Rotating machine control device, machine learning device and inference device

Abstract

A rotating machine control device (20) comprises: a current detector (12) which detects an alternating current flowing in the rotating machine and outputs a current detection value; a power converter (11) which supplies power to the rotating machine by applying an alternating voltage based on a voltage command value; a current controller which adjusts the voltage command value in such a way that the current detection value is consistent with the current command value; an estimator (2) which obtains an estimated value of the amplitude of a magnetic flux vector in the rotating machine, namely, a magnetic flux estimated value; and a magnetic flux controller which adjusts the current command value in such a way that the magnetic flux estimated value is consistent with a set magnetic flux command value during a start-up control period from when the rotating machine is cut off from the power supply of the power converter (11) and becomes in a state of rotating by inertia until the power supply is restored.

Description

Technical Field

[0001] The present invention relates to a rotating machine control device, a machine learning device, and an inference device for controlling a rotating machine. Background Art

[0002] Conventionally, in vector control, which is one of the control methods implemented by a rotating machine control device, a position sensor for detecting the rotor position in an AC rotating machine is used. In addition, as a method for controlling an AC rotating machine without using a position sensor, a sensorless vector control is known, which detects an induced voltage generated by the rotation of a magnetic flux vector in an AC rotating machine, thereby finding the phase of the magnetic flux vector and finding the rotor position synchronized with the phase of the magnetic flux vector. According to the sensorless vector control, the AC rotating machine does not need a position sensor, which is a precision instrument, thereby reducing failures and achieving low costs.

[0003] In sensorless vector control, the rotating machine control device usually estimates the induced voltage based on the voltage command value input to the power converter or the detected value of the current flowing in the AC rotating machine. When the induced voltage is estimated based on the voltage command value or the current detection value, the rotating machine control device cannot detect the phase and frequency of the magnetic flux vector when the AC rotating machine is in a free running state. The free running state is a state in which the rotor continues to rotate by inertia when the power supply from the power converter is cut off while the rotor is rotating. When the AC rotating machine is started from the free running state, the phase and frequency cannot be detected, and proper current control becomes difficult. As a result, the AC rotating machine becomes difficult to start from the free running state due to excessive torque or pulsation in current or frequency. When the torque generated is a regenerative torque, problems sometimes occur in the power converter due to the rapid rise in the DC bus voltage of the power converter.

[0004] Therefore, a technology has been proposed in which an AC voltage is applied to an AC rotating machine according to a certain voltage command value during a startup control period from a free running state to a return to normal control, and the phase and frequency of a magnetic flux vector are estimated based on the voltage command value and a current detection value. Patent document 1 discloses a rotating machine control device that determines a voltage command value by performing current control to set the current command value to zero during startup control.

[0005] Patent Document 1: Japanese Patent Application Publication No. 2004-40837 Summary of the invention

[0006] During the start-up control, it is preferred to set a voltage command value that can minimize the generation of excessive torque, the generation of pulsation, and the rise of the DC bus voltage. Therefore, it is preferred to determine the voltage command value by current control using a small current command value. However, if the current command value during the start-up control is small, the generation of excessive torque can be reduced, and on the other hand, the magnetic flux generated by the excitation in the AC rotating machine also becomes smaller. The smaller the magnetic flux, the more unstable the magnetic flux becomes, and thus a constant time constant is required until the estimated phase or frequency stabilizes, so the start-up control period becomes longer. In addition, the magnetic flux is small and unstable, which easily generates unexpected torque and current or frequency pulsation. According to the prior art involved in Patent Document 1, current control is performed to set the current command value to zero, so the time required for starting from the free running state becomes longer, and there is a problem that the generation of unexpected torque or the generation of current or frequency pulsation cannot be reduced.

[0007] The present invention is proposed in view of the above situation, and its object is to obtain a rotating machine control device that can shorten the time required for starting a rotating machine from a free-running state and can reduce the occurrence of unexpected torque, current or frequency pulsations.

[0008] In order to solve the above-mentioned problems and achieve the purpose, the rotating machine control device involved in the present invention comprises: a current detector, which detects the AC current flowing in the rotating machine and outputs a current detection value; a power converter, which supplies power to the rotating machine by applying an AC voltage based on a voltage command value; a current controller, which adjusts the voltage command value in a manner such that the current detection value is consistent with the current command value; an estimator, which obtains an estimated value of the amplitude of the flux vector in the rotating machine, namely, a flux estimation value; and a flux controller, which adjusts the current command value in a manner such that the flux estimation value is consistent with the set flux command value during the startup control period from the time when the rotating machine is cut off from the power supply of the power converter and becomes a state of rotating by inertia until the power supply is restored.

[0009] Effects of the Invention

[0010] The rotating machine control device according to the present invention has the effects of being able to shorten the time required for starting the rotating machine from a free-running state and being able to reduce the occurrence of unexpected pulsations in torque, current, or frequency. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a diagram showing a configuration example of a rotating machine control device according to Embodiment 1.

[0012] Figure 2 It is a diagram showing a configuration example of a controller included in the rotating machine control device according to the first embodiment.

[0013] Figure 3 It is a diagram showing a configuration example of an estimator included in the rotating machine control device according to the first embodiment.

[0014] Figure 4 This is a diagram for explaining the operation during the startup control of the rotating machine control device according to the first embodiment.

[0015] Figure 5 This is a diagram showing a configuration example of an estimator included in the rotating machine control device according to the second embodiment.

[0016] Figure 6 This is a diagram showing a configuration example of an estimator included in the rotating machine control device according to the third embodiment.

[0017] Figure 7 This is a diagram showing a configuration example of hardware that realizes a voltage command device of a rotating machine control device according to the first to third embodiments.

[0018] Figure 8 This is a diagram showing a machine learning device and a rotating machine control device according to a fourth embodiment.

[0019] Fig. 9 This is a flowchart showing the processing procedure of the machine learning device involved in Implementation Example 4.

[0020] Fig.10 This is a diagram showing an estimation device and a rotating machine control device according to the fifth embodiment.

[0021] Fig.11 This is a flowchart showing the processing procedure of the estimation device involved in the fifth embodiment.

[0022] Fig.12 This is a diagram showing a machine learning device and a rotating machine control device according to a modified example of implementation mode 5.

[0023] Fig.13 This is a diagram showing an estimation device and a rotating machine control device according to a modified example of the fifth embodiment. DETAILED DESCRIPTION

[0024] Hereinafter, a rotating machine control device, a machine learning device, and an estimation device according to the embodiments will be described in detail based on the drawings.

[0025] Implementation method 1.

[0026] Figure 1: is a diagram showing a structural example of a rotating machine control device 20 involved in Embodiment 1. The rotating machine control device 20 controls a synchronous machine 21 as an AC rotating machine. In the following description, the two axes in the fixed coordinate system are referred to as the α-axis and the β-axis, and the two axes in the rotating coordinate system are referred to as the d-axis and the q-axis. In Embodiment 1, a case where the control phase is synchronized with the phase of the primary magnetic flux vector is described. Using the phase of the primary magnetic flux vector, the d-axis and the q-axis in the case of a coordinate transformation to the rotating coordinate of the d-axis based on the direction of the primary magnetic flux vector are represented as the ds-axis and the qs-axis, respectively. The ds-axis is an axis of the direction of the primary magnetic flux vector. The qs-axis is an axis of the direction orthogonal to the primary magnetic flux vector.

[0027] In this specification, the primary magnetic flux refers to the stator magnetic flux. The secondary magnetic flux refers to the magnetic flux corresponding to the excitation of the synchronous machine 21. For example, in the case where the synchronous machine 21 is a non-salient pole machine, the secondary magnetic flux is the magnetic flux generated by the excitation of the rotor. In the case where the synchronous machine 21 is a salient pole machine, the secondary magnetic flux is the magnetic flux that is a combination of the magnetic flux generated by the excitation of the rotor and the magnetic flux generated by the salient polarity of the rotor. In the case where the synchronous machine 21 is a synchronous machine that does not excite the rotor such as a reluctance type synchronous machine, the secondary magnetic flux is the magnetic flux generated by the salient polarity of the rotor.

[0028] The rotating machine control device 20 includes a voltage command device 10 which controls the current detection value i based on the current detection value i. us 、i vs 、i ws For voltage command value v us *、v vs *、v ws * Output; power converter 11, which applies a voltage command value v us *、v vs *、v ws * AC voltage, thereby supplying power to the synchronous machine 21; and a current detector 12, which detects the AC current flowing in the synchronous machine 21 and the current detection value i us 、i vs 、i ws Output.

