Measurement data based method for identifying wind turbine generators which cause sub-synchronous oscillations in complex power system

Abstract

A measurement data based method for identifying wind turbine generators which cause sub-synchronous oscillations in a complex power system has a theoretical foundation of the open-loop modal resonance and the parallel filter design. The advantages of the present invention are as follows. 1) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system using measurement data instead of parametric model. Hence, it simplifies the computation and reduces the modeling cost effectively. 2) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system precisely with reduced amount of measurement data, reducing the cost of hardware and data measurement effectively. 3) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system based on the open-loop modal resonance theory.

Description

CROSS REFERENCE OF RELATED APPLICATION

The present invention claims priority under 35 U.S.C. 119(a-d) to CN 201811320582.6, filed Nov. 7, 2018.

BACKGROUND OF THE PRESENT INVENTION

Field of Invention

The present invention belongs to the technical scope of wind power integration. In specific, a measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in a complex power system is involved.

Description of Related Arts

In recent years, with the development of renewable power generation, the voltage source converters (VSCs) have been widely applied in integrations of wind power and photovoltaic power generation in together with the VSC-based DC transmissions. As the VSCs are of the advantages of flexible controllability, weak AC power system interconnection ability and free of commutation failure, they have become important devices for constructing a smart grid in future. However, the SSOs in power systems as caused by grid-connected wind turbine generators have been a serious threat to the stability of power systems.

In order to develop measures to suppress the SSOs caused by the grid connection of wind turbine generators, first of all, the wind turbine generators which cause the SSOs need to be identified in a complex power system. So far, two main methods of identification have been proposed and applied. They are the simulation model based identification method and frequency scan based identification method. The former identification method is based on simulation platforms and parametric model of power system. After the parametric model of the complex power system is derived, time-domain simulation or/and modal analysis can be conducted to identify the wind turbine generators which cause the SSOs in the complex power system.

Time-domain simulation is based on a conventional simulation platform. By using the modules provided in the simulation platform, operation of the power systems is approximated. During the simulation, the SSOs are reproduced and then the wind turbine generators which cause the SSOs are identified by running simulation and comparing the results of simulation. Nevertheless, main drawbacks of time-domain simulation method are as follows. First, the conventional simulation platforms are of two categories: time-domain software simulation platforms (such as Simulink or PSCAD) and real-time hardware simulation platforms (such as RTDS or RTLAB). No matter which kind of simulation platform is adopted to reproduce the SSOs in the power system, even small error during the simulation process can cause significant difference between the simulation results and field measurement. Reproduction of the SSOs is a challenging and time-consuming job. If the SSOs in a practical power system cannot be reproduced by simulation platforms completely, the simulation results obtained may be meaningless. Second, collecting accurate operational parameters of the practical power system is often very difficult, demanding significant amount of manpower and resources, especially when the scale of the power system is large. Furthermore, considering the complexity of time-variable operational conditions of the practical power system, the time-domain simulation method may very possibly not be able to replicate real operational status of the practical power system to reproduce the SSOs. Those disadvantages have limited the application of time-domain simulation method for identifying the wind turbine generators which cause the SSOs.

Modal analysis is based on the linearized model of the power system at a given operation point of power system, i.e., the state-space model. Power system stability is evaluated by calculating the oscillation modes. Drawbacks of modal analysis method are as follows: (1) Establishment of the linearized state-space model needs complete and accurate data of the power system, including the wind turbine generators, which may not be possible in practice; (2) Dimension of the established linearized state-space model may be too high to be handled with numerically.

Frequency scan based identification method is based on the measurement data. Hence, establishment of parametric model of the power system is not required. This is the main advantage of frequency scan method. Theoretical foundation of frequency scan based identification method is the Nyquist stability criterion. By measuring the impedance of power system and wind turbine generators, system SSO stability can be assessed as to be elaborated briefly as follows.

Normally, resistance of a power system is positive. Resistance of grid-connected wind turbine generators under various frequencies can be measured. If the measured resistance of a wind turbine generator is negative at a frequency, the wind turbine generator may possibly degrade the system stability at the frequency. Frequency scan based identification method is to find the frequency at which the resistance of wind turbine generator is negative such that the instability risk brought about by the wind turbine generator is detected. Drawbacks of the frequency scan based identification method are as follows: (1) The measurement needs to be carried out sequentially by disconnecting each of wind turbine generators with the power system. Thus, system operating points vary when the measurement is conducted for each of wind turbine generators in the power system. None of those operating points can be the real operating point at which all the wind turbine generators are connected with the power system; (2) Positive resistance is the sufficient condition of system stability. Hence, even though the resistance of a wind turbine generator is found being negative, theoretically it cannot conclude that the system is unstable.

SUMMARY OF THE PRESENT INVENTION

In order to solve the drawbacks of conventional methods, a novel measurement data based method for identifying the wind turbine generators which cause the SSOs in a complex power system is proposed. The advantages of the present invention are as follows.

1) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system using measurement data instead of parametric model. Hence, it simplifies the computation and reduces the modeling cost effectively.

2) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system precisely with reduced amount of measurement data, reducing the cost of hardware and data measurement effectively.

3) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system based on the open-loop modal resonance theory, which provides a necessary and sufficient condition for the evaluation of system stability.

The specific technical schemes are demonstrated as follows:

(1) obtaining a set of measurement data when the SSOs occur in the power system; reducing noise by applying a filtering algorithm; identifying an SSO frequency by using a signal processing method; wherein the identified SSO frequency is denoted as f_s.

It should be noted that the signal processing method could be any method which can determine the oscillation frequency (such as the Prony analysis or FFT method).

(2) designing a parallel band-pass filter to ensure that only a signal at frequency f_s passes through to earth and signals at other frequencies are suppressed.

It should be noted that the designed filter is acceptable if its magnitude-frequency characteristic is similar to the one shown in FIG. 6, where (1−n %)f_s and (1+n %)f_s are cutoff frequencies of the designed filter, value of n % is usually less than 30%.

(3) for a power system with n wind turbine generators, installing the parallel band-pass filter designed in the step (2) on an interface between each of the wind turbine generators and the power system; wherein subsequently, dynamic interactions between the wind turbine generators and the power system at the SSO frequency f_s are suppressed and filtered out; whilst those at other frequencies are not affected.

It should be noted that although the point of common connection (PCC) is generally selected as the interface, actually any point which can divide a wind turbine generator from the power system can be chosen as the interface. The points that can be chosen as the interface are shown in FIG. 7.

(4) measuring a response from each of the wind turbine generators by adding a disturbance signal on a side of a wind turbine generator near the band-pass filter.

It should be noted that the injection point of disturbance signal and measurement point should be located on the side of wind turbine generator near the band-pass filter which are shown in FIG. 8.

Moreover, disturbance signal should excite observable dynamic response of the wind turbine generator. FIG. 9 gives an impulse signal as an example of disturbance signal. The time duration and magnitude of the signal are respectively 0.1 second and 2% of the magnitude of terminal voltage at the node where the wind turbine generator is connected with the power system.

Furthermore, wind turbine generator's response is chosen to be the active power output from the wind turbine generator.

(5) from the measurement data of the response of an i-th wind turbine generator, identifying an oscillation frequency, f_ij, and a residue R_ij by applying a signal processing method; wherein in notations above for f_ij and R_ij, subscript i refers to the i-th wind turbine generator and the subscript j refers to the j-th SSO frequency identified;

afterwards, comparing f_ij and f_s; recording the f_ij and R_ij if a following condition is met:

1.05*f_s≥f_ij≥0.95*f_s;

wherein finally, if no f_ij which can meet the above condition, it is concluded that the i-th wind turbine generator is not the wind turbine generator which causes the SSOs in the power system.

