JP2011259637A - Control unit of rotary machine - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress degradation in controllability of control amount because of model error, relating to model prediction control.SOLUTION: Under the condition that an electric angle speed ω is a threshold speed ωF or higher (step S32:YES), d-axis inductance Ld is so operated as to feedback-control an error (predicted error Δiq) between a predicted current of q-axis and an actual current to be zero (step S34). A q-axis inductance Lq is so operated as to feedback-control an error (predicted error Δid) between a predicted current of d-axis and the actual current to be zero(step S36). The d-axis inductance Ld and q-axis inductance Lq at the time when the predicted error Δid, iq come to be zero, are stored as a learning value (step S38).

Description

  The present invention provides a control amount for the rotating machine by operating a power conversion circuit including a plurality of voltage applying means for applying voltages having different values and a switching element for selectively opening and closing a terminal of the rotating machine. The present invention relates to a control device for a rotating machine to be controlled.

  As this type of control device, in order to feedback control the current flowing in each phase of the three-phase motor to the command value, the command voltage of each phase is calculated, and based on the magnitude of the calculated command voltage and the triangular wave carrier A device that performs triangular wave comparison PWM control for operating a switching element of an inverter has been proposed and put into practical use.

  Furthermore, in recent years, for example, as can be seen in Patent Document 1 below, the current flowing through the three-phase motor when the operation state of the inverter is set in various ways is predicted, and the deviation between the predicted current and the command current is minimized. What performs what is called model predictive control which operates an inverter in the operation state which can be made into a thing is proposed. According to this, since the inverter is operated so as to optimize the behavior of the current predicted based on the output voltage of the inverter, the followability to the command current at the time of transient is compared with that by the above-described triangular wave comparison PWM control. improves. For this reason, model predictive control is considered to be highly useful for applications that require particularly high performance as transient tracking characteristics, such as a motor generator control device as an in-vehicle main unit.

  Other examples of the control device for the rotating machine include those described in Patent Document 2 below.

JP 2008-228419 A Japanese Patent Laid-Open No. 11-27997

  However, when the model predictive control is used, the controllability of the control amount depends on the accuracy of the model.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to perform model predictive control, and a rotation that can suitably suppress a decrease in controllability of a control amount due to a model error. It is to provide a control device for a machine.

  Hereinafter, means for solving the above-described problems and the operation and effect thereof will be described.

  According to the first aspect of the present invention, the rotating machine is operated by operating a power conversion circuit including a plurality of voltage applying means for applying voltages having different values and a switching element for selectively opening and closing a terminal of the rotating machine. In the control device for a rotating machine that controls the control amount of the rotating machine, a predicting unit that predicts a control amount of the rotating machine based on a model of the rotating machine when an operation state of the power conversion circuit is set, and the predicted An operation means for determining an actual operation state of the power conversion circuit based on a control amount and a command value of the control amount, and operating the power conversion circuit so as to be the determined operation state; and the prediction means A prediction error detecting means for detecting a prediction error that is an error of a predicted value of the control amount by the update method, and an updater that inputs the detected prediction error and updates the model so as to reduce the prediction error. Characterized in that it comprises and.

  When the model includes an error, an error occurs in the prediction by the prediction unit. In the above invention, in view of this point, by updating the model based on the prediction error, it is possible to suitably suppress a decrease in controllability of the control amount due to the model error.

  According to a second aspect of the present invention, in the first aspect of the invention, the prediction error detection unit detects the prediction error by using the prediction value of the control amount and the detection value of the control amount by the prediction unit as inputs. It is characterized by that.

  According to a third aspect of the invention, in the first or second aspect of the invention, the model is based on a term in which the d-axis current of the rotating machine is proportional to the product of the q-axis inductance and the rotational speed of the rotating machine. The updating means updates the q-axis inductance of the model so as to reduce the prediction error related to the d-axis current of the rotating machine.

  In the case of the above model, the q-axis inductance error tends to significantly affect the d-axis current error under circumstances where the rotational speed is somewhat high. In the above invention, the updating means is configured in view of this point.

  The invention according to claim 4 is the invention according to any one of claims 1 to 3, wherein the model is a product of a q-axis current of the rotating machine, a product of a d-axis inductance and a rotational speed of the rotating machine. The updating means updates the d-axis inductance of the model so as to reduce the prediction error related to the q-axis current of the rotating machine.

  In the case of the above model, the d-axis inductance error tends to significantly affect the q-axis current error under circumstances where the rotational speed is somewhat high. In the above invention, the updating means is configured in view of this point.

  The invention according to claim 5 is the invention according to claim 3 or 4, wherein the model is based on a prediction error detected by the prediction error detection means on condition that the rotational speed of the rotating machine is not less than a specified speed. Storage means for learning the inductance and storing the inductance together with the current or torque of the rotating machine at that time, and the updating means adds the inductance stored in the storage means according to the current or torque. The model used by the prediction means is updated based on the above.

  In the case of the above model, the influence of the inductance error becomes significant in a region where the rotational speed is high. The inductance error depends on the current. In the above invention, in view of this point, in the region where the rotation speed is high, the inductance is learned based on the prediction error, and this is stored together with the current or the torque that is a parameter correlated therewith.

