure 2. Variance of JPEG2026042689000017.jpg87 If JPEG2026042689000018.jpg99 exceeds the threshold α, it is determined that the nonlinearity is high.

[0025] As shown in the lower part of Figure 2, on the Nth day, the characteristics have changed significantly compared to the first day in Figure 2, so a notification is sent to the administrator.

[0026] Thereafter, for example, an administrator may identify the cause of the system change and replace the hardware components of the motor system related to the prediction data, or adjust the controller, which is software.

[0027] FIG. 3A is a flowchart showing the procedure by which the evaluation system of the present invention obtains the variance of predicted input data of an unknown control object from the actual input / output data related to the control object (block diagram shown in FIG. 1), judges the degree of nonlinearity, and determines whether to notify the administrator, and corresponds to the motor system of the first embodiment described above. First, in step S10, actual input / output data is acquired. For example, in the motor system of the first embodiment, it is assumed that actual input / output data (also referred to as initial data in FIG. 3) for any day between the first and Nth days has been obtained. Next, the process proceeds to step S20. In step S20, predicted data for the unknown control object for the same day is generated based on the initial data, and the variance of the predicted data is calculated. JPEG2026042689000019.jpg99 is calculated. Next, the process proceeds to step S30. In step S30, the calculated predicted input Variance of JPEG2026042689000020.jpg87 It is determined whether JPEG2026042689000021.jpg99 is below threshold α. In other words, it is assumed that linearity was exhibited at the time of product delivery (day 1 in Figure 2), and it is determined whether the system characteristics have changed. If it is below threshold α (Y), it is determined that the degree of linearity is high. In this case, since there has been no significant change in the characteristics from day 1 (day 1 in Figure 2), no particular notification is given to the administrator.

[0028] On the other hand, predictive input Variance of JPEG2026042689000022.jpg87 If JPEG2026042689000023.jpg99 exceeds the threshold α (N), it is determined that the degree of nonlinearity is high. In this case, since the characteristics have changed significantly since the first day (Day 1 in Figure 2), the administrator is notified. After that, for example, the administrator can identify the cause of the system change and replace hardware components in the motor system or adjust various software controllers (Figure 1).

[0029] According to this embodiment, by evaluating the degree of linearity or nonlinearity daily, it is possible to determine on what day the nonlinearity became stronger. By understanding the nonlinearity on a daily basis, for example, if the speed of change is very fast, it can be determined that the nonlinearity will become stronger in the future and notify an administrator early on. If the nonlinearity intensity becomes stronger, it is determined that the characteristics of the initial system have changed, and the administrator can be notified. At the administrator's discretion, the hardware components of the motor system can be replaced, or the control parameters of the motor system control unit can be adjusted to improve control performance.

[0030] Next, a second embodiment of the present invention will be described. The evaluation target of the second embodiment is the predicted data of the amount of tasks related to workers (people) and the accuracy of this predicted data. With reference to the block diagram shown in Figure 1, the specific contents of the second embodiment are as follows. JPEG2026042689000024.jpg145148

[0031] FIG. 4 is a graph showing input / output data (database) according to a second embodiment of the present invention, specifically showing the concentration level y(t), which is the actual output data for worker A, and the task amount u(t), which is the calculated input data for the worker. The same applies to workers B and X. t is time (variable). Referring to Figure 4, the concentration level y(t) of a worker can be collected by, for example, monitoring the movement of the worker's pupils with a sensor, assuming that the less the pupil movement, the higher the concentration level. The collected data is used to create actual output data (concentration level y(t)), and control parameters are determined to calculate the task amount u(t) appropriate for each worker so that each worker's concentration level y(t) matches the desired concentration level yref(t). The control parameters are components of the blocks in Figure 1. The task amount u(t) of worker A and the task amount u(t) of worker B are different inputs.

[0032] As shown in Figure 4, suppose that the task amount u(t) for persons A and B is already obtained as data such that the concentration level y(t) follows the desired concentration level yref(t). In this case, since persons A and B have different psychological characteristics (the parameters that configure the control object shown in the block diagram of Figure 1 are different), the task amount u(t) also differs.

[0033] Next, suppose new data for person X is obtained, as shown in Figure 4. Because person X is a new person, the desired concentration level yref(t) cannot be achieved. In this case, if person X is similar to either person A or B (in terms of their psychological characteristics), the control parameters of the similar person can be adopted.