[0029] The voltage command device 10 has a function of generating a voltage command value v us *、v vs *、v ws * The controller 1 and the estimator 2 for estimating the magnetic flux, phase and frequency. The estimator 2 obtains the estimated value of the amplitude of the magnetic flux vector in the synchronous machine 21, that is, the magnetic flux estimated value. The estimator 2 obtains a phase estimated value θ^, which is an estimated value of the phase of the magnetic flux vector. The estimator 2 obtains a frequency estimated value ω^, which is an estimated value of the frequency of the magnetic flux vector. The magnetic flux estimated value output from the estimator 2 is The phase estimated value θ^ and the frequency estimated value ω^ output from the estimator 2 are input to the controller 1 and used for controlling the synchronous machine 21 during the startup control. In addition, the phase estimated value θ^ and the frequency estimated value ω^ output from the estimator 2 can also be used for controlling the synchronous machine 21 when it returns from the free running state to the normal driving state. Variables with "^" added represent estimated values.

[0030] Figure 2 1 is a diagram showing a configuration example of a controller 1 included in a rotating machine control device 20 according to Embodiment 1. The controller 1 includes a three-phase dq converter 31 , a dq three-phase converter 32 , a current controller 33 , a switch 34 , and a magnetic flux controller 35 .

[0031] The current detection value i is input to the three-phase dq converter 31. us 、i vs 、i ws The three-phase dq converter 31 uses the current detection value i based on the phase estimation value θ ^. us 、i vs 、i ws The coordinate transformation of the ds axis current detection value i ds And qs axis current detection value i qs Output.

[0032] The ds-axis current detection value i is input to the current controller 33. ds , qs axis current detection value i qs , ds axis current command value i ds * and qs axis current command value i qs *. The current controller 33 obtains a current detection value i of the ds axis current. ds and ds axis current command value i ds *Consistent, and the qs axis current detection value i qs and the qs axis current command value i qs *The ds-axis voltage command value v ds * and qs axis voltage command value v qs *. The current controller 33 is used to control the ds axis voltage command value v ds * and qs axis voltage command value v qs That is, the current controller 33 adjusts the voltage command value so that the current detection value coincides with the current command value.

[0033] The ds axis voltage command value v is input to the dq three-phase converter 32. ds *、qs axis voltage command value v qs * and the phase estimation value θ ^. The dq three-phase converter 32 generates a ds-axis voltage command value v based on the phase estimation value θ ^. ds * and qs axis voltage command value vqs *Coordinate transformation, for voltage command value v us *、v vs *、v ws The voltage command device 10 outputs a voltage command value v to the power converter 11. us *、v vs *、v ws *.

[0034] The flux estimation value is input to the flux controller 35. and the preset flux command value The magnetic flux controller 35 obtains a magnetic flux estimated value With the flux command value That is, the magnetic flux controller 35 uses the magnetic flux estimated value The set flux command value The magnetic flux controller 35 outputs the excitation current command value of the d-axis.

[0035] The switch 34 switches the d-axis current command value i i, which is a preset excitation current command value. d * and the d-axis excitation current command value output from the magnetic flux controller 35 as the ds-axis current command value i ds * is output. Information indicating the current control period T1 and information indicating the flux control period T2 are input to the switch 34. The switch 34 changes the d-axis current command value i based on the information on the current control period T1 and the information on the flux control period T2. d * and the d-axis excitation current command value output from the flux controller 35. The current control period T1 and the flux control period T2 will be described later. The ds-axis current command value i ds *Input to current controller 33.

[0036] Figure 3 1 is a diagram showing a configuration example of the estimator 2 included in the rotating machine control device 20 according to the first embodiment. The estimator 2 operates during the entire startup control period. A voltage command value v is input to the estimator 2. us *、v vs *、v ws * and current detection value i us 、i vs 、i ws One of the two three-phase dq converters in the estimator 2 is driven by a voltage command value v based on the phase estimated value θ . us *、v vs *、v ws *Coordinate transformation, for the ds-axis voltage command value vds * and qs axis voltage command value v qs The other of the two three-phase dq converters in the estimator 2 is outputted by the current detection value i based on the phase estimation value θ ^. us 、i vs 、i ws The coordinate transformation of the ds axis current detection value i ds And qs axis current detection value i qs Output.

[0037] The ds-axis primary voltage v of the synchronous machine 21 ds And qs axis primary voltage v qs It is represented by the following formula (1). is the primary magnetic flux of the ds axis, is the primary magnetic flux of the qs axis, i ds is the ds axis primary current, i qs is the primary current of the qs axis, R s is the primary phase resistance, ω s is the frequency of the primary flux vector and s is the Laplace operator.

[0038] [Formula 1]

[0039]

[0040] When the control phase is synchronized with the phase of the primary magnetic flux vector, Therefore, the amplitude of the primary flux vector Equal to the primary magnetic flux of ds axis Amplitude of primary flux vector It is expressed by the following formula (2). Figure 3 The structure of the primary magnetic flux vector is calculated by performing the operation shown in formula (2) according to the voltage command value and the current detection value. Make an inference.

[0041] [Formula 2]

[0042]

[0043] The frequency of the primary magnetic flux vector ω s It is expressed by the following formula (3). Figure 3 The structure of the primary magnetic flux vector is calculated by performing the operation shown in equation (3) according to the voltage command value and the current detection value. s Make an inference.

[0044] [Formula 3]

[0045]

[0046] Phase θ of primary magnetic flux vector s It is expressed by the following formula (4). Figure 3 The structure of the primary magnetic flux vector is calculated by performing the operation shown in equation (4) according to the voltage command value and the current detection value. s Make an inference.

[0047] [Formula 4]

[0048]

[0049] Estimation value of the amplitude of the primary magnetic flux vector by estimator 2 That is, the estimated value of magnetic flux Output. Estimator 2 estimates the value ω of the frequency of the primary magnetic flux vector s The estimator 2 outputs the estimated value θ of the phase of the primary magnetic flux vector. s ^, that is, phase estimation value θ^ is output. In the calculation for estimating the magnetic flux in the estimator 2, a term for correcting the voltage error of the power converter 11 may be included in the term of the voltage command value.

[0050] Next, the operation of the rotating machine control device 20 during the startup control will be described. Figure 4 1 is a diagram for explaining the operation during the startup control of the rotating machine control device 20 according to the first embodiment. Figure 4 In the example, the horizontal axis represents time. Figure 4 The details of the transition of the state of the rotating machine control device 20 and the startup control period are shown.

[0051] The driving state refers to a state in which the synchronous machine 21 is driven by normal control performed by the rotating machine control device 20. Figure 4 In the time axis shown, the state of the synchronous machine 21 changes from the driven state to the free-running state, and then returns to the driven state again.

[0052] The start time of the startup control period is t0. At t-1 before t0, the synchronous machine 21 is cut off from the power supply from the power converter 11 and switches from the driving state to the free running state. The startup control period includes the period of time when the synchronous machine 21 is driven by the d-axis current command value i ... d * The current control period T1 during which current control is performed, and the flux control period T2 during which current control is performed based on the d-axis excitation current command value output by the flux controller 35. The current control period T1 is the period from t0 to t1. The flux control period T2 is the period from t1 to t2 when the start control period ends. At t2, the synchronous machine 21 switches from the free running state to the driving state.

[0053] At t0, information about the current control period T1 is input to the switch 34. The switch 34 changes the d-axis current command value i according to the information about the current control period T1. d *Selection. In the current control period T1, the switch 34 selects the d-axis current command value i d *That is, ds-axis current command value i ds *Output. d-axis current command value i d * represents a value of a small excitation current. The rotating machine control device 20 causes a small excitation current to flow through the synchronous machine 21 during the current control period T1 to generate a magnetic flux in the synchronous machine 21. The current control period T1 is a preparatory operation period for smoothly starting the magnetic flux control during the magnetic flux control period T2 located after the current control period T1. Therefore, basically, during the start control period, the magnetic flux control period T2 is set to be a period longer than the current control period T1.

[0054] On the other hand, at t1, information about the flux control period T2 is input to the switch 34. The switch 34 selects the d-axis excitation current command value output by the flux controller 35 according to the information about the flux control period T2. The switch 34 selects the d-axis excitation current command value i output by the flux controller 35, that is, the ds-axis current command value i ds *Output.

[0055] The phase estimation value θ^ obtained by the estimator 2 during the startup control is used as the initial value of the phase in the normal control starting from t2. The frequency estimation value ω^ obtained by the estimator 2 during the startup control is used as the initial value of the frequency in the normal control starting from t2. If the synchronous machine 21 returns to the driving state at t2, the synchronous machine 21 is controlled by an arbitrary method. A known vector control method can be adopted in the control method of the synchronous machine 21 in the driving state. Here, the description related to the control of the synchronous machine 21 in the driving state is omitted. The normal control in the driving state can be achieved by the internal structure of the rotating machine control device 20, and can also be achieved by the external structure of the rotating machine control device 20.

