JP6545428B1 - Control device of power converter - Google Patents

Abstract

When the command value (Xr) of the output value (X) changes, a threshold value (Xth) indicating a condition for switching to feedforward control is set. When the command value (Xr) increases, the threshold value (Xth) is set higher than the command value (Xr) after the change. In a period in which the output value (X) is equal to or less than the threshold (Xth), feedback control is selected, and the control amount of the power converter is set based on the deviation of the output value (X) from the command value (Xr). In a period (Tff) in which the output value (X) is larger than the threshold (Xth), feedforward control is selected, and the control amount of the power converter is set to a predetermined feedforward control amount.

Description

  The present invention relates to a control device of a power converter.

  Japanese Patent No. 5412658 (patent document 1) reduces overshoot and undershoot by combining feedback control and feedforward control by machine learning control such as neuro control in control of a DC / DC converter. The technology is described.

  In Patent Document 1, the feedforward control amount is calculated by multiplying the deviation between the control target value in a certain sampling and the control predicted value calculated from the learning history by the gain α. Specifically, the gain α is decreased by a factor A (positive value other than zero) for suppressing the first (first) undershoot or overshoot in accordance with the decay function. According to λ · n), the gain at the nth sampling is set. Here, λ is a factor (positive value other than zero) to attenuate the second overshoot or undershoot. This achieves both suppression of the first undershoot or overshoot and overcompensation for the subsequent overshoot or undershoot.

Patent No. 5412658 gazette

  In the feed-forward control of Patent Document 1, depending on the selection of the coefficients A and β determined in advance by simulation and control theory, the effect of reducing overshoot or undershoot can not be expected due to inadequate gain. I am concerned.

  Furthermore, if the feedforward control is insufficient and overshoot or undershoot occurs, the overshoot or undershoot may not be suppressed promptly because the control mode is a control mode that is compensated by feedback control.

  Further, in Patent Document 1, since the control prediction value at each sampling is obtained by machine learning, there is also concern that the operation load may affect the response speed of control.

  The present invention has been made to solve such problems, and an object of the present invention is to provide an electric power having a feedforward control function to suppress overshoot or undershoot without increasing computational load. It is providing a control device of a converter.

  According to an aspect of the present invention, a control device of a power converter whose output value changes according to a control amount includes a control switching unit and a control signal generation unit. The control switching unit fixes the control amount to a predetermined feedforward control amount according to a comparison between the output value and a threshold value different from the command value of the output value, feedforward control, and deviation of the output value and the command value It switches to feedback control which calculates a control amount. The control signal generation unit generates a control signal of the power converter based on the control amount set by the control switching unit. At least one of a first period in which the threshold value is set to a value higher than the command value and a second period in which the threshold value is set to a value lower than the command value are provided according to the change in the command value. The control switching unit selects feedback control when the output value is equal to or less than the threshold in the first period, and selects feedforward control when the output value is higher than the threshold in the second period. When the output value is equal to or higher than the threshold value, feedback control is selected, and when the output value is lower than the threshold value, feed forward control is selected.

  According to the present invention, when overshoot or undershoot exceeding the threshold occurs when the command value changes, the feedback control is switched to feedforward control in which the control amount is fixed to a predetermined feedforward control amount. By rapidly changing the control amount, it is possible to realize a feedforward control function for suppressing overshoot or undershoot.

FIG. 2 is a functional block diagram for explaining a control device of the power converter according to the first embodiment. It is a functional block diagram which further demonstrates the structure of the FB control amount calculation part shown by FIG. It is a flowchart explaining operation | movement of the flag determination part shown by FIG. It is a flowchart explaining operation | movement of the control switching part shown by FIG. It is a flowchart explaining the modification of operation | movement of the control switching part shown by FIG. It is a flow chart explaining control operation of a control device at the time of control start of a DC / DC converter. It is a flowchart explaining the control processing using the neural network model by the FF control amount calculation part shown by FIG. It is a conceptual diagram explaining the structural example of the neural network model used by FF control amount calculation part. It is a conceptual diagram explaining the input value and output value in machine learning in the neural network model shown by FIG. It is a flowchart explaining the control processing by the learning data extraction part shown by FIG. It is a flowchart explaining the control processing by the learning data processing part shown by FIG. It is a wave form diagram explaining the control operation example concerning the comparative example at the time of rise of command value. FIG. 7 is a waveform diagram for explaining an example of control operation according to the first embodiment at the time of increase of command value. It is a wave form diagram explaining transition of the control amount (duty ratio) in FIG.12 and FIG.13. It is a wave form diagram explaining the control operation example concerning the comparative example at the time of the fall of command value. FIG. 7 is a waveform chart for explaining an example of control operation according to the first embodiment at the time of decrease of command value. FIG. 10 is a functional block diagram for explaining a controller of a power converter according to a modification of the first embodiment. It is a flowchart explaining the operation | movement at the time of the learning data generation control start by the control apparatus of the power converter which concerns on the modification of Embodiment 1. FIG. 18 is a flowchart illustrating an operation in learning data generation control of the parameter setting unit illustrated in FIG. 17; FIG. 7 is a functional block diagram for explaining a control device of a power converter according to a second embodiment. FIG. 13 is a flowchart for explaining control operation at the start of control of the control device of the power converter according to Embodiment 2. FIG. FIG. 16 is a flowchart illustrating an operation of a control switching unit in a control device of a power converter according to Embodiment 3. FIG. FIG. 5 is a conceptual diagram for explaining switching between feed forward control and feedback control in the first embodiment. FIG. 17 is a conceptual diagram illustrating switching of feed forward control and feedback control in the third embodiment. It is a block diagram explaining the example of application of the control device of the power converter concerning this embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following, the same or corresponding parts in the drawings are denoted by the same reference numerals, and the description thereof will not be repeated in principle.

Embodiment 1
(Basic control configuration)
FIG. 1 is a functional block diagram for explaining a configuration of control device 50 of the power converter according to the first embodiment. Hereinafter, in the present embodiment, a DC / DC converter will be described as an example of a power converter to be controlled.

  Referring to FIG. 1, DC / DC converter 11 shown as an example of a power converter to be controlled by control device 50 supplies DC power indicated by the product of output voltage and output current to a load. .

  Control device 50 controls output value X of DC / DC converter 11 to command value Xr. The controlled output value X is, for example, an output voltage, but can also be an output current, a DC power, or the like. The control device 50 includes an output control unit 100 and a learning control unit 110. The output value X is detected by a sensor (not shown), and the detected value by the sensor is input to the control device 50.

  The output control unit 100 includes a flag determination unit 5, a feedforward (FF) control amount calculation unit 6, a subtraction unit 7, a feedback (FB) control amount calculation unit 8, a control switching unit 9, and a switching control signal generation. And 10. The learning control unit 110 includes a neural network data file 111, a learning data processing unit 112, a learning data file 113, and a learning data extraction unit 114.

  The control device 50 can be configured to include an arithmetic device, a storage device, an input circuit, and an output circuit (not shown) as hardware. For example, the arithmetic device can be configured by a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), or the like.

  Typically, the above-mentioned arithmetic device executes software (program) stored in a storage device such as a ROM (Read Only Memory), as shown in FIG. In addition, each function can be realized by cooperating with other hardware such as a memory device, an input circuit, and an output circuit. Further, some or all of the functions of the functional blocks can be realized by dedicated electronic circuits (hardware). The hardware for realizing the functions of the output control unit 100 and the learning control unit 110 may be mounted on the same chip, or may be divided and mounted on a plurality of chips.

  The output control unit 100 controls the output value X to the command value Xr by selection of feedback control and feedforward control. In particular, in the output control unit 100, feedback control and feedforward control are switched so as to suppress overshoot or undershoot that occurs when controlling the output value X in accordance with a change in the command value Xr.