It should be noted that the signal processing method could be any method which can determine the oscillation frequency (such as the Prony analysis or FFT method).

(6) measuring the response from the power system by adding a disturbance signal on the side of the power system near the band-pass filter for the i-th wind turbine generator.

It should be noted that the injection point of disturbance signal and measurement point should be located on the side of power system near the band-pass filter which are shown in FIG. 10.

Moreover, the disturbance signal should excite the observable dynamic response of the power system. FIG. 9 gives an impulse signal as an example of disturbance signal. The time duration and magnitude of the signal are respectively 0.1 second and 2% of the magnitude of terminal voltage at the node where the i-th wind turbine generator is connected with the power system.

Furthermore, power system response is chosen to be a magnitude of a terminal voltage at a node in the power system where the i-th wind turbine generator is connected to the power system.

(7) from the measurement data of the response, identifying an oscillation frequency f_aij and a residue R_aij by using the signal processing method: wherein in notations of f_aij and R_aij, subscript a indicates that the measurement data is about the power system, subscript i refers to the i-th wind turbine generator and subscript j refers to the j-th oscillation frequency and residue;

afterwards, comparing f_aij and f_s; recording f_aij and R_aij if a following condition is met:

1.05*f_s≥f_aij≥0.95*f_s;

It should be noted that the signal processing method could be any method which can determine the oscillation frequency (such as the Prony analysis or FFT method).

(8) Computing Z_i=|R_ijR_aij|, i=1.2, L n based on recorded results in the step (5) and (7); wherein if Z_k is the largest among Z_i, i=1.2, L n, the i-th wind turbine generator is identified to be the wind turbine generator causing the SSOs in the power system.

Theoretical foundation of the present invention is the open-loop modal resonance and the parallel filter design. They are introduced respectively as follows.

1. Open-Loop Modal Resonance Theory

FIG. 1 shows an AC power system integrated with multiple wind turbine generators, where P_WTGi+_WTGi is the power output of the i-th wind turbine generator (WTG-i); V_cdi+jV_cqi is the terminal voltage of and I_di+jI_qi is the current output from WTG-i; V_pcci and θ_pcci are the magnitude and phase angle of the voltage at the PCC; X_fi is the reactance of the filter. By choosing the PCC of the i-th WTG as the point of interference, the power system can be divided into two subsystems. They are the WTG-i subsystem and subsystem of remainder of power system (ROPS) [1-2].

Normally, the WTG operates with high even unity power factor such that dynamic variation of reactive power output from the i-th WTG is very small or zero [3-6].

Consequently, the input dynamic variable to the WTG-i subsystem is the variation of magnitude of terminal voltage at the PCC. The output dynamic variable from the WTG-i subsystem is the variation of active power output from WTG-i. The input dynamic variable to the ROPS subsystem is the variation of active power output from WTG-i. The output dynamic variable is the variation of magnitude of terminal voltage at the PCC [1-2]. Therefore, linearized models of the WTG-i subsystem and the ROPS subsystem can be described by Eq. (1) and (2) respectively.

$\begin{array}{cc}\frac{d}{\mathrm{dt}}\Delta X_{\mathrm{WTGi}}=A_{\mathrm{WTGi}}\Delta X_{\mathrm{WTGi}}+b_{\mathrm{WTGi}}\Delta V_{\mathrm{pcci}}\Delta P_{\mathrm{WTGi}}=c_{\mathrm{WTGi}}^{\underset{\_}{T}}\Delta X_{\mathrm{WTGi}}& \left(1\right)\\ \frac{d}{\mathrm{dt}}\Delta X_{\mathrm{aci}}=A_{\mathrm{aci}}\Delta X_{\mathrm{aci}}+b_{\mathrm{aci}}\Delta P_{\mathrm{WTGi}}\Delta V_{\mathrm{pcci}}=c_{\mathrm{aci}}^T\Delta X_{\mathrm{aci}}& \left(2\right)\end{array}$

Eq. (1) is the state-space model of WTG-i subsystem. Open-loop oscillation modes of WTG-i subsystem are λ_vn, n=1, 2 . . . N which are the eigenvalues of state-space matrix A_WTGi. Eq. (2) is the state-space model of the ROPS subsystem. Open-loop oscillation modes of ROPS subsystem are λ_acm, m=1, 2 . . . M which are the eigenvalues of the state-space matrix A_aci. From Eq. (1) and (2), following transfer function models of the WTG-i subsystem and the ROPS subsystem are obtained

ΔP_WTGi=c_WTGi^T(sI−A_WTGi)⁻¹b_WTGiΔV_pcci=K_WTGi(s)ΔV_pcci  (3)

ΔV_pcci=c_aci^T(sI−A_aci)⁻¹b_aciΔP_WTGi=K_aci(s)ΔP_WTGi  (4)

For the power system integrated with multiple wind turbine generators, dynamic interactions between the wind turbine generators and the power system are usually weak. Since ΔP_WTGi in (3) is the exhibition of the dynamic interactions, normally it should have ΔP_WTGi≈0. Hence, there should exist a small gain value ε(ε<<1) such that the transfer function of WTG-i subsystem can be expressed as K_WTGi(s)=ε_iG_WTGi(s). Therefore, Eq. (3) can be written as

ΔP_WTGi=K_WTGi(s)ΔV_pcci=ε_iG_WTGi(s)ΔV_pcci  (5)

By combining Eq. (4) and Eq. (5), the power system shown in FIG. 1 can be represented by a closed-loop interconnected model which is displayed in FIG. 2.

Oscillation modes of closed-loop interconnected model thus are referred as closed-loop oscillation modes. Accordingly, oscillation modes of WTG-i subsystem and ROPS subsystem are called open-loop oscillation modes as introduced above Eq. (1) and (2). They are λ_vn, n=1, 2 . . . N and λ_acm, m=1, 2 . . . M. Obviously, the closed-loop oscillation modes are the solutions of K_aci(s)ε_iG_WTGi(s)=1. In addition, ΔP_WTGi, is the exhibition of the dynamic interactions between the WTG-i subsystem and the ROPS subsystem.

As being mentioned before, dynamic interactions between WTG-i and the power system are usually weak such that ΔP_WTGi≈0. Consequently, the closed-loop interconnected model is approximately open such that the closed-loop oscillation modes are approximately equal to the open-loop oscillation modes. This implies that normally integration of the WTG may not affect the stability of closed-loop interconnected system considerably. However, when an open-loop oscillation mode of WTG-i subsystem and an open-loop oscillation mode of ROPS subsystem in FIG. 2 are close to each other on the complex plane, stability of closed-loop system shown by FIG. 2 may be affected considerably. This particular modal condition is examined as follows. The characteristic equation of closed-loop interconnected system shown by FIG. 2 is

εK_ac(s)G_WTG(s)=1  (6)

Without loss of generality, donate λ_vi and λ_aci as open-loop oscillation modes of the WTG-i subsystem and the ROPS subsystem respectively. Transfer functions of two subsystems can be expressed as

$\begin{array}{cc}{K}_{\mathrm{ac}}\left(s\right)=\frac{k_{\mathrm{ac}}\left(s\right)}{s-{\lambda }_{\mathrm{aci}}},G_{\mathrm{WTG}}\left(s\right)=\frac{g_{\mathrm{WTG}}\left(s\right)}{s-{\lambda }_{\mathrm{vi}}}& \left(7\right)\end{array}$

By substituting Eq. (7) into Eq. (6), it can have

εk_ac(s)g_WTG(s)=(s−λ_aci)(s−λ_vi)  (8)

Eq. (8) can be simply expressed as

εf(s)=λ(s)  (9)

where f(s)=k_ac(s)g_WTG(s), λ(s)=(s−λ_aci)(s−λ_vi).