[0056] In the first embodiment, the phase estimated value θ^ and the frequency estimated value ω^ are set as the phase and frequency of the primary magnetic flux vector. In the free running state, the synchronous machine 21 has no load, so in the free running state, the direction of the primary magnetic flux vector becomes the same as the direction of the secondary magnetic flux vector. Therefore, even if a control method based on the secondary magnetic flux vector is adopted in the control in the driving state, there is no problem in using the phase estimated value θ^ and the frequency estimated value ω^ related to the primary magnetic flux vector as the initial value at the moment of change from the free running state to the driving state.

[0057] In the prior art shown in Patent Document 1, the entire start-up control period is a current control period. In this case, in order to minimize the generation of excessive torque, the occurrence of pulsation, and the rise in the DC bus voltage, a value representing a small excitation current must be set in the current command value. In this case, a constant time constant is required until the estimated phase or frequency stabilizes, so the start-up control period becomes longer. In addition, the magnetic flux generated by the small current excitation is small, so the phase and frequency calculated based on the magnetic flux vector of the synchronous machine 21 or the induced voltage generated by the rotation of the magnetic flux vector are prone to become unstable due to the influence of interference, etc. The phase and frequency become unstable, which easily generates unexpected torque and generates current or frequency pulsation. In addition, the DC bus voltage sometimes rises.

[0058] According to the first embodiment, the rotating machine control device 20 provides the current control period T1 before the flux control period T2 during the startup control period, thereby obtaining the flux estimation value generated by the initial excitation at the beginning of the flux control period T2. The rotating machine control device 20 performs a control operation using the flux estimated value from the start of the flux control period T2. The rotating machine control device 20 switches from the current control period T1 to the flux control period T2 at an appropriate timing, and performs a control operation using the flux estimation value from the beginning of the flux control period T2. By controlling the magnetic flux, the magnetic flux vector can be set to a larger value at an early stage and the magnetic flux vector can be stabilized at an early stage.

[0059] Furthermore, according to Embodiment 1, the rotating machine control device 20 uses the d-axis excitation current command value output by the flux controller 35 during the flux control period T2, thereby being able to set the flux vector to a larger value at an early stage and to stabilize the flux vector at an early stage. Therefore, the rotating machine control device 20 can shorten the time until the estimated phase and frequency are stabilized. Thus, the synchronous machine 21 can switch from a free-running state to a driven state through a short startup control period. In addition, the synchronous machine 21 can reduce the variation of the phase and frequency in response to interference, etc. By stabilizing the phase and frequency in advance, it is possible to reduce the generation of unexpected torque and reduce the pulsation of current or frequency. In addition, the rotating machine control device 20 can reduce the rise of the DC bus voltage.

[0060] As described above, the rotating machine control device 20 has the effects of being able to shorten the time required for starting the rotating machine from the free running state and being able to reduce the generation of unexpected torque and the occurrence of pulsation of current or frequency.

[0061] Implementation method 2.

[0062] In the first embodiment, the magnetic flux estimated value is obtained from the primary magnetic flux of the d-axis and the primary magnetic flux of the q-axis. In the second embodiment, the flux estimation value is obtained based on the primary magnetic flux of the α-axis and the primary magnetic flux of the β-axis. The following example is used to illustrate this.

[0063] Figure 5 1 is a diagram showing a configuration example of an estimator 2A included in a rotating machine control device 20 according to Embodiment 2. The configuration of the rotating machine control device 20 according to Embodiment 2 other than the estimator 2A is the same as that of the rotating machine control device 20 according to Embodiment 1. In Embodiment 2, descriptions overlapping with those in Embodiment 1 are omitted.

[0064] The rotating machine control device 20 according to the second embodiment synchronizes the control phase with the phase of the primary magnetic flux vector. Using the phase of the primary magnetic flux vector, the α-axis and the β-axis in the case of coordinate transformation to the fixed coordinate of the α-axis based on the direction of the primary magnetic flux vector are represented as the αs-axis and the βs-axis, respectively. The αs-axis is the axis of the direction of the primary magnetic flux vector. The βs-axis is the axis of the direction orthogonal to the primary magnetic flux vector.

[0065] The voltage command value v is input to the estimator 2A. us *、v vs *、v ws * and current detection value i us 、i vs 、i ws One of the two three-phase αβ converters in the estimator 2A is controlled by the voltage command value v us *、v vs *、v ws * Coordinate transformation, for the αs-axis voltage command value v αs * and βs-axis voltage command value v βs The other of the two three-phase αβ converters in the estimator 2A outputs the current detection value i us 、i vs 、i ws The coordinate transformation of the αs-axis current detection value i αs and βs-axis current detection value i βs Output.

[0066] The following formula (5) is a calculation formula obtained by expressing the above formula (1) by the α-axis and the β-axis. αs is the primary voltage of αs axis, v βs is the primary voltage of βs axis, i αs is the primary current of αs axis, i βsis the βs-axis primary current, is the primary magnetic flux of αs axis, is the primary magnetic flux along the βs axis.

[0067] [Formula 5]

[0068]

[0069] According to formula (5), the primary magnetic flux of the αs axis is and βs-axis primary flux It is represented by the following formula (6).

[0070] [Formula 6]

[0071]

[0072] The operation shown in equation (6) is different from equation (2) above, and becomes the integration of the AC quantity. Therefore, in order to avoid the accumulation of errors in DC quantities such as detection offset, a cutoff frequency ω can be combined in this operation. k High pass filter. Cut-off frequency ω k It is a value smaller than the frequency estimation value ω^ and is determined by multiplying the frequency estimation value ω^ by a constant. The constant is, for example, 1 / 3 or a value smaller than 1 / 3.

[0073] The operation of combining the high-pass filter with equation (6) is expressed by the following equation (7). Figure 5 A configuration example of the estimator 2A in the case of combining a high-pass filter is shown.

[0074] [Formula 7]

[0075]

[0076] Amplitude of primary flux vector It is expressed by the following equation (8). Figure 5 The structure of the present invention performs the calculations shown in equations (7) and (8) according to the voltage command value and the current detection value, and the amplitude of the primary magnetic flux vector is The estimator 2A can estimate the amplitude of the primary magnetic flux vector by performing the calculations shown in equations (6) and (8). Make an inference.

[0077] [Formula 8]

[0078]

[0079] Phase θ of primary magnetic flux vector s It is expressed by the following equation (9). Figure 5The structure of the present invention performs the calculations shown in equations (7) and (9) according to the voltage command value and the current detection value, and the phase θ of the primary magnetic flux vector is s The estimator 2A can estimate the phase θ of the primary magnetic flux vector by performing the calculations shown in equations (6) and (9). s Make an inference.

[0080] [Formula 9]

[0081]

[0082] The frequency of the primary magnetic flux vector ω s It is expressed by the following equation (10). Figure 5 The structure of the present invention performs the calculations shown in equations (7) and (10) according to the voltage command value and the current detection value, and the frequency ω of the primary magnetic flux vector is s The estimator 2A can estimate the frequency ω of the primary magnetic flux vector by performing the calculations shown in equations (6) and (10). s Make an inference.

[0083] [Formula 10]

[0084] ω s =Sθ s …(10)

[0085] Estimation value of the amplitude of the primary magnetic flux vector by the estimator 2A That is, the estimated value of magnetic flux The estimated value ω of the frequency of the primary magnetic flux vector is output by the estimator 2A. s The estimator 2A outputs the estimated value θ of the phase of the primary magnetic flux vector. s ^, that is, phase estimation value θ^ is output. In the calculation for estimating the magnetic flux in the estimator 2A, a term for correcting the voltage error of the power converter 11 may be included in the term of the voltage command value.

[0086] In the second embodiment, the rotating machine control device 20 also has the following effects, similar to the case of the first embodiment, namely, it is possible to shorten the time required to start the rotating machine from the free running state and reduce the generation of unexpected torque and the generation of current or frequency pulsation.

[0087] Implementation method 3.

[0088] In the first and second embodiments, the magnetic flux estimated value is obtained from the primary magnetic flux. In the third embodiment, the flux estimation value is obtained from the secondary magnetic flux. The following example is used to illustrate this.

[0089] Figure 6 1 is a diagram showing a configuration example of an estimator 2B included in a rotating machine control device 20 according to Embodiment 3. The configuration of the rotating machine control device 20 according to Embodiment 3 other than the estimator 2B is the same as that of the rotating machine control device 20 according to Embodiment 1. In Embodiment 3, descriptions overlapping with those in Embodiment 1 are omitted.

