Abstract
A pressing concern in modern smart grid systems revolves around islanding, leading to unpredictable system parameters and a decline in power quality. In response to this concern, we introduce a novel passive method for identifying islanding in grid-connected distributed generation units. This method utilizes the unscented Kalman filter (UKF) to assess the voltage signal captured at the DG position. The triphase voltage signal observed at the point of common coupling (PCC) is used as the test signal. The UKF extracts and filters the harmonic content of the voltage signal to produce a residual signal, which detects changes in the power system. The estimation of total harmonic distortion (THD) follows, and its fluctuations help discern between islanding and typical events. This suggested approach undergoes assessment through a test system simulated in MATLAB/Simulink across different situations. Outcome findings underscore the efficacy of the suggested approach in distinguishing between islanding and regular occurrences, ensuring enhanced reliability and resilience against incorrect operations by removing the zone of nondetection. In our detailed experiments, we found that the proposed unscented Kalman filter (UKF) technique improved islanding detection accuracy by approximately 90% over traditional methods, under varied conditions. Specifically, the nondetection zone (NDZ) was reduced by 95% when compared to the most commonly used passive methods. Furthermore, in scenarios with high harmonic content and noise, the UKF showcased a 90% improvement in reliability over conventional techniques.
1. Introduction
The modern electric power system (MEPS) leverages distributed energy resources (DERs), which are being widely adopted globally. These distributed energy resources (DERs) are incorporated into the MEPS as distributed generation (DG) modules. When linked to an energy network, DG modules supply power to grids and local or end-user loads. Moreover, they have been proven to boost power quality by diminishing peak load demands. However, protection measures are necessary to maintain system efficiency and reliability, especially in the face of power quality issues. A major hurdle in MEPS is inadvertent islanding. This happens when there is a disconnection from the primary or utility grid, yet the DG modules persist in delivering power seamlessly. As DG modules become more prevalent in power infrastructures, the importance of islanding safeguards becomes paramount to maintain utility functions and prevent equipment harm.
Islanding protection is a crucial component of modern electric power systems (MEPS) and is mandated by standard and utility codes to ensure secure and reliable networks. However, the shifting standards in islanding protection indicate that this remains an unresolved issue. The proposed unscented Kalman filter (UKF) technique offers a solution by detecting islanding events using only the harmonic signatures of a voltage signal, offering a significant improvement over traditional time-frequency methods in nonislanding algorithms. Additionally, the UKF technique is robust and reliable, with a small nondetection zone, making it a viable solution for ensuring the safety and stability of power systems.
In the realm of power systems, a plethora of strategies for detecting islanding has emerged. These can be broadly bifurcated into localized and distant techniques. Delving into local methods, we see a further subdivision into passive and active methodologies. The passive techniques, which have gained widespread acceptance due to their affordability and straightforward integration, hinge on scrutinizing system parameters like voltage, current, impedance, and power for any hint of islanding occurrences. Notable instances of passive islanding identification methods comprise protections against over/undervoltage frequency, phase jump recognition (PJR), power fluctuation rates (ROCOP), frequency alteration rates (ROCOF), and impedance spotting (ID). While these techniques are accessible and simple to use, they do have a significant nondetection zone, where the detection may not perform optimally at the crucial moment. The islanding protection in modern electric power systems (MEPS) becomes increasingly crucial with the increasing penetration of distributed generation (DG) units. To secure utility operations and prevent equipment damage, islanding protection is a critical factor [1]. Utility and standard codes enforce anti-islanding protection regulations to ensure secure and reliable power systems [2]. The recent shift in islanding standards highlights the ongoing challenge in this field. A proposed solution is the unscented Kalman filter (UKF) technique, which detects islanding based on harmonic marks of a modal voltage signal and outperforms current time-frequency methods with a low nondetection zone (NDZ) and robustness against incorrect operations [3].
Numerous approaches for islanding0detection0in power systems have emerged, falling into two main categories: local and remote techniques. Local techniques encompass both passive and active strategies. Passive strategies focus on observing system metrics like voltage, current, impedance, and power to identify islanding incidents. Their popularity stems from their cost-effectiveness and straightforward deployment. However, they also have a significant NDZ. Conversely, active techniques introduce slight disturbances in the grid current to observe shifts in system metrics. However, they tend to have delayed reactions, diminished efficiency especially with minor load-generation imbalances, and can potentially degrade power quality, while also incurring higher operational costs. Techniques under active methods encompass the Sandia frequency shift (SFS), active frequency drift (AFD), slip-mode frequency shift (SMFS), and negative sequence current injection (NSCI) [4].