  According to a sixth aspect of the present invention, in the invention according to the third or fourth aspect, the updating means manipulates the inductance of the model so as to control the prediction error detected by the prediction error detecting means to a target value. Features.

  According to a seventh aspect of the invention, in the sixth aspect of the invention, the target is based on a prediction error detected by the prediction error detecting means on condition that the rotational speed of the rotating machine is equal to or lower than a predetermined low speed. Target value output means for learning a value and outputting the learned target value to the update means is provided, and the update means controls the prediction error to the output target value.

  In the case of the above model, it is considered that the influence of the inductance error becomes remarkable in the region where the rotational speed is high, whereas the influence other than the inductance on the prediction error becomes large in the low speed region. In the above invention, in view of this point, the prediction error due to factors other than the inductance error is learned in the low speed region, and the inductance is manipulated to control the prediction error using this as a target value, thereby reducing the inductance error. The inductance can be manipulated.

  According to an eighth aspect of the invention, in the seventh aspect of the invention, the target value output means outputs the target value to be output to the update means, the current of the rotating machine, the torque of the rotating machine, the torque of the rotating machine. A variable setting is made according to at least one of a rotation speed and a temperature of the rotating machine.

  Model parameters vary depending on temperature, current, and the like. For this reason, it is considered that an appropriate value as the target value also changes according to temperature, current, and the like. In the above invention, in view of this point, the target value is variably set.

  According to a ninth aspect of the present invention, the rotating machine is operated by operating a power conversion circuit including a plurality of voltage applying means for applying voltages having different values and a switching element for selectively opening and closing a terminal of the rotating machine. In the control device for a rotating machine that controls the torque of the rotating machine, a predicting unit that predicts a control amount of the rotating machine based on a model of the rotating machine when an operation state of the power conversion circuit is set, and the predicted control An operating means for determining an actual operation state of the power conversion circuit based on the amount and a command value of the control amount, and operating the power conversion circuit so as to be in the determined operation state; Prediction error detection means for detecting a prediction error that is an error in the predicted value of the control amount, and the detected prediction error as input, and the rotation speed of the rotating machine is constant. Characterized in that it comprises an updating means for updating the inductance of the model to reduce the difference between the required torque and the actual torque of the rotating machine when changing the required torque for the machine.

  Even if the rotation speed is constant, if the torque changes, the amount of current changes, so the inductance changes. When the actual inductance corresponding to each torque (current) is different from that assumed by the model, a prediction error occurs, and the difference between the actual torque and the required torque is increased. In this regard, in the above invention, such a problem can be avoided by updating the inductance.

1 is a system configuration diagram according to a first embodiment. FIG. The figure which shows the voltage vector expressing the operation state of an inverter. The flowchart which shows the procedure of the model prediction control concerning the said embodiment. The figure which shows the relationship between a model error and the control amount error. The figure for demonstrating the principle of the said embodiment. The flowchart which shows the procedure of the learning process of the inductance concerning the embodiment. The figure for demonstrating the principle of 2nd Embodiment. The flowchart which shows the procedure of the learning process of the inductance concerning the embodiment. The figure which shows the effect of the same embodiment.

<First Embodiment>
Hereinafter, a first embodiment in which a control device for a rotating machine according to the present invention is applied to a control device for a hybrid vehicle will be described with reference to the drawings.

  FIG. 1 shows the overall configuration of a motor generator control system according to this embodiment. The motor generator 10 is a three-phase permanent magnet synchronous motor. The motor generator 10 is a rotating machine (saliency pole machine) having saliency. Specifically, the motor generator 10 is an embedded magnet synchronous motor (IPMSM).

  The motor generator 10 is connected to the high voltage battery 12 via the inverter IV. The inverter IV includes a series connection body of the switching elements Sup and Sun, a series connection body of the switching elements Svp and Svn, and a series connection body of the switching elements Swp and Swn. The motor generator 10 is connected to the U, V, and W phases, respectively. In the present embodiment, insulated gate bipolar transistors (IGBTs) are used as the switching elements Sup, Sun, Svp, Svn, Swp, and Swn. In addition, diodes Dup, Dun, Dvp, Dvn, Dwp, and Dwn are connected in antiparallel to these.

  In this embodiment, the following is provided as detection means for detecting the state of the motor generator 10 and the inverter IV. First, a rotation angle sensor 14 for detecting the rotation angle (electrical angle θ) of the motor generator 10 is provided. Further, a current sensor 16 that detects currents iu, iv, and iw flowing through the phases of the motor generator 10 is provided. Further, a voltage sensor 18 for detecting an input voltage (power supply voltage VDC) of the inverter IV is provided.

  The detection values of the various sensors are taken into the control device 20 constituting the low voltage system via an interface (not shown). The control device 20 generates and outputs an operation signal for operating the inverter IV based on the detection values of these various sensors. Here, the signals for operating the switching elements Sup, Sun, Svp, Svn, Swp, Swn of the inverter IV are the operation signals gup, gun, gvp, gvn, gwp, gwn.