Obviously, the closed-loop oscillation modes {circumflex over (λ)}_vi={circumflex over (λ)}_vi+Δλ_vi and {circumflex over (λ)}_aci={circumflex over (λ)}_aci+Δλ_aci are the solutions of Eq. (9). By substituting {circumflex over (λ)}_vi into Eq. (9), it can have

εf(λ_vi+Δλ_vi)=λ(λ_aci+Δλ_aci)  (10)

Expand Eq. (10) at λ_vi on the basis of Taylor's theorem, it can have

$\begin{array}{cc}\varepsilon [f\left({\lambda }_{\mathrm{vi}}\right)+f^{\prime }\left({\lambda }_{\mathrm{vi}}\right)\Delta {\lambda }_{\mathrm{vi}}+\frac{f^{″}\left({\lambda }_{\mathrm{vi}}\right)}{2!}\Delta {\lambda }_{\mathrm{vi}}^2+\dots ]=\lambda \left({\lambda }_{\mathrm{vi}}\right)+{\lambda }^{\prime }\left({\lambda }_{\mathrm{vi}}\right)\Delta {\lambda }_{\mathrm{vi}}+\frac{{\lambda }^{″}\left({\lambda }_{\mathrm{vi}}\right)}{2!}\Delta {\lambda }_{\mathrm{vi}}^2+\dots & \left(11\right)\end{array}$

Substitute Δλ_vi=α₁ε+α₂ε²+α₃ε³+ . . . [7] into Eq. (11), it can have

$\begin{array}{cc}\varepsilon [f\left({\lambda }_{\mathrm{vi}}\right)+f^{\prime }\left({\lambda }_{\mathrm{vi}}\right)({\alpha }_1\varepsilon +{\alpha }_2{\varepsilon }^2+{\alpha }_3{\varepsilon }^2+\dots )+\frac{f^{″}\left({\lambda }_{\mathrm{vi}}\right)}{2!}{({\alpha }_1\varepsilon +{\alpha }_2{\varepsilon }^2+{\alpha }_3{\varepsilon }^3+\dots )}^2+\dots ]=\lambda \left({\lambda }_{\mathrm{vi}}\right)+{\lambda }^{\prime }\left({\lambda }_{\mathrm{vi}}\right)({\alpha }_1\varepsilon +{\alpha }_2{\varepsilon }^2+{\alpha }_3{\varepsilon }^3+\dots )+\frac{{\lambda }^{″}\left({\lambda }_{\mathrm{vi}}\right)}{2!}{({\alpha }_1\varepsilon +{\alpha }_2{\varepsilon }^2+{\alpha }_3{\varepsilon }^3+\dots )}^2+\dots & \left(12\right)\end{array}$

As ε is infinitesimal small number, the high-order terms in Eq. (12) can be ignored. Eq. (12) can be rewritten as

$\begin{array}{cc}{\alpha }_1=\frac{f\left({\lambda }_{\mathrm{vi}}\right)}{{\lambda }^{\prime }\left({\lambda }_{\mathrm{vi}}\right)}=\frac{f\left({\lambda }_{\mathrm{vi}}\right)}{{\lambda }_{\mathrm{vi}}-{\lambda }_{\mathrm{aci}}}& \left(13\right)\end{array}$

Substitute Eq. (13) into Δλ_vi=α₁ε+α₂ε²+α₃ε³+ . . . , the approximate solution for the variation of the closed-loop oscillation mode Δλ_vi is obtained to be

$\begin{array}{cc}\Delta {\lambda }_{\mathrm{vi}}={\hat{\lambda }}_{\mathrm{vi}}-{\lambda }_{\mathrm{vi}}\approx {\alpha }_1\varepsilon =\varepsilon \frac{k_2\left({\lambda }_{\mathrm{vi}}\right)g\left({\lambda }_{\mathrm{vi}}\right)}{{\lambda }_{\mathrm{vi}}-{\lambda }_{\mathrm{aci}}}& \left(14\right)\end{array}$

Similarly, the approximate solution for the variation of closed-loop oscillation mode Δλ_aci is obtained to be

$\begin{array}{cc}\Delta {\lambda }_{\mathrm{aci}}={\hat{\lambda }}_{\mathrm{aci}}-{\lambda }_{\mathrm{aci}}\approx {\alpha }_1\varepsilon =\varepsilon \frac{k_2\left({\lambda }_{\mathrm{aci}}\right)g\left({\lambda }_{\mathrm{aci}}\right)}{{\lambda }_{\mathrm{aci}}-{\lambda }_{\mathrm{vi}}}& \left(15\right)\end{array}$

From Eq. (14) and Eq. (15), it can be seen that when the open-loop oscillation modes λ_vi and λ_aci are far away from each other on the complex plane, the dynamic interactions between two subsystems are weak because the difference between the closed-loop and open-loop oscillation modes is small as it is proportional to the small number ε.

However, when the open-loop oscillation modes of those subsystems are close to each other on the complex plane, the dynamic interactions between two subsystems may become significant because Δλ_vi and Δλ_aci may not be small any more as being indicated by (15). In addition, the difference between closed-loop and open-loop oscillation modes for two open-loop oscillation modes are of opposite signs, i.e. Δλ_vi=−Δλ_aci, which implies that two closed-loop oscillation modes are located at the opposite positions on the complex plane in respect to their corresponding open-loop oscillation modes. This can be further elaborated for an extreme open-loop modal condition, λ_vi=λ_aci, as follows.

Under the condition of λ_vi=λ_aci, by substituting Δλ_vi=β₁ε^1/2+β₂ε^2/2+β₃ε^3/2+ . . . into Eq. (11), it can have

$\begin{array}{cc}\varepsilon [f\left({\lambda }_{\mathrm{vi}}\right)+f^{\prime }\left({\lambda }_{\mathrm{vi}}\right)({\beta }_1{\varepsilon }^{\frac{1}{2}}+{\beta }_2{\varepsilon }^{\frac{2}{2}}+{\beta }_3{\varepsilon }^{\frac{3}{2}}+\dots )+\frac{f^{″}\left({\lambda }_{\mathrm{vi}}\right)}{2!}{({\beta }_1{\varepsilon }^{\frac{1}{2}}+{\beta }_2{\varepsilon }^{\frac{2}{2}}+{\beta }_3{\varepsilon }^{\frac{3}{2}}+\dots )}^2+\dots ]=\lambda \left({\lambda }_{\mathrm{vi}}\right)+{\lambda }^{\prime }\left({\lambda }_{\mathrm{vi}}\right)({\beta }_1{\varepsilon }^{\frac{1}{2}}+{\beta }_2{\varepsilon }^{\frac{2}{2}}+{\beta }_3{\varepsilon }^{\frac{3}{2}}+\dots )+\frac{{\lambda }^{″}\left({\lambda }_{\mathrm{vi}}\right)}{2!}{({\beta }_1{\varepsilon }^{\frac{1}{2}}+{\beta }_2{\varepsilon }^{\frac{2}{2}}+{\beta }_3{\varepsilon }^{\frac{3}{2}}+\dots )}^2+\dots & \left(16\right)\end{array}$

Under the condition of λ_vi=λ_aci, it can be obtained that λ(λ_vi)′=0 and λ(λ_vi)″=2. Therefore, by cancelling out high-order terms, Eq. (16) can be simplified as

$\begin{array}{cc}f\left({\lambda }_{\mathrm{vi}}\right)=\frac{{\lambda }^{″}\left({\lambda }_{\mathrm{vi}}\right)}{2!}{\beta }_1^2& \left(17\right)\end{array}$

By expanding open-loop transfer functions of subsystems, G_WTG(s) and K_ac(s) can be expressed as