[0090] The rotating machine control device 20 according to the third embodiment synchronizes the control phase with the phase of the secondary magnetic flux vector. Using the phase of the secondary magnetic flux vector, the α-axis and the β-axis in the case of coordinate transformation of the α-axis based on the direction of the secondary magnetic flux vector are represented as the αr-axis and the βr-axis, respectively. The αr-axis is the axis of the direction of the secondary magnetic flux vector. The βr-axis is the axis of the direction orthogonal to the secondary magnetic flux vector.

[0091] The voltage command value v is input to the estimator 2B. us *、v vs *、v ws * and current detection value i us 、i vs 、i ws One of the two three-phase αβ converters in the estimator 2B is controlled by the voltage command value v us *、v vs *、v ws * Coordinate transformation, for the αs-axis voltage command value v αs * and βs-axis voltage command value v βs * Output. The other of the two three-phase αβ converters in the estimator 2B uses the current detection value i us 、i vs 、i ws The coordinate transformation of the αs-axis current detection value i αs and βs-axis current detection value i βs Output.

[0092] exist Figure 6 The following is a configuration example of an estimator 2B in which a high-pass filter is combined as in Embodiment 2. As a method of extracting components synchronized with the αr axis and the βr axis from the primary magnetic flux vector shown in the above equation (7), a method according to a known technique can be used.

[0093] The following formula (11) shows an example of a method of extracting components synchronized with the αr axis and the βr axis. and βr axis primary flux is obtained by formula (11). qrIt is the inductance of the qr axis component. The d-axis and q-axis in the case of coordinate transformation to the rotating coordinate of the d-axis based on the direction of the secondary magnetic flux vector are represented as the dr-axis and the qr-axis respectively. The method shown in formula (11) is described by "I. Boldea, MC Paicu, and GD Andreescu, Activeflux concept for motion-sensorless unified ac drives, IEEE Transactions on Power Electronics, Vol. 23, No. 5, pp. 2612-2618, 2008".

[0094] [Formula 11]

[0095]

[0096] Amplitude of the secondary flux vector It is expressed by the following equation (12). Figure 6 The structure of the secondary magnetic flux vector Make an inference.

[0097] [Formula 12]

[0098]

[0099] The phase θ of the secondary magnetic flux vector r It is expressed by the following equation (13). Figure 6 The structure of the secondary magnetic flux vector phase θ r Make an inference.

[0100] [Formula 13]

[0101]

[0102] The frequency of the secondary magnetic flux vector ω r It is expressed by the following equation (14). Figure 6 The structure of the secondary magnetic flux vector frequency ω r Make an inference.

[0103] [Formula 14]

[0104] ω r =sθ r …(14)

[0105] Estimation value of the amplitude of the secondary magnetic flux vector by the estimator 2B That is, the estimated value of magnetic flux The estimated value ω of the frequency of the secondary magnetic flux vector is output by the estimator 2B.r The estimator 2B outputs the estimated value θ of the phase of the secondary magnetic flux vector. r ^, that is, phase estimation value θ^ is output. In the calculation for estimating the magnetic flux in the estimator 2B, a term for correcting the voltage error of the power converter 11 may be included in the term of the voltage command value.

[0106] In the third embodiment, the phase estimation value θ^ and the frequency estimation value ω^ are set as the phase and frequency of the secondary magnetic flux vector. In the free running state, the synchronous machine 21 is unloaded, so in the free running state, the direction of the secondary magnetic flux vector becomes the same as the direction of the primary magnetic flux vector. Therefore, even if a control method based on the primary magnetic flux vector is adopted in the control in the driving state, there is no problem in using the phase estimation value θ^ and the frequency estimation value ω^ related to the secondary magnetic flux vector as the initial values ​​at the moment of change from the free running state to the driving state.

[0107] According to the third embodiment, the rotating machine control device 20 can shorten the time until the estimated phase and frequency are stabilized, as in the case of the first embodiment. Therefore, the synchronous machine 21 can switch from the free running state to the driving state through a short startup control period. In addition, the synchronous machine 21 can reduce the fluctuation of the phase and frequency in response to interference, etc. By stabilizing the phase and frequency in advance, the generation of unexpected torque and the pulsation of current or frequency can be reduced. In addition, the rotating machine control device 20 can reduce the rise of the DC bus voltage.

[0108] As described above, the rotating machine control device 20 has the effects of being able to shorten the time required for starting the rotating machine from the free running state and being able to reduce the generation of unexpected torque and the generation of pulsation of current or frequency.

[0109] Next, a hardware configuration of the voltage command device 10 which is a main part realizing the rotating machine control device 20 according to the first to third embodiments will be described. Figure 7 1 is a diagram showing a configuration example of hardware for realizing the voltage command device 10 of the rotating machine control device 20 according to the first to third embodiments. Figure 7 A configuration example is shown in which the controller 1 and the estimators 2 , 2A, and 2B of the voltage command device 10 are realized by a processing circuit 61 including a processor 63 and a memory 64 .

[0110] The processor 63 is a CPU (Central Processing Unit). The processor 63 may be a computing device, a microprocessor, a microcomputer, or a DSP (Digital Signal Processor). The memory 64 is a volatile or non-volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), or an EEPROM (registered trademark) (Electrically Erasable Programmable Read Only Memory).

[0111] The program for operating as the controller 1 and the estimators 2, 2A, and 2B is stored in the memory 64. The controller 1 and the estimators 2, 2A, and 2B can be realized by reading and executing the program by the processor 63. In addition, the program for operating as the controller 1 and the estimators 2, 2A, and 2B stored in the memory 64 can be provided to the user in a state of being written to a storage medium such as a CD (Compact Disc)-ROM, a DVD (Digital Versatile Disc)-ROM, etc., or can be provided via a network. In addition, the processor 63 outputs data such as the calculation results to the volatile memory of the memory 64. Alternatively, the processor 63 outputs data such as the calculation results to the auxiliary storage device via the volatile memory of the memory 64, thereby saving the data.

[0112] The input unit 62 is a circuit that receives an input signal to the voltage command device 10 from the outside. The input unit 62 receives the current detection value i us 、i vs 、i ws The output unit 65 is a circuit that outputs the signal generated by the voltage command device 10 to the outside. The output unit 65 receives the voltage command value v us *、v vs *、v ws *Output.

[0113] Figure 7This is an example of hardware in which the controller 1 and the estimators 2, 2A, and 2B are implemented by a general-purpose processor 63 and a memory 64, but the controller 1 and the estimators 2, 2A, and 2B may be implemented by a dedicated processing circuit instead of the processor 63 and the memory 64. That is, the controller 1 and the estimators 2, 2A, and 2B may be implemented by a dedicated processing circuit. Here, the dedicated processing circuit is a single circuit, a composite circuit, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a circuit composed of a combination thereof. In addition, a part of the controller 1 and the estimators 2, 2A, and 2B may be implemented by the processor 63 and the memory 64, and the remaining part may be implemented by a dedicated processing circuit.

[0114] Implementation method 4.

[0115] In the fourth embodiment, a machine learning device that generates a trained model for estimating the length of the current control period T1 and the length of the magnetic flux control period T2 will be described.

[0116] Figure 8 This is a diagram showing a machine learning device 40 and a rotating machine control device 20A according to Embodiment 4. In Embodiment 4, the same components as those in Embodiments 1 to 3 are denoted by the same reference numerals, and the configurations different from Embodiments 1 to 3 are mainly described. Figure 8 The illustrated machine learning device 40 is provided outside the rotating machine control device 20A.

[0117] The rotating machine control device 20A is obtained by adding a voltage detector 15 to any one of the rotating machine control devices 20 in Embodiments 1 to 3. The voltage detector 15 detects the DC bus voltage of the power converter 11 and generates a bus voltage detection value V dc Output.

[0118] The machine learning device 40 learns the length of the current control period T1 and the length of the flux control period T2 that can achieve the reduction of the amount of increase of the DC bus voltage in the power converter 11, the reduction of the pulsation width of the DC bus voltage, the reduction of the pulsation width of the frequency of the magnetic flux vector in the synchronous machine 21, and the reduction of the pulsation width of the effective current value of the current flowing in the synchronous machine 21, with respect to the rotating machine control device 20A. The machine learning device 40 includes a learning data acquisition unit 41, a feedback calculation data acquisition unit 42, a model generation unit 43, and a trained model storage unit 46. In addition, the machine learning device 40 only needs to learn the length of the current control period T1 and the length of the flux control period T2 that can achieve at least one of the reduction of the amount of increase of the DC bus voltage, the reduction of the pulsation width of the DC bus voltage, the reduction of the pulsation width of the frequency of the magnetic flux vector, and the reduction of the pulsation width of the effective current value.