  The control device 20 operates the inverter IV to control the torque of the motor generator 10 to the required torque Tr. Specifically, the inverter IV is operated so that a command current for realizing the required torque Tr is obtained. That is, in this embodiment, the torque of the motor generator 10 becomes the final control amount, but in order to control the torque, the current flowing through the motor generator 10 is directly controlled as a command current. . In particular, in the present embodiment, in order to control the current flowing through the motor generator 10 to the command current, the current flowing through the motor generator 10 is predicted when the operation state of the inverter IV is set to each of a plurality of ways, and the operation state is Model predictive control is performed in which the predicted current close to the command current is adopted as the actual operation state of the inverter IV.

  Specifically, the phase currents iu, iv, iw detected by the current sensor 16 are converted into actual currents id, iq in the rotating coordinate system by the dq converter 22. Further, the electrical angle θ detected by the angle sensor 14 becomes an input to the speed calculation unit 23, and thereby the rotational speed (electrical angular speed ω) is calculated. On the other hand, the command current setting unit 24 receives the required torque Tr and outputs command currents idr and iqr in the dq coordinate system. The command currents idr and iqr, the actual currents id and iq, and the electrical angle θ are input to the model prediction control unit 30. Based on these input parameters, the model prediction control unit 30 determines a voltage vector Vi that defines the operation state of the inverter IV and inputs it to the operation unit 26. The operation unit 26 generates the operation signal based on the input voltage vector Vi and outputs it to the inverter IV.

  Here, the voltage vectors expressing the operation state of the inverter IV are eight voltage vectors shown in FIG. For example, a voltage vector representing an operation state (indicated as “lower” in the drawing) in which the low-potential side switching elements Sun, Svn, Swn are turned on is the voltage vector V0, and the high-potential side switching elements Sup, A voltage vector representing an operation state (indicated as “upper” in the drawing) in which Svp and Swp are turned on is a voltage vector V7. These voltage vectors V0 and V7 are for short-circuiting all phases of the motor generator 10 and are called zero vectors because the voltage applied to the motor generator 10 from the inverter IV becomes zero. On the other hand, the remaining six voltage vectors V1 to V6 are defined by an operation pattern in which switching elements that are turned on exist in both the upper arm and the lower arm, and are called non-zero vectors. Yes. As shown in FIG. 2B, each of the voltage vectors V1, V3, and V5 corresponds to the positive side of the U phase, the V phase, and the W phase, respectively.

  Next, details of the processing of the model prediction control unit 30 will be described. In the operation state setting unit 31 shown in FIG. 1, the operation state of the inverter IV is set. Here, voltage vectors V0 to V7 shown in FIG. 2 are set as the operation state of inverter IV. The dq conversion unit 32 calculates the voltage vector Vdq = (vd, vq) in the dq coordinate system by performing dq conversion on the voltage vector set by the operation state setting unit 31. In order to perform such conversion, the voltage vectors V0 to V7 in the operation state setting unit 31 are, for example, “upper” as “VDC / 2” and “lower” as “−VDC / 2” in FIG. It can be expressed as above. In this case, for example, the voltage vector V0 is (−VDC / 2, −VDC / 2, −VDC / 2), and the voltage vector V1 is (VDC / 2, −VDC / 2, −VDC / 2). .

In the prediction unit 33, the current id when the operation state of the inverter IV is set to the state set by the operation state setting unit 31 based on the voltage vector (vd, vq), the actual currents id, iq, and the electrical angular velocity ω. , Iq is predicted. Here, the voltage equation expressed by the following (c1) and (c2) is discretized from the following state equations (formulas (c3) and (c4)) obtained by solving the current differential term. Predict.
vd = (R + pLd) id−ωLqiq (c1)
vq = ωLdid + (R + pLq) iq + ωφ (c2)
pid
= − (R / Ld) id + ω (Lq / Ld) iq + vd / Ld (c3)
piq
= −ω (Ld / Lq) id− (Rd / Lq) iq + vq / Lq−ωφ / Lq (c4)
Incidentally, in the above formulas (c1) and (c2), the resistance R, the differential operator p, the d-axis inductance Ld, the q-axis inductance Lq, and the armature linkage flux constant φ are used.

  The prediction of the current is performed for each of a plurality of operation states set by the operation state setting unit 31.

  On the other hand, the operation state determination unit 34 inputs the currents ide and iq predicted by the prediction unit 33 and the command currents idr and iqr, and determines the operation state of the inverter IV. Here, each operation state set by the operation state setting unit 31 is evaluated by the evaluation function J, and the operation state with the highest evaluation is selected. As this evaluation function J, in the present embodiment, a function whose value increases as the evaluation becomes lower is adopted. Specifically, the evaluation function J is calculated based on the inner product value of the difference between the command current vector Idqr = (idr, iqr) and the predicted current vector Idqe = (ide, iqe). This is a technique for expressing that the evaluation is lower as the value is larger in view of the fact that the deviation of each component between the command current vector Idqr and the predicted current vector Idqe can be both positive and negative values. As a result, it is possible to construct an evaluation function J in which the evaluation becomes lower as the difference between the components of the command current vector Idqr and the predicted current vector Idqe is larger.

  FIG. 3 shows a model prediction control processing procedure according to the present embodiment. This process is repeatedly executed at a predetermined cycle (control cycle Tc).

