JP5000460B2 - Parameter identification apparatus and parameter identification program - Google Patents

Description

  The present invention relates to parameter identification in which a plurality of parameters in a physical model is evaluated using an evaluation function, and asymptotically approximated to a true value by repeated correction, to identify the plurality of parameters.

  A method for identifying plant control parameters and the like by simulation is known.

  For example, Patent Document 1 shows that parameters (reaction rate, heat transfer coefficient, etc.) that change due to changes in operating conditions are identified in plant control in the petroleum and chemical industries. Here, if the identification parameter sizes are not uniform among the parameters, the identification accuracy deteriorates. For this reason, in Patent Document 1, a coefficient called a scaling factor is introduced in a linear equation model of a plant so that the contribution of each identification parameter to the evaluation value is approximately the same for each parameter.

  Japanese Patent Laid-Open No. 2004-228561 corrects an error for a plant (process control) in which fluctuations in (i) time-dependent change components and (ii) manufacturing condition-dependent components are the main causes of control model errors. It has been shown. In particular, when one of (i) and (ii) is presumed to change significantly, the error is accurately improved by increasing the learning gain of the greatly changed component and decreasing the learning gain of the other components, It has been shown to correct quickly.

JP 63-206810 A Japanese Patent Laid-Open No. 2005-202803

  Here, in patent document 1, the scaling which paid its attention to the magnitude | size ("absolute value") of a parameter is performed. However, when there is a large non-linear relationship between the parameter and the evaluation value, even if such scaling is performed, the identification is not always accurate.

  In Patent Document 2, when the correction gain (learning gain) of a parameter that is estimated to have a large deviation is increased, the accuracy of the increased parameter is improved, but the accuracy of other parameters is deteriorated.

The present invention targets a nonlinear physical model, and evaluates a plurality of parameters in the physical model using evaluation function values calculated according to a plurality of parameter values using an evaluation function having a plurality of parameters as factors. In the parameter identification device for identifying a plurality of parameters asymptotically approximated to a true value by repeatedly correcting the parameter value according to the evaluation value, the change amount of each of the plurality of parameters to the change amount of the evaluation function value Means for examining the degree of influence for the plurality of parameters, and a coefficient for determining the amount of correction of each parameter in identification so that the degree of influence for the plurality of parameters falls within a predetermined range. Means for correcting a certain correction gain.

  Further, it is preferable that the influence degree is obtained by a one-factor search method.

  Further, it is preferable that the influence degree is obtained using an orthogonal table.

The present invention is directed to a non-linear physical model, evaluating the plurality of parameters in the physical model, the evaluation function value calculated in accordance with the plurality of parameter values using an evaluation function to a plurality of parameters and factors And a parameter identification program for identifying a plurality of parameters asymptotically approaching a true value by repeatedly correcting a plurality of parameter values in accordance with the evaluation value, and a computer for changing the amount of each of the plurality of parameters, Means for examining the degree of influence on the amount of change in the evaluation function value for the plurality of parameters, and the amount of correction of each parameter in identification so that the degree of influence for the plurality of parameters falls within a predetermined range. It functions as a means for correcting a correction gain which is a coefficient for determining the thickness.

  According to the present invention, the correction gain of each parameter in the identification is corrected so that the degree of influence of the plurality of parameters falls within a predetermined range. Therefore, the plurality of parameters can be effectively converged. Identification can be performed.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows a schematic configuration of a hydraulic system to be identified in the present embodiment. An automatic transmission is connected to the engine, and the rotation of the engine output shaft is changed by this automatic transmission and transmitted to the tire. The automatic transmission receives a hydraulic pressure and performs a shift operation. For example, when the shift timing is reached in accordance with the running state of the vehicle, an ECU (electronic control unit) detects this, supplies a current according to the shift command to the hydraulic control circuit, and the clutch pack in the automatic transmission. The speed change operation is performed by controlling the hydraulic pressure to.

  Here, the hydraulic system of the automatic transmission has a hydraulic pump driven by an engine, and oil from the hydraulic pump is supplied to the hydraulic line. A hydraulic pressure adjusting valve is provided near the hydraulic pump in the hydraulic line, and thereby the line pressure (original pressure) in the hydraulic line is controlled to a predetermined value. A linear solenoid valve is provided in the hydraulic line, and the supply of the line pressure of the hydraulic line to the clutch pack is controlled. Note that an orifice is provided in the flow path from the linear solenoid valve to the clutch pack to limit the supply of hydraulic pressure to the clutch pack. The clutch pack controls the engagement / release of the input side disk and the output side disk by moving the piston against the spring pressure by hydraulic pressure. For example, when shifting from the first speed to the second speed, the first-speed output disk originally engaged with the input-side disk is released, and the second-speed output disk is engaged.