  First, a basic control operation by the output control unit 100 will be described. In the following, it is assumed that the control operation by the output control unit 100 is executed every fixed time interval Δt, that is, every control cycle, and the nth control cycle (n: natural number) and the previous control cycle It is written as the (n-1) -th. In the following, by appending subscripts (n) and (n-1), it is shown that the value is at a specific control cycle.

  The subtraction unit 7 calculates the deviation ΔX by subtracting the detected output value X from the command value Xr in each control cycle. Accordingly, ΔX (n) = Xr (n) −X (n). The calculated deviation ΔX (n) is input to the FB control amount calculator 8.

FIG. 2 is a functional block diagram further illustrating the configuration of the FB control amount calculation unit 8. As shown in FIG.
Referring to FIG. 2, the FB control amount calculation unit 8 includes a proportional operation unit 8A, an integration operation unit 8B, a differentiation operation unit 8C, an addition unit 8D, and a limiter 8E.

  The proportional operation unit 8A outputs the product of the proportional gain Kp and the deviation ΔX (n) as a proportional control amount. The integral operation unit 8B outputs the product of the integral value of the deviations ΔX (1) to ΔX (n) and the integral gain Ki as an integral control amount. The differential operation unit 8C outputs the product of the difference value between the deviations ΔX (n) and ΔX (n-1) and the differential gain Kd as a differential control amount.

  The addition unit 8D adds the proportional control amount, the integral control amount, and the differential control amount to calculate the FB control amount DTfb. The FB control amount DTfb is limited within the range of 0 to 1.0 by the limiter 8E. That is, when the output value of the adding unit 8D is larger than 1, it is corrected to DTfb = 1.0. Similarly, if the output value of the adding unit 8D is negative, it is corrected to DTfb = 0, but if the output value of the adding unit 8D is within the range of 0 to 1.0, the output value is directly changed to DTfb and Be done.

  The FB control amount calculation unit 8 calculates the FB control amount DTfb in each control cycle (0 ≦ DTfb ≦ 1.0). The FB control amount DTfb (n) is input to the control switching unit 9 of FIG.

  In the configuration example of FIG. 2, the FB control amount calculating unit 8 realizes feedback control by so-called PID (Proportional Integral Differential) control, but feedback control may be realized by other methods. The proportional gain Kp, the integral gain Ki, and the differential gain Kd are read into the FB control amount calculator 8 at a start of operation of the DC / DC converter 11, as described later.

  Referring again to FIG. 1, the flag determination unit 5 sets a plurality of flags used for switching between feed forward control and feedback control.

  According to the flag Fff set by the flag determination unit 5, the FF control amount calculation unit 6 generates the FF control amount DTff and the threshold value Xth used for determining the operation of the FF control.

  The control switching unit 9 compares the FF control amount DTff from the FF control amount calculation unit 6 and the FB control from the FB control amount calculation unit 8 based on the comparison between the output value X and the threshold value Xth from the FF control amount calculation unit 6. One of the amounts DTfb is selectively output as the control amount DT in the control period.

  Switching control signal generation unit 10 generates a control signal of DC / DC converter 11 in accordance with control amount DT output from control switching unit 9 in each control cycle. For example, the DC / DC converter 11 controls the output value X by the duty ratio of a switching element (not shown) such as a transistor, that is, the ratio of the on time in each switching cycle of the switching element turned on and off at a constant frequency. Configured to Therefore, the control signal generated by the switching control signal generation unit 10 is an on / off control signal of the switching element, more specifically, a voltage pulse signal having a change in duty ratio, having a duty ratio indicated by the control amount DT. be able to. In the present embodiment, the output value X is increased when the duty ratio of the switching element is increased, and the output value X is decreased when the duty ratio is decreased.

  FIG. 3 is a flowchart for explaining the operation of the flag determination unit 5. The control process according to the flowchart shown in FIG. 3 is executed every control cycle.

  Referring to FIG. 3, flag determination unit 5 determines which of flag Fff for extracting a control cycle in which command value Xr has changed, a period after increase of command value Xr, and a period after decrease Control the setting of the flag Fw for determining the Furthermore, the flag Fq is used to limit the operation period of feed forward control described later.

  When the flag determination unit 5 receives the command value Xr in the control cycle in step (hereinafter, simply referred to as "S") 100, in step S110, the command value Xrp in the previous control cycle (hereinafter, the previous command) It is determined whether or not Xr> Xrp is established between the value Xrp and the value Xrp.

  If Xr> Xrp (YES in S110), that is, if the command value Xr rises in the current control cycle, the flags Fff = 1, Fw = 1, and Fq = 0 are set in S120. On the other hand, when Xr ≦ Xrp (at the time of NO determination at S110), it is determined at S150 whether or not Xr <Xrp holds, contrary to S110.

  When Xr <Xrp (when YES in S150), that is, when command value Xr decreases in the current control cycle, flag determination unit 5 causes flag Fff = 1, Fw = −1, Fq in S120. Set to = 0.

  When Xr = Xrp, that is, when the command value Xr does not change from the previous control cycle, NO is determined in both S110 and S150, so Fff = 0 is set in S170. In S170, the flags Fw and Fq are maintained at the current values.

  After the process of S120, S160 or S170, the flag determination unit 5 outputs the flags Fff, Fw, Fq, the command value Xr, and the previous command value Xrp in S130. Furthermore, in S140, Xrp = Xr is set in preparation for the next control cycle.

  As a result, the flag Fff is set to "1" only in the control cycle in which the command value Xr has changed (increased or decreased), and is set to "0" in the other control cycles. The flag Fq is cleared to "0" in a control cycle in which the command value Xr has changed (increased or decreased).

  The flag Fw is set to “1” in the control cycle in which the command value Xr has risen, and is set to “−1” after the control cycle in which the command value Xr has fallen. The value of the flag Fw does not change in a control cycle in which the command value Xr does not change. Therefore, when Fw is set to 1 when command value Xr is set from 0 (initial value) to a positive value at the start of operation of DC / DC converter 11, Fw is subsequently reduced until command value Xr decreases. It is maintained at = 1. When the command value Xr is lowered, it is updated to Fw = -1 and is maintained at Fw = -1 until the command value Xr is raised. Thus, after the control of the DC / DC converter 11 is started, the flag Fw is set to “1” or “−1”.

  Referring back to FIG. 1, FF control amount calculation unit 6 generates FF control amount DTff and threshold value Xth in the control cycle of flag Fff = 1 set by flag determination unit 5. That is, the FF control amount calculation unit 6 calculates the FF control amount DTff and the threshold value Xth in a control cycle in which the command value Xr changes (increases or decreases). The FF control amount DTff is a fixed value maintained until the value of the flag Fw changes, and is calculated from at least the command value Xr and the previous command value Xrp. For example, it is possible to create in advance a look-up table or the like that outputs the FF control amount DTff and the threshold value Xth, using the command value Xr and the previous command value Xrp as variables.

  Alternatively, in a control cycle in which command value Xr changes (increases or decreases), an allowable limit of overshoot or undershoot, ie, an allowable upper limit of an absolute value of deviation ΔX (ΔX = Xr−X) It is also possible to set The allowable chute amount Xmsh indicates the allowable upper limit of the absolute value of the deviation of the output value X from the command value Xr. Therefore, the threshold value Xth is set such that the absolute value of the difference between the threshold value Xrh and the command value Xr is smaller than the allowable chute amount Xmsh. The FF control amount DTff may be calculated from the command value Xr, the previous command value Xrp, and the allowable chute amount Xmsh. Also in this case, it is possible to create in advance a lookup table or the like in which the command value Xr, the previous command value Xrp, and the allowable chute amount Xmsh are variables.

  In the control cycle in which the command value Xr does not change, the values of the FF control amount DTff and the threshold value Xth are maintained, so the FF control amount calculation unit 6 needs to execute calculation processing of the FF control amount DTff in all control cycles. Absent.