$\begin{array}{cc}{G}_{\mathrm{WTG}}\left(s\right)=\frac{R_{\mathrm{vi}}}{\left(s-{\lambda }_{\mathrm{vi}}\right)}+\sum _{j=2}^n\frac{R_{\mathrm{vj}}}{\left(s-{\lambda }_{\mathrm{vj}}\right)}K_{\mathrm{ac}}\left(s\right)=\frac{R_{\mathrm{aci}}}{\left(s-{\lambda }_{\mathrm{aci}}\right)}+\sum _{j=2}^m\frac{R_{\mathrm{acj}}}{\left(s-{\lambda }_{\mathrm{acj}}\right)}& \left(18\right)\end{array}$

where R_vi=1, 2, . . . , n are the residues corresponding to λ_vi=1, 2 . . . n; R_aci, i=1, 2, . . . m are the residues corresponding to λ_aci, i=1, 2 . . . m. Therefore, it can have

f(λ_vi)=(s−λ_vi)(s−λ_aci)G_WTG(s)K_ac(s)|_s=λvi=R_viR_aci  (19)

By combining Eq. (17) and Eq. (19), it can have

β₁=±√{square root over (R_viR_aci)}  (20)

From Eq. (20), it can have

{circumflex over (λ)}_vi≈λ_vi+√{square root over (εR_viR_aci)}

{circumflex over (λ)}_aci≈λ_aci−√{square root over (εR_viR_aci)}  (21)

It can be concluded from Eq. (21) that, if the open-loop subsystems are stable, the closed-loop system is prone to be destabilized with larger modal residues when the open-loop modal resonance happens, i.e., λ_vi≈λ_aci. Particularly, the system may be unstable under the condition that |Real(√{square root over (εR_viR_aci)})|>|Real(λ_vi)|.

2. Parallel Filtering Method

According to the theory of open-loop modal resonance, a wind turbine generator which causes the SSOs can be detected by identifying the open-loop SSO modes and comparing the associated residues of wind turbine generator and power system. When the identification is conducted by using the measurement data, the measurement data of open-loop subsystems, i.e., the wind turbine generator and power system, need to be obtained. The SSOs are caused by the grid connection of wind turbine generator. This means that the SSOs occur in the power system when the wind turbine generator is connected. For the detection of wind turbine generator to cause the SSOs, open-loop SSO modes and residues of wind turbine generator and power system need to be identified when the wind turbine generator is connected. If the measurement is conducted by disconnecting the wind turbine generator, though the measurement data obtained are about the open-loop wind turbine generator and power system such that the open-loop SSO modes and associated residues are obtained, it is not ensured that the open-loop SSO modes and associated residues are equal to the required open-loop SSO modes and residues when the wind turbine generator is connected to the power system. This problem is solved by parallel filtering method to be introduced as follows.

Parallel filtering method is to identify the open-loop SSO frequency and associated residues of the wind turbine generators and the power system when the wind turbine generator is connected to the power system. That is to identify the open-loop SSO modes and associated residues of open-loop subsystems on the basis of measurement data obtained from the closed-loop interconnected system

FIG. 2 shows the closed-loop interconnected model with the wind turbine generator subsystem (represented by G_WTGi(s)) and the subsystem of remainder of power system (ROPS, represented by K_aci(s)). When the subsystems are expressed in the form of modal decomposition, configuration of closed-loop interconnected model is shown by FIG. 3. For the convenience of introduction, λ_aci, R_aci and λ_vi, R_vi are denoted as the SSO modes and associated residues of ROPS subsystem and wind turbine generator subsystem to be identified respectively. Frequency of SSOs is known to be f_s. The wind turbine generator which causes the SSOs needs to be detected by using the measurement data. Procedure to apply the parallel filtering method to obtain the data is as follows.

(1) Design a parallel band-pass filter to ensure that only the signal at frequency f_s can pass through to earth, the signals at other frequencies are suppressed. The ideal magnitude-frequency characteristic of the filter is shown in FIG. 4.

(2) With the parallel band-pass filter being installed, configuration of the equivalent closed-loop interconnected model of the power system with the wind turbine generator is shown by FIG. 5. Subsequently, dynamic interactions between the wind turbine generators and the power system at the SSO frequency f_s are suppressed and filtered out, whilst at other frequencies are not affected.

(3) From FIG. 5, it can be seen that the parallel filter suppresses the component in ΔP_WTGi and ΔV_pcci at the SSO frequency, f_s. This implies that the dynamic interactions between the wind turbine generator and power system at the SSO frequency, f_s, are suppressed. Since the frequencies of SSO modes, λ_aci and λ_vi, are near f_s. Hence, with the band-pass filter being installed, the closed-loop interconnected system becomes effectively open. Consequently, the modal information of two open-loop subsystems, i.e., R_aci, λ_aci, R_vi and λ_vi, can be measured successfully without having to physically disconnect the wind turbine generator from the power system.

The measurement data based method for identifying the wind turbine generators which cause the SSOs in the complex power system proposed in the present invention is of the following advantages.

1) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system by using the measurement data without having to establish the parametric model.

2) The present invention can identify the wind turbine generators which cause the SSOs in the complex power system by using small amount of measurement data, reducing the cost of hardware and data measurement.

3) The present invention identifies the wind turbine generators which cause the SSOs in the complex power system based on the necessary and sufficient condition of system SSO stability, which is derived from the open-loop modal resonance theory

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 illustrates configuration of the power system integrated with multiple WTGs;

FIG. 2 illustrates closed-loop transform function of the power system integrated with multiple WTGs;

FIG. 3 illustrates expansion of power system by multiple oscillation modes and corresponding residues;

FIG. 4 illustrates magnitude-frequency characteristic of the ideal band-pass filter;

FIG. 5 illustrates expansion of the power system installed with the band-pass filter;

FIG. 6 illustrates magnitude-frequency characteristic of the ideal band-pass filter;

FIG. 7 illustrates points could be chosen to install filter in the step (3);

FIG. 8 illustrates points could be chosen to inject disturbance signal and monitor signal on the side of the i-th wind turbine generator in the step (4);

FIG. 9 illustrates the typical impulse signal in the step (4);

FIG. 10 illustrates points could be chosen to inject disturbance signal and measuring signal on the side of the power system in the step (6);

FIG. 11 illustrates technical schemes of the present invention;

FIG. 12 illustrates configuration of 4-machine 2-area power system integrated with four wind turbine generators;

FIG. 13(a) illustrates magnitude-frequency characteristic of the low-pass filter;

FIG. 13(b) illustrates active power output from PMSG-1 before filtering;

FIG. 13(c) illustrates active power output from PMSG-1 after filtering;

FIG. 14 illustrates FFT analysis results of the measurement data in FIG. 13;

FIG. 15 illustrates magnitude-frequency and phase-frequency characteristics of the high-pass filter;

FIG. 16(a) illustrates configuration of single-machine infinite power system integrated with a wind turbine generator;

FIG. 16(b) illustrates configuration of single-machine infinite power system integrated with a wind turbine generator and the parallel band-pass filter;

FIG. 17(a) illustrates active power output from wind turbine generator without the parallel band-pass filter;

FIG. 17(b) illustrates active power output from wind turbine generator with the parallel band-pass filter;

FIG. 18 illustrates installation location (1) of the high-pass filter;

FIG. 19 illustrates typical disturbance signal;

FIG. 20(a) illustrates active power output from the wind turbine generator before filtering;

FIG. 20(b) illustrates active power output from the wind turbine generator after filtering;

FIG. 21 illustrates point for injecting the disturbance signal and measuring signal;

FIG. 22(a) illustrates active power output from AC system before filtering;

FIG. 22(b) illustrates active power output from AC system after filtering;

FIG. 23 illustrates installation location (2) of the high-pass filter;

FIG. 24(a) illustrates active power output from the wind turbine generator before filtering;

