1. Introduction
The environmental pollution resulting from fossil fuels is posing a great challenge for human society. To address this problem, wind farms and plug-in electric vehicles (PEVs) are increasingly employed and integrated into the power system as a result of their green characteristics. However, the randomness associated with wind power outputs and power demands of PEVs could have significantly negative impacts on the secure, stable, and economic operation of the power system concerned, especially in the scenarios with high-level penetrations. As a result, the stochastic stability analysis of a power system with uncertainties has attracted much attention in the academic as well as industrial communities. Given this background, power system stochastic small signal stability is investigated in this work.
Up to now, much effort has been dedicated to power system stability analysis with uncertainties, and some methods have been proposed. Generally, there are two main categories; i.e., probabilistic small-signal stability [
1,
2] and probabilistic transient stability [
3,
4,
5]. The index for identifying stochastic stability is related to involved random factors through the algebraic relationship. By assuming that random factors follow certain probability distributions, the probability density function (PDF) of the index can be derived by the well-established Monte Carlo simulation. On the other hand, power system stabilizers (PSSs) are applied for enhancing stochastic small-signal stability via eigenvalue sensitivity analyses [
6]. Furthermore, the adaptive control based strategies [
7,
8], frequency-domain approach [
9], and particle swarm optimization-based methods [
10] are adopted to enhance power system stability with stochastic disturbances. In [
11], uncertainties are characterized by a compact set, then robust optimization is carried out for the worst-case scenario, or the random factors are measured in real time in the context of the adaptive control. However, these approaches rely on a deterministic modeling of the power system, and this means that the electromechanical transient of the power system is still governed by ordinary differential equations, and hence, the stochastic nature of the system dynamic process is neglected. Besides, the PDF is obtained by the Monte Carlo simulation, and the curse of computing is involved.
To capture the stochastic nature, the stochastic differential equation (SDE) theory [
12,
13] is employed to model a power system and investigate the system stability. In [
14], a general stochastic modeling approach for a power system is described based on SDE. In [
15], system transient stability is assessed by the trajectories obtained from stochastic calculus. In [
16], the evolution of the PDF of power system state variables is solved by the Fokker–Planck equation for examining the impacts of random load disturbances on the system stability. In [
17], the singular perturbation theory is employed for accurate stability assessment with variable wind power outputs taken into account. In [
18], mean-value stability and mean-square stability are utilized to assess the stochastic small-signal stability in a power system. The stochastic counterpart of the direct method using energy functions for power system stability analysis has also been investigated. The direct method is a common analytical approach for deterministic transient stability assessment [
19,
20,
21]. In [
20], it is pointed out that limited scalability and conservativeness limit the applicability of the direct method, and the energy function represents a specific form of the Lyapunov function. The Lyapunov function approach is proposed to mitigate the drawbacks of the classical energy method [
20]. In [
22], the structure-preserved model is employed for power system stochastic transient stability using a stochastic energy function. However, these existing publications compute the stability probability of the power system through the Monte Carlo simulation approach, and heavy computation burden is involved especially for large-scale power systems. In light of the quasi-Hamiltonian form, the stochastic energy function is utilized to analyze power system transient stability [
23] and dynamic performance [
24], but this method needs to solve a partial differential equation (PDE) for attaining the stability probability and hence involves a very hard solving procedure. In addition, in the mentioned methods, the impacts of uncertainties are only investigated through numerical simulations, and these methods fail to explicitly consider the magnitude of uncertainties in the stability index. Moreover, based on our knowledge, few publications have focused on power system small signal stability analysis using the SDE theory. Even though the work [
18] focuses on this topic, the method in [
18] cannot provide the information regarding the stability probability.
There is another body of publications relevant to this work. For example, in [
25], unit commitment is used to enhance the capability of accommodating a high penetration level of wind power outputs, and the maximum penetration level is determined by taking the probabilistic transient stability and frequency security into account. In [
26], the impacts of various wind power penetration levels in a power system are investigated by evaluating the eigenvalue sensitivity with respect to generator inertia. However, these approaches are based on the deterministic model, and uncertain factors are not taken into account.
The noise-to-state stability (NSS) is first proposed for the problem of stochastic disturbance attenuation [
27]. This approach makes the system states bounded by a monotone function of the norm of the disturbance magnitude. Besides, this notion can deal with the case where disturbances do not vanish at the origin of the state space. These characteristics are suitable for analyzing the stability of a power system with uncertainties. NSS is related to the concept of the input-to-state stability [
28,
29,
30]. Local input-to-state stability has been applied for voltage stability analysis of an isolated power system [
31].