  In this series of processing, first, in step S10, the electrical angle θ (n) and the actual currents id (n) and iq (n) are detected, and the voltage vector V (n) determined in the previous control cycle. Is output. In subsequent step S12, the current (ide (n + 1), iqe (n + 1)) in one control cycle ahead is predicted. This is a process of predicting what will happen to the current one control cycle ahead based on the voltage vector V (n) output in step S10. Here, the currents ide (n + 1) and iqe (n + 1) are calculated by using the model expressed by the above equations (c3) and (c4) discretized by the forward difference method with the control cycle Tc. . At this time, the actual currents id (n) and iq (n) detected in step S10 are used as the initial value of the current. Further, the voltage vector on the dq axis is obtained by dq-transforming the voltage vector V (n) of the fixed coordinate system based on an angle obtained by adding “ωTc / 2” to θ (n) detected in step S10. .

  In subsequent steps S14 to S22, a process of predicting a current two control cycles ahead is performed for each of cases where a plurality of voltage vectors are set in the next control cycle. That is, first, in step S14, the number j that defines the voltage vector is set to “0”. In subsequent step S16, voltage vector Vj is set as voltage vector V (n + 1) in the next control cycle. In subsequent step S18, predicted currents ide (n + 2) and iqe (n + 2) are calculated in the same manner as in step S12. However, here, the predicted currents ide (n + 1) and iqe (n + 1) calculated in step S12 are used as the initial values of the currents. Further, the voltage vector on the dq axis is obtained by dq-transforming the voltage vector V (n + 1) of the fixed coordinate system based on the angle obtained by adding “3ωTc / 2” to the electrical angle θ (n) detected in step S10. And

  In a succeeding step S20, it is determined whether or not the number j is “7”. This process is for determining whether or not the current prediction process has been completed for all of the voltage vectors V0 to V7 that determine the operation state of the inverter IV. If a negative determination is made in step S20, the number j is incremented in step S22, and the process returns to step S16. On the other hand, when a positive determination is made in step S20, the process proceeds to step S24.

  In step S24, a process for determining the voltage vector V (n + 1) in the next control cycle is performed. Here, a voltage vector that minimizes the evaluation function J is a final voltage vector V (n + 1). That is, at the time when an affirmative determination is made in step S20, predicted currents ide (n + 2) and iqe (n + 2) are calculated for each of the voltage vectors V0 to V7. Therefore, eight values of the evaluation function J can be calculated using these eight predicted currents ide (n + 2) and iqe (n + 2). In the subsequent step S26, the voltage vectors V (n) and V (n + 1) are set to the voltage vectors V (n−1) and V (n), respectively, and the electrical angle θ (n) is the electrical angle θ (n−1). And real currents id (n) and iq (n) are assumed to be real currents id (n-1) and iq (n-1), respectively.

  In addition, when the process of step S26 is completed, this series of processes is once complete | finished.

  By the way, the controllability of the controlled variable such as current in the model predictive control depends on the accuracy of the model expression expressed by the above expressions (c1) and (c2). FIG. 4A shows an error ΔLd of the d-axis inductance Ld, an error ΔLq of the q-axis inductance Lq, an error ΔR of the resistance R, an error Δφ of the armature linkage flux constant φ, and a prediction error of the predicted currents ide and iqe. The relationship between Δid and Δiq is shown. The prediction error Δid is “ide (n) −id (n)”, and the prediction error Δiq is “iqe (n) −iq (n)”. The table in FIG. 4A is calculated by discretizing the model equations (c3) and (c4) and including the above errors in each parameter.

  4B shows the current error when the q-axis inductance Lq includes an error, and FIG. 4C shows the current error when the d-axis inductance Ld includes an error. FIG. 4D shows a current error when the resistor R includes an error, and FIG. 4E shows a current error when the armature flux linkage constant φ includes an error. However, these are data in a high rotation region. As shown in the drawing, when there is an error in the q-axis inductance Lq and the d-axis inductance Ld, the current error tends to increase. As can be seen from FIG. 4A, it is considered that the d-axis current and the q-axis current are proportional to the product of the q-axis inductance Lq and the d-axis inductance Ld and the electrical angular velocity ω.

  For this reason, if the prediction errors Δid and Δiq are detected in the high rotation speed region, even if the error factors are considered to be only the q-axis inductance Lq and the d-axis inductance Ld, these true values are learned with high accuracy. It is considered possible.

  FIG. 5 shows measurement results of current errors when the q-axis inductance Lq and the d-axis inductance Ld learned from such a viewpoint are used in various rotational speed regions. Specifically, as shown in FIG. 5A, the q-axis inductance Lq and the d-axis inductance Ld are learned for each torque value (here, 6 points). The q-axis inductance Lq learned in this way is shown in FIG. Accordingly, as shown in FIGS. 5C to 5E, if the learned q-axis inductance Lq and d-axis inductance Ld are used for each torque, the current error is suitably suppressed regardless of the rotation speed. I understand that I can do it. This is considered to mean that the q-axis inductance Lq and the d-axis inductance Ld significantly affect the prediction error of the control amount such as current. Incidentally, the reason why the q-axis inductance Lq and the d-axis inductance Ld are learned for each torque is that the q-axis inductance Lq and the d-axis inductance Ld change according to the current.

  FIG. 6 shows the procedure of the inductance learning process according to this embodiment. This process is repeatedly executed at a predetermined cycle, for example.