  The threshold value Xth is set to Xth> Xr in the control cycle in which the command value Xr is increased, and the value is maintained during the flag Fw = 1. Conversely, the threshold value Xth is set to Xth <Xr in a control cycle in which the command value Xr is lowered, and the value is maintained during the flag Fw = -1. That is, the period of Fw = 1 corresponds to the “first period”, and the period of Fw = −1 corresponds to the “second period”.

  FIG. 4 is a flowchart illustrating the operation of the control switching unit 9. The control process according to the flowchart shown in FIG. 4 is executed every control cycle.

  Referring to FIG. 4, control switching unit 9 performs FF control amount DTff and threshold value Xth from FF control amount calculation unit 6, FB control amount DTfb from FB control amount calculation unit 8, and flag Fw from flag determination unit 5 in S 200. Receive Furthermore, the output value X in the current control cycle is also input to the control switching unit 9 at S200.

  At S210, the control switching unit 9 determines whether the flag Fw = 1. In the period of Fw = 1 (at the time of YES determination in S210), whether or not the output value X is higher than the threshold Xth in S220, ie, over the threshold Xth set higher than the command value Xr as described above. It is determined whether a shoot has occurred.

  On the other hand, in the period of not Fw = 1, that is, in the period of Fw = -1 (at the time of NO determination of S210), whether or not the output value X is lower than the threshold Xth according to S260 It is determined whether an undershoot below the threshold value Xth set lower than the command value Xr has occurred.

  At the time of detection of X> Xth at S220 (at the time of YES determination at S220), feedforward control is selected, and at S230, the control amount DT = DTff is set. Similarly, at the time of detection of X <Xth at S260 (at the time of YES determination at S260), feedforward control is selected, and at S270, the control amount DT = DTff is set.

  On the other hand, feedback control is selected in each of non-detection of X> Xth in S220 (NO determination in S220) and non-detection of X <Xth in S260 (NO determination in S260), respectively. Thus, the control amount DT = DTfb is set.

  Furthermore, the control switching unit 9 outputs the control amount DT (duty ratio) set in S230, S250, or S270 in S240 to the switching control signal generation unit 10 (FIG. 1), and End the processing of.

  According to the control process of FIG. 4, in a period after the command value Xr rises, while feedforward control is selected in an overshoot occurrence period in which the output value X is higher than the threshold value Xth (Xth> Xr), In other periods, feedback control is selected. On the contrary, in the period after the command value Xr decreases, feedforward control is selected in the occurrence period of the undershoot where the output value X is lower than the threshold value Xth (Xth <Xr), while in the other period , Feedback control is selected.

  As a result, in the control device according to the first embodiment, even when the command value Xr changes, the feedback control is initially coped with, but when overshoot or undershoot occurs that exceeds the threshold value Xth, feedback control is performed. Can be switched to feed forward control. Therefore, when overshoot or undershoot occurs exceeding the threshold value Xth as compared with the case where the feedforward control and the feedback control are added when changing the command value Xr as in Patent Document 1, the control amount DT is increased. By changing the value promptly, the output value X can be made to approach the command value Xr quickly. In addition, the occurrence of the deviation ΔX of the output value X immediately after the change of the command value Xr can be prevented by the fact that the feedforward control is excessive due to a factor such as a nonconformity of gain.

  As described above, in the control cycle at the start of operation of DC / DC converter 11, the change in command value Xr is a positive value corresponding to the original target value from command value Xr from 0 (initial value) Or, the change set to negative value is included. That is, even when the command value Xr is held constant after the start of control of the DC / DC converter 11, the above-described feedback according to the FF control amount DTff and the threshold value Xth set in the control cycle at the start of the operation By switching between control and feedforward control, output value X can be controlled to suppress overshoot or undershoot with respect to command value Xr.

  Alternatively, it is possible to further restrict the selection of feedforward control by operating the control switching unit 9 according to the flowchart of FIG.

  FIG. 5 is a flowchart for explaining a modification of the operation of the control switching unit 9. The control process according to the flowchart shown in FIG. 5 can be executed every control cycle, instead of the control process of FIG.

  Referring to FIG. 5, in addition to FIG. 4, control switching unit 9 further executes S205, S225, S255, S265, and S275 related to flag Fq. As described in FIG. 3, the flag Fq is cleared to “0” in the control cycle in which the command value Xr has changed.

  The control switching unit 9 sets the flag Fq = 1 in S225 and S275 when YES is determined in S220 or when YES is determined in S260, that is, when feedforward control is selected. On the other hand, at the time of NO determination of S220 or at NO determination of S260, that is, at the time of selection of feedback control, the control switching unit 9 determines whether or not the flag Fq = 1 in S265. If (YES at S265), Fq is set to 2 at S255.

  Therefore, the flag Fq is set to "1" when the feedforward control is operated due to the occurrence of overshoot or undershoot larger than the threshold value Xth after the flag Fq is set to "0" in the control cycle in which the command value Xr changes. . Thereafter, when the overshoot or undershoot is settled and feedback control is selected again by application of feed forward control, the flag Fq is set to 2. When the flag Fq = 2, S265 returns NO, and the value of the flag Fq does not further increase.

  When the control switching unit 9 receives the flag Fq in addition to the FF control amount DTff, the threshold value Xth, the FB control amount DTfb, the flag Fw, and the output value X similar to S200 by S202 executed instead of S200. First, at S205, it is determined whether or not the flag Fq = 2. If Fq = 2 (YES in S205), feedback control is selected regardless of the output value X and the threshold value Xth, and the control amount DT = DTfb is set in S245.

  When the flag Fq is not 2 (when the determination is NO in S205), the processing of S210 onward similar to that of FIG. 4 is executed with the addition of S225, S255, S265, and S275 for controlling the flag Fq.

  As described above, according to the control process shown in FIG. 5, the control switching unit 9 responds to the overshoot or undershoot that first occurs in the period after the command value Xr changes (increases or decreases). Only feedforward control is applied, and thereafter, even if overshoot of X> Xr (S220) or undershoot of X <Xr (S260) occurs, reselection of feedforward control is inhibited, and feedback is performed. Control will be selected.

(Setting of FF control amount using learning)
In the control device of the power converter according to the first embodiment, the threshold value Xth and the FF control amount DTff set in the control cycle in which the command value Xr changes (increases or decreases) is maintained, and thus updating is performed every control cycle. While setting the threshold Xth and the FF control amount DTff to appropriate values affects the control accuracy while avoiding an increase in calculation load due to the above. In the following, a method of setting the threshold value Xth and the FF control amount DTff with high accuracy by learning of the past results (representative example, learning by a neural network) will be described.

  FIG. 6 is a flowchart for explaining the control operation of the control device 50 at the start of control of the DC / DC converter 11.

  Referring to FIG. 6, control device 50 initializes the previous command value Xrp = 0 used by flag determination unit 5 in FIG. 3 in S300, and initializes control amount DT = 0 in S310. In S310, a parameter α for setting the allowable chute amount Xmsh is further set. The allowable chute amount Xmsh is set by the product of the command value Xr and the parameter α (Xmsh = Xr · α).

  Furthermore, in S310, Xth = X0 is set as an initial value of the threshold Xth. As described later, since the threshold value Xth is newly set by the FF control amount calculating unit 8 when the command value Xr is changed, the period in which the initial value X0 is valid is limited. Therefore, it is preferable to set the initial value X0 to a value larger than the range in which the threshold Xth is normally set so that only feedback control is used until the command value Xr changes.

  In S320, control device 50 initializes a chute amount Xsh = 0 obtained by learning data extraction unit 114 described later, and in S330, proportional gain Kp used in FB control amount calculation unit 8 (FIG. 2), integral gain The values of Ki and the differential gain Kd are set. Furthermore, at S340, control device 50 determines the structure of a neural network model for FF control amount calculation unit 6 to set FF control amount DTff and threshold value Xth as initial values for learning data processing unit 112 described later. Set the parameter value. Specifically, the number of neurons (for example, three) in the input layer of the neural network model, the number of intermediate layers (hidden layers), the number of neurons in each layer of the intermediate layers (hidden layers), and the number of neurons in the output layer ( For example, 2) is set.