Given the above stated background, this paper proposes an analytical method for stochastic small signal stability analysis of a power system with respect to the magnitudes of uncertainties. The power system is considered as a set of SDEs. Then the norm of covariance is utilized to quantify the impacts of the magnitudes of stochastic variations. Next, the stochastic stability index of the power system is given with respect to this supremum through NSS. NSS-LF is next employed to deduce the algebraic conditions for stochastic small signal stability of the power system. The proposed method has three advantages:
- (1)
First, the probability of the state deviations falling within an interval is provided in an analytical way, so it is easier than the existing methods by solving a partial differential equation;
- (2)
Secondly, only some algebraic conditions need to be examined for identifying stochastic stability and this can be carried out efficiently compared with the Monte Carlo based approaches for practical applications;
- (3)
Thirdly, since the proposed system stability index explicitly takes the uncertainty magnitudes into consideration, the dynamic process of the system states with uncertainties can be theoretically explained.
The organization of this paper is as follows:
Section 2 describes necessary notions of the NSS and stochastic Lyapunov function.
Section 3 is devoted to the application of the NSS-LF for stochastic small signal stability analysis.
Section 4 presents the simulation results to demonstrate the analytical results. Conclusions are given in
Section 5.
2. Mathematical Background
An operating power system is continually subject to uncertainties. A class of typical stochastic variations involves a sequence of switching events (e.g., a device is disconnected, and then reconnected after a while). These switching events usually lead to the changes in the power network topology during the fault-on stage, and hence the power system suffers an immediate state change. In this case, power system transient stability is identified by comparing the system energy at the fault clearing time with the critical potential energy. Another class of stochastic variations is generation and/or load level fluctuation, which generally cause slight changes of the system state but may also lead to the small signal stability problem. This work focuses on power system small signal stability with respect to various degrees of uncertainty magnitudes. To formulate this problem, some stochastic analysis tools are given first below.
A nonlinear stochastic dynamic procedure of a power system can be expressed as follows:
where
x(
t)
Rn is the state vector;
W(
t)
Rr denotes the stochastic continuous variations, and its elements are independent standard Wiener processes; Σ
Rr×r is a non-negative-definite matrix, and its elements denote the variation magnitudes. Assume that the vector field
f(
x):
Rn→
Rn and matrix function
g(
x):
Rn→
Rn×r are both locally Lipschitz continuous.
Unlike the mean stability [
18], the probability stability is employed for the stochastic small signal stability analysis.
Definition 1. [32,33]The equilibrium point x = x*Rn of the system (1) is locally stable in probability if, for every ε > 0 and δ > 0, there exists a number r > 0 such that ||x0 − x*|| < r implieswhere x(t,x0) is a solution of the system (1) starting from x0 at t = 0. This definition means that, given a fixed value of ε, the system with the initial state x0 is stable if the event ||x(t) − x*|| ≤ δ happens with a probability larger than 1 − ε. Note that the larger the parameter ε is, the smaller the stability probability will be. So ε indicates the risk level.
In the literature, the traditional probabilistic stability index is extensively used for power system stochastic stability analysis, see [
3,
4,
5,
6,
14,
15,
16]. The existing methods require carrying out Monte Carlo simulation or solving a PDE. Thus, the use of this formulation of the probability stability is limited due to heavy computation burden. Besides, the magnitudes of the stochastic power fluctuations are not explicitly considered in the probabilistic stability index and only investigated by numerical simulations. In [
22,
25,
26], it is pointed out that the uncertainty magnitudes can affect the system stochastic stability. To theoretically investigate the system probabilistic stability with the consideration of uncertainty magnitudes, an additional term needs to be incorporated into the probabilistic stability index.
To overcome the abovementioned drawbacks, the NSS based probabilistic stability index is proposed in this paper. The probabilistic stability index in this paper is formulated through the class
function [
28]. This formulation facilitates the probabilistic stability analysis of a power system without the use of Monte Carlo simulation or solving a PDE. Moreover, it is convenient for the incorporation of uncertainty magnitudes into the index. The definition of the NSS based probabilistic stability index is given as follows.
Definition 2. [27] The equilibrium point x* of the nonlinear stochastic system (1) is called locally noise-to-state stable (NSS) if, for any ε > 0, there exist βand γsuch thatfor any x0Ω,ΣUr×r, and t ≥ 0. In comparison with Definition 1, the term β(||x0||,t) in (3) corresponds to the quantity δ (2), which represents the impact of the initial state on the system stability. The additional term γ(||ΣΣT||) in (3) quantifies the impact of the uncertainty magnitude. Since the parameter Σ in the stochastic model (1) denotes the uncertainty magnitude, it is natural to consider how this magnitude affects the probabilistic stability of the power system. The Expression (3) means that the system is probabilistically stable if and only if ||x(t) − x*|| is not larger than the sum of these two terms with a probability P ≥ 1 − ε. It is obvious that the impacts of stochastic continuous variations are explicitly taken into account in the stability index.