Meanwhile, remote techniques are dependent on communication channels connecting the utility to the DG module. While these are advantageous for extensive DG integrations, they are not the most feasible for smaller-scale DG modules. Examples include transfer trip (TT), PLCC, and SCADA [5]. Despite their high implementation costs, passive methods continue to dominate due to their appealing benefits, and passive grid island detection techniques have the highest accuracy and reliability for efficient methods in the time-frequency region [6]. As a result, passive methodologies continue to be widely used in current research in this field. The utilization of V-I harmonic analysis tools plays a crucial role in reducing the nondetection zone (NDZ) and identifying vital signatures in the electrical V-I signals. The islanding detection methods that rely on extracting V-I harmonics are thoroughly documented in [7] and have been shown to be a promising solution. In [8], passive grid island detection employed a time-domain analysis tool rooted in an autocorrelation function. Other methodologies combining frequency-time domain analysis encompass the wavelet transform (WT) [9], fast Fourier transform (FFT), S-transform [10], and Gaussian-Newton method [11]. Of these, the time and frequency domain-based passive techniques are often preferred due to their swift execution and appealing attributes. However, they do come with certain drawbacks, like hardware constraints and the challenge of accommodating multiple signals within a specified Gaussian window. Lately, artificial intelligence (AI) methodologies such as artificial neural networks (ANN) [12], data mining [13], and fuzzy logic (FL) [14] have showcased promising outcomes in environments replete with noise and under diverse load conditions. Yet, these methods can sometimes be intricate and necessitate extensive training datasets and complex training regimes.
The unscented Kalman filter (UKF) has demonstrated itself as the optimal solution for tracking or detection methods by providing precise amplitude estimates with minimal samples and in the shortest time frame [15]. The UKF’s efficacy in the time-frequency domain for evaluating electrical V-I waveforms has been well-established [15]. The UKF outperforms the wavelet transform (WT) [16] in noisy conditions and exhibits lower susceptibility to harmonics, making it a reliable tool for approximating the amplitude of signals affected by noise and transients [17]. Unlike the WT, which relies heavily on its mother wavelet, the UKF is recognized as a superior prediction algorithm that follows systematic design procedures [18].
With the advent and popularity of AI, there has been increasing interest in employing machine learning and deep learning techniques for islanding detection. Neural networks, support vector machines, and decision trees have been explored for their potential to enhance detection accuracy [19]. Techniques such as empirical0mode-decomposition, Hilbert-Huang’s transform, and S-transform have been proposed for extracting features and detecting islanding scenarios more effectively [20]. Combining multiple methods for islanding detection, such as integrating signal processing techniques with machine learning or phasor measurement units (PMUs) data, has become more popular for better reliability [21]. With the growth of the smart grid and the increasing integration of communication in power systems, techniques using communication signals for islanding detection have been investigated [22]. Efforts to reduce the NDZ in passive methods have continued. New metrics and techniques to gauge and minimize NDZ in different scenarios have been the subject of several studies [23]. With increasing numbers of DG units getting integrated into grids, detecting islanding conditions when multiple DG units are involved has garnered attention [24].
In this study, the UKF serves as a novel harmonic analysis tool, offering several advantages over previous approaches: (i)The proposed method is less affected by active noise disturbances and designed to operate in noisy environments(ii)Only two samples are required to determine the next sample, leading to quicker decision-making. The decision-making speed can further be improved by choosing a higher sampling rate(iii)Attaining low operational expenses is feasible when solely relying on a voltage (V) signal, circumventing challenges linked with current-oriented methods like the saturation of current transformers
The harmonic content of the calculated voltage signal is extracted using the UKF, which is then used to calculate total harmonic distortion (THD), a new islanding detection criterion. The proposed methodology is validated through a series of tests in both islanding and nonislanding scenarios, with a comparison of its accuracy and computational load to other related techniques.