  FIG. 7 is a flow chart for explaining control processing using a neural network model by the FF control amount calculation unit 6. The control process shown in FIG. 7 is performed in each control cycle.

  Referring to FIG. 7, the FF control amount calculation unit 6 receives the flags Fff and Fw, and the command value Xr and the previous command value Xrp of the current control cycle from the flag determination unit 5 at S400. Further, in S410, the allowable chute amount Xmsh = Xr · α is set from the parameter α set in S310 of FIG. 6 and the command value Xr read in S400. Furthermore, in S420, it is determined whether or not the flag Fff = 1.

  When Fff = 1 (YES in S420), that is, in the control cycle in which the command value Xr changes (increases or decreases), the FF control amount calculation unit 6 determines the neural network model from the neural network data file 111 in S430. Load

  FIG. 8 shows a conceptual diagram for explaining a configuration example of the neural network model used in the FF control amount calculating unit 8. As shown in FIG.

  Referring to FIG. 8, neural network model 120 includes K (K: an integer of 2 or more) neurons forming an input layer, L (L: natural numbers) neurons forming an output layer, and an input layer. And a plurality of neurons that constitute a hidden layer connected between the output layers. Furthermore, the hidden layer is configured by interconnecting up to J (M, J: integers of 2 or more) neurons across M layers. By determining the above-mentioned number parameters K, L, M, J, the structure of the neural network model 120 is set. Thus, the structure of the neural network model 120 can be arbitrarily set according to the number of input layers, hidden layers, output layers, and the number of neurons in each layer. An activation function is input to each neuron represented by a circle symbol in FIG. For example, a sigmoid function can be used for the activation function, but any known activation function can be applied.

  In the example of FIG. 8, K = 3 and L = 2, and three neurons N11 to N13 are disposed in the input layer, and neurons N21 and N22 are disposed in the output layer. The number of input layers and output layers, and the number of hidden layers and neurons in each phase are set in S340 of FIG. 6 to set the structure of the neural network model 120. Furthermore, in S430, the weight coefficient between each neuron is read out from the neural network data file 111, whereby the neural network model 120 is determined.

  As described later, each weighting factor is determined by machine learning using a plurality of learning data obtained from past actual values, and the value of the neural network data file 111 is stored.

  Referring again to FIG. 7, the FF control amount calculation unit 8 sets the command value Xr, the previous command value Xrp, and the allowable chute amount Xmsh in the input layer of the neural network model 120 at S440. Specifically, in the neural network model 120 of FIG. 8, the command value Xr is input to the neuron N11, the previous command value Xrp is input to the neuron N12, and the allowable shoot amount Xmsh calculated in S410 is input to the neuron N13. It is input.

  In S450, the FF control amount calculation unit 8 inputs the command value Xr in S440, the previous command value Xrp, and the allowable chute amount Xmsh with respect to the neural network model 120 in which the weighting factor read in S430 is set. By setting in the layer, the threshold Xth and the FF control amount DTff are obtained in the output layer. Specifically, in the neural network model 120 of FIG. 8, the threshold value Xth is obtained for the neuron N21, and the FF control amount DTff is obtained for N22.

  In S460, the FF control amount calculation unit 8 outputs the threshold Xth and the FF control amount DTff obtained in S450 to the control switching unit 9.

  On the other hand, when Fff = 0 (NO in S420), that is, in a control cycle in which the command value Xr does not change, the FF control amount calculation unit 6 skips S430 to S450 and performs the process in S460. Advance. Therefore, the process using the neural network model is not executed, and the threshold Xth and the FF control amount DTff obtained in the first control cycle after the change to the current command value Xr are not updated, and control is performed according to S460. It is output to the switching unit 9.

  Next, neural network learning by the learning control unit 110 will be described. FIG. 9 shows a conceptual diagram for explaining input values and output values in machine learning in the neural network model 120.

  Referring to FIG. 9, every time the command value Xr changes, the chute amount Xsh in the actual control result is obtained. Then, the command value Xr, the command value Xr before change, the threshold value Xth used for feedforward control, and the FF control amount DTff in the control result are extracted as learning data together with the chute amount Xsh.

  Then, the shoot amount Xsh is taken as an input value to the neuron N13 of the input layer. The command value Xr, the previous command value Xrp, the threshold value Xth, and the FF control amount DTff are respectively set to the values of the neurons N11, N12, N21 and N22 as in FIG.

  FIG. 10 is a flowchart illustrating control processing by the learning data extraction unit 114. The control process shown in FIG. 10 is performed every control cycle.

  Referring to FIG. 10, in S500, the learning data extraction unit 114 uses the command value Xr, the previous command value Xrp, the flags Fff and Fw, the threshold value Xth, and the FF control amount DTff used by the output control unit 100 in the control cycle. When the output value X is received, it is determined in S510 whether or not the flag Fff = 0.

  In a control cycle of Fff = 0 (when YES in S510), S520 to S540 for extracting the shoot amount, that is, the maximum value (absolute value) of the deviation ΔX with respect to the command value Xr, are executed. The learning data extraction unit 114 obtains a product of a value obtained by subtracting the command value Xr from the output value X and the flag Fw as a variable PT for extracting a shoot amount (PT = Fw · (X−Xr)). As described above, when the command value Xr rises, the flag Fw is set to “1” until the command value Xr falls. Therefore, in the period of Fw = 1, Fw · (X−Xr) = X−Xr, and the overshoot amount with respect to the command value Xr is set as the variable PT.

  On the other hand, the flag Fw is set to a period "-1" until the command value Xr rises after the command value Xr decreases. Therefore, in the period of Fw = −1, Fw · (X−Xr) = Xr−X, and the undershoot amount with respect to the command value Xr is set as the variable PT.

  In S530, the learning data extraction unit 114 compares the variable PT calculated in S520 with the shoot amount Xsh. As described later, the shoot amount Xsh is initialized to 0 (Xsh = 0) according to the control cycle of the flag Fff = 1, that is, the change (increase or decrease) of the command value Xr in S580. Then, if PT> Xsh (YES in S530), Ssh is updated to Xsh = PT. On the other hand, when PT ≦ Xsh (when NO in S530), the current value of the shoot amount Xsh is maintained.

  On the other hand, in the control cycle of the flag Fff = 1 (at the time of NO determination at S550), it is determined at S550 whether or not the shoot amount Xsh> 0. If Xsh> 0 (YES in S550), the command value Xr, the previous command value Xrp, the threshold value Xth, the FF control amount DTff, and the shoot amount Xsh received in S500 in S570 as learning data in S570. It is output. The output learning data is separately written for each set into the learning data file 113. Then, at S580, the shoot amount Xsh is initialized to zero.

  As a result, when the threshold Xth and the FF control amount DTff are set in response to the increase or decrease of the command value Xr in the control cycle of the flag Fff = 1, then the flag Fff = 1 is set to the threshold Xth and the FF In the period until the control amount DTff is updated, the maximum value of the overshoot amount (X-Xr) or the maximum value of the undershoot amount (Xr-X) is extracted as the actual shoot amount Xsh by S520 to S540. Be done. Then, the obtained shoot amount Xsh and the command value Xr, the previous command value Xrp, the threshold value Xth, and the FF control amount DTff used for control are extracted as one set of learning data. The extracted learning data is stored in the learning data file 113 for each set.

  The learning data processing unit 112 shown in FIG. 1 uses the learning data stored in the learning data file 113 to determine the weighting factor between each neuron of the neural network model 120 shown in FIG.