[0034] 5 is a flowchart showing the evaluation performed by the second embodiment. First, in step S60, predicted input data is calculated using the initial actual output data (degree of concentration y(t)) of person X in FIG. JPEG2026042689000025.jpg913 is calculated, and the process proceeds to the next step S70. Variance of JPEG2026042689000026.jpg913 JPEG2026042689000027.jpg99 is generated, and the process proceeds to the next step S80. In step S80, the variance of the predicted input calculated in step S70 is calculated. It is determined whether JPEG2026042689000028.jpg99 is below the threshold value α. If it is above the threshold value α (N), the process proceeds to step S90. In step S90, the calculated predicted input The accuracy of JPEG2026042689000029.jpg913 is determined to be low. Using JPEG2026042689000030.jpg913, it is not possible to determine whether person X is similar to person A or person B. Therefore, in step S90, it is determined that the psychological characteristics of person X are different from those of persons A and B, and control parameters unique to person X are calculated. Here, the control parameter calculation method may use a model-based control approach or a data-driven control approach.

[0035] On the other hand, in step S80, the distribution of the predicted input JPEG2026042689000031.jpg99 is the threshold If it is lower than JPEG2026042689000032.jpg87 (Y), proceed to the next step S100 and use the calculated predicted input JPEG2026042689000033.jpg812 is judged to be highly accurate. Using JPEG2026042689000034.jpg812, the task volume of person A and person B Based on the degree of similarity with JPEG2026042689000035.jpg913, it is determined whether the image resembles person A or person B. JPEG2026042689000036.jpg813 and For example, the degree of coincidence of JPEG2026042689000037.jpg912 may be determined by determining whether the time integral of the absolute value of the difference between the two or the correlation coefficient is equal to or less than a threshold value. JPEG2026042689000038.jpg1113 and input JPEG2026042689000039.jpg913 is compared with the concentration level y(t) and task amount u(t), which are the input / output data of person A, and the concentration level y(t) and task amount u(t), which are the input / output data of person B. If it is determined that the predicted data of person X is similar to the input / output data of person A, the process proceeds to the next step S110, where the control parameters applied to person A are adopted for person X. Alternatively, if it is determined in step S100 that the predicted data of person X is similar to the input / output data of person B, the process proceeds to the next step S120, where the control parameters applied to person B are adopted for person X.

[0036] 5B is a diagram showing the processing of steps S70 and S100 in FIG. 5A together with data. The variance generated in step S70 and shown in the lower right map of FIG. 5B is Regarding the comparison of JPEG2026042689000040.jpg99 and the threshold value α, The degree of nonlinearity is determined based on JPEG2026042689000041.jpg99.

[0037] 6 is a diagram showing the processing of steps S60 to S80 in FIG. 5A together with data. JPEG2026042689000042.jpg1013 and predictive input JPEG2026042689000043.jpg913 is compared with the input output data in step S100.

[0038] According to the second embodiment described above, the psychological characteristics of unknown person X can be obtained with a fairly high degree of accuracy.

[0039] Next, a third embodiment of the present invention will be described. The third embodiment is a modification of the motor system according to the first embodiment described above, and is the same as the description of the block diagram in FIG. 1 described above. The third embodiment assumes a case in which the motor load (motor rotation speed y(t)) changes during motor operation. FIG. 7 is a block diagram of the third embodiment and a graph showing actual input / output data divided into cases 1 to 3.

[0040] When the motor load changes during the operation time t, the control parameters (PID control parameters), which are components of the block diagram in Figure 1, are adjusted appropriately. JPEG2026042689000044.jpg1013 and When JPEG2026042689000045.jpg913 is not generated, the PID gain (control parameter) does not change before and after the load change, as shown in the time chart for Case 1 in the left column in FIG.

[0041] It is predictive data JPEG2026042689000046.jpg1013 and The case where JPEG2026042689000047.jpg913 is generated is shown in the time chart for Case 2 in FIG. 7 and the time chart for Case 1 in the right column in FIG.

[0042] Generated The variance of JPEG2026042689000048.jpg913 is calculated, and if the variance is less than a threshold, it is determined that the motor load characteristics are linear (Case 2), and if the variance is greater than or equal to the threshold, it is determined that the motor load characteristics are nonlinear (Case 3).

[0043] If the motor load characteristics are linear, good control results can be obtained by utilizing input / output data predicted in advance, as in Case 2.

[0044] If the motor load characteristics are nonlinear, the accuracy of the predicted input-output data will be poor, as in Case 3. In other words, the predicted input-output data cannot be used. Therefore, we aim to improve control performance by using a technology that does not use the predicted input-output data.