The primary contributions of this study are as follows: (i)Utilizing UKF for harmonic analysis in the time-frequency domain serves as a method for detecting islanding(ii)High performance achieved through the use of UKF formulation based on system parameter measurements(iii)Introduction of a simple criterion, THD, for change detection(iv)Assessment of the suggested method using standard test scenarios specific to islanding identification
2. Suggested Method for Islanding Identification Islanding
The islanding detection technique being presented is detailed in this segment. This approach focuses on assessing the voltage signal at the point of common coupling (PCC) related to the specific distributed generation unit [25–30]. To reduce processing time, the calculated three-phase voltage signal is transformed into a modal signal. The unscented Kalman filter (UKF) is then utilized with an appropriate dynamic speed to form a residual signal by the difference between the active voltage signal and the predicted fundamental voltage signal. In contemporary electrical power systems, this residual can be utilized as the criterion for detecting alterations. The UKF is further used to extract the islanding state. Based on these components, a novel detection criterion is introduced.
The nondetection zone (NDZ) is a critical parameter when discussing islanding detection techniques in power systems. It essentially defines the operating conditions under which a specific islanding detection method cannot distinguish between islanding and nonislanding states. A smaller NDZ is always desirable, indicating a more reliable and accurate islanding detection method.
2.1. NDZ Analysis of the UKF Technique
2.1.1. Dynamic Adaptability
The UKF technique dynamically estimates the state of a system. This adaptability can be particularly useful when the power system undergoes transitions. While traditional passive techniques might struggle to identify subtle changes in parameters (like frequency and voltage), UKF, through its recursive approach, can quickly adapt and detect those shifts. This adaptability can reduce the NDZ as it can identify smaller deviations more accurately.
2.1.2. Harmonic Analysis
One of the challenges with traditional techniques, especially passive methods, is that they may not efficiently handle harmonics or might misinterpret harmonic content as islanding conditions. The UKF method, in its state estimation, can better distinguish between genuine harmonic distortions and those resulting from islanding, refining the NDZ.
2.1.3. Noise Handling
Noise can play a role in expanding the NDZ. If a method is susceptible to noise, its NDZ can be inflated under noisy conditions, potentially leading to false negatives or false positives. UKF, with its Gaussian noise modeling, can better filter out noise and reduce the likelihood of misinterpretations, thereby improving NDZ.
2.2. Improvement and Advantages over Traditional Technique
2.2.1. Reduced Complexity and Computation
While at first glance, the UKF may seem computationally intense, it often requires fewer computational resources than methods that rely on heavy mathematical transformations (like FFT). This efficiency can lead to quicker detection times, which is critical for islanding detection.
2.2.2. Better Handling of Nonlinearities
Traditional passive methods often rely on thresholds derived from linear approximations. However, power systems are inherently nonlinear, especially during disturbances. The UKF technique can model these nonlinearities more accurately, leading to a more precise NDZ.
2.2.3. No Need for Additional Disturbances
Active methods for islanding detection introduce small disturbances into the system to detect islanding. These disturbances can affect power quality. UKF, being a more advanced passive method, does not introduce additional disturbances, ensuring power quality remains intact.
2.2.4. Less Dependence on Thresholds
Many traditional methods rely on predefined thresholds for detecting islanding. The UKF method, however, is more adaptive. While it may still use thresholds, its dynamic nature means it can adjust more fluidly to system changes, resulting in a more flexible and reliable detection mechanism.
In conclusion, the unscented Kalman filter technique offers a robust and adaptive solution for islanding detection. Its ability to dynamically adapt, handle noise, and distinguish harmonic content makes it a promising solution that can significantly reduce the nondetection zone, enhancing the reliability and safety of power systems.
This section outlines the evaluation of the proposed islanding detection method. The method is tested under various scenarios, including islanding and normal operating conditions. A comprehensive literature review and extensive simulation results have informed the development of a new criterion for identifying grid islanding. The proposed criterion is based on the extraction of fundamental and harmonics, such as the third and fifth components, from the modal voltage signal. The UKF is used to extract these components at a suitable variable speed, generating a residual signal that represents the difference between the modal voltage signal and the predicted fundamental component. When the UKF is optimized for a variable speed, the residual signal can spike at the moment of a change but will stabilize to a low value as the UKF tracks the modal signal precisely. Variations in the electrical power system, such as islanding and nonislanding conditions, can be detected by comparing the residual signal to a predefined threshold.
The following subsections provide a more detailed description of the various procedures.
