Digital control of voltage harmonic distortion and overload current protection for inverters

Abstract

A method reducing transient overshoot of an output signal of an uninterruptable power supply, the uniterruptable power supply including: (a) an inverter having at least one power switch; (b) a digital controller that generates a pulse width modulated (PWM) signal for controlling the power switch, the PWM signal having a fixed period TPWM and (c) an analog to digital signal corresponding to the sampled output signal, the method comprising the steps of (1) operating the A/D converter to sample the output signal at a first time to generate a first sampled output value; (2) operating the A/D converter to sample the output signal at a second time to generate a second sampled output value, the period between the second time and the first time being equal to TPWM; (3) generating a control input value that is equal to 1.5 times the second sampled output value minus 0.5 times the first sampled output value; (4) computing a PWM signal at a third time, the PWM pulse having a duration equal to the PWM pulse duration, the period between the second time and the third time being equal to 0.5 TPWM.

Description

BACKGROUND OF THE INVENTION

Techniques for producing low Total Harmonic Distortion in PWM inverters (single-phase or three-phase) have been known to exist in several prior works. In the early days, the carrier-modulated PWM techniques such as the triangular wave comparison type PWM were very popular. For example such techniques are disclosed in: A. Schonung and Stemmler, “Static frequency changes with subharmonic control in conjunction with reversible variable speed drive,” BBC RevI., pp. 555-557, August/September 1964; P. D. Ziogas, “Optimum voltage and harmonic control of PWM techniques for three phase static UPS systems,” IEEE Trans. Ind. Appl., vol. IA-16, no. 4, pp. 542-446, July/August 1980; and B. K. Bose and H. A. Sutherland, “A high performance pulse-width modulator for an inverter-fed drive system using a microcomputer,” in Conf. Rec. 1982 Ann. Meet. Ind. Appl. Soc., pp. 847-853.

Microcomputer based techniques using preprogrammed PWM patterns have also been utilized. For example in: H. S. Patel and R. G. Hoft, “Generalized technique of harmonic elimination and voltage control in thyristor inverter, Part I,” IEEE Trans. Ind. Appl. vol. IA-9, no. 3, pp. 310-316, March/April 1973; and “Part II, “ IEEE Trans. Ind. Appl., vol. LA-10, no. 5, pp. 666-673, September/October 1973; F. G. Turnbull, “Selected harmonic reduction in static dc-ac inverters, “ IEEE Trans. Commun. Elec., vol. 83, pp. 374-478, July 1964; I. J. Pitel, S. N. Talukdar, and P. Wood, “Characterization of programmed-waveform pulse-width modulation,” IEEE Trans. Ind. Appl., vol. IA-16, no. 5, pp. 707-715, September/October 1980; and G. S. Buja, “Optimum output waveforms in PWM inverters, “ IEEE Trans. Ind. Appl., vol. IA-16, no. 6, pp. 830-836, November/December 1980.

In these techniques, the switching edges of the PWM pattern are computed to satisfy certain performance requirements; the most common of which is controlling the fundamental component and eliminating specified harmonics. Two main disadvantages of these techniques are: slow voltage regulation response due to average voltage control, and phase displacement between the reference sine wave and the filter output varies with the load.

More recent techniques include the time optimal response switching PWM, disclosed in, for example: A. Kernik, D.L. Stechshulte, and D.W. Shireman, “Static Inverter with synchronous output waveform synthesized by time-optimal response feedback” IEEE Trans. IECI, vol. IECI-24, no. 4, pp. 297-305, 1977; A. Kawamura and R. G. Hoft, “Instantaneous feedback controlled PWM inverter with adaptive hysteresis,” IEEE Trans. Ind. Appl., vol. IA-20, no. 4, pp. 769-775, July/August 1984; and P. Ziogas, “Delta modulation technique in static PWM inverters,” IEEE Trans. Ind. Appl., vol. IA-17, pp. 289-295, 1981. Other recent techniques include the real-time deadbeat-controlled PWM, disclosed in, for example: K. P. Gokhale, A. Kawamura, and R. G. Hoft, “Dead beat microprocessor control of PWM interter for sinusoidal output waveform synthesis,” in Conf. Rec. IEEE Power Elect. Spec. Conf., 1985, pp. 28-36 and A. Kawamura, T. Haneyoshi, and R. G. Hoft, “Deadbeat controlled PWM inverter with parameter estimation using only voltage sensor,” in Conf Rec. IEEE Power Elec. Spec. Conf, 1986. These techniques have very fast response for load disturbances, but it is also known that these systems have a high THD for non-linear load (crest-load).

The voltage control technique proposed in the present invention uses the perfect control of robust servomechanism problem (Perfect RSP) theory developed in Davison, E. J. and Scherzinger, B., “Perfect Control of the robust servomechanism problem”, IEEE Trans. On Automatic Control, 32 (8), 689-702, 1987 (“Davison et al”) to ensure perfect tracking of the output voltages under unknown load by providing means for eliminating errors at specified harmonic and at the same time ensuring good transient response. The theory is based on the internal model principle, proposed by Francis, B. A., and Wonham, W. M., “The internal mode principle for linear multivariable regulators”, Appl. Math. Opt., vol. 2, pp. 170-194, which states that asymptotic tracking of controlled variables toward the corresponding references in the presence of disturbances (zero steady state tracking error) can be achieved if the models that generate these references and disturbances are included in the stable closed loop systems. In other words, if we include the frequency modes of the references and the disturbances to be eliminated in the control loop, then the steady state error will not contain these frequency modes.

Applying the internal model principle into the output voltages control in a three-phase PWM inverter means that the fundamental frequency mode (50 Hz or 60 Hz) has to be included in the controller since the references vary at this frequency. Elimination of the voltages errors due to the load currents at other harmonics frequency can then be achieved by including the frequency modes of these harmonics into the controller. The perfect RSP theory combines this internal model principle with the optimal state feedback to guarantee stability of the closed loop system and providing arbitrary good transient response.

Similar techniques that use the internal model principle to achieve very low THD output voltage in single-phase PWM inverters have been reported recently in T. Haneyoshi, A. Kawamura, and R. G. Hoft, “Waveform compensation of PWM inverter with cyclic fluctuating loads, IEEE Trans. Ind. Appl., vol. 24, pp. 582-589, 1988 (“Haneyoshi et al”) and Y. Tzou, S. Jung, and H. Yeh, “Adaptive Repetitive Control of PWM Inverters for Very Low THD AC-Voltage Regulation with Unknown Loads (“Tzou et al”). In these papers, the control development follows the repetitive control theory developed in T. Inoue, M. Nakano, and S. Iwai, “High accuracy control of servomechanism for repeated contouring,” in Proc 10 th Annu. Symp. Incremental Motion Contr. Syst. Devices, 1981, pp. 258-292; T. Inoue and M. Nakano, “High accuracy control of a proton synchroton magnet power supply,” in IFAC, vol. 20, 1981, pp. 216-221; and S. Hara, Y. Yamammoto, T. Omata, and M. Nakano, “Repetitive control system: A new type servo system for periodic exogenous signals,” IEEE Trans. Automatic Contr., vol. 33, no. 7, pp. 659-666, 1988. Unlike the technique based on the perfect RSP that provides zero steady state error for references or disturbances only at finite specified frequencies, the repetitive control guarantees zero steady state error at all the harmonic frequencies less than half of the sampling period. However, the repetitive control is not easy to stabilize for all unknown load disturbances and cannot obtain very fast response for fluctuating load. In Haneyoshi et al, the latter problem is solved by including a “one sampling ahead preview controller”, and Tzou et al enhances the stability result of Haneyoshi et al by providing an adaptive mechanism for unknown load disturbances.