FIG. 24(b) illustrates active power output from the wind turbine generator after filtering;

FIG. 25 illustrates points for injecting the disturbance signal and measuring signal;

FIG. 26(a) illustrates active power output from AC system before filtering;

FIG. 26(b) illustrates active power output from AC system after filtering;

FIG. 27 illustrates installation location (3) of the high-pass filter;

FIG. 28(a) illustrates active power output from the wind turbine generator before filtering;

FIG. 28(b) illustrates active power output from the wind turbine generator after filtering;

FIG. 29 illustrates points for injecting the disturbance signal and measuring signal;

FIG. 30(a) illustrates active power output from AC system before filtering;

FIG. 30(b) illustrates active power output from AC system after filtering;

FIG. 31 illustrates installation location (4) of the high-pass filter;

FIG. 32(a) illustrates active power output from the wind turbine generator before filtering;

FIG. 32(b) illustrates active power output from the wind turbine generator after filtering;

FIG. 33 illustrates points for injecting the disturbance signal and measuring signal;

FIG. 34(a) illustrates active power output from AC system before filtering;

FIG. 34(b) illustrates active power output from AC system after filtering;

FIG. 35 illustrates configuration of the power system without DFIG-1

FIG. 36 illustrates active power output from SG-1 without DFIG-1;

FIG. 37 illustrates active power output from SG-2 without DFIG-1;

FIG. 38 illustrates active power output from SG-3 without DFIG-1;

FIG. 39 illustrates active power output from SG-4 without DFIG-1;

FIG. 40 illustrates configuration of the complex power system integrated with four wind turbine generators;

FIG. 41(a) illustrates magnitude-frequency characteristic of the low-pass filter;

FIG. 41(b) illustrates frequency response of the power system before filtering;

FIG. 41(c) illustrates frequency response of the power system after filtering;

FIG. 42 illustrates FFT analysis results of the measurement data in FIG. 41;

FIG. 43 illustrates magnitude-frequency and phase-frequency characteristics of the high-pass filter;

FIG. 44 illustrates installation location (1) of the high-pass filter;

FIG. 45 illustrates the typical disturbance signal;

FIG. 46(a) illustrates active power output from PMSG-1 before filtering;

FIG. 46(b) illustrates active power output from PMSG-1 after filtering;

FIG. 47 illustrates points for injecting the disturbance signal and measuring signal;

FIG. 48(a) illustrates active power output from SG-1 before filtering;

FIG. 48(b) illustrates active power output from SG-2 after filtering

FIG. 49 illustrates installation location (2) of the high-pass filter;

FIG. 50(a) illustrates active power output from PMSG-2 before filtering;

FIG. 50(b) illustrates active power output from PMSG-2 after filtering;

FIG. 51 illustrates points for injecting the disturbance signal and measuring signal;

FIG. 52(a) illustrates active power output from SG-8 before filtering;

FIG. 52(b) illustrates active power output from SG-8 after filtering;

FIG. 53 illustrates installation location (3) of the high-pass filter;

FIG. 54(a) illustrates active power output from PMSG-3 before filtering;

FIG. 54(b) illustrates active power output from PMSG-3 after filtering;

FIG. 55 illustrates points for injecting the disturbance signal and measuring signal;

FIG. 56(a) illustrates active power output from SG-10 before filtering;

FIG. 56(b) illustrates active power output from SG-10 after filtering;

FIG. 57 illustrates installation location (4) of the high-pass filter;

FIG. 58(a) illustrates active power output from PMSG-4 before filtering;

FIG. 58(b) illustrates active power output from PMSG-4 after filtering;

FIG. 59 illustrates points for injecting the disturbance signal and measuring signal;

FIG. 60(a) illustrates active power output from SG-7 before filtering;

FIG. 60(b) illustrates active power output from SG-7 after filtering;

FIG. 61 illustrates configuration of the power system without PMSG-1;

FIG. 62(a) illustrates active power output from PMSG-2 without PMSG-1;

FIG. 62(b) illustrates active power output from PMSG-3 without PMSG-1;

FIG. 62(c) illustrates active power output from PMSG-4 without PMSG-1.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

The flowchart of the present invention shown in FIG. 11 is demonstrated by following two examples.

EXAMPLE 1

FIG. 12 shows a 4-machine 2-area power system integrated with two PMSGs and two DFIGs, which represent four grid-connected wind farms. When the SSOs occur, the wind turbine generators which cause the SSOs in the power system are identified by the present invention without using the parametric model of the power system. Detailed steps of identification are as follows.

1. Determine the SSO Frequency

In order to identify a wind turbine generator which causes the SSOs, the SSO frequency in the power system needs to be determined firstly. Afterwards, the band-pass filter can be designed near this frequency.

Step 1: reducing the noise in the measurement data.

A low-pass filter depicted in FIG. 13(a) is designed to reduce the harmonic components in the measurement data.

Step 2: determining the SSO frequency by the signal processing method.

The SSOs are observed in the active power output from wind turbine generators, for example that from PMSG-1 shown in FIG. 13 (c). The FFT analysis results are displayed in FIG. 14. From FIG. 14, it can be concluded that the SSO frequency in the power system is 15 Hz, i.e. f_s=15 Hz

2. Design a Parallel Band-Pass Filter

Step 3: designing a parallel band-pass filter.

Design a parallel band-pass filter to ensure that the signal at frequency 15 Hz can pass through to earth, the signals at other frequencies are suppressed. Subsequently, dynamic interactions between the wind turbine generators and the power system at the SSO frequency f_s are suppressed and filtered out; whilst at other frequencies are not affected. A 4th order band-pass filter is designed by Besself method as shown by Eq. (22), where the cut-off frequencies are 14 Hz and 16 Hz.

$\begin{array}{cc}{G}_T\left(s\right)=\frac{4s^2}{s^4+3.4641s^3+452s^2+775.96s+50176}& \left(22\right)\end{array}$

Results of magnitude-frequency and phase-frequency characteristics of the filter designed by Eq. (22) are shown in FIG. 15. From FIG. 15, it can be concluded that the filter designed by Eq. (22) can meet the requirement.

Step 4: verifying the effectiveness of the parallel band-pass filter designed in the step 3.

A single-machine infinite-bus power system installed with PMSG-1 is shown by FIG. 16(a). It is used to verify the effectiveness of the parallel band-pass filter. FIG. 16(b) shows that the parallel band-pass filter is installed at the interface between the PMSG-1 and remainder of the power system (ROPS).

When the SSOs occur in the power system of FIG. 16(a), the SSO frequency is calculated to be 15 Hz from the measurement data of active power output from PMSG-1. After the parallel band-pass filter is installed, the active power output from PMSG-1 is measured which is shown in FIG. 17(b). By comparing FIG. 17(b) and FIG. 17(a), it can be seen that the 15 Hz component in the variation of active power output from PMSG-1 is reduced effectively. Therefore, the effectiveness of the parallel band-pass filter is verified.

3. Measure Open-Loop Residues

Open-loop residues of each wind turbine generator represented by either PMSG or DFIG and the ROPS can be measured by installing a parallel band-pass filter at the PCC of each wind turbine generator.

Step 5: installing a parallel band-pass filter on the side of PMSG-1 and measure the residues.

The installing location of the band-pass filter, the place to inject the disturbance signal and to gain the measurement data are indicated in FIG. 18. The disturbance signal is 2% increase of magnitude of PCC voltage for 0.1 second as shown by FIG. 19. The measurement data gained on the side of PMSG-1 is shown by FIG. 20(a). With the noise being reduced, the same measurement data is shown in FIG. 20(b).

Prony analysis is applied to the measurement data shown by FIG. 20(b) to obtain the residue of PMSG-1 to be −0.0001+0.0003j.