  Considering such a hydraulic system to be controlled, the input is a hydraulic pressure command value, and the output is the hydraulic pressure supplied to the clutch pack.

  In FIG. 2, the outline | summary of the identification method of this embodiment is shown. It shows that the parameters (1) to (4) in the detailed hydraulic model are identified based on the error between the hydraulic system output of the actual machine and the model output. In this example, the parameters are (1) delay of the driver circuit, (2) return spring set load, (3) spring constant of the return spring, and (4) clutch piston stroke. The parameter values of these search parameters change sequentially according to the number of searches. (1) The delay of the driver circuit is a (1), a (2)..., (2) The return spring set load is b (1). , B (2), ..., (3) The spring constant of the return spring is c (1), c (2), ..., (4) The clutch piston stroke is d (1), d (2 ), ... and change.

  Hereinafter, a specific description will be given according to the flowchart of the identification method in FIG. In the description, the output of the hydraulic detail model in which the values of the parameters (1) to (4) are appropriately changed is used instead of the output data of the actual machine.

  First, in step 1, an initial (nominal) value of a model parameter to be identified (optimized) is determined. The hydraulic system parameters to be identified this time are indicated by broken circles in FIG.

  In step 2, the model is simulated, and in step 3, an error (sum of squares of error, evaluation value) between the model output and the output of the actual machine is calculated by equation (1) or equation (2).

  Subsequently, steps 4 to 8 will be described. It is determined whether the error (evaluation value) of Expression (1) or Expression (2) is within a threshold (Step 4). If this determination is within the threshold value, the identification step is completed.

However, if the error is larger than the threshold value in the determination in step 4, in step 5, the model parameter initial value appropriately shifted is set as a new parameter (formula (3)).
[a (2) b (2) c (2) ...] = [a (1) + δ ₁ b (1) + δ ₂ c (1) + δ ₃ ...] (3)

  Then, the model simulation of step 6 is performed in the same manner as in step 2 by using the parameter of equation (3), and in step 7, the error is calculated again by equation (1) or equation (2). Next, in step 8, it is determined whether the error is within a threshold value.

  If it is determined in step 8 that the error is within the threshold value, the identification step is completed. On the other hand, if the error is larger than the threshold value, the process proceeds to step 9.

  In step 9, a gradient vector is calculated from the factor effect of each parameter. To this end, first, a total of 16 combinations of two-level parameters are prepared for the four identification parameters shown in FIGS. 2 and 4 as shown in Table 1, and the error sum of squares J1 to J16 for each combination. Calculate

  As shown in Table 2, a factor effect (average value for each level) is calculated for each parameter level (2 levels) from the error thus obtained.

  From the factor effects in Table 2, the gradient vector defined by Equation (4) is obtained.

In step 10, the parameter correction amount is calculated based on the gradient vectors Sa (i) to Sd (i). The next model parameters a (i + 1) to d (i + 1) are calculated by the equation (5) using the correction gains k a to k d.

  When the next model parameter is calculated, the process returns to step 6 and the same steps are repeated until the error is less than or equal to the threshold value.

  In this way, each identification parameter is identified.

"Correction of correction gain"
Next, correction gain correction according to the present embodiment will be described with reference to the flowchart of FIG. That is, the correction gain of equation (5) is set as follows.

  First, in step 11, the relationship between the “change amount” of the identification parameter and the “change amount” of the evaluation function value is investigated. This relationship is the degree of influence of the change amount of the plurality of identification parameters on the change amount of the evaluation function value, and it is examined how each change affects the evaluation function value for the plurality of identification parameters.

  FIG. 6 shows the results of investigating the relationship between the change amount of the identification parameter and the change amount of the evaluation function value (integral value of the square error) using a hydraulic system model by a one-factor search method. Here, the horizontal axis of FIG. 6 is obtained by scaling (normalizing) each identification parameter value. “0%” indicates the nominal value (design value) of the parameter, and “−10%” indicates each parameter. Represents −10% change from the design value. The reason why the horizontal axis is scaled in this way is because it is assumed that the identification is performed by scaling the parameters. When parameter scaling is not performed, a parameter value that is not scaled may be taken as the horizontal axis of FIG.

  In this example, it is assumed that the identification parameter changes within a range of ± 10% from the nominal value (design value). From the results of FIG. 6, it can be seen that the return spring set load greatly affects the evaluation function value.