  FIG. 11 is a flowchart illustrating control processing by the learning data processing unit 112. The control process shown in FIG. 11 is executed each time a machine learning execution instruction is generated. For example, the execution instruction can be generated at startup of the power converter (DC / DC converter 11). Alternatively, it is also possible to generate execution instructions periodically (every day, every week, etc.). As a result, it becomes possible to update the weighting coefficients in the neural network model 120 by adding learning data newly accumulated after the previous machine learning.

  Referring to FIG. 11, the learning data processing unit 112 sets the structure of the neural network model 120 shown in FIG. 9 at S600. As described above, in the input layer of the neural network model 120, the command value Xr, the previous command value Xrp before change, and the chute amount Xsh obtained as the control result are set in the input layer. In the output layer, the threshold value Xth and the FF control amount DTff used when the shoot amount Xsh is generated are set.

  At S610, the learning data processing unit 112 sets an initial value of a weighting factor among the plurality of neurons in the neural network model 120 using, for example, a random number. As described above, for example, a sigmoid function can be used for the function of each neuron.

  At S620, the learning data processing unit 112 reads from the learning data file 113 a command value Xr constituting one set of learning data, a previous command value Xrp, a chute amount Xsh, a threshold value Xth, and an FF control amount DTff. In S630, the threshold value Xth and the FF control amount DTff obtained in the output layer of the neural network model 120 by giving the command value Xr, the previous command value Xrp, and the shoot amount Xsh to the input layer for each set of learning data. The weighting factor of the neural network model 120 is corrected using, for example, an error back propagation method so that an error between the learning data threshold value Xth and the FF control amount DTff is eliminated.

The learning data processing unit 112 executes convergence determination based on the root mean square value of the error (root mean square error) in S640. For example, for the threshold Xth and the FF control amount DTff in the learning data of the i-th set (i: 1 ≦ i ≦ N natural number) of the total N sets of learning data (N: natural number of 2 or more), the output layer If the square error Ei = (Yi−βi) ² is defined using the calculated value βi obtained in and the actual learning data Yi, the average value of EEi obtained by integrating Ei from i = 1 to i = N If (ΣEi) / N is smaller than a predetermined determination value ε (for example, 0.001), S640 can be determined as YES.

  When the root mean square error does not converge, if (.SIGMA.Ei) / N> .epsilon., S640 is determined as NO, and the weighting factor is corrected in S630. The processes of S630 and S640 are repeated until a weighting factor is obtained at which the mean squared error converges.

  The weighting factor when the root mean square error converges (at the time of YES determination at S640) is written out to the neural network data file 111 at S650. The weighting factor is used for setting the threshold value Xth and the FF control amount DTff as the FF control amount calculation unit 6 reads it from the neural network data file 111 in S430 of FIG.

  As described above, in the control device of the power converter according to the present embodiment, the threshold value Xth and the FF control amount DTff fixedly used after setting in the control cycle in which the command value Xr changes (increases or decreases) It can be set to an appropriate value using learning (neural network learning) using actual results data of. As a result, it is possible to avoid an increase in calculation load due to updating for each control cycle and to improve control accuracy.

(Description of control operation example)
Next, an example of the control operation by the control device for the power converter according to the first embodiment will be described using FIGS. 12 to 16.

  FIGS. 12 and 13 show an example of the control operation when suppressing the overshoot in accordance with the increase of the command value Xr. FIG. 12 shows an operation example corresponding to only feedback control as a comparative example, while FIG. 13 shows an operation example corresponding to switching between feedback control and feedforward control according to the first embodiment. . The vertical axis in FIG. 12 and FIG. 13 also shows a value (Xr + Xsmh) obtained by adding the allowable chute amount Xmsh to the command value Xr after the increase.

  Referring to FIG. 12, in accordance with the increase of command value Xr, output value X is controlled using FB control amount DTfb calculated by FB control amount calculation unit 8. In FIG. 12, while feed-back control is used continuously, feed-forward control is not used. The output value X settles to the command value Xr after rising after occurrence of an overshoot larger than the allowable chute amount Xmsh.

  Referring to FIG. 13, in the first embodiment, feedforward control is selected in a period of X> Xth with respect to threshold value Xth set in a control cycle in which command value Xr is increased. Therefore, the control switching unit 9 (FIG. 1) sets the FF control amount DTff as the control amount DT in a period Tff in the drawing, and sets the FB control amount DTff as the control amount DT in a period other than Tff.

  FIG. 14 shows a waveform chart for comparing the transition of the control amount DT set by the control switching unit 9 in FIG. 12 and FIG.

  The controlled variable (ie, the duty ratio) DT is shown on the vertical axis of FIG. The horizontal axis is a time axis obtained by enlarging the horizontal axes of FIGS. 12 and 13. Further, in FIG. 14, the dotted line indicates the transition of the control amount DT in FIG. 12 (comparative example), and the solid line indicates the transition of the control amount DT in FIG. 13 (first embodiment).

  When command value Xr is increased at time t0, also in the first embodiment, control of the output value is started by feedback control on command value Xr after the increase. Therefore, a period occurs in which the control amount DT matches between the solid line and the dotted line. Further, as described above, in the control cycle corresponding to time t0, the FF control amount calculation unit 8 calculates the threshold Xth and the FF control amount DTff for corresponding to the command value Xr after the increase.

  In the first embodiment (solid line), a period Tff in which feed forward control is applied is provided as X> Xth. Thereby, particularly at the start of the period Tff, the control amount DT is rapidly reduced to the FF control amount DTff in order to reduce the output value X, that is, to suppress an overshoot. As a result, as shown in FIG. 13, the rise of the output value X is rapidly suppressed. On the other hand, in the feedback control, the decrease of the control amount DT is also gradual, so that the overshoot amount also becomes relatively large as shown in FIG.

  Since the FF control amount DTff is maintained at the value set at the control cycle at time t0 throughout the period Tff, the calculation process for calculating the FF control amount DTff is unnecessary. Also, by setting the threshold value Xth using the neural network model 120 including the allowable chute amount Xmsh in the input value to the input layer, the overshoot amount is allowed chute by starting application of the feedforward control according to the threshold value Xth. It is understood that it can be made smaller than the quantity Xmsh.

  When the output value X decreases due to the application of feedforward control and X <Xth, the period Tff in which the feedforward control is applied is ended. Then, feedback control is resumed according to the deviation ΔX (ΔX = Xr−X) at that time. Thereafter, as shown in FIG. 13, the output value X is settled to the command value Xr after the increase by feedback control.

  FIGS. 15 and 16 show an example of the control operation when suppressing the undershoot according to the decrease of the command value Xr. Similar to FIG. 13, FIG. 15 shows an operation example in the case of dealing with only feedback control as a comparative example. On the other hand, FIG. 16 shows an operation example corresponding to switching between feedback control and feedforward control according to the first embodiment, as in FIG. The vertical axis in FIG. 15 and FIG. 16 also shows a value (Xr−Xsmh) obtained by subtracting the allowable chute amount Xmsh from the command value Xr after the increase.

  Referring to FIG. 15, in accordance with the decrease of command value Xr, output value X is controlled using FB control amount DTfb calculated by FB control amount calculation unit 8. In FIG. 15, feedback control is continuously used as in FIG. As a result, the output value X settles to the command value Xr after reduction after the occurrence of an undershoot larger than the allowable chute amount Xmsh.

  Referring to FIG. 16, in the first embodiment, feedforward control is selected in a period of X <Xth with respect to threshold value Xth set in the control cycle in which command value Xr is decreased. Therefore, the control switching unit 9 (FIG. 1) sets the FF control amount DTff as the control amount DT in the period Tff in the drawing, and sets the FB control amount DTfb as the control amount DT in the period other than Tff.

  The reduction of the output value X is rapidly suppressed by the application of feedforward control, and the output value X turns to increase in the period Tff. Thus, it is understood that the undershooting amount can be made smaller than the allowable chute amount Xmsh, as described with reference to FIGS. 13 and 14.