The inventor would like to point out here that, although the perfect RSP used in this disclosure only eliminates voltage harmonic at a finite specified frequencies, the perfect RSP control is still a very suitable control for a three-phase PWM inverters. It is to be emphasized that in a three-phase system, most of the voltage harmonics, like the triplen harmonics (3 rd , 6 th , 9 th , etc.), are either non-existent, uncontrollable, and/or negligible in values. Therefore, not too many harmonics are left for the control to handle. Moreover, the closed loop stability under unknown load is easier to achieve with the perfect RSP than the repetitive control, and the perfect RSP by itself already provides a good transient response.

The inventor has also successfully combined the perfect RSP control of the voltages harmonics with a fast current controller using a discrete time sliding mode controller, as disclosed in Utkin, V., Guldner, J., Shi, J., Sliding Mode Control in Electromechanical Systems, Taylor & Franci, Philadelphia, Pa., 1999, (“Utkin et al”) for limiting the inverter currents under overload condition. This is one of the important features necessary for a UPS, which is not addressed in the repetitive control work of Haneyoshi et al and Tzou et al. The discrete sliding mode controller has been chosen because of the fast and no-overshoot response that it provides. The current controller acts as an inner loop to the perfect RSP control of the output voltages in the outer loop. In this case, the perfect RSP voltage control has been designed by accounting for the extra dynamic introduced by the discrete time sliding mode controller. This way, the stability and robustness of the overall control system are still guaranteed.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 diagrammatically illustrates a general controller that solves the robust servomechanism problem.

FIG. 2 illustrates a power converter system used with the present invention.

FIG. 3 illustrates each phase of the delta-wye transformer model used in the present invention.

FIGS. 4A-4E illustrate equivalent circuits of the inverter in the DQ 0 stationary reference frame.

FIG. 5 illustrates a control strategy using a two-loop control structure, an inner inverter currents loop and an outer load voltages loop, in accordance with the present invention.

FIG. 6 shows the timing diagram of the PWM gating signals generation in relation with the A/D sampling time of the DSP.

FIG. 7 illustrates the robust servomechanism controller structure during current limit saturation.

FIGS. 8A-8D shows the waveforms of the load currents, load voltages, and inverter voltages under various linear loads: resistive, inductive, balanced, and unbalanced load.

FIG. 9 shows the load voltages, load currents, inverter voltages, and inverter currents under non-linear loads.

FIG. 10 illustrates a 0% to 100% resistive load transient.

FIG. 11 illustrates the load voltages when subjected to a 100% to 0% purely resistive load transient.

FIG. 12 illustrates the inverter currents during a short circuit.

FIG. 13 illustrates a schematic of a power system controlled in accordance with the present invention.

FIG. 14 illustrates the discrete state space implementation of the harmonic-eliminating transfer function for the fundamental frequency and the fifth and seventh harmonics.

FIG. 15 illustrates the overall sum for each axis derived by multiplying each state of the servo compensator states (x servo (k)) with the corresponding servo compensator gains.

FIG. 16 illustrates the current limiting accomplished by limiting the magnitude of the inverter current commands vector and maintaining its original direction.

SUMMARY OF THE INVENTION

The present invention relates to control techniques, and more particularly to techniques for digitally controlling uninterruptible power supplies. Specifically, the present invention relates to a method for reducing transient overshoot of the output of an uninterruptible power supply.

Particularly, the method of the claimed invention involves operating an analog to digital converter to sample the output signal, e.g., a voltage or a current, at a first time to generate a first sampled output value. The claimed invention further involves operating the analog to digital converter to sample the output signal at a second time to generate a second sampled output value. The period between the first operation of the analog to digital converter and the second analog to digital converter is equal the PWM period of the inverter; i.e., T PWM .

The method then involves generating a control input value for the inverter that is equal to 1.5 times the second sampled output value minus 0.5 times the first sampled output value. Once the control input value is determined, a PWM pulse duration is computed as a function of the control input value. A pulse having the computed pulse duration (i.e., a PWM signal) is initiated at a third time. The third time is one-half PWM period after the second sampling time.

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT

Perfect Control of Robust Servomechanism Problem

Finding a solution to a Robust Servomechanism Problem (RSP) for a system involves finding a controller for the system such that: given a modeled class of unstable tracking/disturbance signals, exact asymptotic tracking/regulation occurs for all plant perturbations that do not produce instability. A perfect controller for the RSP solves the RSP and also provides arbitrarily good transient error, with no unbounded peaking in the error response of the system. Below some of the main results of the perfect RSP discussed in Davison et al are summarized.

Consider the plant to be regulated described by the following equations:

{dot over (x)}=Ax+Bu+Ed y=Cx+Du+Fd y m =C m x+D m u+F m d e=y ref −y   (1)

where xεR n , xεR m are the inputs, yεR r are the outputs to be regulated, y m εR rm are the measurable outputs, dεR δ are the disturbances, y ref εR r are the reference input signals, and eεR r are the error in the system. It is assumed that the disturbance d arises from the following class of systems:

 {dot over (η)} 1 =Ψ 1 η 1 , d=C 1 η 1 , η 1 εR n1   (2)

and the reference input signals y ref arise from the following:

{dot over (η)} 2 =Ψ 2 η 2 , y ref =C 2 η 2 , η 2 εR n2   (3)

with eig(Θ 1 )⊂C + and eig(Θ 2 )⊂C + , where eig( ) denotes the eigenvalues and C + denotes the closed right complex half plane. This class of signals includes most classes of signals that occur in application problems, e.g., constant, polynomial, sinusoidal, polynomial-sinusoidal, etc. Let Λ={(λ 1 , λ 2 , . . . , λ p } be the zeros of the least common multiple of minimal polynomial of Ψ 1 Ψ 2 (multiplicities repeated), then a linear controller that solves the RSP for (1), consists of the following structure:

u=ξ+K 1 η  (4)

where ηεR rp is the outputs of a servo-compensator and ξεR m the output of a stabilizing-compensator S. The servo compensator has the form:

{dot over (η)}= A c η+B c e   (5)

with

A c =block diag (Ω,Ω, . . . ,Ω)

B c =block diag(γ,γ, . . . ,γ)

and $$ Ω = [ 0 1 0 … 0 0 0 1 … 0 ⋮ ⋮ ⋮ ⋮ ⋮ - σ 1 - σ 2 - σ 3 … - σ p ] , γ = [ 0 0 ⋮ 0 1 ]

where the coefficients σ i ,i=1,2, . . . , p are given by the coefficients of the polynomial $$ ∏ i = 1 P ⁢   ⁢ ( λ - λ i ) ; i . e . , ⁢ λ p + σ p ⁢ λ p - 1 + ⋯ ⁢   + σ 2 ⁢ λ + σ 1 := ∏ i = 1 P ⁢   ⁢ ( λ - λ 1 ) ( 6 )

The servo compensator gain K 1 in (4) and the stabilizing-compensator S should be found to stabilize and “give desired behavior” to the following stabilizable and detectable system: $$ [ x . n . ] = [ A 0 - B c ⁢ C A c ] ⁡ [ x n ] + [ B - B c ⁢ C ] ⁢ u ⁢ [ y m n . ] = [ C m 0 0 I ] ⁡ [ x n ] + [ D m 0 ] ⁢ u ( 7 )

A general controller that solves the robust servomechanism problem can be diagrammatically shown as in FIG. 1 .