[0119] The learning data acquisition unit 41 acquires a value indicating the length of the current control period T1, a value indicating the length of the magnetic flux control period T2, a frequency estimation value ω^, and a current effective value I rms Hereinafter, the value indicating the length of the current control period T1 is referred to as the value of the current control period T1. The value indicating the length of the flux control period T2 is referred to as the value of the flux control period T2. The learning data acquisition unit 41 acquires the values ​​of the current control period T1 and the flux control period T2, the frequency estimation value ω^, and the current effective value I rms As learning data.

[0120] Current effective value I rms According to the current detection value i us 、i vs 、i ws The current effective value is calculated based on the current detection value i us 、i vs 、i ws The three-phase dq converter 31 in the power converter 11 is configured to detect the effective value of at least one of the currents i us 、i vs 、i ws At least one of them has an effective current value I rms The learning data acquisition unit 41 acquires the effective current value I calculated by the three-phase dq converter 31. rms The three-phase dq converter 31 can be based on the ds axis primary current i ds Or the primary current i of the qs axis qs The effective value of current I rms Perform calculations.

[0121] In addition, the effective current value I rmsThe calculation is performed by the three-phase dq converter 31, but the present invention is not limited to this. rms The calculation may be performed by any of the internal factors of the rotating machine control device 20A and the external factors of the rotating machine control device 20A. The internal factors of the machine learning device 40 may be used to calculate the effective current value I rms In this case, the learning data acquisition unit 41 can acquire the current detection value i us 、i vs 、i ws 、ds axis primary current i ds Or the primary current i of the qs axis qs , for the effective value of current I rms Perform calculations.

[0122] The learning data acquisition unit 41 operates when the synchronous machine 21 is in the driving state and acquires the learning data. In addition, the learning data acquisition unit 41 creates a learning data by combining the values ​​of the current control period T1 and the flux control period T2, the frequency estimated value ω^, and the current effective value I rms That is, the learning data is a data set in which the learning data is collected. The learning data acquisition unit 41 sends the created data set to the model generation unit 43 .

[0123] The feedback calculation data acquisition unit 42 acquires the frequency estimation value ω^ and the current effective value I from the voltage command device 10. rms The feedback calculation data acquisition unit 42 acquires the bus voltage detection value V from the voltage detector 15. dc The report calculation data acquisition unit 42 operates from when the synchronous machine 21 is in the driving state to when the synchronous machine 21 is in the free running state.

[0124] When the synchronous machine 21 is in the driving state, the feedback calculation data acquisition unit 42 acquires the frequency estimated value ω^ and the current effective value I rms And bus voltage detection value V dc The feedback calculation data acquisition unit 42 obtains the bus voltage detection value V dc The bus voltage detection value V dc The pulsation width V dc _rip. Pulse width V dc _rip is the bus voltage detection value V dc The amplitude of the waveform that changes with time.

[0125] The report calculation data acquisition unit 42 obtains the ripple width ω^_rip of the frequency estimation value ω^ based on the obtained frequency estimation value ω^. The ripple width ω^_rip is the amplitude of the waveform indicating the time series change of the frequency estimation value ω^. The report calculation data acquisition unit 42 obtains the ripple width ω^_rip of the frequency estimation value ω^ based on the obtained current effective value I rms The effective value of the current Irms Pulsation width I rms _rip. Pulse width I rms _rip is the effective value of current I rms The amplitude of the waveform changes in time. Each pulse width V dc _rip、ω^_rip、I rms _rip is set to an absolute value.

[0126] The report calculation data acquisition unit 42 uses an arbitrary filtering method such as a high-pass filter to obtain the bus voltage detection value V dc , frequency estimation value ω ^ and current effective value I rms The pulsation width V dc _rip、ω^_rip、I rms _rip is extracted. Thus, the report calculation data acquisition unit 42 obtains the pulsation width V dc _rip、ω^_rip、I rms _rip.

[0127] The feedback calculation data acquisition unit 42 obtains the bus voltage detection value V dc and pulsation width V dc _rip、ω^_rip、I rms The feedback calculation data acquisition unit 42 stores the feedback calculation data at a preset timing when the synchronous machine 21 is in the driving state. The feedback calculation data acquisition unit 42 can acquire the bus voltage detection value V at multiple timings. dc and pulsation width V dc _rip、ω^_rip、I rms _rip, the average value of the acquired data is stored. dc and pulsation width V dc _rip、ω^_rip、I rms _rip are respectively represented by V dc _drv, V dc _rip_drv、ω^_rip_drv、I rms _rip_drv.

[0128] In addition, during the startup control period when the synchronous machine 21 is in the free-running state, the feedback calculation data acquisition unit 42 acquires the frequency estimated value ω^ and the current effective value I rms And bus voltage detection value V dc The feedback calculation data acquisition unit 42 obtains the bus voltage detection value V during the startup control period. dc The maximum value V dcThe report calculation data acquisition unit 42 uses an arbitrary filtering method such as a high-pass filter to obtain the bus voltage detection value V dc The bus voltage detection value V during the startup control period is obtained dc The maximum value of the pulsation width V dc The feedback calculation data acquisition unit 42 uses an arbitrary filtering method such as a high-pass filter to obtain the maximum value ω^_rip_rstmax of the pulsation width of the frequency estimation value ω^ during the startup control period based on the obtained frequency estimation value ω^. The feedback calculation data acquisition unit 42 uses an arbitrary filtering method such as a high-pass filter to obtain the maximum value ω^_rip_rstmax of the pulsation width of the frequency estimation value ω^ during the startup control period based on the obtained frequency estimation value ω^. rms The effective current value I during the startup control period is calculated rms The maximum value of the pulsation width I rms _rip_rstmax.

[0129] Then, the report calculation data acquisition unit 42 obtains the pulsation width V dc _rip increase V dc _rip_dat, pulsation width ω^_ripIncrease ω^_rip_dat, pulsation width I rms _rip increase I rms _rip_dat and bus voltage detection value V dc The increase in V dc _dat. Increase amount V dc _rip_dat、ω^_rip_dat、I rms _rip_dat, V dc _dat is the increase amount when the synchronous machine 21 is in the driving state.

[0130] The reward calculation data acquisition unit 42 calculates the increase V by performing the calculation shown in the following formula (15): dc _rip_dat.

[0131] V dc _rip_dat=V dc _rip_rstmax-V dc _rip_drv· · · (15)

[0132] The reward calculation data acquisition unit 42 obtains the increase amount ω^_rip_dat through the calculation shown in the following equation (16).

[0133] ω^_rip_dat=ω^_rip_rstmax-ω^_rip_drv· · · (16)

[0134] The reward calculation data acquisition unit 42 calculates the increase I by performing the calculation shown in the following formula (17): rms _rip_dat.

[0135] I rms _rip_dat=I rms _rip_rstmax-I rms _rip_drv· · · (17)

[0136] The reward calculation data acquisition unit 42 calculates the increase V by performing the calculation shown in the following formula (18): dc _dat.

[0137] V dc _dat=V dc _rstmax-V dc _drv· · · (18)

[0138] The reward calculation data acquisition unit 42 obtains the reward calculation data, that is, the increase amount V dc _rip_dat、ω^_rip_dat、I rms _rip_dat, V dc _dat is sent to the model generation unit 43.

[0139] The model generation unit 43 uses the values ​​of the current control period T1 and the flux control period T2, the frequency estimation value ω^, and the current effective value I rms The model generation unit 43 generates a trained model based on the frequency estimated value ω^ and the current effective value I rms The reduction in the amount of increase in the DC bus voltage in the power converter 11, the reduction in the pulsation width of the DC bus voltage, the reduction in the pulsation width of the frequency of the magnetic flux vector in the synchronous machine 21, and the current effective value I of the current flowing in the synchronous machine 21 can be achieved. rms The trained model infers the length of the current control period T1 and the length of the flux control period T2 of the reduction of the pulsation width.

[0140] In addition, the model generation unit 43 only needs to generate a model for reducing the amount of increase in the DC bus voltage, reducing the pulsation width of the DC bus voltage, reducing the pulsation width of the frequency of the magnetic flux vector, and reducing the effective current value I rms The trained model can be used to estimate the length of at least one of the current control period T1 and the flux control period T2 of the reduction in the pulsation width. The reward calculation data only needs to be the increase V dc _rip_dat, increase ω ^_rip_dat, increase I rms _rip_dat and increase Vdc When the pulsation width of the DC bus voltage is reduced, the feedback calculation data includes the increase amount V dc _rip_dat. When the pulsation width of the frequency of the magnetic flux vector is reduced, the reward calculation data includes the increase amount ω^_rip_dat. When the effective current value I rms In the case of a decrease in the pulsation width, the increase I is included in the feedback calculation data rms When the amount of increase in the DC bus voltage is reduced, the feedback calculation data includes the increase amount V dc _dat.