  In step 12, correction gain correction amounts ha to hd are determined for each parameter so that the relationship between each parameter and the evaluation function value is a predetermined reference relationship. In FIG. 6, the return spring set load line is used as a reference relationship, and each line is multiplied by a constant so that the other lines match the reference relationship. The results are shown in FIG. In the figure, the delay (time constant) line of the driver circuit is doubled (ha = 2), the spring constant line of the return spring is quadrupled (hc = 4), and the clutch piston stroke line is tripled (hd = By setting to 3), it can be seen that the four lines almost overlap each other within the range of the parameter change width of ± 10%.

  Here, as a method for setting the reference relationship, for example, an appropriate correction gain ka = kb = kc = kd = k is set and identification is performed according to FIG. A parameter that changes stably (for example, a return spring set load) is selected. Note that once the reference relationship is set, it is not necessary to set it again unless the k value is changed.

Finally, in step 13, the correction amount [ha hb hc hd] corresponding to each obtained parameter is applied to the correction gain (scalar value) k before correction (formula (6)).
ka = k × ha
kb = k × hb
kc = k × hc
kd = k × hd (6)
Here, ha = 2, hb = 1 (reference), hc = 4, hd = 3

  FIG. 8 shows the result of identification using the corrected gain obtained by Expression (6). It can be seen that both parameters quickly converge to the true value.

In the conventional concept as shown in Patent Document 1, the identification parameters shown in the left column of Table 3 are scaled (normalized) by Expression (7) using the initial parameter values in the right column of the same table. The correction gain (learning gain) at the time of identification uses the same value for all four parameters.
X1 ′ = X1 / x1 (7)
Here, X1 ′: parameter after scaling, X1: parameter before scaling, x1: initial parameter (scaling parameter, values in Table 3).

  The result of identification in this way is shown in FIG. It can be seen that the convergence of identification parameters other than the return spring set load is poor. This is because even if scaling is performed, the influence of the change in the return spring set load on the evaluation value (square error) is larger than other parameters.

  In the present embodiment, as described above, the delay (time constant) line of the driver circuit is doubled (ha = 2), the spring constant line of the return spring is quadrupled (hc = 4), and the clutch piston stroke line. FIG. 8 shows the result of the identification performed by multiplying the ratio by 3 (hd = 3). Thus, it is understood that all the identification parameters converge toward the true value.

  Next, step 11 will be described with reference to FIG. First, a variable i = 1 is set (step 111), and initial (nominal) values a (1), b (1), c (1), and d (1) of parameters a, b, c, and d are set. (Step 112). Next, a δ value indicating the amount of parameter change is set according to Table 4 indicating the setting for one-factor search (step 113).

  That is, if i = 1, δ1 = δ1 (1), δ2 = 0, δ3 = 0, δ4 = 0, and based on this, new parameters a (2), b (2), c (2) and d (2) are obtained from a (1) + δ1, b (1) + δ2, c (1) + δ3, d (1) + δ4, respectively (step 114). That is, a (2) changes to a (1) + δ1 (1), but other parameters, b (2), c (2), d (2) are b (1), c (1), It is the same as d (1).

  Then, using this new parameter, the model is simulated (step 115), error calculation is performed (step 116), i = i + 1 is set (step 117), and the process returns to step 113. In this way, for each parameter, the change in evaluation value (the magnitude of error) is examined when only that parameter is changed within a predetermined range.

  Next, step 12 will be described with reference to FIG. 11A. First, the relationship between the change amount of each identification parameter and the evaluation function value is displayed (illustrated) (step 121). For example, a diagram like FIG. 6 is displayed. One “reference relationship” is determined in each line in step 121 (step 122). Next, the parameter change width is determined (step 123). For example, as shown in FIG. 7, an assumed change width is determined for each parameter after normalization. This may be a user's input as a simulation condition. Then, a constant multiple value (correction amount) [ha, hb, hc...] Of each line is obtained so that each line in step 121 matches the “reference relationship” in the change width in step 123. In this way, the correction amount [ha, hb, hc...] Is obtained.