  Although illustration is omitted, also in FIG. 16, since the FF control amount DTff is maintained at the value set at the time of decrease of the command value Xr, the FF control amount DTff is calculated in the period Tff. No arithmetic processing is required.

  When the output value X rises, that is, X> Xth due to the application of the feedforward control, the period Tff in which the feedforward control is applied is ended. Then, feedback control is resumed according to the deviation ΔX (ΔX = Xr−X) at that time. Thereafter, the output value X settles to the command value Xr after reduction by feedback control.

  As understood from the comparison of FIG. 12 and FIG. 13 and the comparison of FIG. 15 and FIG. 16, according to the first embodiment, both the increase of the command value Xr and the decrease of the command value Xr The overshoot amount and the undershoot amount can be suppressed to the allowable shoot amount Xmsh or less by switching between feedforward control and feedback control according to the threshold value Xth set corresponding to the allowable chute amount Xmsh. In addition, since it is not necessary to calculate the FF control amount for each control cycle as in Patent Document 1 in the application period of the feedforward control, the calculation processing can also be reduced.

Modification of Embodiment 1
In the first embodiment, the FF control amount calculation unit 8 sets the threshold value Xth and the FF control amount DTff by machine learning using learning data based on the control result of the DC / DC converter 11 during operation. In the modification of the first embodiment, learning data generation control for collecting initial learning data before the start of use of the control device will be described.

  FIG. 17 is a functional block diagram for explaining a configuration of a control device of a power converter according to a modification of the first embodiment.

  Referring to FIG. 17, control device 51 according to the modification of the first embodiment includes an output control unit 100A, a learning control unit 110A, and a learning data generation control unit 130.

  The learning control unit 110 </ b> A includes the learning data file 113 and the learning data extraction unit 114 of the learning control unit 110 in FIG. 1. The learning data file 113 has a plurality of sets of test parameters configured by the command value Xr, the threshold value Xth, and the FF control amount DTff for executing learning data generation control, that is, a plurality of test patterns, It is stored in advance.

  The learning data generation control unit 130 includes a parameter setting unit 12. The parameter setting unit 12 sequentially reads the plurality of test patterns from the learning data file 113, and outputs the DC / DC converter 11 according to the command value Xr, the threshold value Xth, and the FF control amount DTff in the read test pattern. Control. Then, in the control operation, as the learning data extraction unit 114 operates as described in the first embodiment, one set of learning data is extracted corresponding to one test pattern. The extracted learning data is written to the learning data file 113.

  That is, the output control unit 100A differs from the output control unit 100 of FIG. 1 in that the arrangement of the FF control amount calculating unit 6 is omitted. The threshold value Xth read from the learning data file 113 by the parameter setting unit 12 and the FF control amount DTff are supplied to the control switching unit 9.

  Similarly, as for the command value Xr, the value of the test parameter read out from the learning data file 113 by the parameter setting unit 12 is input to the subtraction unit 7 and the flag determination unit 5. That is, in the control device 51, each time the parameter setting unit 12 reads the next test pattern from the learning data file 113, the command value Xr changes (increases or decreases).

  FIG. 18 is a flowchart for explaining an operation at the start of learning data generation control by the control device 51 of the power converter according to the modification of the first embodiment.

  Referring to FIG. 18, control device 51 sets an interval time Tint used by parameter setting unit 12 in S710. The interval time Tint corresponds to the test pattern switching cycle described above. The interval time Tint is predicted and set in advance so as to be longer than the time required for the output value X to settle to the command value Xr by the control operation using the test parameter.

  Further, the control device 51 initializes the previous command value Xrp = 0 used by the flag determination unit 5 in S720, as in S300 of FIG. Also, in S730, as in S320 of FIG. 6, the chute amount Xsh relating to the learning data extraction unit 114 is initialized to 0. Furthermore, in S740, the controller 51 sets values of the proportional gain Kp, the integral gain Ki, and the differential gain Kd used in the FB control amount calculation unit 8 (FIG. 2) as in S330 of FIG.

  In FIG. 18, unlike the normal control start time (FIG. 6), the FF control amount DTff and the threshold value Xth are previously determined for each test pattern, and thus the processing of S310 and S340 is not executed. Thus, the interval time Tint for specifying the test pattern switching timing is set.

  FIG. 19 is a flow chart for explaining the operation in the learning data generation control of the parameter setting unit 12. The control process shown in FIG. 19 is activated in response to the start of learning data generation control.

  Referring to FIG. 19, the parameter setting unit 12 acquires the current time Tx in S800, and updates the old time Tp to Tx in S810. Step S810 corresponds to a state in which a timer value for detecting the elapse of the interval time Tint is initialized. Furthermore, the parameter setting unit 12 reads test parameters for one test pattern of the plurality of test patterns described above from the learning data file 113. Thereby, the command value Xr, the threshold value Xth, and the FF control amount DTff for one pattern are read out.

  After reading one test pattern in S820, the parameter setting unit 12 determines in S830 whether all the plurality of test patterns stored in advance in the learning data file 113 have been read.

  The parameter setting unit 12 uses the test parameters (the command value Xr, the threshold value Xth, and the FF control amount DTff) read out at S820 until the reading of all the test patterns is completed (at the time of NO determination at S830). Execute control operation of DC converter.

  The parameter setting unit 12 acquires the current time Tx in S850, and compares the time difference between the old time Tp set in S810 and the current time Tx (S860) in S860 with the interval time Tint set in S710. . The command value Xr, the threshold value Xth, and the FF control amount DTff read in S820 are output from the parameter setting unit 12 while | Tx−Tp | <Tint (when YES in S860).

  The output control unit 100A controls the output value X of the DC / DC converter 11 using the command value Xr, the threshold value Xth, and the FF control amount DTff output in S870. At this time, in the learning data extraction unit 114 described in the first embodiment, the control process of FIG. 10 is executed.

  DC / DC converter using test parameters (command value Xr, threshold value Xth, and FF control amount DTff) read out at S820 when | Tx−Tp | ≧ Tint (when NO at S860) When the control of is continued for the interval time Tint, the process is returned to S800.

  Then, at S800 and S810, the old time Tp is updated so as to indicate the start time of the control operation according to the next test pattern, and at S820, the next test pattern is read from the learning data file 113. Thereby, the command value Xr, the threshold value Xth, and the FF control amount DTff for one pattern are read out. In response to the switching of the test pattern in S820, the output control unit 100A generates a control cycle in which the flag Fff is set to 1 in accordance with the change in the command value Xr. Therefore, S570 in FIG. 10 (learning data extraction unit 114). Thus, one set of learning data is written to the learning data file 113.

  The processes of S850 to S870 and S800 to S830 are repeated until reading of all the test patterns is completed (at the time of NO determination of S830). Thus, control of the output value X corresponding to the change (rising or lowering) of the command value Xr is executed each time the test pattern is switched, and one set of learning data is written to the learning data file 113.

  When the reading of all the test patterns from the learning data file 113 is completed (YES in S830), the parameter setting unit 12 instructs the DC / DC converter 11 to stop in S840. Thus, learning data generation control for learning data collection using a plurality of tester patterns is ended.

  As described above, according to the control device of the power converter according to the modification of the first embodiment, the FF control amount calculation unit 8 sets the threshold value Xth and the FF control amount DTff for machine learning using a neural network model. It is possible to collect initial learning data of

Second Embodiment
In the first embodiment, the learning data based on the control result is extracted during the operation of the DC / DC converter 11, but the learning data may not be accumulated during the operation.

  FIG. 20 is a functional block diagram for explaining a configuration of a control device of a power converter according to the second embodiment.

  Referring to FIG. 20, control device 52 according to the second embodiment includes an output control unit 100 similar to that of FIG. 1 and a learning control unit 110B. The learning control unit 110 </ b> B includes the neural network data file 111, the learning data processing unit 112, and the learning data file 113 in the learning control unit 110 of FIG. 1.