Note that, given a modeled class of references/disturbances, the servo-compensator is unique within the class of coordinate transformations and nonsingular input transformations. The stabilizing compensators S are, however, not unique; there are various classes of stabilizing compensators that can be used in a robust servomechanism controller. One of them is the complementary controller where the stabilizing compensator is given by:

S:ξ=K 0 x   (8)

where x can either be the measured states of the system, or the estimates found using an observer (if not all states are measurable). In this case, the control u is given by state feedback found to stabilize the augmented system of the plant and the servo-compensator: $$ x ∼ . = A ∼ · x ∼ + B ∼ · u A ∼ = [ A 0 - B c ⁢ C A c ] ⁢ B ∼ = [ B - B c ⁢ D ] ⁢ x ∼ = [ x n ] , u = K 0 ⁢ x + K 1 ⁢ η = [ K 0 ⁢ K 1 ] ⁢ x ∼ := K · x ∼ ( 9 )

A perfect controller for the robust servomechanism problem can be constructed using Cheap Control method, where the gains K=[K 0 K 1 ] are found by minimizing the following performance index: $$ J ε = ∫ 0 ∞ ⁢ ( z ′ ⁢ z + ε · u ′ ⁢ u ) ⁢   ⁢ ⅆ τ , z = C ∼ · x ∼   ( 10 )

where γ>0 is an arbitrarily small scalar. The optimal control gain K=K opt , which minimizes this performance index, is given by: K opt =−1/ε{tilde over (B)}′P opt , where P opt is the unique positive semi-definite solution to the algebraic Riccati equation. $$ A ′ ∼ ⁢ P opt + P opt ⁢ A ∼ + C ′ ∼ ⁢ C ∼ - 1 ε ⁢ P opt ⁢ B ∼ ⁢ B ′ ∼ ⁢ P opt = 0 ( 11 )

Discrete-Time Sliding Mode Control

Discrete time sliding mode controller is an approach in the sliding mode control theory, which is suitable for digital implementation since it does not exhibit the chattering phenomena due to direct digital implementation of continuous time sliding mode control. See Utkin et al.

Consider a continuous linear time invariant system

{dot over (x)} ( t )= Ax ( t )+ Bu ( t )+ Ed ( t ) y ( t )= Cx ( t ) e ( t )= y ( t )− y ref ( t )  (12)

with state vector x(t)εR n , control u(t)εR m , output to be regulated y(t), reference input y ref (t), and disturbance d(t). System (12) can be transformed to discrete time system with sampling time Ts to yield.

x ( k+ 1)= A*x ( k )+ B*u ( k )+ E*d ( k ) y ( k )= Cx ( k ) e ( k )= y ( k )− y ref ( k )  (13)

where:

A*=exp(A·T S ) B*=∫ 0 T s e A·(Ts−τ)B dτ

E*=∫ 0 T s e A·(T s −τ)E dτ

It is desired to control the output y(k) to follow the reference y ref (k). For this purpose we can choose a sliding mode manifold in the form of: s(k)=Cx(k)−y ref (k) such that when discrete-time sliding mode exists we have y(k)→y ref (k). Discrete-time sliding mode exists if the control input u(k) is designed as the solution of:

s ( k+ 1)= CA*x ( k )+ CB*u ( k )− y ref ( k+ 1)=0  (14)

The control law that satisfies (13) is called ‘equivalent control’ and is given by:

u eq ( k )=−( CB *) −1 ( CA*x ( k )− y ref ( k+ 1)  (15)

If the control is limited to a value u 0 i.e. ∥u(k)∥≦u 0 then the following modified control law can be applied: $$ u ⁡ ( k ) = { u ⇀ eq ⁡ ( k ) for &LeftDoubleBracketingBar; u ⇀ eq ⁡ ( k ) &RightDoubleBracketingBar; ≤ u 0 u 0 ⁢ u eq ⁡ ( k ) &LeftDoubleBracketingBar; u eq ⁡ ( k ) &RightDoubleBracketingBar; for &LeftDoubleBracketingBar; u ⇀ eq ⁡ ( k ) &RightDoubleBracketingBar; > u 0 ( 16 )

With control law (16), the discrete-time sliding mode exists after a finite number of steps.

The Power Converter System

A power converter system to which the controller of the present invention may be applied comprises a typical three-phase PWM voltage inverter with LC output filter (Linv and Cinv) and a delta-wye transformer that act both as a potential transformer and electrical isolation to the load. FIG. 2 shows a circuit diagram of the system. Notice that the delta-wye transformer converts a three-wire (UVW) power system of the inverter to a four-wire (XYZ-N) system for the load. Small capacitors (denoted as Cgrass in FIG. 2 ) are added at the load side of the transformer to provide further harmonics filtering and stabilization of the load voltages. A DSP (Digital Signal Processor) system controls the operation of the power converter, providing required PWM gating signals to the power II devices. Voltages and currents measured by the DSP system for control purposes are shown labeled in FIG. 2 . The line-to-neutral load voltages (at points xyz-n in FIG. 2 ) are denoted as: Vload an , Vload bn , and Vload cn , the load phase currents as: Iload a , Iload b , and Iload c , the line-to-line inverter capacitor voltages (at points uvw in FIG. 2 ) as Vinv ab , Vinv bc , and Vinv ca ., the inverter phase currents as: Iinv a , Iinv b , and Iinv c .

For development of the control algorithm, a state space model of the system is needed. Each phase of the delta-wye transformer has been modeled as an ideal transformer with leakage inductance Ltrans and series resistance Rtrans on the secondary winding as shown in FIG. 3 . The secondary transformer currents are denoted as Isnda, Isnd b , and Isnd c .

Using the transformer model in FIG. 3 , the dynamic equations of the output filter circuit in FIG. 2 can be written as in equations (17.a)-(17.d): $$ ⅆ V ⇀ ⁢   ⁢ inv abc ⅆ t ⁢ 1 3 · C inv ⁢ I ⇀ ⁢ inv abc - 1 3 · C inv ⁢ Tr i · I ⇀ ⁢ snd abc (17.a) $$ ⅆ I ⇀ ⁢   ⁢ inv abc ⅆ t = 1 L inv ⁢ V ⇀ ⁢ pwm abc - 1 L inv ⁢ V ⇀ ⁢ inv abc (17.b) ⅆ V ⇀ ⁢   ⁢ load abc ⅆ t = 1 C load ⁢ I ⇀ ⁢   ⁢ snd abc - 1 C load . I ⇀ ⁢   ⁢ load abc (17.c) ⅆ I ⇀ ⁢   ⁢ snd abc ⅆ t = R trans L tran ⁢ I ⇀ snd + 1 L tran ⁢ Tr v · V ⇀ ⁢   ⁢ inv abc - 1 L tran ⁢ V ⇀ ⁢   ⁢ load abc (17.d)

where the voltages and currents vectors are defined as in (18).

{right arrow over (V)} inv abc =[V inv ab V inv bc V inv ca ] T

{right arrow over (V)} load abc =[V load a V load b V load c ] T ,

{right arrow over (I)} load abc =[I load a I load b I load c ] T ,

{right arrow over (I)} snd abc =[I snd a I snd b I snd c ] T ,

{right arrow over (I)} inv abc =[I inv ab I inv bc I inv ca ] T ,=[I inv a −I inv b I inv c I inv c −I inv a ] T   (18)

Matrices Tr i and Tr v in equation (17.a) and (17.d) denote the currents and voltages transformations of the delta-wye transformer. Denoting the transformer's turn ratio as tr, these matrices are given by (19): $$ Tr i = tr · [ 1 - 2 1 1 1 - 2 - 2 1 1 ] , Tr v = tr · [ 0 0 - 1 - 1 0 0 0 - 1 0 ] ( 19 )

To obtain a state space model of the system, the dynamic equations in (17) are transformed to DQ 0 stationary reference frame using the transformation:

{right arrow over (f)} qd0 =K S ·{right arrow over (f)} abc ,  (20)

with $$ K S = 2 3 [   ⁢ 1 - 0.5 - 0.5 0 - 3 / 2 3 / 2 0.5 0.5 0.5 ⁢   ] , ⁢ f qd 0 = [ f q , f d , f 0 ] T , f abc = [ f a , f b , f c ] T

where {right arrow over (f)} qd0 denotes the abc voltages and currents defined in (18), and {right arrow over (f)} qd0 the corresponding DQ 0 stationary reference frame variables. The circuit dynamics can then be written as in (21): $$ ⅆ V → ⁢ inv qd ⅆ t = 1 3 · C inv ⁢ I → ⁢ inv qd - 1 3 · C inv ⁢ Tri qd0 · I → ⁢ snd qd0 ( 21. ⁢ a ) ⅆ I → ⁢ inv qd ⅆ t = 1 L inv ⁢ V → ⁢ pwm qd - 1 L inv · V → ⁢ inv qd ( 21. ⁢ b ) ⅆ V → ⁢ load qd0 ⅆ t = 1 C load ⁢ I → ⁢ snd qd0 - 1 C load · I → ⁢ load qd0 ( 21. ⁢ c ) ⅆ I → ⁢ snd qd0 ⅆ t = R tran L tran ⁢ I → ⁢ snd qd0 + 1 L tran ⁢   ⁢ Trv qd ·   ⁢ V → ⁢ inv qd - 1 L tran ⁢ V → ⁢ load qd0 ( 21. ⁢ d )

where the matrices Tri qd0 and Trv qd are defined as: $$ Tri qd0 = [ K s · Tr i · K S - 1 ] row ⁢   ⁢ 1 , 2 = tr · 3 2 [   ⁢ 1 3 0 - 3 1 0 ⁢   ] ( 22. ⁢ a ) Trv qd = [ K s · Tr v · K S - 1 ] col ⁢   ⁢ 1 , 2 = tr · 1 2 [   ⁢ 1 - 3 - 3 1 0 0 ⁢   ] ( 22. ⁢ b )

Notice that, due to the three-wire system of the inverter and filter, the zero components of the inverter voltages ({right arrow over (V)}inv qd ), the inverter currents ({right arrow over (I)}inv qd ) and the input PWM voltages ({right arrow over (V)}pwm qd ) are trivial and they do not appear in (21).

Using the dynamic equations in (21), equivalent circuits can be drawn in the DQ 0 stationary reference frame as shown in FIGS. 4A-4E . From FIGS. 4A-4E it can be seen that the 0-axis equivalent circuit of the load-side of the transformer is completely de-coupled from the qd-axis equivalent circuits. This shows that the zero components of the load voltages and the secondary transformer currents (Vload 0 and Isnd 0 ) are uncontrollable by the input PWM voltages ({right arrow over (V)}pwm qd ). A linear controllability analysis performed to (21) can also be used to show this fact. The 0-component of the load currents will be non-zero under unbalanced load condition or under non-linear load in the form of triplen harmonics. Because these 0-components are uncontrollable, no action can be done by the control to compensate them. However, as is apparent from FIGS. 4A-4E , the small capacitor at the output along with the leakage inductance of the transformer act as an LC filter which can be tuned to provide attenuation to the effect of the 0-component of the load voltages. From FIGS. 4A-4E , the amount of this attenuation at steady state can be calculated from: $$ &LeftBracketingBar; Vload 0 ⁡ ( j ⁢   ⁢ ω ) &RightBracketingBar; = &LeftBracketingBar; jω · L trans + R trans 1 - ω 2 ⁢ L trans ⁢ C grass + j ⁢   ⁢ ω · R trans ⁢ C grass &RightBracketingBar; · &LeftBracketingBar; Iload 0 ⁡ ( j ⁢   ⁢ ω ) &RightBracketingBar; ( 23 )

where ω=2 πf and f is a harmonic frequency of interest.

Control System Development

To achieve fast current limiting capability for the inverter, the control strategy uses a two-loop control structure; an inner inverter currents loop and an outer load voltages loop as shown in FIG. 5 . The outer loop regulates the load voltages ({right arrow over (V)}load qd ) to follow 50/60 Hz balanced three-phase voltages references ({right arrow over (V)}ref qd ) and generates the inverter currents commands ({right arrow over (l)}cmd qd ), which are limited. The inner loop in turn generates the PWM voltage commands to regulate the inverter currents to follow the inverter current command. A standard voltage space vector algorithm is then used to realize these PWM command voltages. Notice that the 0-components of the load voltages are not regulated by the control, since it is uncontrollable.

FIG. 6 shows the timing diagram of the PWM gating signals generation in relation with the A/D sampling time of the DSP. It can be seen that there is a one-half PWM period delay between the time the signals are sampled by the A/D and the time the PWM control action is applied.

The next two subsections summarize the development of the two control loops, the inner current loop using the discrete sliding mode controller and the voltage control loop using the robust servomechanism principles.

Discrete-Time Sliding Mode Current Controller

For designing the discrete time sliding mode current controller, consider the intverter and filter subsystem with no transformer and load dynamics: $$ ⅆ V → ⁢ inv qd ⅆ t ≐ 1 3 · C inv ⁢ I → ⁢ inv qd - 1 3 · C inv ⁢ Tri qd0 · I → ⁢ snd qd0 ( 24. ⁢ a ) ⅆ I → ⁢ inv qd ⅆ t = 1 L inv ⁢ Vpwm qd - 1 L inv ⁢ V → ⁢ inv qd . ( 24. ⁢ b )

Assuming the secondary transformer currents {right arrow over (I)}snd qd0 as disturbances, this subsystem can be written in state space form as: $$ x → . = A 1 ⁢ x → 1 + B 1 ⁢ u → + E → 1 ⁢ d ⁢ ⁢ A → 1 = [   ⁢   ⁢ 0 → 2 × 2 ( 3 · C inv ) - 1 ·   ⁢ I → 2 × 2 ) - ( L inv ) - 1 ·   ⁢ I → 2 × 2   ⁢ 0 → 2 × 2 ] ,   ⁢ ⁢ B → 1 = [   ⁢ 0 → 2 × 2 ( L inv ) - 1 ·   ⁢ I → 2 × 2 ] ,   ⁢ ⁢ E → 1 = [   ⁢ - ( 3 · C inv ) - 1 ·   ⁢ Tri qd0   ⁢ 0 → 2 × 3 ] , ⁢   ( 25 )

where the states are {right arrow over (x)}=[{right arrow over (V)}inv qd ,{right arrow over (I)}inv qd ] the inputs {right arrow over (u)}={right arrow over (V)}pwm qd and disturbances {right arrow over (d)} 1 ={right arrow over (I)}snd qd .

The discrete form 25 can be calculated as:

{right arrow over (x)} 1 ( k+ 1)= A 1 *{right arrow over (x)} 1 ( k )+ B 1 *{right arrow over (u)}( k )+ E 1 *{right arrow over (d)} 1 ( k )  (4.2)

where

A 1 *=exp( A·T S ) B 1 *=∫ 0 Ts e A 1 ·T s B 1 dτE 1 *=∫ 0 T e A 1 ·T, E 1 dτ

and T S is the A/D sampling time, which in this case is equal to the PWM period T pwm

To force the inverter currents to follow their commands, the sliding mode surface is chosen as: {right arrow over (s)}(k)=C 1 ·{right arrow over (x)} 1 (k)−{right arrow over (I)}cmd(k) where C 1 ·{right arrow over (x)} 1 (k)={right arrow over (I)}inv qd (k), so that when discrete sliding mode occurs, we have {right arrow over (s)}(k)=0 or {right arrow over (I)}inv(k)={right arrow over (I)}cmd(k). The existence of the discrete sliding mode can be guaranteed if the control is given: $$ u ⁡ ( k ) = {   ⁢ u → eq ⁡ ( k ) for ⁢   ⁢ &LeftBracketingBar; &RightBracketingBar; ⁢ u → eq ⁡ ( k ) ⁢ &LeftBracketingBar; &RightBracketingBar; ≤ u 0 ⁢   u 0 ⁢ u → eq ⁡ ( k ) &LeftBracketingBar; &RightBracketingBar; ⁢ u → eq ⁡ ( k ) ⁢ &LeftBracketingBar; &RightBracketingBar; for ⁢   ⁢ &LeftBracketingBar; &RightBracketingBar; ⁢ u → eq ⁡ ( k ) ⁢ &LeftBracketingBar; &RightBracketingBar; > u 0 ⁢   ⁢   ( 26 )

where the equivalent control input {right arrow over (u)} eq (k) is calculated from:

{right arrow over (u)} eq ( k )=( C 1 B 1 *) −1 ( {right arrow over (I)} cmd qd −C 1 A 1 *{right arrow over (x)} 1 ( k )− C 1 E 1 *{right arrow over (d)} 1 ( k ))  (27)

and u 0 denotes the maximum value of the PWM voltage command realizable by the space vector algorithm.