Note that both
β(||
x0||,
t) and
γ(||ΣΣ
T||) in (3) need to be identified to reach the conclusion regarding the system probabilistic stability. For power system small signal stability,
β(||
x0||,
t) is related to the eigenvalues of the system matrix. When all the eigenvalues are located within the left-half complex plane, the term
β(||
x0||,
t) will decay to zero. Then the probabilistic stability of the power system is identified, and ||
x(
t) −
x*|| is bounded above by the terms
γ(||ΣΣ
T||). This means that, after a small disturbance occurs, when
t→∞, the state deviations of the system will fall within a range [−
γ(||ΣΣ
T||),
γ(||ΣΣ
T||).] with a probability 1 −
ε. This agrees with the conclusions in [
18]. Unlike the mean stability index in [
18], the NSS based probabilistic stability index can provide the probability information regarding the system stability.
Based on the above discussions, the proposed analytical method for power system stochastic small signal stability analysis can be presented as follows:
Theorem 1. (NSS based Probabilistic Stability) For a stochastic system (1), if the following conditions are satisfied: (1) the stochastic system (1) is locally NSS; (2) β(||x0||,t)→0 as t→∞, then the system is stable in probability and P{||x(t) − x*|| ≤ γ(||ΣΣT||)} ≥ 1 − ε.
The formal proof of this theorem is provided in the
Appendix A.
NSS-LF is a useful tool to verify condition (1) in Theorem 1. It is defined as follows:
Definition 3. [27] Suppose there exists a proper function V: Rn → R+, such that classfunctions α1, α2, α3, and α4 satisfyand for all ΣUr×r Then, V(x) is called a NSS-LF.
Theorem 2. [27] If a nonlinear stochastic system admits a NSS-LF, then the system is locally NSS. Note that the NSS-LF is given for nonlinear systems. It can also be applied to linear systems for small signal stability analysis such as linearized power systems. When the term
α4(||ΣΣ
T||) is sufficiently small, the inequality
≤ −
α3(||
x||) +
α4(||ΣΣ
T||) in Definition 3 reduces to
≤ −
α3(||
x||) which implies
. In this case, the function
V(
x) is the stochastic Lyapunov function (sLF) as in [
22]. In this sense, the NSS-LF is a generalization of the sLF. Hence, NSS-LF is adopted to identify the probabilistic stability of power systems in this work.
3. Noise-to-State Stability Based Probabilistic Stability Analysis
In this section, the idea for NSS based probabilistic stability assessment is presented. First, the supremum of the norm of the covariance of power stochastic variations is given for characterizing the impacts of uncertainty magnitudes. Then, the NSS-LF is employed to implement the NSS based probabilistic stability assessment. Finally, the proposed algorithm is applied to a typical power system.
3.1. The Stochastic Model of a Power System
A renewable energy system (RES), such as wind power generation and solar power generation, can be integrated to a power system concerned. To study the impacts of the stochastic variations from a RES on the system stability, the RES can be introduced into the power balance equation. By transformations in [
34], these stochastic continuous variations can be manifested through the dynamic model of a power system. Then the approaches for the stochastic stability problem of a power system can be derived considering generation and/or load level fluctuation.
In this work, the proposed method aims at analyzing power system stability with the stochastic continuous variations of generation, so the power system is modeled as a set of SDEs. Synchronous generators are modeled using the traditional swing equation model. Loads are assumed to be constant impedances. When the losses in the transmission network are ignored, the load nodes can be eliminated by transformations, this leads to a reduced network with only generator nodes contained. The stochastic model of a power system can be expressed as follows [
16,
23]:
Here
Σi(
t)
dWi(
t) represents stochastic continuous variations;
Wi(
t) represents independent standard Wiener processes;
Σi represents uncertainty magnitudes;
δi,
ωi,
Hi,
Pmi,
Ei, and
Di are the rotor angle, rotor speed deviation, inertia coefficient, mechanical power, internal voltage magnitude and damping coefficient of the
ith generator, respectively;
Bij is the element in the
ith row and
jth column of Kron-reduced susceptance matrix;
ωN is the synchronous machine speed. The quantity
yi is the output.
c1 and
c2 are the weighted coefficients of state variables. e.g., if we want to examine the stability of the angle,
c1 can be set to 1 and
c2 can be set to 0. This input-output formulation is a standard paradigm in robust control theory [
35] and has been adopted in power systems in [
31,
36].