  The learning data file 113 stores a plurality of sets of learning data in advance. For example, a plurality of sets of learning data can be obtained by executing the learning data generation control described in the modification of the first embodiment before factory shipment.

  The learning data processing unit 112 calculates the weight coefficient of the neural network model 120 shown in FIG. 8 and FIG. 9 by executing the control process of FIG. 11 and writes it to the neural network data file 111. On the other hand, in the learning control unit 110B of the second embodiment, the learning data extraction unit 114 is not disposed, and thus no learning data is accumulated during the operation of the DC / DC converter 11.

  FIG. 21 is a flow chart for explaining the control operation at the start of control of the control device 52 of the power converter according to the second embodiment.

  Referring to FIG. 21, control device 52 initializes the previous command value Xrp = 0 used by flag determination unit 5 in FIG. 3 according to S300 similar to FIG. 6, and performs control according to S310 similar to FIG. Initialize the quantity DT = 0. Further, in S310, a parameter α for setting the allowable chute amount Xmsh and an initial value (Xth = X0) of the threshold value Xth are set.

  Further, the control device 52 sets values of the proportional gain Kp, the integral gain Ki, and the differential gain Kd used in the FB control amount calculation unit 8 in S330 similar to FIG. 3. On the other hand, in the second embodiment, since the learning data extraction unit 114 is unnecessary, S320 in FIG. 3 for initializing the shoot amount Xsh = 0 is not executed.

  In S340, the control device 52 instructs the learning data processing unit 112 to use the parameter values for determining the structure of the neural network model 120, specifically, the number of neurons in the input layer of the neural network model (for example, three). The number of intermediate layers (hidden layers), the number of neurons in each layer of intermediate layers (hidden layers), and the number of neurons in the output layer (for example, 2) are set. Furthermore, by generating an instruction to execute machine learning using learning data stored in advance in the learning data file 113 to the learning data processing unit 112, a weighting factor between each neuron of the neural network model 120 is set. Ru.

  As described above, in the control device of the power converter according to the second embodiment, the feedforward control similar to that of the first embodiment is performed without accumulating learning data based on the control result during the operation of the DC / DC converter 11. Undershoot and overshoot can be suppressed by controlling the output value X by switching feedback control.

  In the controller of the power converter according to the second embodiment, the learning data stored in the learning data file 113 does not increase, so the start instruction of the control process of FIG. Can be generated each time the controller (DC / DC converter 11) starts up. Alternatively, by configuring the neural network data file 111 with a non-volatile memory, it is possible to directly store the weighting coefficient obtained by the learning data processing unit 112 in the neural network data file 111. In this case, it is also possible to omit the arrangement of the learning data processing unit 112 and the learning data file 113 by reading out the weighting factors in addition at S340 in FIG.

Third Embodiment
In the third embodiment, a modification of the switching between the feedforward control and the feedback control described in the first embodiment will be described. In the third embodiment, in the configuration shown in FIG. 1, the operation of the control switching unit 9 is changed from the first embodiment (FIG. 4). The other control functions are the same as in the first embodiment, and thus the detailed description will not be repeated.

  FIG. 22 is a flowchart for explaining the operation of the control switching unit in the control device of the power converter according to the third embodiment. The control process according to the flowchart shown in FIG. 22 is executed every control cycle.

  Referring to FIG. 22, at S900, control switching unit 9 receives FF control amount DTff, threshold value Xth, FB control amount DTfb, flags Fw and Fff, and output value X.

  Control switching portion 9 determines whether or not flag Fff = 1 in S910, and in the control cycle of Fff = 1 (when YES in S910), variable Xp indicating the output value in the previous cycle in S920. (Hereinafter, also referred to as previous output value Xp) is initialized (Xp = 0). In the control cycle of Fff = 0 (when NO in S910), S920 is skipped and the value of the previous output value Xp is maintained.

  The control switching unit 9 compares the threshold value Xth with the threshold value Xth and the output value X based on S930, S940, and S992 similar to S210, S250, and S260 in FIG. I do.

  When Fw = 1 (when YES in S930) and X> Xth (when YES in S940), the control switching unit 9 compares the output value X with the previous output value Xp in S950. When X> Xp (YES in S950), that is, while the output value X is rising, the control amount DT = DTff is set in S960. That is, feed forward control is selected. On the other hand, when X ≦ Xp (NO at S950), that is, when the output value X turns from rising to decreasing, the control amount DT is obtained from S945 even if an overshoot exceeding the threshold Xth occurs. It is set to = DTfb. That is, feedback control is selected.

  Similarly, when Fw = −1 (when NO at S930) and X <Xth (when YES at S992), control switching unit 9 compares output value X with previous output value Xp at S994. Do. When X <Xp (YES in S994), that is, while the output value X is decreasing, the control amount DT = DTff is set in S996. On the other hand, when X ≧ Xp (at the time of NO determination at S994), that is, when the output value X turns from a decrease to a rise, the control amount DT is obtained at S945 even if an undershoot exceeding the threshold Xth occurs. It is set to = DTfb.

  Furthermore, after setting the control amount DT in S960, S945, or S996, the control switching unit 9 sets the previous output value Xp used in the next control cycle as Xp = X in S970. The control switching unit 9 outputs the control amount DT (duty ratio) to the switching control signal generation unit 10 (FIG. 1) in S980, and ends the process in the control cycle.

  Next, the operation of the control switching unit 9 will be compared between Embodiment 1 and Embodiment 3 with reference to FIGS. 23 and 24. FIG. FIG. 23 shows the operation of the control switching unit 9 in the first embodiment for the overshoot, and FIG. 24 shows the operation of the control switching unit 9 in the third embodiment for the overshoot.

  As shown in FIG. 23, in the first embodiment, control amount DT = DTff is set in the period from time tx to ty where X> Xth by comparison of output value X and threshold value Xth, and the feedforward is performed. Control is selected. On the other hand, in the period up to time tx where X ≦ Xth and the period after time ty, the control amount DT = DTfb is set, and feedback control is selected.

  On the other hand, as shown in FIG. 24, in the third embodiment, of the period from time tx to ty where X> Xth, only in the period from time tx to ty # in which the output value X is rising. The control amount DT = DTff is set, and feedforward control is selected. On the other hand, in the period from time t # to ty, in the period from time tx to ty, when the output value X turns to decrease, the control amount DT =, similarly to the period up to time tx and the period after time ty. It is set as DTfb, and feedback control is selected.

  Therefore, the control device of the power converter according to the third embodiment overshoots the application period of feedforward control to which the fixed control amount (DTff) is applied without depending on the deviation ΔX from the command value Xr. Alternatively, it can be limited to a period in which the undershoot is widening. As a result, the effect of suppressing overshoot and undershoot quickly by switching from feedforward control to feedback control as described in the first embodiment is maintained, and command value Xr after suppression of overshoot or undershoot is suppressed. Control stability can be enhanced.

  The control device of the power converter described in the first to third embodiments can be applied to, for example, a power system.

  FIG. 25 is a block diagram for explaining an application example of a control device of a power converter according to the present embodiment.

  Referring to FIG. 25, DC / DC converter 11 whose output value is controlled by the control device according to Embodiment 1 and its modification, Embodiment 2 or Embodiment 3 corresponds to wind power generator 710. , DC / DC power conversion with the DC transmission line 720. In general, DC power collected from a plurality of wind power generators 710 is input to the DC / DC converter 11. Furthermore, a DC / AC converter (inverter) 720 is disposed between the DC power transmission line 720 and the AC power transmission line 740. For example, an alternating current transmission line 740 is arranged for on-shore power transmission of the power system. That is, the DC / DC converter 11 is connected to the power system.

  Although the wind power generator 710 is arranged at the ocean or the like, the generated power changes depending on the weather condition (wind power), so the input voltage fluctuation to the DC / DC converter 11 is relatively large. If the voltage fluctuation of the DC transmission line 720 becomes large according to the input voltage fluctuation to the DC / DC converter 11, there is a concern that the operation of the electric power system becomes unstable due to the voltage fluctuation of the AC transmission line 740.