[0045] FIG. 8A is a time chart showing the evaluation of the third embodiment. In the motor system, first, in step S130, the actual motor input voltage uo(t) and the actual rotation speed yo(t) output by the motor are acquired as initial data. Such data is the motor input voltage u(t) and output rotation speed y(t) from the start of work until the load changes in case 1 in FIG. 7. At this time, it is assumed that the motor load (output rotation speed y(t)) changes at a certain time. And it is assumed that the motor rotation speed y(t) is oscillating. Since the motor load is changing, it is desirable to adjust the PID gains in accordance with this change. Therefore, in the next step S140, the initial data of case 1 in FIG. 7 is used to calculate predicted input / output data, and the variance of the predicted input is calculated. JPEG2026042689000049.jpg109 is calculated, and the process proceeds to the next step S150.

[0046] In step S150, the predicted input calculated in step S140 is Variance of JPEG2026042689000050.jpg913 Determine whether JPEG2026042689000051.jpg99 is below a threshold α. If it is above the threshold (N), The accuracy of JPEG2026042689000052.jpg913 is judged to be low. JPEG2026042689000053.jpg913 Because the accuracy is low, proceed to the next step S170 and perform predicted input. JPEG2026042689000054.jpg913 is not stored or utilized, and the process proceeds to the next step S180. In step S180, for example, using a database-driven control method shown in the block diagram on the right side of FIG. 7, only the initial data (the actual motor input voltage uo(t) and the actual motor output rotation speed yo(t) in Case 1 of FIG. 7) is stored in a database, and the motor system is controlled using the database-driven control method (Case 3 of FIG. 7). The database-driven control method is as described in, for example, Japanese Patent No. 4274046. As a result, better control results can be obtained in Case 3 than in Case 1, in which no prediction data is generated.

[0047] Returning to step S150 described above, the distribution of the predicted input If JPEG2026042689000055.jpg99 is below the threshold α (Y), the generated predicted input It is determined that the accuracy of JPEG2026042689000056.jpg913 is high, and the process proceeds to the next step S160. JPEG2026042689000057.jpg913 is stored in the database, and the process proceeds to the next step S180. In step S180, the motor system is controlled using the database-driven control method shown on the right of Figure 7 and described in Japanese Patent No. 4274046 (Case 2 in Figure 7). As a result, better control results can be obtained in Case 2 than in Case 1, in which no prediction data is generated.

[0048] FIG. 8B is a graph illustrating the processing of steps S130 to S150 in the flowchart of FIG. 8A described above.

[0049] FIG. 9 is a graph and block diagram illustrating the processing of steps S140 and S180 of the flowchart of FIG. 8A described above.

[0050] FIG. 10 is a graph and a block diagram illustrating the process of Case 2 of the time chart of FIG. 7 and step S180 of the flowchart of FIG. 8, which is the best mode of the third embodiment.

[0051] According to the method for evaluating predicted data shown in FIG. 3A, as shown in FIG. 1, a target r(t) is externally applied to the control object related block diagram, and the input u(t) and output y(t) of the control object are monitored. In step S10, the actual input uo(t) and output yo(t) of the control object are collected to create input-output data. In step S20, the input of the unknown control object is calculated based on the input-output data. Generate JPEG2026042689000058.jpg87 (prediction data for the input) and calculate the variance of the prediction data. JPEG2026042689000059.jpg99 is calculated and distributed in step S30. The degree of linearity or nonlinearity of the predicted data is determined based on JPEG2026042689000060.jpg99. If JPEG2026042689000061.jpg99 is smaller than the threshold α, it is determined that the degree of linearity is stronger than the degree of nonlinearity, and the accuracy of the predicted data is evaluated as high. If JPEG2026042689000062.jpg99 is equal to or greater than α, the degree of nonlinearity is determined to be stronger than the degree of linearity, and the accuracy of the predicted data is evaluated as low. This allows for the creation of a highly accurate overall plan, contributing to improved work efficiency.

[0052] 1 to 10, the evaluation system of this embodiment includes a generation unit (steps S10 and S20) that generates prediction data for judging accuracy and calculates the variance of the prediction data, a linearity judgment unit (step S30) that judges whether the variance is less than a predetermined threshold value α and judges the degree of linearity or nonlinearity of the prediction data based on the judgment result, and a design method selection unit (step S40) that selects (determines) a control system design method for the control object by notifying the administrator based on the judged degree of linearity or nonlinearity, such as by prompting the administrator to replace parts of the control object or to adjust control parameters (PID gains, etc.) or variable parameters (system parameters, etc.) of the control object. This contributes to improving work efficiency.

[0053] Furthermore, according to this embodiment, in the evaluation system for the database-driven design method shown on the right side of FIG. As shown in FIG. 8A, in step S150, If it is determined that JPEG2026042689000063.jpg99 is less than the threshold value α (Y), that is, the predicted input of the predicted data If JPEG2026042689000064.jpg813 is determined to be linear, the predicted input JPEG2026042689000065.jpg813 is stored in the database of the database-driven design method.