  In this series of processes, first, in step S30, it is determined whether or not a learning completion flag F indicating that the learning is performed in the entire torque region is “1”. If the learning completion flag F is not “1”, the process proceeds to step S32 on the assumption that the torque learning has not been completed in the entire region. In step S32, it is determined whether the logical product of the electrical angular velocity ω being equal to or higher than the threshold velocity ωF, the steady state, and the unlearned region is true. This process is for determining whether or not to perform inductance learning. Here, the threshold speed ωF is set to a speed at which the inductance error is assumed to be dominant for the prediction errors Δid and Δiq. The steady state means a state in which the control amount of the motor generator 10 is stable, for example, a case where the fluctuation amount of the required torque Tr is a predetermined value or less. Further, the unlearned region means a region where learning is not performed among a plurality of regions divided by torque as a region where inductance is learned.

  When a positive determination is made in step S32, the d-axis inductance Ld is learned based on the prediction error Δiq in step S34. In the present embodiment, this process is performed by manipulating the d-axis inductance Ld so as to feedback control the prediction error Δiq to zero using the prediction error Δiq as a control amount. More specifically, this feedback controller is constituted by an integration element that receives the prediction error Δiq. Note that the learning value of the d-axis inductance Ld may be a value when the output of the integration element is stabilized.

  In the subsequent step S36, the q-axis inductance Lq is learned based on the prediction error Δid. In this embodiment, this process is performed by manipulating the q-axis inductance Lq so as to feedback control the prediction error Δid to zero using the prediction error Δid. More specifically, this feedback controller is constituted by an integral element that receives the prediction error Δid. Note that the learning value of the q-axis inductance Lq may be a value when the output of the integration element is stabilized.

  In the subsequent step S38, the d-axis inductance Ld and the q-axis inductance Lq learned together with the required torque Tr are stored. Next, in step S40, it is determined whether or not learning has been completed in all regions. If it is determined that learning has been completed, the learning completion flag F is set to “1” in step S42.

  If a positive determination is made in step S30, a negative determination is made in steps S32 and S40, or if the process in step S42 is completed, the series of processes is temporarily terminated.

  According to the embodiment described in detail above, the following effects can be obtained.

  (1) The d-axis inductance Ld and the q-axis inductance Lq are updated so as to reduce the prediction errors Δid and Δiq. Thereby, the fall of the controllability of the control amount by a model error can be suppressed suitably.

  (2) The q-axis inductance Lq was manipulated to reduce and control the d-axis prediction error Δid, and the value at the time when the prediction error Δid became zero was learned. As a result, the q-axis inductance Lq can be learned in consideration of the fact that the error of the q-axis inductance Lq significantly affects the error of the d-axis current.

  (3) The d-axis inductance Ld was manipulated to reduce and control the q-axis prediction error Δiq, and the value when the prediction error Δiq became zero was learned. As a result, the d-axis inductance Ld can be learned in consideration of the fact that the error of the d-axis inductance Ld significantly affects the error of the q-axis current.

(4) On the condition that the electrical angular velocity ω is equal to or higher than the threshold velocity ωF, the inductance is manipulated to control the prediction errors Δid and Δiq to zero, and the value is learned. Thereby, learning of inductance can be performed with high accuracy.
<Second Embodiment>
Hereinafter, the second embodiment will be described with reference to the drawings with a focus on differences from the first embodiment.

  In this embodiment, the q-axis inductance Lq is operated to feedback control the prediction error Δid each time, and the d-axis inductance Ld is operated to feedback control the prediction error Δiq each time. At this time, the target values of the prediction errors Δid and Δiq as the control amounts are set based on the prediction errors Δid and Δiq in the low rotation speed region. This is due to the following reason.

  FIG. 7A shows the prediction error Δid when the electrical angular velocity ω is zero and the torque is relatively large. As illustrated, in this case, the prediction error Δid is not zero. As can be seen from the relationship shown in FIG. 4A, the main factor is not the error of the q-axis inductance Lq but the error ΔR of the resistor R and the like. For this reason, when operating the q-axis inductance Lq to feedback control the prediction error Δid to zero, as shown in FIG. 7B, the operated q-axis inductance Lq has an error in the low rotation speed region. Become. On the other hand, if the target value of the prediction error Δid is the value when the electrical angular velocity ω is zero, the q-axis inductance Lq can be set with high accuracy in the entire region.

  However, as shown in FIG. 4A, for the prediction error Δiq, in addition to the term that depends on the error ΔLd of the d-axis inductance Ld and the term that depends on the error ΔR of the resistance R, the armature linkage It is influenced by the term depending on the error Δφ of the magnetic flux constant φ. Therefore, in the present embodiment, the target value of the prediction error Δiq is set in consideration of the amount of the prediction error Δiq due to the error Δφ of the armature flux linkage constant φ.

  FIG. 8 shows a procedure of inductance learning processing according to the present embodiment. This process is repeatedly executed at a predetermined cycle, for example. Note that the predetermined period here may coincide with the control period Tc of the process shown in FIG.