Note that the secondary transformer currents are needed for the control, but these currents are not measured in the system (see FIG. 2 ). A linear Luenberger observer can be easily designed to estimate these currents for control purposes. However, in most practical cases we can approximate these currents with the load currents (i.e., {right arrow over (I)}snd qd ={right arrow over (I)}load qd ) because the currents through the output capacitor filters are small. According to the inventor's experience, the effect of using this approximation is unnoticeable in the control performance.

Due to the computation delay of the DSP, the control action given by (27) will result in undesirable overshoots during transients. This effect can be minimized, however, if the states {right arrow over (x)} 1 (k) and disturbances {right arrow over (d)} 1 (k) are replaced with their first order one-half step ahead predicated values given by:

{right arrow over (x)} 1 p ( k )=1.5· {right arrow over (x)} 1 ( k )−0.5· {right arrow over (x)} 1 ( k− 1) {right arrow over (d)} 1 p ( k )=1.5· {right arrow over (d)} 1 ( k )−0.5· {right arrow over (d)} 1 ( k− 1)  (28)

The equivalent control input {right arrow over (u)} eq (k) then becomes:

{right arrow over (u)} eq ( k )=( C 1 B 1 *) −1 ( {right arrow over (I)} cmd qd −C 1 A 1 *{right arrow over (x)} 1 p ( k )− C 1 E 1 *{right arrow over (d)} 1 p ( k ))  (29)

Voltage Controller Design Using Discrete Perfect RSP

The voltage control loop designed in this research is based on the discrete form of the technique developed in Davison et al discussed above. To design the load voltages controller, first consider the entire plant system with the 0-components of the voltages and currents omitted as given in (30). As explained in above, these 0-components are completely de-coupled and uncontrollable from the inputs, and therefore are not useful to be included in the design. In the system (30), an input delay of one-half the PWM period (0.5T pwm ) has been explicitly included to account for the computation delay of the DSP. $$ x → p ⁡ ( t ) = A p ⁢ x → p ⁡ ( t ) + B p ⁢ u → ⁡ ( t - 0.5 ⁢ T pwm ) , ⁢ A ⇀ ⁢ p = [   ⁢   ⁢ 0 → 2 × 2 ( 3 · C inv ) - 1 · I → 2 × 2   ⁢ 0 → 2 × 2 ( 3 · C inv ) - 1 · T ^ ⁢ ri qd ( - L inv ) ) - 1 · I → 2 × 2   ⁢ 0 → 2 × 2   ⁢ 0 → 2 × 2   ⁢ 0 → 2 × 2   ⁢ 0 → 2 × 2   ⁢ 0 → 2 × 2   ⁢ 0 → 2 × 2 ( C load ) - 1 · I → 2 × 2 ( L inv ) - 1 · T ^ ⁢ rv qd   ⁢ 0 → 2 × 2 ( - L inv ) - 1 · I → 2 × 2 - R trans ⁡ ( L trans ) - 1 · I → 2 × 2 ⁢   ) ⁢ ⁢ Bp = [   ⁢   ⁢ 0 → 2 × 2 ( L inv ) - 1 · I → 2 × 2   ⁢ 0 → 2 × 2   ⁢ 0 → 2 × 2 ] ,   ⁢ T ^ ⁢ ri qd = tr · 3 2 [   ⁢ 1 3 - 3 1 ⁢   ] ⁢   ⁢ T ^ ⁢ rv qd = tr · 1 2 [   ⁢ 1 - 3 3 1 ⁢   ] ( 30 )

The states variables for the system (30) are chosen as {right arrow over (x)} p =[{right arrow over (V)}inv qd ,{right arrow over (I)}inv qd , {right arrow over (V)}load qd , {right arrow over (I)}snd qd ], with the inputs as u={right arrow over (V)}pwm qd . System (30) can be transformed to a discrete-time system with sampling time Ts=T pwm to yield:

{right arrow over (x)} p ( k+ 1)=Φ· {right arrow over (x)} p ( k )+Γ 1 ·u ( k− 1)+Γ 2 ·{right arrow over (u)} ( k )  (31)

where $$ Φ = ⅇ A p ⁢ T S ,   ⁢ Γ 1 = ∫ 0.5 ⁢ T S T S ⁢ ⅇ A p ⁢ τ ⁢ B p ⁢   ⁢ ⅆ τ ,   ⁢ Γ 2 = ∫ 0 0.5 ⁢ T s ⁢ ⅇ A p ⁢ τ ⁢ B p ⁢   ⁢ ⅆ τ

Discrete time system (31) can be written in a standard discrete time state space equations by adding the extra states: {right arrow over (x)} a (k)={right arrow over (u)}(k−1)={right arrow over (V)}pwm qd (k−1) to yield: $$ [   ⁢ x → p ⁡ ( k + 1 ) x → a ⁡ ( k + 1 ) ⁢   ] = [ Φ Γ 1   ⁢ 0 → 2 × 2   ⁢ 0 → 2 × 2 ⁢   ] · [   ⁢ x → p ⁡ ( k ) x → a ⁡ ( k ) ] + [ Γ 2   ⁢ I → 2 × 2 ] · u → ⁡ ( k ) ( 32 )

so that the system can be written as:

{right arrow over (x)} p *( k+ 1)= A p *{right arrow over (x)} p *( k )+ B p *{right arrow over (u)} ( k )  (33)

where $$ x → p * ⁡ ( k ) = [   ⁢ x → p ⁡ ( k ) x → p ⁡ ( k ) ] ,   ⁢ A p * = [   ⁢ Φ Γ 1 0 → 2 × 2 0 → 2 × 2 ] , B p * = [ Γ 2 I → 2 × 2 ⁢   ]

To design the voltage controller, we need to consider the true plant (33) and the discrete time sliding mode current controller as the equivalent ‘plant’ as seen by the outer voltage loop. Using equation (27) and (33) the augmented true plant and discrete sliding mode current controller can be found as in (34).

{right arrow over (x)} p *( k+ 1)= A d {right arrow over (x)} p *( k )+ B d {right arrow over (u)} 1 ( k )  (34)

with {right arrow over (u)} 1 (k)={right arrow over (I)}cmd qd *(k), and $$ A d = A p * - B p * ⁡ ( C 1 ⁢ B 1 * ) - 1 ⁢ ( C 1 ⁢ A 1 * ⁢ C 11 + C 1 ⁢ E 1 * ⁢ C 12 ) $$ B d = B p * ⁡ ( C 1 ⁢ B 1 * ) - 1 $$ C 11 = [   ⁢   ⁢ I → 2 × 2   ⁢ 0 → 2 × 2 0 → 2 × 2 0 → 2 × 2 0 → 2 × 2   ⁢ 0 → 2 × 2   ⁢ I → 2 × 2 0 → 2 × 2 0 → 2 × 2 0 → 2 × 2 ⁢   ] $$ C 12 = [   ⁢ 0 → 2 × 2 0 → 2 × 2 0 → 2 × 2 I → 2 × 2 0 → 2 × 2 ⁢   ]

Note that the augmented system given in (34) was found assuming the approximation {right arrow over (I)}snd qd ≈{right arrow over (I)}load qd has been used.