  However, for example, by controlling the output voltage of the DC / DC converter 11 as the output value X using the control devices 50 to 52 described above, the DC / DC converter can be operated even if the amount of power generation by the wind power generator 710 fluctuates. Since it is possible to suppress the overshoot and undershoot of the output voltage from 11 to the DC transmission line 720, the power system can be stabilized. In addition, if an accident occurs in the AC transmission line 740, there is a concern that the DC voltage of the DC transmission line 720 may rise sharply, but the output value X of the converter 11 with respect to the command value Xr The voltage fluctuation can be suppressed by controlling DC / DC by the control device according to the present embodiment so as to suppress the overshoot of the output voltage. Since the capacity of the smoothing capacitor disposed in DC power transmission line 720 can be reduced by controlling voltage fluctuation in DC power transmission line 720 by control including such a situation, downsizing of the device and It is also possible to reduce costs.

  In the control device for a power converter according to the present embodiment, a DC / DC converter has been exemplified as a control target in the first embodiment and its modifications, and in the second and third embodiments. The invention is not limited to the DC / DC converter, and it is also possible to control other power converters such as an AC / DC converter and a DC / AC converter. That is, in the case of a power converter whose output value changes according to the control amount, the output value (voltage, current, power, etc.) of the power converter of any circuit configuration and input / output characteristics (including non-linear characteristics) Can be controlled by the control device according to the first embodiment and its modification, the second embodiment, or the third embodiment.

  Further, in the present embodiment, the feedforward control is applied by setting the threshold value Xth for both overshoot and undershoot, but the present embodiment can be applied to only one of the overshoot or the undershoot. It is also possible to apply switching between feedforward control and feedback control similar to the above. For example, the operations by the flag determination unit 5 and the FF control amount calculation unit 6 can be fixedly selected without setting the threshold Xth and the FF control amount DTff when the command value Xr decreases. It is also possible to transform the present embodiment into control corresponding to overshoot only. Similarly, when the command value Xr rises, the feedback control is fixedly selected without setting the threshold Xth and the FF control amount DTff, so that the control according to the present embodiment corresponds to only the undershoot. It is also possible to transform into

  It should be understood that the embodiments disclosed herein are illustrative and non-restrictive in every respect. The scope of the present invention is indicated not by the above description but by the scope of the claims, and is intended to include all modifications within the scope and meaning equivalent to the scope of the claims.

  5 flag determination unit, 6 FF control amount calculation unit, 7 subtraction unit, 8 FB control amount calculation unit, 8A proportional operation unit, 8B integration operation unit, 8C differentiation operation unit, 8D addition unit, 8E limiter, 9 control switching unit, DESCRIPTION OF SYMBOLS 10 switching control signal generation part, 11 converter, 12 parameter setting part, 50-52 control apparatus, 100, 100A output control part, 110, 110A, 110B learning control part, 111 neural network data file, 112 learning data processing part, 113 Learning data file, 114 learning data extraction unit, 120 neural network model, 130 initial learning data generation control unit, 710 wind power generator, 720 DC transmission line, 740 AC transmission line, DT control amount (duty ratio), DTfb FB control amount , DTff FF control amount, Xth threshold , Fff, Fq, Fw flag, Kd differential gain, Ki integral gain, Kp proportional gain, N11 to N13, N21, N22 neurons, Tff feedforward application period, Tint interval time, X output value, Xmsh permissible shoot amount, Xp previous time Output value, Xr command value, Xrp previous command value, Xsh chute amount.

Claims (12)

A control device for a power converter, the output value of which changes in accordance with a control amount,
Feedforward control for fixing the control amount to a predetermined feedforward control amount according to a comparison between the output value and a threshold value different from the command value of the output value, and the deviation from the output value and the command value A control switching unit that switches between feedback control that calculates a control amount and
A control signal generation unit that generates a control signal of the power converter according to the control amount set by the feedback control or the feedforward control by the control switching unit;
According to a change of the command value, at least a first period in which the threshold is set to a value higher than the command value, and a second period in which the threshold is set to a value lower than the command value One is provided,
The control switching unit selects the feedback control when the output value is less than or equal to the threshold in the first period, and the feedforward control when the output value is higher than the threshold. In the second period, the feedback control is selected when the output value is the threshold or more in the second period, and the feedforward control is selected when the output value is lower than the threshold. Power converter control unit. The first period is provided until the command value turns to decrease after the command value increases.
The control device of the power converter according to claim 1, wherein the second period is provided until the command value turns to increase after the command value decreases.   In the first or second period, when the output value approaches the command value and the feedback control is selected after the selection of the feedforward control in the first or second period, the control switching unit performs the first or the second operation. The control device of the power converter according to claim 2, wherein reselection of the feedforward control is inhibited until the period 2 ends.   The control switching unit is configured to output the output value after the selection of the feedforward control in the first period, when the output value changes from rising to decreasing or after selecting the feedforward control in the second period. The control device of the power converter according to any one of claims 1 to 3, wherein the feedforward control is switched to the feedback control when the value changes from a decrease to an increase. The threshold is set such that an absolute value of a difference from the command value is smaller than a preset allowable chute amount.
5. The feedforward control amount according to claim 1, wherein the feedforward control amount is set before the start of the feedforward control from a value before change and a value after change of the command value and the allowable chute amount. The control device of the power converter according to claim 1. From the threshold used for control, the feedforward control amount, the value before change of the command value and the value after change, and the command value generated for the output value in the first or second period A learning control unit that learns a correspondence between learning data including a shoot amount which is a maximum value of deviation;
According to the start of the first or second period, using the correspondence relationship learned by the learning control unit, a value before change and a value after change of the command value, and a preset allowance The control device of the power converter according to any one of claims 1 to 4, further comprising: a feedforward control amount calculation unit that calculates the threshold value and the feedforward control amount from a chute amount. The learning control unit outputs a value before the change of the command value and a value after the change, an input layer to which the chute amount is input, an output to which the threshold value used for control and the feedforward control amount are input. Learning the correspondence by obtaining weighting coefficients between neurons of a neural network model having a layer;
The feedforward control amount calculation unit may calculate a value before the change of the command value in the input layer of the neural network model learned by the learning control unit in response to the start of the first or second period. The control device of a power converter according to claim 6, wherein the threshold value and the feedforward control amount are calculated using an output value in the output layer when a value after change and the allowable chute amount are input. It further comprises a learning data extraction unit that extracts the learning data in the first or second period according to the end of the first or second period,
The learning control unit accumulates the learning data extracted by the learning data extraction unit during operation of the power converter, and adds the accumulated learning data to update the learning of the correspondence relationship. Item 8. A control device of a power converter according to item 6 or 7. A learning data extraction unit that extracts the learning data in the first or second period according to the end of the first or second period;
The system further includes a learning data generation control unit that executes learning data generation control in which a plurality of test patterns in which the command value, the threshold value, and the feedforward control amount are predetermined are sequentially read at time intervals.
The learning data extraction unit controls the learning data each time the first or second period provided in response to a change in the command value due to switching between the plurality of test patterns is completed in the learning data generation control. The control apparatus of the power converter of Claim 6 or 7 which extracts.   The learning control unit according to claim 6 or 7, wherein, during operation of the power converter, extraction of the learning data is not performed, and the correspondence relationship is learned using the learning data accumulated in advance. Power converter control device. The power converter includes a DC / DC converter having a switching element,
The output value is an output voltage of the DC / DC converter,
The control amount is a duty ratio indicating an on / off period ratio of the switching element,
The control device according to any one of claims 1 to 10, wherein the control signal is an on / off control signal of the switching element.   The control device of the power converter according to any one of claims 1 to 11, wherein the power converter is connected to a power system.

Images