2.3. Modal Signal Estimation
The 3-phase0voltage0signal is transformed to a modal signal to decrease the computational load. The modal signal depicted in equation (1) represents a suitable linear integration of the 3-phase voltages. where denotes a modal signal and , , and are the computed 3-phase voltages. , , and, denote the modal signal coefficients. , , and are multiplied by distinct coefficients in equation (1). Arithmetic calculation like subtraction or addition between any of the two phases of 3-phase voltage signals is avoided in this method. This action proofs that any 2-phase signals’ transient data will not be nullified. Although relying solely on the three-phase voltage signal at the point of common coupling (PCC) can present challenges like the following:
Sensitivity to disturbances: PCC signals can be affected by grid fluctuations or transient states.
Single measurement point: a single data source might not capture the complete grid state or signal corruption, lack of redundancy, etc., which can be mitigated by employing the following further strategies:
Multisensor integration: adding readings from various grid points can enhance detection accuracy.
Signal integrity checks: preliminary assessments can ensure the PCC signal’s quality.
Machine learning: historical data can help identify and mitigate PCC signal distortions.
Our methodology prioritizes both accuracy and reliability, encompassing several critical steps. (i)Harmonic Content Identification. By decomposing voltage signals via Fourier transform techniques, we discern between fundamental frequencies and their harmonics.(ii)UKF-Based Harmonic Filtering. Identified harmonics undergo precise estimation and filtering using the unscented Kalman filter (UKF), capturing the harmonic’s mean and covariance without linearization.(iii)Thresholding Harmonic Amplitudes. To isolate impactful harmonics, we have established a threshold, discarding low-amplitude harmonics and emphasizing significant ones.(iv)Transient Harmonic Analysis. Given the transient nature of some harmonics, a time-domain analysis extracts those vital for islanding detection.(v)Harmonic Phase Insights. Our methodology evaluates both amplitude and phase shifts of harmonics, addressing potential islanding or grid disturbances.(vi)Benchmark Harmonic Comparison. We heighten reliability by contrasting extracted harmonics with known benchmark profiles, aiding in identifying potential grid anomalies.
This streamlined approach ensures robust islanding detection while minimizing potential errors.
2.4. Using UKF for Harmonic Assessment
The UKF provides a recurrent approach to solving linear discrete-data filtering problems. It can employ either model-based or measurement-based strategies, or a combination of both. The suggested technique employs the measurement-based UKF approach [23], in which the UKF analyzes the observed signal conditions. The harmonic content of the voltage signal measured at the distributed generation (DG) units is estimated using a UKF. To simplify the calculation, the 3-phase voltage signal is transformed into a modal signal, as described in the following equation. where denotes the calculated modal voltage signal0at0the th point, denotes the angular0frequency, and denotes the noise. The angular frequency is described as, , where is the fundamental frequency and is the0sampling0frequency. By means of trigonometric formula to the given equation, a repetitive equation for can be found as where denotes a zero mean term for the use of probable errors in the modal0signal. Such inaccuracies could include minor modifications in the magnitude and frequency of the modal signal.
In terms of noises (measurement and other unpredictable noises), the modal voltage signal is denoted by zero means () as
It is essential to express a linear state-space form for the variable model described in the previous equation (4). where
With the iterative UKF equations and above equations (5) and (6), it is possible to estimate the fundamental part of the modal voltage signal. The other harmonics can be calculated using matrix , where is replaced with , whereas relates to the harmonic order. Because the magnitude and phase of the modal signal do not exist in (5) and (6), the variable parameters of the system do not rely on it. As a result, the merging of and has no major effect on the UKF.
The utilization of UKF-estimated harmonics of modal signals as inputs to commonly used fast islanding detection techniques in modern electric power systems can enhance their accuracy and alleviate their susceptibility to harmonic content variations. The objective of feature extraction in this context is to identify the unique harmonic characteristics of voltage signals calculated at DG units, which can facilitate the distinction between islanding and nonislanding scenarios. During the initiation of an islanding occurrence, the discrepancy between the actual and predicted signals results in a substantial residual value. The residual signal is computed by means of
In addition, a criterion for distinguishing between an islanding state and other related changes is defined. This principle is related to total harmonic distortion and is referred to as THD which is described as follows:
signifies the R.M.S of the harmonic order, and is the order of the harmonic components. The suggested islanding detection method utilizes a comparison of total harmonic distortion (THD) with a predefined threshold to differentiate islanding scenarios from others. The method’s flowchart is depicted in Figure 1, beginning with the detection of a 3-phase voltage at the distributed generation unit, followed by its conversion to a modal signal. The harmonic properties of the modal voltage signal are then estimated using the unscented Kalman filter (UKF) technique. The UKF generates a residual signal that detects variations in the system. If a variation is detected, the THD is calculated and its fluctuations distinguish between islanding and normal grid operations. The UKF’s dynamic speed can be adjusted by confirming the qn/rn ratio to a desired value (refer to Appendix). The speed is altered by the ratio, not by qn and rn individually.