Similarly, in order to measure the residue of the ROPS subsystem, the parallel band-pass filter is installed on the side of power system near PMSG-1. The installing location of the band-pass filter, the place to inject the disturbance signal and to gain the measurement data are indicated in FIG. 21. The disturbance signal is the same to that shown by FIG. 19. The measurement data gained on the side of power system and noise reduced data are shown in FIG. 22(a) and FIG. 22(b), respectively.

From the measurement data displayed in FIG. 22(b), the residue of the ROPS subsystem is estimated by the Prony analysis to be −0.0441+0.0473j.

Step 6: installing a parallel band-pass filter on the side of PMSG-2 and measure the residues.

The procedure to measure the residues for PMSG-2 is as same as that presented above in the step 5 for PMSG-1. The measurement data on the side of PMSG-2 and the noise reduced data are shown in FIG. 24(a) and FIG. 24(b). The residue of PMSG-1 is obtained to be −0.0001+0.0003j.

The measurement data and noise reduced data on the side of power system are shown by FIG. 26(a) and FIG. 26(b) respectively. The residue of the ROPS subsystem is obtained to be −0.0820+0.0405j.

Step 7: installing a parallel band-pass filter on the side of DFIG-2 and measure residues.

The procedure to measure the residues for DFIG-2 is as same as that presented above in the step 5 for PMSG-1. The measurement data on the side of DFIG-2 and the noise reduced data are shown in FIG. 28(a) and FIG. 28(b). The residue of DFIG-2 is obtained to be 0.0003+0.0006j.

The measurement data and noise reduced data on the side of power system are shown by FIG. 30(a) and FIG. 30(b) respectively. The residue of the ROPS subsystem is obtained to be −0.0502+0.0413j.

Step 8: installing a parallel band-pass filter on the side of DFIG-1 and measure residues.

The procedure to measure the residues for DFIG-1 is as same as that presented above in the step 5 for PMSG-1. The measurement data on the side of DFIG-1 and the noise reduced data are shown in FIG. 32(a) and FIG. 32(b). The residue of DFIG-1 is obtained to be −0.0078+0.0050j.

The measurement data and noise reduced data on the side of power system are shown by FIG. 34(a) and FIG. 34(b) respectively. The residue of the ROPS subsystem is obtained to be −0.0976+0.0805j.

4. Identify the Wind Turbine Generators which Cause the SSOs According to the Index

Step 9: calculating the indexes.

The indexes Z_i=|√{square root over (R_iR_ai)}|, i=1, 2, 3, 4 are computed based on the residues obtained above, where R_i and R_ai, i=1, 2, 3, 4 are the residues of each wind turbine generator and the corresponding ROPS subsystem, respectively. The computational results are presented in Tab. 1-1.

TABLE 1-1
Results of residues and corresponding index obtained by using the
present invention
	Index	Value	The cause of the SSOs
	Z4 (DFIG-1)	0.0291 − 0.0016i	Yes
	Z3 (DFIG-2)	0.0054 − 0.0038i	No
	Z1 (PMSG-1)	0.0036 + 0.0052i	No
	Z2 (PMSG-2)	0.0055 + 0.0031i	No

Step 10: identifying the wind turbine generator which causes the SSOs.

According to the open-loop modal resonance theory, the wind turbine generator corresponding to the maximum index among all Zi is identified to be the wind turbine generator causing the SSOs. Therefore, DFIG-1 is identified to be the cause of the SSOs in the four-machine two-area power system.

5. Verify the Correctness of Result Obtained by the Present Invention

The linearized model of FIG. 9 is used in this section to verify the correctness of the above identification made by the present invention.

Step 11: establishing the parametric model of the power system integrated with the wind turbine generators.

The open-loop linearized models of each wind turbine generator subsystem and the corresponding ROPS subsystem described by Eq. (1) and Eq. (2) are respectively derived. Following notations are used to denote various results of derived parametric models and computational results.

A_dfi, b_dfi, c_dfi, i=1, 2 are the state matrix, control vector and output vector of open-loop state-space models of DFIG-1 and DFIG-2.

p_dfi, r_dfi, i=1, 2 are the left and right eigenvectors corresponding to the open-loop SSO modes of DFIG-1 and DFIG-2.

A_dfact, b_dfact, c_dfact, i=1, 2 are the state matrix, control vector and output vector of open-loop state-space models of the ROPS subsystem without DFIG-1 and DFIG-2 being included, respectively;

p_dfact, r_dfact, i=1.2 are the left and right eigenvectors corresponding to the open-loop SSO modes of ROPS subsystem without DFIG-1 and DFIG-2 being included, respectively;

A_pmi, b_pmi, c_pmi, i=1, 2 are the state matrix, control vector and output vector of open-loop state-space models of PMSG-1 and PMSG-2.

p_pmi, r_pmi, i=1, 2 are the left and right eigenvectors corresponding to the open-loop SSO modes of PMSG-1 and PMSG-2.

A_pmaci, b_pmaci, c_pmaci, i=1, 2 are the state matrix, control vector and output vector of the ROPS subsystem without PMSG-1 and PMSG-2 being included, respectively.

p_pmaci, r_pmaci, i=1, 2 are the left and right eigenvectors corresponding to the open-loop SSO modes of ROPS subsystem without PMSG-1 and PMSG-2 being included, respectively.

Step 12: calculating the residues of each open-loop wind turbine generator subsystem and corresponding open-loop ROPS subsystems.

From the results obtained in the step 11, the open-loop residues of DFIG-1 and the corresponding ROPS subsystem are calculated to be R_df1=p_df1^Tb_df1c_df1r_df1=−1.28+164.55i and R_dfac1=p_dfac1^Tb_dfac1c_dfac1r_dfac1=−0.019+0.009i, respectively. Thus, the index associated with DFIG-1 is calculated to be Z_r4=√{square root over (R_df1R_dfac1)}=1.01-1.57i.

Similarly, the open-loop residues of DFIG-2 and the corresponding ROPS subsystem are calculated to be R_df2=p_df2^Tb_df2c_df2r_df2=−1.38+130.01i and R_dfac2=p_dfac2^Tb_dfac2c_dfac2r_dfac2=−0.0061-0.002i, respectively. The index associated with DFIG-2 is calculated to be Z_r3=√{square root over (R_df2R_dfac2)}=0.27-0.79i.

The open-loop residues of PMSG-1 and the corresponding ROPS subsystem are calculated to be R_pm1=p_pm1^Tb_pm1c_pm1r_pm1=−3.03+17.2i and R_pmac1=p_pmac1^Tb_pmac1c_pmac1r_pmac1=−0.011+j0.006, respectively. The index associated with PMSG-1 is calculated to be Z_r1=√{square root over (R_pm1R_pmac1)}=0.38+0.273i.

The open-loop residues of PMSG-2 and the corresponding ROPS subsystem are calculated to be R_pm1=p_pm2^Tb_pm2c_pm2r_pm2=−25.78+23.33i and R_pmac2=p_pmac2^Tb_pmac2c_pmac2r_pmac2=−0.009+j0.002, respectively. The index associated with PMSG-2 is calculated to be Z_r2=√{square root over (R_pm2R_pmac2)}=0.26+0.503i

Step 13: verifying the correctness of result obtained in the step 10 by the present invention.

Computational results of indexes above by using the parametric model are listed in Table 1-2. It can be seen that DFIG-1 causes the SSOs, confirming the correctness of identification made previously from Table 1-1 by using the present invention.