Now, assume ω i =2 πf i i=1,2, . . . n are frequencies of the reference voltages and harmonics to be eliminated. For a 60-Hz UPS system with desire to eliminate 5 th and 7 th harmonics, for example, we use ω 1 =2π·60, ω 2 =2π·5·60, and ω 3 =2π 7·60. We can then choose the servo-compensator to be of the form (35):

{right arrow over ({dot over (η)})}= A c {right arrow over (η)}+B c e Vqd {right arrow over (e)} Vqd ={right arrow over (V)} ref qd −{right arrow over (V)} load qd   (35)

where

 {right arrow over (η)}=[{right arrow over (η)} 1 ,{right arrow over (η)} 2 , . . . {right arrow over (η)} n ] T {right arrow over (η)} i εR 4 , i= 1,2, . . . n A c =block diag[Ac 1 ,Ac 2 , . . . ,Ac n ]B c =[Bc 1 ,Bc 2 , . . . ,Bc n ] T

with $$ Ac i = (   ⁢   ⁢ 0 → 2 × 2   ⁢ I → 2 × 2 - ω i 2 ⁢ I → 2 × 2   ⁢ 0 → 2 × 2 ⁢   ) ,   ⁢ i = 1 , 2 , … ⁢   ⁢ n $$ Bc i = ( 0 → 2 × 2   ⁢ I → 2 × 2 ) T ⁢   ⁢ i = 1 , 2 ,   ⁢ … ⁢   ⁢ n ⁢  

Note that each of the blocks {right arrow over ({dot over (η)})} i =Ac 1 {right arrow over (η)} i +Bc i {right arrow over (e)} Vqd represents a state space implementation of the continuous transfer function 1/(s 2 +ωi 2 ) for each of the qd-axis voltages errors.

The servo compensator (35) can be transformed to a discrete time system to yield:

{right arrow over (η)}( k+ 1)= A c *{right arrow over (η)}( k )+ B c *{right arrow over (e)} Vqd ( k ), {right arrow over (e)} Vqd ( k )= {right arrow over (V)} ref qd ( k )− {right arrow over (V)} load qd ( k )  (36)

where

A c *=exp( A c ·T s ) B c *=∫ 0 T s e A 1 ·(T S −τ) B c dτ

Now that we have determined the ‘plant’ and the servo compensator, the control input for the perfect robust servomechanism controller is given by:

{right arrow over (u)} 1 ( k )= {right arrow over (I)} cmd qd *( k )= K x p *( k )+ K 1 η( k )  (37)

where the gains K [K 0 K 1 ] are found by minimizing the discrete performance index: $$ J ∈ = ∑ k = 0 ∞ ⁢   ⁢ ( z ⁡ ( k ) ′ ⁢ z ⁡ ( k ) + ε · u ⁡ ( k ) ′ ⁢ u ⁡ ( k ) ) , ⁢ z = [ x p * η ] ( 38 )

for the augmented ‘equivalent plant’ (34) and the servo compensator (36): $$ [   ⁢ x → p * ⁡ ( k + 1 ) η ⁡ ( k + 1 ) ] = [   ⁢ Ad 0 - B c * ⁢ C A c * ⁢   ] ⁢   [ x → p * ⁡ ( k ) η ⁡ ( k ) ] + [ B d - B c * ⁢ D ] ⁢ u 1 ⁡ ( k ) ( 39 )

where ε>0 is an arbitrarily small scalar.

Current Limit and Control Saturation Handling

The current command {right arrow over (I)}cmd qd *(k) generated by the robust servomechanism voltage controller above is limited in magnitude as in (40) to yield the current command {right arrow over (I)}cmd qd *(k), which will be implemented by the inner loop current controller. $$ I → ⁢ cmd qd ⁡ ( k ) = {   ⁢ I → ⁢ cmd qd * ⁡ ( k ) if ⁢   ⁢ &LeftBracketingBar; I → ⁢ cmd qd * ⁡ ( k ) &RightBracketingBar; ≤ I max I → ⁢ cmd qd * ⁡ ( k ) I → ⁢ cmd qd * ⁡ ( k ) ⁢ I max if ⁢   ⁢ &LeftBracketingBar; I → ⁢ cmd qd * ⁡ ( k ) &RightBracketingBar; > I max ⁢   ( 40 )

I max represents the maximum allowable magnitude of the inverter currents. Equation (40) limits the magnitude of the current commands but maintains their vector directions in the qd space.

The states {right arrow over (η)} i of the servo-compensator can be seen as sine wave signal generators that get excited by the harmonic contents of the error signals at frequency ω i . When the control inputs of the robust servomechanism voltage controller saturate i.e., |{right arrow over (I)}cmd qd *(k)|>I max the servo-compensator states will grow in magnitude due to the break in the control loop. This problem is similar to the integrator windup problem that occurs in an integral type controller. To prevent this, the servo-compensator in (36) can be modified as follows: $$ η → ⁡ ( k + 1 ) = A c * ⁢ η → ⁡ ( k ) + B c * ⁢ e → 1 ⁡ ( k ) , ⁢ e → 1 ⁡ ( k ) = {   ⁢ e → Vqd ⁡ ( k ) if ⁢   ⁢ &LeftBracketingBar; I → ⁢ cmd qd * &RightBracketingBar; ≤ I max 0 if ⁢   ⁢ &LeftBracketingBar; I → ⁢ cmd qd * &RightBracketingBar; < I max ⁢   ( 41 )

Using (41), during the current limit saturation, the servo compensator states will continue to oscillate at the harmonic frequency with constant magnitude. The resulting robust servomechanism controller structure is shown in FIG. 7 .

Experimental Results

The effectiveness the proposed control strategy has been verified on an 80 kVA UPS unit with system parameters shown in Table 1. The DC bus voltage is obtained from a 6-pulse thyristor controlled rectifier in parallel with a 480V battery system. The DSP control system used is based on the TMS320F240 fixed point DSP with control timing diagram as given in FIG. 6 . The PWM timing is calculated through a standard space vector PWM with switching frequency of 3.2 kHz T pwm =T s =320 μsec.

The experimental results presented in this report have been obtained using the proposed control strategy with the 5 th and 7 th harmonics being eliminated. Table 2 gives the steady state RMS output voltages regulation under different types of loads. It can be seen that the control strategy provides good output voltages regulation in all cases. Note that the deviations in the output voltages for unbalanced load are due to the uncontrollable 0-component of the load currents. However, as can be seen the effect is minimal showing the effectiveness of the LC filter at the output side of the transformer.

FIG. 8 shows the waveforms of the load currents, load voltages, and inverter voltages under various linear loads: resistive, inductive, balanced, and unbalanced load. FIG. 9 shows the load voltages, load currents, inverter voltages, and inverter currents under non linear crest load. The THD of the load voltages in FIG. 9 was measured at 2.8% with the load currents of 3:1 crest factor. Table 3 summarizes the output voltages THD under different types of loads showing the superior THD performance of the proposed control strategy. Better results may be obtained by including more harmonies to be eliminated into the robust servomechanism controller.