  In this series of processing, first, in step S50, it is determined whether or not a learning completion flag F1 indicating that learning of the target value of the prediction error based on the error ΔR of the resistor R has been completed is “1”. If it is determined that the learning completion flag F1 is not “1”, whether or not the logical product between the electrical angular velocity ω is equal to or lower than the threshold velocity ωL and the steady state is true in step S52. Judging. This process is for determining whether or not the error ΔR of the resistor R is a region having a dominant influence on the prediction errors Δid and Δiq. When an affirmative determination is made in step S52, the target value Δiq1 based on the error R of the resistance R related to the prediction error Δiq and the target value Δid1 based on the error ΔR of the resistance R related to the prediction error Δid are learned. These target values may be the values when the prediction errors Δiq and Δid are stabilized, respectively. Target value Δiq1 is stored for each of a plurality of regions divided by temperature TH and current (command current iqr) of motor generator 10, and target value Δid1 is divided by temperature TH and current (command current idr). Stored for each of a plurality of areas. Incidentally, the temperature TH is a parameter for changing the value of the resistance R.

  When the process of step S54 is completed, it is determined in step S56 whether learning has been completed in all the regions divided by the temperature TH and the current. If it is determined that learning has been performed in all regions, the learning completion flag F1 is set to “1” in step S58.

  On the other hand, if an affirmative determination is made in step S50 or a negative determination is made in step S52, learning completion indicating that learning of the target value according to the error Δφ of the armature flux linkage constant φ is completed in step S60. It is determined whether or not the flag F2 is “1”. When a negative determination is made in step S60, in step S62, the absolute value of the q-axis command current iqr is less than or equal to the threshold current Iqth, and the absolute value of the d-axis command current idr is less than or equal to the threshold current idth. It is determined whether or not the logical product with this is true. This process is for determining a situation where the error Δφ of the armature flux linkage constant φ is dominant to the q-axis prediction error Δiq. If an affirmative determination is made in step S62, the target value Δiq2 is learned based on the prediction error Δiq in step S64. Specifically, this learning is performed for each of a plurality of regions divided by the electrical angular velocity ω. This is because, as can be seen from FIG. 4A, among terms that generate the prediction error Δiq, terms that are proportional to the electrical angular velocity ω also include terms that are proportional to the error ΔLd. In a succeeding step S66, it is determined whether learning is completed in all the regions divided by the electrical angular velocity ω. If an affirmative determination is made in step S66, the learning completion flag F2 is set to “1” in step S68.

  On the other hand, if an affirmative determination is made in step S60 or a negative determination is made in step S62, the d-axis inductance is used to feedback control the prediction error Δiq to the sum of the target value Δiq1 and the target value Δiq2 in step S70. Ld is operated, and q-axis inductance Lq is operated to feedback control the prediction error Δid to the target value Δid1. Here, as the target value Δiq1 and the target value Δid1, values in regions where the current and the temperature TH correspond to the current values are used. The target value Δiq2 uses a value in a region where the electrical angular velocity ω corresponds to the current value. Incidentally, in the figure, Ld0 and Lq0 are initial values recognized by the control device 20 for the d-axis inductance and the q-axis inductance, respectively. When learning of the target value Δiq1, the target value Δid1, and the target value Δiq2 is not completed, these may be set to “0”. However, if there are a plurality of learned regions, the target value may be calculated by interpolation.

  When a negative determination is made in steps S56 and S66, or when the processes in steps S58, S68, and S70 are completed, the series of processes is temporarily ended.

  Thus, the current error can be suitably reduced by performing feedback control on the prediction errors Δid and Δiq each time. 9 (a) and 9 (b), currents when the d-axis inductance Ld is corrected in the high rotation speed region and when the q-axis inductance Lq is corrected when the target value is not yet learned. Indicates an error.

  According to the present embodiment described above, the following effects can be obtained in addition to the effects (1) to (3) of the first embodiment.

  (5) The d-axis inductance Ld and the q-axis inductance Lq are manipulated to control the prediction error Δiq and the prediction error Δid to target values. Thereby, the error of a model formula can be reduced.

  (6) The target values of the prediction error Δiq and the prediction error Δid are learned by the prediction error Δiq and the prediction error Δid in the region where the electrical angular velocity ω is equal to or lower than the threshold velocity ωL. Thereby, it can suppress suitably that an inductance is updated by prediction error (DELTA) id and (DELTA) iq by factors other than an inductance error.

(7) The target values of the prediction error Δiq and the prediction error Δid are variably set according to the current flowing through the motor generator 10, the temperature TH of the motor generator 10, and the electrical angular velocity ω. As a result, the target value can be set appropriately, and as a result, the inductance update accuracy can be improved.
<Other embodiments>
Each of the above embodiments may be modified as follows.
"About storage means"
The storage means is not limited to storing the inductance in accordance with the required torque Tr. For example, the inductance may be stored according to the predicted torque Te. Further, the inductance may be stored according to the current. In this case, the d-axis inductance is stored in accordance with the q-axis current, considering that the relational expressions between the prediction errors ΔId and ΔIq and the inductance errors ΔLq and ΔLd shown in FIG. 4 are dependent on the current iq and id, respectively. It is desirable to store the q-axis inductance according to the d-axis current.