[0141] Among the learning algorithms used by the model generation unit 43, well-known algorithms such as teacher learning, teacherless learning, and reinforcement learning can be used. As an example, the case where reinforcement learning is applied is described. Reinforcement learning is an intelligent agent, i.e., an acting subject, in a certain environment observing the current state and deciding the action to be taken. The intelligent agent obtains rewards from the environment by selecting actions, and learns the countermeasures that obtain the most rewards through a series of actions. As representative methods of reinforcement learning, Q-learning, TD-learning, etc. are known. For example, in the case of Q-learning, the general update formula of the action value function Q(s, a), i.e., the action value table, is expressed by the following formula (19). The action value function Q(s, a) represents the value of the action of selecting action "a" based on the environment "s", i.e., the action value Q.

[0142] [Formula 15]

[0143] Q(s t , a t )←Q(s t , a t )+α(r t+1 +γmax a Q(s t+1 , a t )-Q(s t , a t ))…(19)

[0144] “s t " represents the state of the environment at time "t". t " represents the action at time t. Through the action "a t ”, the status changes from “s t ”To”s t+1 "Change." t+1 " indicates that the state is from "st ”To”s t+1 "The reward brought by the change. "γ" represents the discount rate, which satisfies 0<γ≤1. "α" represents the learning coefficient, which satisfies 0<α≤1. In implementation mode 4, action "a t " is the value of the current control period T1 and the flux control period T2. State "s t ” is the frequency estimation value ω^ and the current effective value I rms The model generation unit 43 calculates the state "s" at time "t". t "The best action" t "To study.

[0145] The update formula represented by the above formula (19) is that if the action value of the best action "a" at time "t+1" is greater than the action value Q of the action "a" executed at time "t", the action value Q is increased, and in the opposite case, the action value Q is decreased. In other words, the action value function Q (s, a) is updated in such a way that the action value Q of the action "a" at time "t" is close to the best action value at time "t+1". In this way, the best action value in a certain environment is continuously propagated to the action value in its previous environment.

[0146] The model generation unit 43 includes a reward calculation unit 44 and a function update unit 45. The reward calculation unit 44 calculates the value of the reward based on the increase V dc _rip_dat、ω^_rip_dat、I rms _rip_dat, V dc _dat, calculate the values ​​for the current control period T1 and the flux control period T2, the frequency estimated value ω^ and the current effective value I rms The function updating unit 45 generates a report based on the frequency estimated value ω^ and the current effective value I according to the report. rms The function for obtaining the values ​​of the current control period T1 and the flux control period T2 is updated. The function updating unit 45 outputs the trained model created by updating the function to the trained model storage unit 46. The trained model storage unit 46 stores the trained model.

[0147] The feedback calculation unit 44 calculates the current value I based on the values ​​of the current control period T1 and the flux control period T2, the frequency estimation value ω^, and the current effective value I rms and the increase V dc _rip_dat、ω^_rip_dat、I rms _rip_dat, V dc _dat calculates the feedback "r". For example, if the value of the current control period T1 or the value of the flux control period T2 is changed, the result is that the increase V dc_rip_dat、ω^_rip_dat、I rms _rip_dat, V dc When at least one of _dat decreases, the reward calculation unit 44 increases the reward "r". The reward calculation unit 44 increases the reward "r" by assigning a reward value of "1". The reward value is not limited to "1".

[0148] On the other hand, the value of the current control period T1 or the value of the flux control period T2 is changed. As a result, the increase V dc _rip_dat、ω^_rip_dat、I rms _rip_dat, V dc When at least one of _dat increases, the reward calculation unit 44 decreases the reward "r". The reward calculation unit 44 gives the reward value "-1" to increase the reward "r". The reward value is not limited to "-1".

[0149] The reward calculation data can be selected according to the installation environment of the synchronous machine 21. dc _rip_dat、ω^_rip_dat、I rms _rip_dat can be appropriately omitted when the pulsation width is sufficiently lower than the required standard of the system and no measures for reduction are required.

[0150] The function updating unit 45 updates the function for determining the values ​​of the current control period T1 and the flux control period T2 according to the reward calculated by the reward calculating unit 44. The machine learning device 40 repeatedly performs the learning described above. The function updating unit 45 outputs the trained model created by updating the function to the trained model storage unit 46.

[0151] For example, in the case of Q learning, the machine learning device 40 converts the action value function Q(s) represented by the above equation (19) into t , a t ) is used as a function for determining the values ​​of the current control period T1 and the flux control period T2. The trained model storage unit 46 stores the action value function Q(s) updated by the function update unit 45. t , a t ), that is, the trained model is stored.

[0152] Fig. 9 4 is a flowchart showing the processing procedure of the machine learning device 40 according to the fourth embodiment. Fig. 9 Flowchart of , illustrating the reinforcement learning method for updating the action-value function Q(s, a).

[0153] In step S1, the machine learning device 40 acquires the values ​​of the current control period T1 and the magnetic flux control period T2, the frequency estimated value ω^, and the current effective value I rms That is, the machine learning device 40 acquires the learning data. In addition, the machine learning device 40 acquires the reward calculation data, that is, the increase amount V, through calculation in the reward calculation data acquisition unit 42. dc _rip_dat、ω^_rip_dat、I rms _rip_dat, V dc _dat.

[0154] In step S2, the machine learning device 40 calculates the feedback using the feedback calculation unit 44. The feedback calculation unit 44 calculates the values ​​of the current control period T1 and the flux control period T2, the frequency estimation value ω^, and the current effective value I rms The reward calculation unit 44 increases or decreases the reward based on the comparison result between the acquired reward calculation data and the previous reward calculation data.

[0155] In step S3, the machine learning device 40 updates the action value function Q(s, a) based on the reward calculated in step S2. The machine learning device 40 updates the action value function Q(s, a) stored in the trained model storage unit 46. t , a t ) to update.

[0156] In step S4, the machine learning device 40 determines whether the action-value function Q(s, a) has converged. The machine learning device 40 determines that the action-value function Q(s, a) has converged by not updating the action-value function Q(s, a) in step S3.

[0157] If it is determined that the action value function Q(s, a) does not converge (step S4, No), the machine learning device 40 returns the sequence to step S1. If it is determined that the action value function Q(s, a) converges (step S4, Yes), the machine learning device 40 ends the process. Fig. 9 The learning is performed in the order shown. In addition, the machine learning device 40 may not perform the judgment of step S4, but return the order from step S3 to step S1 to continue learning. The trained model storage unit 46 stores the generated action-value function Q(s, a), that is, the trained model.

[0158] In Embodiment 4, the case where reinforcement learning is applied to the learning algorithm used by the machine learning device 40 is described, but learning other than reinforcement learning may also be applied to the learning algorithm. The machine learning device 40 may perform machine learning using a well-known learning algorithm other than reinforcement learning, such as deep learning, neural network, genetic programming, inductive logic programming, or a learning algorithm such as a support vector machine.

[0159] Figure 8 The machine learning device 40 shown is provided outside the rotating machine control device 20A. The machine learning device 40 may also be a device built into the rotating machine control device 20A. The machine learning device 40 may also be a device that can be connected to the rotating machine control device 20A via a network. The machine learning device 40 may also be a device existing on a cloud server.

[0160] The machine learning device 40 can learn the respective values ​​of the current control period T1 and the flux control period T2 according to the data set created by the plurality of rotating machine control devices 20A. The machine learning device 40 can obtain the learning data from the plurality of rotating machine control devices 20A used in the same place, or can also obtain the learning data from the plurality of rotating machine control devices 20A used in different places. The learning data can be data collected from the plurality of rotating machine control devices 20A operating independently of each other in the plurality of places. After the collection of learning data from the plurality of rotating machine control devices 20A starts, a new rotating machine control device 20A can be added to the object from which the learning data is collected. In addition, after the collection of learning data from the plurality of rotating machine control devices 20A starts, a part of the plurality of rotating machine control devices 20A can be excluded from the object from which the learning data is collected.

[0161] The machine learning device 40 that has performed learning on a certain rotating machine control device 20A can perform learning on other rotating machine control devices 20A other than the rotating machine control device 20A. The machine learning device 40 that has performed learning on the other rotating machine control device 20A can update the trained model by re-learning on the other rotating machine control device 20A.