TABLE 1
System Parameters
DC Bus Voltages	Vdc	540	V (nom.) 390 V (min)
AC Output voltage	Vload	208	V (LL-RMS), 120 V (LN)
	F	60	Hz
Inverter filters	C inv	540	μF
	L inv	300	μH
Delta-Wye Transformer	L trans	48	μH (≈0.03 p.u)
	R trans	0.02	ohm
Output filter	C grass	90	μF

TABLE 2
Output voltages regulation
TYPE OF LOAD	% Reg	N LD OUTPUT V	AVG	FLD OUTPUT V	AVG
V REG (100% RES LD)	0.031	120.23	120.07	120.16	120.15	120.17	120.03	120.15	120.12
V REG (100% 0.8 PF LD)	0.033	120.23	120.07	120.16	120.15	120.20	120.00	120.14	120.11
V REG (100% 2.75 ICF LD)	0.019	120.23	120.07	120.16	120.15	120.14	120.11	120.14	120.13
V REG, 100% UNBAL RES	0.019	120.23	120.07	120.16	120.15	120.59	119.59	120.21	120.13
LD(A)
V REG 100% UNBAL RES LD	0.028	120.23	120.07	120.16	120.15	120.67	119.97	119.72	120.12
AAB)
TYPE OF LOAD	% Reg	540 VDC, OUTPUT V	AVG	390 VDC, OUTPUT	AVG
V REG. 100% RES LD ON	−0.089	120.17	120.03	120.15	120.12	120.26	120.12	120.29	120.22
BATT. @ 390 VDC

TABLE 3
Output voltages THD
TYPES OF LOAD                    Output voltages THD
No load                          0.90%
100% balanced resistive load     1.30%
100% 0.8 pf load                 1.32%
100% unbal. Resistive (ph. A)    1.70%
100% unbal. Resistive (ph A & B) 1.89%
Crest load (3:1)                 2.7%

FIGS. 8A-8D the load currents, load voltages and inverter voltages for the experimental system loaded at steady state linear load as follows: (a) 100% resistive balanced; (b) 100% 0.8 pf load; (c) 100% resistive unbalanced (phase A unloaded) and; (d) 100 % resistive unbalanced (phase A&B). The top illustrates load currents. The middleindicates load voltages. The bottom indicates inverter voltages.

FIG. 9 illustrates the load voltages and load currents for a steady state non-linear load of 100% having 3:1 crest factor. The topillustrates phase A of load voltages and load currents, while the bottomillustrates phase A of inverter voltages and inverter currents.

FIG. 10 illustrates the load currents and load voltages for a 0 to 100% resistive load transient the topillustrates three-phase load currents, while the bottom illustrates three-phase load voltages.

FIG. 11 illustrates the load currents and load voltages for a 100% to 0% resistive load transient. The top illustrates three-phase load currents, and the bottom illustrates three-phase load voltages.

FIGS. 9 and 10 show the responses of the load voltages on resistive load transients. 0-100% and 100%-0 respectively. It can be seen that the load voltages recover within less than a cycle after the load is applied. In both cases, the output voltages magnitudes deviate less than 5% of the nominal. These result show that the perfect RSP controller provides fast transient recovery under load transients with minimal overshoot in the response. Finally, the effectiveness of the current controller was verified by applying a sudden three-phase short circuit on the output load terminals. The current limit was set at 300% level, and the inverter was shutdown deliberately after ten cycles of short circuit condition. The results are shown in FIG. 12 . It can be seen that the discrete sliding mode controller provides a fast and minimal overshoot on the inverter currents. The top illustrates inverter currents; the middle illustrates load voltages; and the bottom illustrates inverter voltages.

Step-By-Step Control Flow for One Embodiment

The steps in operating the disclosed control system will now be described in sequence. A first step is, at each control sampling time k, sampling the following voltage and current signals: (1)the line-to-neutral voltages (at points xyz-n in FIG. 13 ) denoted as Vload an , Vload bn , and Vload cn ; (2) the load phase currents, denoted as Iload b , Iload b , and Iload c ; (3) the line-to-line inverter filter capacitor voltages (at points uvw in FIG. 13 ), denoted as Vinv ab , Vinv bc , and Vinc ca ; and (4) the inverter phase currents, denoted as Iinv a , Iinv b , and Iinv c .

From the sampled values, the inverter voltage, inverter current, load voltage, and load current vectors are formed in abc reference frame:

{right arrow over (V)} inv abc =[V inv ab V inv bc V inv ca ] T

{right arrow over (V)} load abc =[V load a V load b V load c ] T

{right arrow over (I)} load abc =[I load a I load b I load c ] T

{right arrow over (I)} inv abc =[I inv ab I inv bc I inv ca ] T =[I inv a −I inv b I inv b −I inv c I inv c −I inv a ] T

Note that the inverter currents vectors are represented as line-to-line quantities. This is done to match the chosen state space equations and state variables used to derive the control. Inverter phase currents can also be used here with proper modification to the state space equations used.

The third step is to transform the sampled voltages and currents to the DQ 0 stationary reference frame using the transformation:

$$ f ϖ qd0 = K - S · f → abc ,

with $$ K s = 2 3 [   ⁢ 1 - 0.5 - 0.5 0 - 3 / 2 3 / 2 0.5 0.5 0.5 ⁢   ] , ⁢ f qd0 = [ f q , f d , f 0 ] T , f abc = [ f a , f b , f c ] T

The following voltages and currents variables in DQ stationary reference frame are obtained:

{right arrow over (V)} inv qd0 ={right arrow over (K)} S ·{right arrow over (V)} inv abc =[V inv q V inv d V inv 0 ] T

{right arrow over (l)} inv qd0 ={right arrow over (K)} S ·{right arrow over (l)} inv abc =[l inv q l inv d l inv 0 ] T

{right arrow over (V)} load qd0 ={right arrow over (K)} S ·{right arrow over (V)} load abc =[V load q V load d V load 0 ] T

{right arrow over (l)} load qd0 ={right arrow over (K)} S ·{right arrow over (l)} load abc =[l load q l load d l load 0 ] T

The 0-component of the inverter voltages and currents will always be zero and can be discarded due to the three-wire system of the inverter. The 0-components of the load voltages and currents can be non-zero, however, these components are non-controllable and have no use for control. Therefore only the DQ components need to be used and calculated: $$ V → ⁢ inv qd = [ Vinv q ( Vinv d ] ) T ⁢   ⁢ ⁢   ⁢ V → ⁢ load qd = [ Vload q ( Vload d ] ) T ⁢   ⁢ ⁢ l → ⁢ inv qd = [   ⁢ linv q linv d ] T ⁢   ⁢ ⁢ l → ⁢ load qd = [   ⁢ lload q ( lload d ] ) T ⁢  

Note that the DQ 0 transformation used here is a standard transformation commonly used in three-phase power applications. The transformation puts the three-phase variables in the system into an orthogonal coordinate system. The use of an orthogonal coordinate system allows the identification of state variables and creation of the state space equations for control purposes.

The fourth step is to generate the sine waves references at discrete time k for the load voltages in the abc reference frame and perform the DQ transformation on them to obtain the references in DQ.

Vref an =V md ·sin(2·π·f ref ·k·T S )

Vref bn =V mB ·sin(2·π·f ref ·k·T S −2π/3)

Vref cn =V mC ·sin(2·π·f ref ·k·T S +2π/3)

where V md is the voltage magnitude of reference AN, VmB is the voltage magnitude of reference BN, VmC is the voltage magnitude of reference CN, T S is the sampling period, and f ref is the fundamental reference frequency.

These voltages are then transformed to DQ reference frame:

{right arrow over (V)} ref abc =[V ref an V ref bn V ref cn ] T

{right arrow over (V)} ref qd0 ={right arrow over (K)} S ·{right arrow over (V)}ref abc =[V ref q V ref d V ref 0 ] T

Again only the DQ components are needed for control:

{right arrow over (V)}ref qd =[V ref q Vref d ]

The fifth step is then to compute the load voltage errors to be used by the voltage servo compensator, which are given by:e q =Vref q −Vload q and e d =Vref d −Vload d .