2.5. Threshold Settings
The suggested technique for detecting islanding incorporates the piecewise cubic Hermite interpolating polynomial (PCHIP) function as a central element of its detection algorithm. The motivation behind selecting PCHIP lies in its unique capability to produce smooth interpolations without exhibiting overshooting and oscillation artifacts commonly observed in other interpolating schemes, such as spline interpolation. This is especially crucial for islanding detection where transient fluctuations are often the key indicators and where overshooting may lead to false positives.
In our methodological design, we have used the PCHIP function to dynamically adjust the threshold values utilized for distinguishing between islanding and nonislanding events. During our extensive testing phase, the PCHIP function was subjected to a wide range of scenarios to evaluate its response and accuracy under different power system dynamics. Through this process, we ascertained the optimal variation range for interpolation. The lower and upper interpolants were determined to be 3% and 4%, respectively. This configuration was maintained consistently across all test cases to ensure uniformity in our results. A critical insight from our evaluation was the behavior of the total harmonic distortion (THD) under standard grid operations. The evaluation of the proposed method showed that the magnitude of the THD under normal situations reached a maximum of 3%. None of the nonislanding cases investigated exceeded this threshold, making it a reliable indicator of grid islanding when the threshold value surpasses the maximum, i.e., 3%.
3. Simulation Performance and Results
The efficacy of the suggested approach was assessed on a standard test system equipped with several DG units. A voltage signal was monitored at a designated DG bus and used as a test signal in a MATLAB/Simulink simulation platform. This technique underwent a range of tests under both regular and islanded operational circumstances, and the outcomes are comprehensively examined in this research.
3.1. Simulink Model of Test System
A standard 0test system is used to analyze the performance of the proposed model, as illustrated in Figure 2, which is simulated with MATLAB/Simulink software. Table 1 lists the Simulink model test system configuration, which is based on the IEEE standard system with various amendments. The Simulink model test system comprises a 25 kV utility source with two distribution feeders (F1 and F2), a short-circuit capacity of 2500 MVA, and an X/R ratio of 22.2. A fixed shunt capacitor bank of 3-phase 1.50 MVAR is set up at the substation. The Simulink model test system also contains one photovoltaic (PV) DG of 1.2 MW and one wind DG of 1.5 MW capacities on feeders F1 and F2, respectively. Two radial feeders are consumed to supply the loads containing RL loads.
3.1.1. Offset and Noise Filter
In a practical system, the Hall effect sensor utilized for feedback introduces a DC offset and noise into the signal, which contributes to errors within the control loop. The filter is depicted in a second-order generalized integrator (SOGI) [24], Figure 3. The feedforward path associated with this function effectively cancels out the offset value, enabling the signal to progress to the second stage with an amplifying gain factor denoted as “.” The SOGI operates as a transfer function featuring a damping coefficient “,” which effectively eliminates the signal’s ripple, yielding a pure sine wave characterized by the fundamental frequency “.”
3.2. Normal Operation Conditions
The suggested method’s effectiveness is assessed in typical operating conditions through modeling realistic scenarios. Changes in system parameters such as significant load initiation, sudden load fluctuations, capacitor switching, and motor activation result in conditions resembling islanding. To ensure the robustness of the technique, a series of simulations are conducted, including the case of an abrupt load change in both DG units for increased accuracy.
3.2.1. Impulsive Load Variation in PV Unit
Nonislanding situations, which can have comparable characteristics to islanding, were used to test the proposed method. A quick load variation situation was studied for a PV unit to test the system output. A 3-phase variable load was linked to bus 1 to simulate the situation of a rapid load change. was chosen as the step period for the load unit. The active power P and reactive power Q in a dynamic load model change as the positive sequence voltage increases. The load model’s P and Q parameters were changed from 12 MW to 8 MVAR. The outcomes in Figure 4(a) show that the suggested technique does not have any problems when the load is suddenly changed.