[0054] Conversely, in step S150, If it is determined that JPEG2026042689000066.jpg99 is equal to or greater than the threshold value α (N), that is, the predicted input of the predicted data If JPEG2026042689000067.jpg913 is determined to be nonlinear, the predicted input JPEG2026042689000068.jpg913 is not stored in the database of the database-driven design method.

[0055] This allows predictive input The prediction accuracy of JPEG2026042689000069.jpg913 can be improved.

[0056] The left side of Fig. 11 is a graph illustrating the initial data and the subsequent predicted data for the controlled object, and the right side of Fig. 11 is a graph illustrating the system parameters of the controlled object calculated from the initial data and the predicted data. First, the input uo(t) and output yo(t) are actually measured as the initial data for the controlled object. Next, the predicted output uo(t) is calculated from the controlled object to become the predicted data. JPEG2026042689000070.jpg1013 and predictive input JPEG2026042689000071.jpg913 is generated. Based on the input uo(t) and output yo(t) that serve as initial data, the system parameters of the controlled object, specifically the time constant To and the system gain Ko, are calculated. Similarly, the predicted output JPEG2026042689000072.jpg1013 and predictive input JPEG2026042689000073.jpg913 based on the system parameters to be controlled, specifically the time constant JPEG2026042689000074.jpg97 and system gain JPEG2026042689000075.jpg97 is calculated.

[0057] Referring to Example 1 in the upper left of Figure 11, it can be seen that there are few sudden increases or decreases in the initial data (input uo(t) and output yo(t)...vertical axis) over time (horizontal axis), so it is likely that this is close to a linear system. Also, referring to Example 2 in the lower left of Figure 11, it can be seen that there are sudden increases or decreases in the initial data (input uo(t) and output yo(t)...vertical axis) over time (horizontal axis), so it is likely that this is close to a linear system.

[0058] Next, referring to Example 1 in the upper left of Figure 11, the predicted data (vertical axis) over time (horizontal axis) is as follows: It can be seen that there are few sudden increases or decreases in JPEG2026042689000076.jpg1013, It is clear that JPEG2026042689000077.jpg913 has sudden increases and decreases, so it is likely close to a nonlinear system. In contrast, referring to Example 2 in the lower left of Figure 11, the predicted data ( JPEG2026042689000078.jpg1013 and JPEG2026042689000079.jpg913···vertical axis), it can be seen that there are few sudden increases or decreases on either side, so it is probably close to a linear system.

[0059] In Examples 1 and 2 on the right side of Figure 11, the horizontal axis represents the time constant T and the vertical axis represents the system gain K. Such a coordinate system can show the distribution (or variation) of the system parameters. In Example 1, the system parameters (time constant To and system gain Ko) of the controlled object based on the initial data (input uo(t) and output yo(t)) are shown by black circles. Also, the predicted data (predicted output JPEG2026042689000080.jpg1013 and predictive input The system parameters (time constant) of the control target based on JPEG2026042689000081.jpg913 JPEG2026042689000082.jpg97 and system gain JPEG2026042689000083.jpg107) is indicated by a star. In the T, K coordinate system of Example 1, the black circles and stars do not match, and the distance is large, so it is determined that the predicted data has large variations and is highly nonlinear.

[0060] In contrast, in the T,K coordinate system of Example 2, the black circles and stars coincide, so it is determined that there is no variance in the predicted data and no nonlinearity.Alternatively, although not shown as an example, even if the black circles and stars are close to each other in the T,K coordinate system, the distance is small, so it is determined that there is little variance in the predicted data and weak nonlinearity.

[0061] 12 is a graph showing the initial data, the predicted data, and the strength of the nonlinearity of the predicted data. The degree of nonlinearity increases as the distance from the system parameters of the initial data to the system parameters of the predicted data increases, or as the variance of the system parameters of the predicted data increases with respect to the system parameters of the initial data.

[0062] Therefore, as shown on the right side of Figure 12, the horizontal axis represents the degree of nonlinearity, and the vertical axis represents d TK are the distances and variances of the system parameters described above. A threshold value α is set on the vertical axis, and if the distances or variances of the system parameters are less than the threshold value α, it is determined that the nonlinearity is weak. Conversely, if they are equal to or greater than the threshold value α, it is determined that the nonlinearity is strong. Such a determination result is used in the control system design of the controlled object, such as whether to store it in a database, as in (Y) or (N) in step S150 of FIG. 8B.