TABLE 1-2
Results of residues and index obtained using the parametric model for
confirmation
	Method in		Parametric	The cause
	the present	The cause of	model	of
Index	invention	the SSOs	method	the SSOs
Z4 (DFIG-1)	Zb4 = 1	Yes	Zbr4 = 1	Yes
Z3 (DFIG-2)	Zb3 = 0.2268	No	Zbr3 = 0.4521	No
Z1 (PMSG-1)	Zb1 = 0.2165	No	Zbr1 = 0.2550	No
Z2 (PMSG-2)	Zb2 = 0.2165	No	Zbr2 = 0.3063	No

DFIG-1 is disconnected from the example power system as shown by FIG. 35. Simulation results are shown from FIG. 36 to FIG. 39. It can be seen that no SSOs occur in the example power system, confirming that DFIG-1 causes the SSOs as identified by the present invention.

EXAMPLE 2

FIG. 40 shows a 10-machine 39-node New England power system integrated with four PMSGs, which represent four grid-connected wind farms. When the SSOs occur, the wind turbine generators which cause the SSOs in the power system are identified by the present invention without using the parametric model. Detailed steps of identification are as follows.

1. Determine the SSO Frequency

In order to identify the wind turbine generators which cause the SSOs, the SSO frequency in the power system needs to be determined firstly. Afterwards, the band-pass filter can be designed near this frequency.

Step 1: reducing the noise in the measurement data.

A low-pass filter depicted in FIG. 41 is designed to reduce the harmonic components in the measurement data.

Step 2: determining the SSO frequency by the signal processing method.

The SSOs can be observed in frequency response curve in FIG. 41(a)-(c). The FFT analysis results are summarized in FIG. 42. From FIG. 42, it can be concluded that the SSO frequency in the power system is 14.5 Hz, i.e. f=14.5 Hz

2. Design a Parallel Band-Pass Filter

Step 3: designing a parallel band-pass filter.

Design a parallel band-pass filter to ensure that the signal at frequency 14.5 Hz can pass through to earth, the signals at other frequencies are suppressed. Subsequently, dynamic interactions between the wind turbine generators and the power system at the SSO frequency f_s are suppressed and filtered out; whilst at other frequencies are not affected. A 4th order band-pass filter is designed by Besself method, where the cut-off frequencies are 14 Hz and 16 Hz. The analysis results of magnitude-frequency characteristics of the filter are shown in FIG. 43. From FIG. 43, it can be concluded that the designed filter can meet the requirement.

Step 4: verifying the effectiveness of the parallel band-pass filter designed in the step 3.

The procedure to verify the effectiveness of the parallel band-pass filter is the same as that presented above in the step 4 of example 1.

3. Measure Open-Loop Residues

Open-loop residues of each wind turbine generator represented by PMSG and the remainder of the ROPS can be measured by installing a parallel band-pass filter at the PCC of each wind turbine generator.

Step 5: installing a parallel band-pass filter on the side of PMSG-1 and measure the residues.

The installing location of the band-pass filter, the place to inject the disturbance signal and to gain the measurement data are indicated in FIG. 44. The disturbance signal is 10% increase of magnitude of PCC voltage for 0.1 second as shown by FIG. 45. The measurement data gained on the side of PMSG-1 is shown by FIG. 46(a). With the noise being reduced, the same measurement data is shown in FIG. 46(b).

Prony analysis is applied to the measurement data shown by FIG. 46(b) to obtain the residue of PMSG-1 to be 0.0183+0.0497.

Similarly, in order to measure the residue of the ROPS subsystem, the parallel band-pass filter is installed on the side of power system near PMSG-1. The installing location of the band-pass filter, the place to inject the disturbance signal and to gain the measurement data are indicated in FIG. 47. The disturbance signal is the same as that shown by FIG. 45. The measurement data gained on the side of power system and noise reduced data are shown in FIG. 48(a) and FIG. 48(b) respectively.

From the measurement data displayed in FIG. 48(b), the residue of the ROPS subsystem is estimated by the Prony analysis to be 0.0020+0.0026j.

Step 6: installing a parallel band-pass filter on the side of PMSG-2 and measure the residues.

The procedure to measure the residues for PMSG-2 is the same as that presented above in the step 5 for PMSG-1. The measurement data on the side of PMSG-2 and the noise reduced data are shown in FIG. 50(a) and FIG. 50(b). The residue of PMSG-2 is obtained to be 0.0003+0.0000j.

The measurement data and noise reduced data on the side of power system are shown by FIG. 52(a) and FIG. 52(b) respectively. The residue of the ROPS subsystem is obtained to be −0.0000+0.0007j.

Step 7: installing a parallel band-pass filter on the side of PMSG-3 and measure the residues.

The procedure to measure the residues for PMSG-3 is the same as that presented above in the step 5 for PMSG-1. The measurement data on the side of PMSG-3 and the noise reduced data are shown in FIG. 54(a) and FIG. 54(b). The residue of PMSG-3 is obtained to be 0.0001+0.0000j.

The measurement data and noise reduced data on the side of power system are shown by FIG. 56(a) and FIG. 56(b) respectively. The residue of the ROPS subsystem is obtained to be −0.0000+0.0000j.

Step 8: installing a parallel band-pass filter on the side of PMSG-4 and measure the residues.

The procedure to measure the residues for PMSG-4 is the same as that presented above in the step 5 for PMSG-1. The measurement data on the side of PMSG-4 and the noise reduced data are shown in FIG. 58(a) and FIG. 58(b). The residue of PMSG-4 is obtained to be 0.0006+0.0005j.

The measurement data and noise reduced data on the side of power system are shown by FIG. 60(a) and FIG. 60(b) respectively. The residue of the ROPS subsystem is obtained to be −0.0000+0.0007j.

4. Identify the Wind Turbine Generators which Cause the SSOs According to the Indexes

Step 9: calculating the indexes.

The indexes Z_i=|√{square root over (R_iR_ai)}|, i=1, 2, 3, 4 are computed based on the residues obtained above, where R_i and R_ai, i=1, 2, 3, 4 are the residues of each wind turbine generator and the corresponding ROPS subsystem, respectively. The computational results are presented in Tab. 2-1.

TABLE 2-1
Results of residues and corresponding index obtained by using the
present invention
	Index	Value	The cause of the SSOs
	Z1 (PMSG-1)	0.0020 + 0.0130i	Yes
	Z2 (PMSG-2)	0.0003 + 0.0003i	No
	Z3 (PMSG-3)	≈0	No
	Z4 (PMSG-4)	≈0	No

Step 10: identifying the wind turbine generator which causes the SSOs.

According to the open-loop modal resonance theory, the wind turbine generator corresponding to the maximum index among all Zi is identified to be the wind turbine generator causing the SSOs. Therefore, PMSG-1 is identified to be the cause of the SSOs in the 10-machine 39-node power system.

5. Verify the Correctness of Result Made by the Present Invention

The linearized model of FIG. 40 is used in this section to verify the correctness of the above identification made by the present invention.

Step 11: establishing the parametric model of the power system integrated with wind turbine generators.

The open-loop linearized models of each wind turbine generator subsystem and the corresponding ROPS subsystem described by Eq. (1) and Eq. (2) are respectively derived. Following notations are used to denote various results of derived parametric models and computational results

A_pmi, b_pmi, c_pmi, i=1, 2, 3, 4 are the state matrix, control vector and output vector of open-loop state-space models of PMSG-1, PMSG-2, PMSG-3, and PMSG-4, respectively;

p_pmi, r_pmi, i=1, 2, 3, 4 are the left and right eigenvectors corresponding to the open-loop SSO modes of PMSG-1, PMSG-2, PMSG-3, and PMSG-4.

A_pmaci, b_pmaci, c_pmaci, i=1, 2, 3, 4 are the state matrix, control vector and output vector of open-loop state-space models of the ROPS subsystem without PMSG-1, PMSG-2, PMSG-3, and PMSG-4 being included, respectively;

p_pmaci, r_pmaci, i=1, 2, 3, 4 are the left and right eigenvectors corresponding to the open-loop SSO modes of ROPS subsystem without PMSG-1, PMSG-2, PMSG-3, and PMSG-4 being included, respectively;

Step 12: calculating the residues of each open-loop wind turbine generator subsystem and corresponding open-loop ROPS subsystems.