As in [
18,
22], the terms
Σi(
t)
dWi(
t) are added to the rotor motion equations of synchronous generators. By doing this, it can be used to represent the stochastic variations generated from generation and/or load.
The stochastic continuous variation is usually modeled as a stochastic process. Naturally, the covariance of the stochastic process is considered as the degree of the impact. To see this point, it is helpful to note that in (6) Σ(
t)
dW satisfies the following equations:
since
dW~
N (0,
dt). Here,
E(∙) is the mean value function, and
D(∙) is the infinitesimal covariance function.
As in (3), the system probabilistic stability index requires the computation of the term
γ(||Σ(
t)Σ(
t)
T||). Based on the above analysis, Σ(
t)Σ(
t)
T is the covariance. According to [
27], the impact of the magnitudes of uncertainties is quantified as the norm of the covariance in that the stochastic continuous variation may be unbounded.
Both the Wiener process and the supremum of the norm of the covariance are plotted in
Figure 1. It can be seen that Σ(
t)
dW is a time-dependent stochastic process which is used in the numerical integration for the SDE model, as in [
22,
23]. For the model of sup||Σ(
t)Σ(
t)
T||, it only gives an upper bound of the uncertainty magnitude and has the potential for the analytical method of the probabilistic stability analysis. Thus, the latter is adopted in this paper.
3.2. Stochastic Small Signal Stability Analysis
In this paper, Σ is assumed to be independent of time [
16,
22,
23]. To apply the proposed method in
Section 2 for the stochastic small signal stability, the linearization of the stochastic nonlinear system (6) at a steady state operating point yields to:
where
A Rn×n and Σ
Rn×n. Δ
x denotes the state deviation from the steady state point.
Since the two conditions in Theorem 1 need to be satisfied, they are separately illustrated in the remaining part of this subsection.
For the condition (1) in Theorem 1, the existence of the NSS-LF ensures that the system is NSS based on the Theorem 2. Hence, the NSS-LF is proposed to construct the function γ.
Theorem 3. The term γ can be obtained by the following expression:if the system (9) admits a NSS-LF. To evaluate the condition (2) in Theorem 1, the following linear matrix inequality (LMI) is proposed
Here, P is a positive definite and symmetric matrix. The superscript T denotes the transpose operator. c is a scalar.
Theorem 4. If the inequality (11) holds, then the probability of the state deviation for the system (9) satisfies: The feasibility of (11) can be guaranteed by solving the following semi-definite programming (SDP) problem:
where superscript
T denotes the transpose operator.
P is a symmetric matrix and
P =
PT.
Here the optimization is carried over the matrix
P. The constraint (14) ensures that
P is a positive definite matrix. The constraint (15) guarantees that the NSS-LF exists. Whenever the value
c is positive, the NSS-LF for the system (9) can be constructed by
P (see
Appendix A).
3.3. Application of Noise-to-State Stability Based Probabilistic Stability Analysis
To qualitatively illustrate the application of the proposed approach, a stochastic single-machine infinite-bus (sSMIB) power system is considered in this subsection. The sSMIB system comprises of a classical SMIB system with the integration of stochastic power fluctuations; i.e., active power outputs from a wind power plant (WPP) and/or load fluctuation, as shown in
Figure 2.
Assume that the system operates in the steady-state state with the initial values of
δ* and
ω*. When a small disturbance occurs, the system model (6) is linearized around the equilibrium point, and rewritten as the form in (9) with
Based on
Section 3.2,
P can be obtained by solving the linear matrix inequality (LMI) (11). Then the Lyapunov function can be selected as
V(
x) =
xTPx. Suppose the matrix
P is [
a d;
d b], then
α4 can be chosen as
b such that
Let α1 and α2 be the minimum eigenvalue and the maximum eigenvalue of P, respectively. Then (12) holds for the system (9) and the proposed probabilistic stability index (3) is identified.
To examine the probabilistic stability of the rotor angle, we can choose
c1 = 1 and
c2 = 0 in (6) and then attain
By similar procedures, the following expression can be derived:
3.4. Algorithm of Noise-to-State Stability Based Probabilistic Stability Analysis
The flowchart of the proposed NSS based approach is depicted in
Figure 3.
Step 1: Set the values of Σ, ε, and c. A smaller value ε amounts to a more reliable result from the proposed method.
Step 2: Choose the NSS-LF V(x).
Step 3: Select the class
functions
α1,
α2,
α3 and
α4 based on the method presented in
Appendix A. As shown in
Section 3.2, the selection principle is to satisfy the conditions in Theorem 1.
Step 4: Compute the value of the function γ(||ΣΣT||).
Step 5: Set the weighted coefficient to obtain the stability probability of the power system state.