Further, the value stored in the storage means is not limited to the value learned in the above-described mode, and may be a value calculated by, for example, the process in step S70 of FIG.
"Target value output means"
As the target value output means, the target values Δiq1, Δid1 are calculated for each temperature TH and currents idr, iqr, or the target value Δiq2 is calculated for each electrical angular velocity ω in the process shown in FIG. Not limited to things. For example, the target values Δiq1, Δid1 may be calculated for each of the currents idr, iqr regardless of the temperature. Further, for example, the target value Δiq2 may not be calculated. Further, for example, it is not necessary to calculate only the target value Δiq2 and calculate the target values Δiq1 and Δid1. For example, the single target values Δiq1 and Δid1 calculated in the low rotation speed region may be used as fixed values in the entire region.

  Further, considering that the target values Δiq1 and Δid1 depend on the q-axis current and the d-axis current, respectively, a target value calculation coefficient is obtained by dividing the target values Δiq1 and Δid1 by the command currents iqr and idr and the actual currents iq and id. May be stored as Thus, the target value is calculated each time by the product of the command current iqr, idr or the actual current id, iq and the target value calculation coefficient. The target value calculation coefficient is preferably calculated and stored for each temperature TH or is corrected according to the temperature.

Similarly, in view of the fact that the target value Δiq2 depends on the electrical angular velocity ω, a value obtained by dividing the target value Δiq2 by the electrical angular velocity ω may be stored as a target value calculation coefficient. At this time, the target value calculation coefficient is desirably calculated based on a prediction error when the electrical angular velocity ω is in the low rotation speed region. However, if the d-axis current is zero, the prediction error Δiq does not depend on the inductance ΔLd, so that the target value can be calculated with high accuracy even when the electrical angular velocity ω is in the high rotation speed region.
“Target value”
The target value is not limited to that output by the target value output means. For example, it may be a predetermined fixed value. Here, the target value may be simply set to “0”. In particular, when the rotational speed is equal to or higher than the specified speed, the inductance can be updated with high accuracy by feedback control of the prediction error even if the target value is a fixed value (such as “0”). Further, not limited to a fixed value, for example, relationship information that defines a relationship between a target value and at least one parameter of the current of the rotating machine, the torque of the rotating machine, the rotational speed of the rotating machine, and the temperature of the rotating machine in advance. By storing in the control device 20, the target value may be variably set according to the at least one parameter.
About prediction error detection means
The predicted currents ide and iq and the actual currents id and iq are not limited to inputs. For example, in an operation state in which the difference between the actual currents id, iq controlled by the model predictive control and the command currents idr, iqr is assumed to be small, the command currents idr, iqr and the actual currents id, iq are used as inputs. An error between the command currents idr and iqr and the actual currents id and iq may be used as a prediction error.

The prediction error is not limited to a current error. For example, when a means for detecting the actual torque is provided, an error between the predicted torque Te and the actual torque T may be used.
"About update means"
The update process of the q-axis inductance Lq is not limited to the process based on the d-axis current error. For example, it may be based on an error between the predicted torque Te and the actual torque T.

The model formula updating method is not limited to the d-axis inductance Ld and the q-axis inductance Lq. For example, the resistance R may be updated by calculating the error ΔR of the resistance R from the target values Δid1 and Δiq1 calculated in step S54 of FIG.
"Initial value of inductance"
In each of the above embodiments, a value of 1 is assumed as the initial value of each of the d-axis inductance and the q-axis inductance, but the present invention is not limited to this. In view of the fact that these change according to the current, different initial values may be set for each current. Even in this case, if the initial value includes an error, it is effective to update the model formula by learning the error.
About prediction means
In each of the above embodiments, the control amount due to the operation of the inverter IV at the next update timing (the timing one control cycle ahead) of the operation state of the inverter IV is predicted, but this is not limitative. For example, the operation state at the update timing one control cycle ahead may be determined by sequentially predicting the control amount by the operation of the inverter IV at the update timing several control cycles ahead.

  The process for predicting the initial value of the current prediction associated with the setting of the operation state at the next update timing need not be performed. However, in this case, since the prediction accuracy of the control amount one control cycle ahead decreases from the update timing of the operation state of the inverter IV, the prediction error used for updating the model formula may be the prediction error several control cycles ahead. desirable.

  The model used for predicting the current is not limited to a model that ignores iron loss, and may be a model that takes this into consideration.

In each of the above-described embodiments, all of the possible operation states (voltage vectors V0 to V7) are temporarily set, and the control amount is predicted for all of them, but the present invention is not limited to this. For example, the zero vector may be a prediction target only in one of V0 and V7. Further, for example, in order to reduce the number of terminals of the motor generator 10 whose switching states are switched at the same time, only the operation state where the number of switching terminals in the switching state from the current operation state is a specified value (≦ 2) or less is temporarily set Also good.
"How to determine the operation state"
The evaluation function is not limited to the one that the evaluation is higher as the control amount deviation is smaller. In short, what is necessary is just to operate the inverter IV so that the deviation between the predicted control amount and its command value does not become excessively large. At this time, the predicting means is not necessarily limited to predicting control amounts for a plurality of operation states. For example, if the current operation state is maintained on condition that the difference between the control amount and its command value is equal to or less than the threshold value, the control amount for the current operation state is predicted until the difference exceeds the threshold value. Only processing is required.
"About controlled variables"
The control amount used for determining the operation of the inverter IV based on the command value and the predicted value is not limited to the current. For example, torque or magnetic flux may be used. Here, when the control amount is other than the current, the direct detection target by the sensor may be other than the current. However, the predicted currents ide and iqe may be calculated in the manner exemplified in the above embodiments, and converted into predicted values of torque and magnetic flux.