[0063] Next, an embodiment will be described in which a control parameter of a controlled object is calculated depending on the degree of nonlinearity.

[0064] The degree of nonlinearity w is JPEG2026042689000084.jpg2040 or This can be calculated using JPEG2026042689000085.jpg1942. where: JPEG2026042689000086.jpg1427 and JPEG2026042689000087.jpg1328 is the maximum value set by the user of the design system as appropriate, and can be changed as appropriate at the user's discretion.

[0065] Next, the degree of nonlinearity w is used to calculate the control parameter K in the PID control shown in Fig. 7. P (t), K I (t), K D (t) (hereinafter simply referred to as K(t)).

[0066] Specifically, the control parameter based only on the initial data is defined as K_ini(t). K_ini(t)=[K Pini , K. Iini , K. Dini ] The control parameter based only on the predicted data is defined as K_cal(t) (hereinafter also referred to as the control parameter after learning). The control parameter K(t) taking into account the degree of nonlinearity w is given by K(t)=w K_ini(t)+(1-w) K_cal(t) is. As described above, according to the design system of the embodiment, the degree of nonlinearity w of the predicted data is calculated, and the control parameter K(t) of the controlled object is determined based on the calculated degree of nonlinearity w, thereby improving system identification of the controlled object.

[0067] In addition, in the control system design method and design system according to this embodiment, in step S130 of FIG. 8A, a target is given to the control object from the outside, the actual inputs and outputs of the control object are collected, and input-output data (input uo(t) and output yo(t)) is created as initial data shown on the left side of FIG. 11. Based on this initial data, in step S140, predicted data (predicted output JPEG2026042689000088.jpg1013 and predictive input JPEG2026042689000089.jpg813) and the system parameters of the controlled object based on the predicted data (time constants shown in FIG. 11 JPEG2026042689000090.jpg97 and system gain JPEG2026042689000091.jpg97) is calculated. The system further comprises a linearity determination unit that determines whether the predicted data is a linear system or a nonlinear system based on the variance of the calculated system parameters or the distance from the calculated system parameters to the system parameters of the controlled object (time constant To and system gain Ko calculated from (input uo(t) and output yo(t))), and a design method selection unit that selects (determines) a control system design method for the controlled object in accordance with the determination result of the linearity determination unit as to whether the system is a linear system or a nonlinear system.

[0068] Although the embodiments of the present invention have been described above with reference to the drawings, the present invention is not limited to the illustrated embodiments. Various modifications and variations can be made to the illustrated embodiments within the same scope as the present invention or within an equivalent scope. For example, some components may be extracted from one embodiment described above, and other components may be extracted from another embodiment described above, and these extracted components may be combined. [Industrial Applicability]

[0069] The present invention is advantageously used in work planning.

Claims

1. giving a target to a controlled object from the outside, collecting actual inputs and outputs of the controlled object, and creating input-output data; generating prediction data regarding inputs and outputs of a controlled object based on the input / output data; calculating the variance of the predicted data; A method for evaluating predicted data, which determines the degree of linearity or nonlinearity of the predicted data based on the variance.

2. determining whether the variance is less than a predetermined threshold; The method for evaluating predicted data according to claim 1 , further comprising determining accuracy of the predicted data based on a result of the determination.

3. giving a target to a controlled object from the outside, collecting actual inputs and outputs of the controlled object, and creating input-output data; generating prediction data regarding inputs and outputs of a controlled object based on the input / output data; calculating the variance of the predicted data; a linearity determination unit that determines whether the variance is less than a predetermined threshold and determines whether the prediction data is a linear system or a nonlinear system based on the determination result; a design method selection unit that selects a control system design method for the controlled object in accordance with the determination result of whether the system is a linear system or a nonlinear system determined by the linearity determination unit.

4. A database-driven control method design system, comprising: The design system according to claim 3 , wherein the design method selection unit determines whether or not to store the prediction data in the database of the controlled object based on the determination result.

5. the linearity determination unit calculates a degree of nonlinearity of the prediction data; The design system according to claim 3 , wherein the design method selection unit determines a control parameter of the controlled object based on the calculated degree of nonlinearity.

6. giving a target to a controlled object from the outside, collecting actual inputs and outputs of the controlled object, and creating input-output data; generating prediction data regarding inputs and outputs of a controlled object based on the input / output data; calculating a system parameter of the controlled object from the predicted data, and determining whether the predicted data is a linear system or a nonlinear system based on a variance of the calculated system parameter or a distance from the calculated system parameter to the system parameter of the controlled object; A control system design method for selecting a control system design method for the controlled object according to the determination result of whether the system is a linear system or a nonlinear system.

Images