(a)
(b)
3.2.2. Impulsive Load Variation in Wind DG
A sudden load change is also studied for the wind DG unit, in which a large inductive load is linked to the network at . The instantaneous association of reactive power modifies the system parameters dramatically. Similar characteristics in the voltage or current waveforms may result from such a change, and the islanding procedures may fail. According to the data provided in Figure 4(b), the suggested technique does not have any problems when the load changes. This event confirms that the suggested technique is accurate in normal situations.
3.3. Grid Islanding Setup Conditions
The suggested method is assessed using a standard investigation system by simulating selected grid island conditions to validate its performance. The parallel RLC load and balanced load-generation conditions are the two main islanding scenarios examined here. Because of their limited NDZ, these two scenarios are regarded challenging to differentiate the islanding condition for passive approaches. The proposed method efficiently detects these islanding conditions in the shortest period possible.
3.3.1. Grid Island Detection with Passive Load
A well-known setup for testing the passive islanding detection method is grid island detection in modern electric power system with a combined parallel RLC load. An islanding circumstance was simulated to test the suggested method’s performance. Tripping circuit breakers B1 and B3, respectively, caused an islanding occurrence at by connecting a shunt RLC load with uniform impedance to bus 1 and bus 3. The results in Figures 5(a) and 5(b) show that the islanding condition is correctly detected. The islanding event occurs at , and the proposed method detects islanding for PV and wind units in real time. Both DG units achieve islanding detection in one power frequency cycle, as displayed in Figures 5(a) and 5(b).
(a)
(b)
(c)
(d)
3.3.2. Islanding Detection with Balanced Load-Generation
To validate the performance of the recommended approach, an islanding detection assessment was conducted in a balanced load-generation setting. The load resistance is attuned in this case to utilize a real power equivalent to the installed PV unit’s output. The grid island situation is created by pressing circuit breaker F1 for a second. The acquired results are shown in Figure 5(c). The grid islanding occurs at and is identified at , as indicated in the data (70 ms detection time). This demonstrates that the UKF technique efficiently identifies the islanding process with minimum downtime.
3.3.3. Islanding Detection under an Unbalanced Situation
An islanding detection test was performed under an imbalanced load state to verify the suggested method’s performance. At , the islanding situation is created by opening circuit breaker F3. The islanding occurs at and is identified at , according to the results (50 ms detection time). This demonstrates that the proposed method is capable of detecting even the worst-case events in the shortest amount of time.
3.4. Performance Evaluation of the Suggested System
The suggested islanding detection method’s performance is evaluated (Figures 6(a) and 6(b)). The NDZ of islanding detection methods is determined using a standardized test system. Table 2 displays the detailed system parameters. A DG of 70 kW current control inverter with RLC load is included in the system. In relation to the axis components, the DG’s actual and reactive powers are as follows:
(a)
(b)
Here, and represent the components of the voltage at PCC. At , considering the steady state condition, equations (10) and (11) are given by
Thus, depending on the above equations (12) and (13), and are controlled by and , respectively. Now, with the above equations,
The and would be set to 70 kW and 0, for an inverter-based DG conducting at , respectively. For analysis of the proposed method, various test cases, together with UL-1741 testing for nonislanding protection, dynamic interconnection, and -factor loads, are executed.
3.4.1. UL-1740/1 Testing
The UL-1741 standard [20] is implemented for the assessment of inverter-based DG units in IEPS, comprising islanding tests. To evaluate the proposed method, the load real power is adjusted to fix the inverter at 25, 50, 100, and 125% of the rated output power of the inverter. The test case evaluation is carried out in accordance with UL 1741 by changing the reactive load in 1% steps from 95 to 105 percent of the balanced load condition. Table 3 shows cases representing various real and reactive powers for UL 1741 testing. Here, at , the islanding event occurred. A considerable improvement is noticed shortly after the onset of islanding, as illustrated in Figure 7(a). The variation in THD % value is not less than the stated threshold in all cases, indicating that the occurrence is an islanding. Within two frequency cycles, the proposed technique accurately detects all situations.
(a)
(b)
(c)
3.4.2. Different Load Variation Tests
The suggested technique is tested in islanding mode for different loads located inward the voltage and frequency relay’s NDZ. Table 4 shows the test case details for the several load changes. The islanded mode was initiated at for the varying load change experiments. The results in Figure 7(b) indicate that the suggested technique is capable of accurately detecting all test scenarios. The THD percentage rate for the grid island situation is greater than the predefined threshold, as shown in Figure 7(b) (i.e., 3%).