[0162] According to the fourth embodiment, the machine learning device 40 generates a frequency estimation value ω^ and a current effective value I rmsThe trained model is used to estimate the values ​​of the current control period T1 and the flux control period T2. The machine learning device 40 can generate a model for obtaining a current effective value I of the current flowing in the synchronous machine 21 and a reduction in the increase amount of the DC bus voltage in the power converter 11, a reduction in the pulsation width of the DC bus voltage, a reduction in the pulsation width of the frequency of the flux vector in the synchronous machine 21, and a reduction in the DC bus voltage. rms The trained model of the various values ​​of the current control period T1 and the flux control period T2 for the reduction of the pulsation width.

[0163] Implementation method 5.

[0164] In the fifth embodiment, an estimation device for estimating the length of the current control period T1 and the length of the magnetic flux control period T2 using the trained model generated by the machine learning device 40 according to the fourth embodiment will be described.

[0165] Fig.10 This is a diagram showing an estimation device 50 and a rotating machine control device 20 according to Embodiment 5. In Embodiment 5, the same components as those in Embodiments 1 to 4 are denoted by the same reference numerals, and the configurations different from Embodiments 1 to 4 will be mainly described. Fig.10 The estimation device 50 shown is provided outside the rotating machine control device 20. Fig.10 Only the trained model storage unit 46 in the machine learning device 40 is shown.

[0166] The rotating machine control device 20 is any one of the rotating machine control devices 20 in Embodiments 1 to 3. In Embodiment 5, instead of any one of the rotating machine control devices 20 in Embodiments 1 to 3, a rotating machine control device 20A in Embodiment 4 may be applied.

[0167] The estimation device 50 estimates, with respect to the rotating machine control device 20, the length of the current control period T1 and the length of the flux control period T2 that can achieve a reduction in the amount of increase in the DC bus voltage in the power converter 11, a reduction in the pulsation width of the DC bus voltage, a reduction in the pulsation width of the frequency of the magnetic flux vector in the synchronous machine 21, and a reduction in the pulsation width of the effective current value of the current flowing in the synchronous machine 21. The estimation device 50 includes an estimation data acquisition unit 51 and an estimation unit 52.

[0168] The estimation data acquisition unit 51 acquires the frequency estimation value ω^ and the current effective value I ω from the voltage command device 10 of the rotating machine control device 20. rms The estimation data acquisition unit 51 sends the acquired estimation data to the estimation unit 52 .

[0169] The estimation unit 52 reads the trained model generated by the machine learning device 40 according to the fourth embodiment from the trained model storage unit 46. The estimation unit 52 stores the frequency estimated value ω^ and the current effective value I rms The trained model is inputted, thereby estimating the length of the current control period T1 and the length of the flux control period T2. The trained model outputs the values ​​of the current control period T1 and the flux control period T2 as the estimation results. The estimation unit 52 sends the values ​​of the current control period T1 and the flux control period T2 outputted from the trained model to the rotating machine control device 20.

[0170] Fig.11 1 is a flowchart showing the processing procedure of the estimation device 50 according to the fifth embodiment. In step S11, the estimation device 50 acquires the frequency estimation value ω^ and the current effective value I , which are estimation data, in the estimation data acquisition unit 51. rms In step S12, the estimation device 50 inputs the frequency estimation value ω^ and the current effective value I into the trained model in the estimation unit 52. rms , find the length of the current control period T1 and the length of the flux control period T2.

[0171] In step S13, the estimation device 50 outputs the values ​​of the current control period T1 and the flux control period T2 from the estimation unit 52 to the rotating machine control device 20. Fig.11 The processing is performed in the order shown.

[0172] Fig.10 The estimation device 50 shown is provided outside the rotating machine control device 20. The estimation device 50 may be a device built into the rotating machine control device 20. The estimation device 50 may be a device connectable to the rotating machine control device 20 via a network. The estimation device 50 may be a device existing on a cloud server.

[0173] In the fifth embodiment, the estimation device 50 estimates the length of the current control period T1 and the length of the flux control period T2 based on the model trained based on the learning data obtained from the rotating machine control device 20A, and transmits the values ​​of the current control period T1 and the flux control period T2 to the rotating machine control device 20 other than the rotating machine control device 20A. The rotating machine control device 20, 20A that transmits the values ​​of the current control period T1 and the flux control period T2 may be the rotating machine control device 20, 20A that has obtained the learning data, or may be a rotating machine control device 20, 20A other than the rotating machine control device 20, 20A that has obtained the learning data.

[0174] The estimation device 50 can estimate the reduction in the increase amount of the DC bus voltage, the DC bus voltage, the frequency, and the effective current value I by using the trained model for estimation. rms The length of the current control period T1 and the length of the flux control period T2 of each pulsation reduction are estimated.

[0175] The rotating machine control device 20 receives the values ​​of the current control period T1 and the flux control period T2 transmitted from the estimation device 50. The rotating machine control device 20 determines the length of the current control period T1 and the length of the flux control period T2 during the start control period based on the values ​​of the current control period T1 and the flux control period T2. The rotating machine control device 20 can reduce the amount of rise of the DC bus voltage in the power converter 11, reduce the pulsation width of the DC bus voltage, reduce the pulsation width of the frequency of the flux vector in the synchronous machine 21, and reduce the effective current value I of the current flowing in the synchronous machine 21. rms As a result, the rotating machine control device 20 reduces the rise in the DC bus voltage, and the DC bus voltage, frequency and current effective value I rms In a state where each pulsation is reduced, the synchronous machine 21 can be stably started from the free running state. The rotating machine control device 20 can start the synchronous machine 21 stably.

[0176] The rotating machine control device 20 can reduce the fluctuation of the DC bus voltage, thereby reducing the generation of unexpected torque. The rotating machine control device 20 can start the synchronous machine 21 from the free running state in a state where the generation of unexpected torque is reduced.

[0177] Next, a modification of the machine learning device and the estimation device will be described. In the modification of the fifth embodiment, the learning data and the estimation data each include the bus voltage detection value V dc . Fig.12 It is a diagram showing a machine learning device 40A and a rotating machine control device 20A according to a modification of the fifth embodiment. Fig.13 It is a diagram showing an estimation device 50A and a rotating machine control device 20 according to a modification of the fifth embodiment.

[0178] Fig.12 The learning data acquisition unit 41A of the machine learning device 40A shown in FIG. Figure 8 Similarly, the learning data acquisition unit 41 shown in FIG. 1 acquires the values ​​of the current control period T1 and the magnetic flux control period T2, the frequency estimated value ω^, and the current effective value I rms The learning data acquisition unit 41A also acquires the bus voltage detection value V from the voltage detector 15. dcThe learning data acquired by the learning data acquisition unit 41A includes the values ​​of the current control period T1 and the magnetic flux control period T2, the frequency estimated value ω^, the current effective value I rms And bus voltage detection value V dc The learning data acquisition unit 41A creates a learning data set by combining the values ​​of the current control period T1 and the magnetic flux control period T2, the frequency estimated value ω^, and the current effective value I rms And bus voltage detection value V dc Aggregated data sets.

[0179] The model generation unit 43 uses the values ​​of the current control period T1 and the flux control period T2, the frequency estimation value ω^, the current effective value I rms And bus voltage detection value V dc The machine learning device 40A generates a trained model based on the frequency estimated value ω ^ and the current effective value I rms And bus voltage detection value V dc A trained model is used to estimate the values ​​of the current control period T1 and the flux control period T2.

[0180] Fig.13 The estimation data acquisition unit 51A of the estimation device 50A shown in FIG. Fig.10 Similarly, the estimation data acquisition unit 51 shown in FIG. 1 acquires the frequency estimation value ω^ and the current effective value I ω from the rotating machine control device 20. rms The estimation data acquisition unit 51A further acquires the bus voltage detection value V from the voltage detector 15 of the rotating machine control device 20. dc The estimation data acquired by the estimation data acquisition unit 51A includes the frequency estimation value ω ^, the current effective value I rms And bus voltage detection value V dc .

[0181] The estimation unit 52 uses the trained model generated by the machine learning device 40A to estimate the frequency estimated value ω^ and the current effective value I rms And bus voltage detection value V dc The length of the current control period T1 and the length of the magnetic flux control period T2 are estimated. The estimation unit 52 uses the frequency estimation value ω^ and the current effective value I rms And bus voltage detection value V dc The trained model is inputted to estimate the length of the current control period T1 and the length of the magnetic flux control period T2.

[0182] According to the modification of the fifth embodiment, the machine learning device 40A adds the bus voltage detection value V to the learning data. dc, so that a higher-precision trained model can be generated. The inference device 50A uses the trained model to perform inference, thereby reducing the amount of increase in the DC bus voltage and the DC bus voltage, frequency, and current effective value I rms The length of the current control period T1 and the length of the flux control period T2 of each pulsation reduction can be estimated with high accuracy.