From the results obtained in the step 11, the open-loop residues of PMSG-1 and the corresponding ROPS subsystem are calculated to be R_pm1=p_pm1^Tb_pm1c_pm1r_pm1=−3.02+134.24i and R_pmac1=p_pmac1^Tb_pmac1c_pmac1r_pmac1=−0.01+j0.002, respectively. Thus, the index associated with PMSG-1 is calculated to be Z_r1=√{square root over (R_pm1R_pmac1)}=0.63+0.64i

Similarly, the open-loop residues of PMSG-2 and the corresponding ROPS subsystem are calculated to be R_pm2=p_pm2^Tb_pm2c_pm2r_pm2=0 and R_pmac2=p_pmac2^Tb_pmac2c_pmac2r_pmac2=−0.006+j0.004, respectively. The index associated with PMSG-2 is calculated to be Z_r2=√{square root over (R_pm2R_pmac2)}=0.

The open-loop residues of PMSG-3 and the corresponding ROPS subsystem are calculated to be R_pm3=p_pm3^Tb_pm3c_pm3r_pm3=0 and R_pmac3=p_pmac3^Tb_pmac3c_pmac3r_pmac3=−0.011+j0.001, respectively. The index associated with PMSG-3 is calculated to be Z_r3=√{square root over (R_pm3R_pmac3)}=0.

The open-loop residues of PMSG-4 and the corresponding ROPS subsystem are calculated to be R_pm4=p_pm4^Tb_pm4c_pm4r_pm4=0 and R_pmac4=p_pmac4^Tb_pmac4c_pmac4r_pmac4=−0.008+j0.002, respectively. The index associated with PMSG-4 is calculated to be Z_r4=√{square root over (R_pm4R_pmac4)}=0.

Step 13: verifying the correctness of result obtained in the step 10 by the present invention.

Computational results of indexes above by using the parametric model are listed in Table 2-2. It can be seen that PMSG-1 causes the SSOs, confirming the correctness of identification made previously from Table 2-1 by using the present invention.

TABLE 2-2
Results of residues and index obtained using the parametric model for
confirmation
	Method in		Parametric	The cause
	the present	The cause of	model	of
Index	invention	the SSOs	method	the SSOs
Z1 (PMSG-1)	Zb1 = 1	Yes	Zbr1 = 1	Yes
Z2 (PMSG-2)	Zb2 = 0.0321	No	Zbr2 = 0.0000	No
Z3 (PMSG-3)	Zb3 = 0.0000	No	Zbr3 = 0.0000	No
Z4 (PMSG-4)	Zb4 = 0.0000	No	Zbr4 = 0.0000	No

PMSG-1 is disconnected from the example power system as shown by FIG. 61. Simulation results are shown from FIG. 62(a) to FIG. 62(c). It can be seen that no SSOs occur in the example power system, confirming that PMSG-1 causes the SSOs as identified by the present invention.

Claims

1. A measurement data based method for identifying wind turbine generators which cause sub-synchronous oscillations (SSOs) in a complex power system, comprising following steps of: (1) obtaining a set of measurement data when the SSOs occur in the power system; reducing noise by applying a filtering algorithm; identifying an SSO frequency by using a signal processing method; wherein the identified SSO frequency is denoted as fs;(2) designing a parallel band-pass filter to ensure that only a signal at frequency fs passes through to earth and signals at other frequencies are suppressed;(3) for a power system with n wind turbine generators, installing the parallel band-pass filter designed in the step (2) on an interface between each of the wind turbine generators and the power system; wherein subsequently, dynamic interactions between the wind turbine generators and the power system at the SSO frequency fs are suppressed and filtered out; whilst those at other frequencies are not affected;(4) measuring a response from each of the wind turbine generators by adding a disturbance signal on a side of a wind turbine generator near the band-pass filter;(5) from the measurement data of the response of an i-th wind turbine generator, identifying an oscillation frequency, fij, and residuing Rij by applying a signal processing method; wherein in notations above for fij and Rij, subscript i refers to the i-th wind turbine generator and the subscript j refers to the j-th SSO frequency identified;afterwards, comparing fij and fs; recording the fij and Rij if a following condition is met: 1.05*fs≥fij≥0.95*fs;wherein finally, if no fij which can meet the above condition, it is concluded that the i-th wind turbine generator is not the wind turbine generator which causes the SSOs in the power system;(6) measuring the response from the power system by adding a disturbance signal on the side of the power system near the band-pass filter for the i-th wind turbine generator;(7) from the measurement data of the response, identifying an oscillation frequency faij and a residue Raij by using the signal processing method; wherein in notations of faij and Raij, subscript a indicates that the measurement data is about the power system, subscript i refers to the i-th wind turbine generator and subscript j refers to the j-th oscillation frequency and residue;afterwards, comparing faij and fs; recording faij and Raij if a following condition is met: 1.05*fs≥faij≥0.95*fs and(8) Computing Zi=|√{square root over (RijRaij)}|, i=1, 2, L n based on recorded results in the step (5) and (7); wherein if Zk is the largest among Zi, i=1, 2, L n, the i-th wind turbine generator is identified to be the wind turbine generator causing the SSOs in the power system.

2. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 1, wherein the interface in the step (3) adopts any point between the i-th wind turbine generator and the power system; at the point, the power system is divided to two subsystems; one subsystem consists of the i-th wind turbine generator; the other subsystem is comprised of remainder of the power system excluding the i-th wind turbine generator.

3. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 1, wherein the disturbance signal in the step (4) excites an observable dynamic response of the i-th wind turbine generator.

4. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 1, wherein the response of the wind turbine generator in the step (4) is an active power output from the wind turbine generator.

5. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 1, wherein the disturbance signal in the step (4) excites an observable dynamic response of the power system.

6. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 1, wherein the power system response in the step (6) is a magnitude of a terminal voltage at a node in the power system where the i-th wind turbine generator is connected to the power system.

7. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 1, wherein the disturbance signal in the step (4) or step (6) is an impulse signal; a time duration and a magnitude of the signal are respectively 0.1 second and 2% of a magnitude of a terminal voltage at a node where the i-th wind turbine generator is connected with the power system.

8. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 2, wherein the disturbance signal in the step (4) or step (6) is an impulse signal; a time duration and a magnitude of the signal are respectively 0.1 second and 2% of a magnitude of a terminal voltage at a node where the i-th wind turbine generator is connected with the power system.

9. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 3, wherein the disturbance signal in the step (4) or step (6) is an impulse signal, a time duration and a magnitude of the signal are respectively 0.1 second and 2% of a magnitude of a terminal voltage at a node where the i-th wind turbine generator is connected with the power system.

10. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 4, wherein the disturbance signal in the step (4) or step (6) is an impulse signal; a time duration and a magnitude of the signal are respectively 0.1 second and 2% of a magnitude of a terminal voltage at a node where the i-th wind turbine generator is connected with the power system.

11. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 5, wherein the disturbance signal in the step (4) or step (6) is an impulse signal; a time duration and a magnitude of the signal are respectively 0.1 second and 2% of a magnitude of a terminal voltage at a node where the i-th wind turbine generator is connected with the power system.

12. The measurement data based method for identifying the wind turbine generators which cause the sub-synchronous oscillations in the complex power system, as described in claim 6, wherein the disturbance signal in the step (4) or step (6) is an impulse signal, a time duration and a magnitude of the signal are respectively 0.1 second and 2% of the magnitude of the terminal voltage at the node where the i-th wind turbine generator is connected with the power system.