In each of the above embodiments, the ultimate control amount of the rotating machine (the control amount that is ultimately required to be a desired amount regardless of whether or not it is a prediction target) is the torque. For example, the rotation speed may be used.
"Power conversion circuit"
The power conversion circuit including a switching element that selectively opens and closes between a plurality of voltage applying units that apply voltages having different values and each terminal of the rotating machine is not limited to the inverter IV. For example, three or more voltage applying means for applying different values of voltage and a switching element for selectively opening and closing each terminal of the rotating machine may be provided. An example of a power conversion circuit for applying three or more voltages having different values to each terminal of a rotating machine is exemplified in Japanese Patent Application Laid-Open No. 2006-174697.
"Other"
The rotating machine is not limited to an embedded magnet synchronous machine, and may be an arbitrary synchronous machine such as a surface magnet synchronous machine or a field winding type synchronous machine. Furthermore, it is not limited to a synchronous machine, but may be an induction rotating machine such as an induction motor.

  -As a rotary machine, not only what is mounted in a hybrid vehicle but the thing mounted in an electric vehicle may be sufficient. Further, the rotating machine is not limited to the one used as the main machine of the vehicle.

  The DC power source is not limited to the high voltage battery 12 and may be, for example, an output terminal of a converter that boosts the voltage of the high voltage battery 12.

  DESCRIPTION OF SYMBOLS 10 ... Motor generator, 12 ... High voltage battery (one Embodiment of DC power supply), 14 ... Control apparatus (One Embodiment of the control apparatus of a rotary machine).

Claims (9)

A rotating machine that controls a control amount of the rotating machine by operating a power conversion circuit including a plurality of voltage applying means for applying voltages having different values and a switching element that selectively opens and closes a terminal of the rotating machine. In the control device of
Predicting means for predicting a control amount of the rotating machine based on a model of the rotating machine when an operation state of the power conversion circuit is set;
Operating means for determining an actual operation state of the power conversion circuit based on the predicted control amount and a command value of the control amount, and operating the power conversion circuit so as to be in the determined operation state; ,
A prediction error detecting means for detecting a prediction error which is an error of a predicted value of the control amount by the prediction means;
An apparatus for controlling a rotating machine, comprising: an update unit configured to input the detected prediction error and update the model so as to reduce the prediction error.   The control apparatus for a rotating machine according to claim 1, wherein the prediction error detection means detects the prediction error by using the predicted value of the control amount and the detected value of the control amount by the prediction means as inputs. The model determines the d-axis current of the rotating machine based on a term proportional to the product of the q-axis inductance and the rotational speed of the rotating machine,
3. The control device for a rotating machine according to claim 1, wherein the updating unit updates the q-axis inductance of the model so as to reduce the prediction error related to the d-axis current of the rotating machine. The model determines the q-axis current of the rotating machine based on a term proportional to the product of the d-axis inductance and the rotational speed of the rotating machine,
The said update means updates the d-axis inductance of the said model in order to reduce the said prediction error regarding the q-axis current of the said rotary machine, The rotary machine of any one of Claims 1-3 characterized by the above-mentioned. Control device. On the condition that the rotational speed of the rotating machine is equal to or higher than a specified speed, the inductance of the model is learned based on the prediction error detected by the prediction error detecting means, and the inductance is calculated based on the current of the rotating machine. Or further storing means for storing together with the torque,
5. The control device for a rotating machine according to claim 3, wherein the updating unit updates a model used by the prediction unit based on the inductance stored in the storage unit in accordance with the current or torque. .   5. The control device for a rotating machine according to claim 3, wherein the updating means manipulates the inductance of the model so as to control the prediction error detected by the prediction error detecting means to a target value. The target value is learned based on the prediction error detected by the prediction error detection means on the condition that the rotation speed of the rotating machine is equal to or lower than a predetermined low speed, and the learned target value is stored in the update means. A target value output means for outputting,
The rotating machine control device according to claim 6, wherein the updating unit controls the prediction error to the output target value.   The target value output means can vary the target value output to the update means according to at least one of the current of the rotating machine, the torque of the rotating machine, the rotational speed of the rotating machine, and the temperature of the rotating machine. 8. The control device for a rotating machine according to claim 7, wherein the control device is set. A rotating machine that controls a torque of the rotating machine by operating a power conversion circuit that includes a plurality of voltage applying means for applying voltages having different values and a switching element that selectively opens and closes a terminal of the rotating machine. In the control device,
Predicting means for predicting a control amount of the rotating machine based on a model of the rotating machine when an operation state of the power conversion circuit is set;
Operating means for determining an actual operation state of the power conversion circuit based on the predicted control amount and a command value of the control amount, and operating the power conversion circuit so as to be in the determined operation state; ,
A prediction error detecting means for detecting a prediction error which is an error of a predicted value of the control amount by the prediction means;
Using the detected prediction error as an input, the difference between the actual torque of the rotating machine and the required torque when the required torque for the rotating machine is changed under a situation where the rotational speed of the rotating machine is constant. An updater for updating the inductance of the model so as to reduce it.

Images