  Next, step 122 in FIG. 11A will be described based on FIG. 11B. Steps 1221 to 1226 are the same as steps 1 to 3 and 5 to 7 in FIG. That is, the determination whether or not the error in step 4 is within the threshold is merely omitted. Accordingly, in step 1226, an error between the model output and the output of the actual machine (optimization target) is calculated. Next, it is determined whether the number of corrections is N or more (step 1227). The N times are determined in advance in consideration of the error change state. If NO in this determination, the gradient vector is calculated from the factor effect of each parameter as in step 9 (step 1228), and the parameter correction amount is calculated based on the gradient vector S (i) (step 1229). ). Here, in the calculation of step 1229, the correction gains ka, kb,... Of each parameter are all set to the same value k. If the parameter correction amount is obtained in step 1229, the process returns to step 1225 and the same processing is repeated. If the number of corrections becomes N in step 1227, the process proceeds to step 122a. . In step 122a, the correction history of the value of each parameter is examined, and the parameter that changes fastest and stably is selected. In the example in FIG. 7, the return spring set load is selected as the reference relationship. Then, the relationship in step 11 corresponding to the selected parameter is set as a reference relationship (step 122b).

"Other examples"
In the above example, the relationship between the identification parameter and the evaluation function value is examined using the one-factor search method shown in Table 4, but an orthogonal table may be used. For example, using the L18 orthogonal table shown in FIG. 12, the relationship between the parameter and the evaluation function value is examined under 18 conditions. The factor-effect diagram shown in FIG. 13 is drawn from the results of FIG. 12, and the correction amount hi of the correction gain is obtained so that the relationship between each identification parameter and the evaluation function value becomes a reference relationship. The following procedure is the same as the description after Formula (6).

"Other"
According to the present embodiment, the parameter identification program is executed by the computer, the model is simulated, and the parameters are identified. Then, a program for performing a simulation with the model using the identified parameters is stored in a vehicle-mounted memory and used for vehicle control. That is, it is a model of a transmission as described above, and stores a simulation program using the identified parameters in the memory of the in-vehicle computer, and by executing this program, the operation of various devices is simulated. The actual device control command value is generated to control the shift. As a result, it is possible to perform appropriate control according to each condition regarding the shift operation under various conditions.

  Furthermore, in actual travel, it is also preferable that a true value for parameter identification is obtained by various measurements, the above-described parameter identification program itself is executed by the in-vehicle computer, and parameter identification is also executed at any time in the vehicle. The in-vehicle program may be provided on a DVD or the like, may be loaded online at the time of vehicle manufacturing inspection, or may be loaded by communication.

  Identification of parameters in a factory or the like at the time of vehicle manufacture may be performed by a computer in the factory, and various methods can be used as appropriate for installing (loading) the program.

It is a figure which shows the outline | summary of a hydraulic control system. It is a figure shown about the identification method of a parameter. It is a flowchart which shows the whole operation | movement of parameter identification. It is a figure which shows the parameter to identify. It is a flowchart which shows correction | amendment of the correction gain in embodiment. It is a figure which shows the relationship (before correction) of a parameter and an evaluation function. It is a figure which shows the relationship (after correction) of a parameter and an evaluation function. It is a figure which shows the identification result in a correction gain. It is a figure which shows the identification result by a conventional method. It is a flowchart which shows the operation | movement about a relationship investigation. It is a flowchart which shows the operation | movement of gain correction. It is a flowchart which shows the setting operation | movement of the reference | standard relationship. It is a figure which shows a L18 orthogonal table. It is a figure which shows a factor effect.

Claims (4)

Targeting a nonlinear physical model, multiple parameters in the physical model are evaluated with evaluation function values calculated according to multiple parameter values using an evaluation function with multiple parameters as factors. In the parameter identification device for identifying a plurality of parameters asymptotically approaching the true value by repeatedly correcting according to the evaluation value ,
Means for examining the degree of influence of each change amount of the plurality of parameters on the change amount of the evaluation function value for the plurality of parameters;
Means for correcting a correction gain that is a coefficient for determining the magnitude of the correction amount of each parameter in identification so that the degree of influence of the plurality of parameters falls within a predetermined range;
A parameter identification device characterized by comprising: The parameter identification device according to claim 1,
The parameter identification device characterized in that the influence degree is obtained by a one-factor search method. The parameter identification device according to claim 1,
The parameter identification device characterized in that the influence degree is obtained using an orthogonal table. Intended for non-linear physical model, a plurality of parameters in the physical model, and evaluated by the evaluation function value calculated in accordance with the plurality of parameter values using an evaluation function to a plurality of parameters and factors, multiple parameter values Is a parameter identification program for identifying a plurality of parameters asymptotically approaching a true value by repeatedly correcting according to an evaluation value ,
Computer
Means for examining the degree of influence of each change amount of the plurality of parameters on the change amount of the evaluation function value for the plurality of parameters;
Means for correcting a correction gain that is a coefficient for determining the magnitude of the correction amount of each parameter in the identification so that the degree of influence of the plurality of parameters falls within a predetermined range;
Parameter identification program characterized by functioning as