[0183] The machine learning device 40 or 40A according to the fourth or fifth embodiment is configured to Figure 7 The learning data acquisition unit 41, 41A, the reward calculation data acquisition unit 42, and the model generation unit 43 are implemented by a processing circuit 61 having a processor 63 and a memory 64. The learning data acquisition unit 41, 41A, the reward calculation data acquisition unit 42, and the model generation unit 43 can be implemented by a dedicated processing circuit.

[0184] The inference device 50, 50A according to the fifth embodiment is implemented by Figure 7 The estimation data acquisition units 51, 51A and the estimation unit 52 are realized by a processing circuit 61 having a processor 63 and a memory 64. The estimation data acquisition units 51, 51A and the estimation unit 52 can be realized by a dedicated processing circuit.

[0185] The machine learning device 40, 40A can include the increase in the torque detection value of the synchronous machine 21 in the report calculation data. The report calculation data acquisition unit 42 of the machine learning device 40, 40A acquires the torque detection value from the torque detector provided in the synchronous machine 21. The report calculation data acquisition unit 42 calculates the increase in the torque detection value during the start control period. The increase is the increase when the synchronous machine 21 is in the driving state. The report calculation unit 44 increases the report when the increase in the torque detection value decreases, and decreases the report value when the increase in the torque detection value increases. In this case, the machine learning device 40, 40A can generate a trained model, which is used to obtain the values ​​of the current control period T1 and the flux control period T2 that can reduce the unexpected increase in torque.

[0186] Furthermore, the torque detection value may be added to each of the learning data and the inference data. The learning data acquisition unit 41, 41A of the machine learning device 40, 40A acquires the torque detection value from the torque detector. The inference data acquisition unit 51, 51A of the inference device 50, 50A acquires the torque detection value from the torque detector. The machine learning device 40, 40A can generate a more accurate trained model by adding the torque detection value to the learning data. The inference device 50, 50A uses the trained model generated by the machine learning device 40, 40A to estimate the length of the current control period T1 and the length of the flux control period T2 that can reduce the increase in torque.

[0187] The structures shown in the above embodiments illustrate an example of the content of the present invention. The structures of the embodiments can be combined with other known technologies. The structures of the embodiments can also be appropriately combined with each other. Part of the structure of the embodiments can be omitted or changed without departing from the scope of the gist of the present invention.

[0188] Description of the label

[0189] 1 controller, 2, 2A, 2B estimator, 10 voltage command device, 11 power converter, 12 current detector, 15 voltage detector, 20, 20A rotating machine control device, 21 synchronous machine, 31 three-phase dq converter, 32 dq three-phase converter, 33 current controller, 34 switch, 35 flux controller, 40, 40A machine learning device, 41, 41A learning data acquisition unit, 42 feedback calculation data acquisition unit, 43 model generation unit, 44 feedback calculation unit, 45 function update unit, 46 trained model storage unit, 50, 50A inference device, 51, 51A inference data acquisition unit, 52 inference unit, 61 processing circuit, 62 input unit, 63 processor, 64 memory, 65 output unit.

Claims

1. A rotating machine control device, characterized in that: have: a current detector that detects an alternating current flowing in the rotating machine and outputs a current detection value; a power converter for supplying electric power to the rotating machine by applying an AC voltage based on a voltage command value; a current controller that adjusts the voltage command value in such a way that the current detection value is consistent with the current command value; an estimator for obtaining a magnetic flux estimated value which is an estimated value of the amplitude of the magnetic flux vector in the rotating machine; as well as A flux controller adjusts the current command value in such a way that the flux estimation value is consistent with the set flux command value during the startup control period from the time when the rotating machine is cut off from the power supply of the power converter and becomes a state of rotation by inertia to the time when the normal driving state achieved by vector control is restored.

2. The rotating machine control device according to claim 1, characterized in that: The current command value output by the magnetic flux controller is a current command value of the d-axis, which is one of two axes in the rotating coordinate system of the rotating machine.

3. The rotating machine control device according to claim 1 or 2, characterized in that: The magnetic flux vector is a vector of a primary magnetic flux of the rotating machine or a vector of a secondary magnetic flux of the rotating machine.

4. The rotating machine control device according to any one of claims 1 to 3, characterized in that: The startup control period is a flux control period during which current control is performed based on the current command value output by the flux controller and a period before the flux control period, and includes a current control period during which current control is performed based on a preset current command value, that is, a predetermined command value.

5. The rotating machine control device according to claim 4, characterized in that: During the startup control period, the magnetic flux control period is longer than the current control period.

6. The rotating machine control device according to claim 4 or 5, characterized in that: Also features: an estimation data acquisition unit that acquires an effective current value obtained from the frequency estimation value of the magnetic flux vector and the current detection value; and An estimating unit estimates the length of the current control period and the length of the magnetic flux control period by inputting the frequency estimation value and the current effective value into a model trained by machine learning.

7. The rotating machine control device according to claim 6, characterized in that: Also features: a learning data acquisition unit that acquires a value indicating the length of the current control period, a value indicating the length of the magnetic flux control period, the frequency estimation value, and the current effective value; as well as A model generation unit generates the trained model using a data set created based on the value indicating the length of the current control period, the value indicating the length of the flux control period, the frequency estimation value, and the current effective value.

8. A machine learning device, which, with respect to the rotating machine control device described in claim 4 or 5, learns the length of a current control period and the length of a flux control period that can achieve at least one of a reduction in the amount of increase of a DC bus voltage in a power converter, a reduction in the pulsation width of the DC bus voltage, a reduction in the pulsation width of the frequency of a magnetic flux vector in the rotating machine, and a reduction in the pulsation width of the effective current value of a current flowing in the rotating machine, The machine learning device is characterized by having: a learning data acquisition unit that acquires a value indicating the length of the current control period, a value indicating the length of the magnetic flux control period, a frequency estimation value of the magnetic flux vector, and the current effective value; a model generation unit that generates a trained model using a data set created based on the value indicating the length of the current control period, the value indicating the length of the flux control period, the frequency estimation value, and the current effective value; and The trained model storage unit stores the trained model.

9. The machine learning device according to claim 8, characterized in that Also features: A voltage detector detects the DC bus voltage and outputs a bus voltage detection value; a report calculation data acquisition unit, which receives the bus voltage detection value, the frequency estimation value, and the current effective value as input, and obtains at least one of the increase in the pulsation width of the bus voltage detection value, the increase in the pulsation width of the frequency estimation value, the increase in the pulsation width of the current effective value, and the increase in the bus voltage detection value, i.e., the report calculation data; The model generation unit has: a report calculation unit that calculates a report for a combination of a value indicating the length of the current control period, a value indicating the length of the magnetic flux control period, the frequency estimation value, and the current effective value based on the report calculation data; as well as a function updating unit that updates, according to the feedback, a function for obtaining a value indicating the length of the current control period and a value indicating the length of the flux control period from the frequency estimated value and the current effective value, The function updating unit outputs the trained model generated by updating the function.

10. The machine learning device according to claim 9, characterized in that: The learning data acquisition unit further acquires the bus voltage detection value. The model generation unit generates the trained model using the data set created based on the value indicating the length of the current control period, the value indicating the length of the flux control period, the frequency estimation value, the current effective value, and the bus voltage detection value.

11. An inference device, characterized in that: Regarding the rotating machine control device described in claim 4 or 5, the length of the current control period and the length of the flux control period that can achieve at least one of the reduction in the amount of increase of the DC bus voltage in the power converter, the reduction in the pulsation width of the DC bus voltage, the reduction in the pulsation width of the frequency of the magnetic flux vector in the rotating machine, and the reduction in the pulsation width of the effective current value of the current flowing in the rotating machine are inferred, The inference device is characterized by having: an estimation data acquisition unit that acquires the frequency estimation value of the magnetic flux vector and the effective current value obtained from the current detection value; and An inference unit inputs the frequency estimation value and the current effective value into a trained model for inferring the length of the current control period and the length of the flux control period based on the frequency estimation value and the current effective value, thereby inferring the length of the current control period and the length of the flux control period.

12. The inference device according to claim 11, characterized in that: The estimation data acquisition unit further acquires a bus voltage detection value which is a detection value of the DC bus voltage. The inference unit uses a trained model for inferring the length of the current control period and the length of the flux control period based on the frequency estimation value, the current effective value and the bus voltage detection value, and infers the length of the current control period and the length of the flux control period based on the frequency estimation value, the current effective value and the bus voltage detection value.

Images