The sixth step is to calculate the next discrete states (time k+l) of the servo compensator for each of the harmonic frequency to be eliminated and the fundamental frequency and for each Q and D axis voltages errors. FIG. 14 shows the case when the 5 th and 7 th are to be eliminated. Each harmonic servo compensator block in FIG. 2 is the discrete state space implementation of the analog transfer function: $$ T ⁢ ( s ) = 1 s 2 + ϖ n 2

where T 1 =2Bθiθf,f is the fundamental frequency and i−f is the harmonic frequency (i.e., in the case of FIG. 14 , i=1, 5, and 7). Specifically, the following continuous state space implementation of T(s) is used for each harmonic servo compensator block: $$ ρ n i = (   ⁢ 0 1 ω i 2 0 ⁢   ) ⁢ n ρ i + ( 0 1 ) ⁢ e

which is then discretized to obtain the a's and b's coefficients shown in FIG. 14. A standard discretization method like ZOH (Zero order hold) method is used in this case. Each harmonic servo compensator block in FIG. 14 ensures the elimination of the corresponding harmonic contents of the errors signals. This condition is guaranteed by the internal model principle that states that if frequency modes (poles) of the references and the disturbances to be eliminated are included in the control loop, then the steady state error will not contain these frequency contents. Each of these harmonic compensators can be viewed as signals generators that vary at the harmonic frequency and reacts only to the existence of the corresponding harmonic in the errors signals. It provides adjustment to the inverter current commands generated by the voltage control loop to force the corresponding harmonic content of the errors to zero.

The seventh step is to multiply each state of the servo compensator states (x servo (k)) with the corresponding servo compensator gain and get the overall sum for each axis as shown in FIG. 15 . The plant states (consisting of the inverter voltages, inverter currents, load voltages, load currents, and the previous PWM voltage commands) are multiplied with the stabilizing compensator gains to get the summation for each axis. The sum of the servo compensator and the stabilizing compensator for each axis constitutes the inverter current commands generated by the voltage control loop i.e., Icmd q and Icmd d .

The servo compensator and the stabilzing compensator gains are designed to ensure the stability of the overall control loop i.e., the combined voltage control loop and the current control loop. The perfect RSP theory provides a way of calculating these gains using an OPTIMAL Control Technique. Using this technique, a cost function consisting of the weighted sum of squares of the servo compensator states, the plant states, and the control inputs (inverter current commands) is created. An optimization routine is then used to seek for the gains that minimize this cost function. The weighted sum of squares of the servo compensator states in this case can be viewed as the energy in the errors signals, while those of the plant states and the control inputs can be viewed as the fuel energy needed. Different weighting used can be used to obtain different behavior of the control system.

The eighth step in the control method is to perform the current limiting function. The purpose of this current limiting function is to limit the inverter currents commands generated by the voltage controller. At each sampling time the inverter currents commands generated by the voltage controller will be implemented by the inner current loop which will compute the necessary PWM command.

The current limiting is accomplished by limiting the magnitude of the inverter current commands vector and maintaining its original direction as shown in FIG. 16 . The current limit appears as a circle in the DQ stationary reference frame. If the command vector generated by the voltage controller goes outside of this circle, its magnitude is limited to be within the circle with its direction unaltered. Limiting the inverter currents in this way ensures that at each time the magnitudes of the 3-phase inverter currents in the abc reference frame remain within the limit value but still provides the phase relations required by the voltage control loop. Mathematically, this limiting function can be written as: $$ l → ⁢ cmd qd ⁡ ( k ) = { l → ⁢ cmd qd * ⁡ ( k ) l → ⁢ cmd qd * ⁡ ( k ) l → ⁢ cmd qd * ⁡ ( k ) ⁢ I max ⁢   ⁢ if if ⁢ &LeftBracketingBar; l → ⁢ cmd qd * ⁡ ( k ) l → ⁢ cmd qd * ⁡ ( k ) &RightBracketingBar; ⁢ ≤ I max ≤ I max

where absolute value indicates square root of sum of the squares.

The ninth step in the control method is to compute the errors between the limited inverter current commands and the actual inverter commands as follows: el q =Icmd q −linv q and el d =Icmd d −linv d . The discrete sliding mode current controller will force these errors to zero at the next sampling time (k+1) by computing the necessary PWM voltage command.

The tenth step is to compute the one-half step ahead prediction of the inverter voltages and currents and the load currents. These predicted values of the voltages and currents are necessary to account for the delay in the application of the PWM voltage due to DSP computation time.

{right arrow over (x)} 1 p ( k )=1.5· {right arrow over (x)} 1 ( k )−0.5· {right arrow over (x)} 1 ( k− 1) {right arrow over (d)} 1 p ( k )=1.5· {right arrow over (d)} 1 ( k )−0.5· {right arrow over (d)} 1 ( k− 1)  (28)

The eleventh step is to compute the necessary PWM voltages to implement the inverter current command using the Discrete sliding mode current controller. The Discrete Sliding Mode current controller ensures that at the next sampling time (k+1), the inverter current commands obtain the values of the command using the available DC bus voltage at the time. The necessary PWM voltages commands are given by: $$ V → ⁢ pwm qd = u ⁡ ( k ) = { u → eq ⁡ ( k ) for ⁢   ⁢ &LeftDoubleBracketingBar; u → eq ⁡ ( k ) &RightDoubleBracketingBar; ≤ u 0 u 0 ⁢ u → eq ⁡ ( k ) &LeftDoubleBracketingBar; u → eq ⁡ ( k ) &RightDoubleBracketingBar; for ⁢   ⁢ &LeftDoubleBracketingBar;   ⁢ u → eq ⁢ ( k ) &RightDoubleBracketingBar; ≤ u 0   {right arrow over (u)} eq ( k )=( C 1 B 1 *) −1 ( {right arrow over (I)} cmd qd )− C 1 A 1 *{right arrow over (x)} 1 p ( k )− C 1 E 1 *{right arrow over (d)} 1 p ( k )

Note the PWM voltage commands are limited by the amount of available DC bus voltage at that time. This limit value is denoted as u 0 in the above equations.

The twelfth and final step involves implementation of the PWM voltage commands into an actual PWM voltages using a standard Space Vector PWM technique. The Space Vector PWM technique takes the PWM voltages commands Vpwm q and Vpwm d , and computes the necessary timing to switch on and off the power switches.

Having shown and described a preferred embodiment of the invention, those skilled in the art will realize that many variations and modifications may be made to II affect the described invention and still be within the scope of the claimed invention. Thus, many of the elements indicated above may be altered or replaced by different elements which will provide the same result and fall within the spirit of the claimed invention.

For example, it is appreciated that the present invention may be used with single-phase systems. It is also appreciated that various inverter systems may be used as well as other Uninterruptible Power Supplies. It is the intention, therefore, to limit the invention only as indicated by the scope of the claims.

Claims

1. A method reducing transient overshoot of an output signal of an uninerruptable power supply, the uniterruptable power supply including: (a) an inverter having at least one power switch; (b) a digital controller that generates a pulse width modulated (PWM) signal for controlling the power switch, the PWM signal having a fixed period TPWM and (c) an analog to digital signal corresponding to the sampled output signal, the method comprising the steps of:i) operating the A/D converter to sample the output signal at a first time to generate a first sampled output value; ii) operating the A/D converter to sample the output signal at a second time to generate a second sampled output value, the period between the second time and the first time being equal to Tpwm; (iii) generating a control input value that is equal to 1.5 times the second sampled output value minus 0.5 times the first sampled output value; (iv) computing a PWM pulse duration as a function of the control input value; (v) initiating a pulse of the PWM signal at a third time, the PWM pulse having a duration equal to the PWM pulse duration, the period between the second time and the third time being equal to 0.5 TPWM.