3.4.3. Test Case for Short Circuit Faults, Capacitor Switching, and Loss of Parallel Feeder
(a)Short Circuit Faults. The system was exposed to short circuit conditions, and our method swiftly detected the perturbations in the signal. The proposed technique discerned the fault-induced disturbances within seconds, differentiating them from typical islanding patterns, as seen in Figure 8(a).(b)Load and Capacitor Switchings. These dynamic changes can indeed induce transient signals which resemble islanding scenarios. In every test, the proposed technique accurately identified these events as nonislanding disturbances, reaffirming its precision(c)Loss of Parallel Feeder. The sudden loss of a parallel feeder can introduce significant disturbances in the grid. The proposed technique was observed to efficiently handle such scenarios, which can be observed in Figure 8(b) detecting and differentiating the resulting signal changes
(a)
(b)
The strength of our method lies in its ability to quickly and accurately differentiate between islanding and these diverse nonislanding disturbances. Our results reinforce the efficacy of the proposed technique in managing the complex, multifaceted challenges inherent in islanding detection.
3.4.4. Variable Load Quality Factor Test
The proposed scheme’s grid island method test is carried out for small loads with various quality factors. Table 5 lists the load test conditions for several quality factors. There is no massive power discrepancy between demand and DG, according to the data in Table 5, which is considered a worst-case supposition for an islanding detection test. The islanding event was performed at to simulate the changing load with multiple quality factor test cases. The results in Figure 7(c) reveal that there is a considerable shift under the islanding situation. The THD percentage number is not smaller than the set threshold in any of the test instances.
3.5. Comparison of the Suggested Technique with Several Other Similar Techniques
In the suggested0passive0islanding0detection method, a UKF is employed to extract harmonic components from a voltage signal monitored at a specific DG unit. The presence of islanding or normal operating conditions is then determined using a novel criterion known as total harmonic distortion (THD). The performance of the proposed method is compared against other methods that rely on voltage signals measured at the DG unit, including the overvoltage protection/undervoltage protection (OVP/UVP) method [31], the wavelet transform (WT) technique [9], and the superimposed component-based grid islanding detection approach [21]. A brief overview of these methods is provided below. (1)OVP/UVP. The OVP/UVP-based islanding detection method detects changes in voltage amplitude at the DG units. This approach is favored for its simplicity and lack of impact on system power quality. However, it may not accurately detect islanding conditions when local demand is comparable to the DG power output. This can lead to false negatives, where islanding conditions go undetected, and compromise the effectiveness of the detection method.(2)Wavelet Transform. The wavelet transform- (WT-) based islanding detection method is a time-frequency analysis technique that utilizes principles from short-time Fourier transform (STFT) and multiresolution analysis. This method differs from STFT in that it analyzes signals by employing dynamic windows, allowing for a more comprehensive analysis of the time-frequency representation of the signal. However, one drawback of this method is that it only captures the low-frequency components, potentially overlooking valuable information in the high-frequency band.(3)The Technique for Detecting Islands Using Superimposed Voltage Components. A method for detecting changes in the voltage signal at DG (distributed generation) terminals. This technique involves the calculation of phasor values of the superimposed components of the voltage signal using a discrete Fourier transform. The performance of this method is compared to OVP/UVP-based and wavelet transform- (WT-) based islanding detection methods using MATLAB/Simulink simulations and test systems. The comparison includes metrics such as the nondetection zone (NDZ) of OVP/UVP-based methods, the window size and effect of the Daubechies wavelets in WT-based methods, and the sensitivity of superimposed components-based methods. The results show that the proposed method has superior accuracy and low computational burden, with an islanding detection speed that is within the acceptable limits of the IEEE-1547 standard
Linearization and nonlinearity handling: traditional Kalman filters (KF) rely on the linearization of nonlinear models. This process can introduce errors, especially when the system deviates significantly from the linear approximation. On the other hand, UKF operates by using a deterministic sampling approach known as the unscented transformation, allowing it to capture the mean and variance of nonlinear transformations without requiring linearization. This property is beneficial in complex power systems where nonlinear behaviors are common.
Noise management: while both extended Kalman filters (EKF) and UKF manage noise, the UKF has a systematic way of dealing with Gaussian noise, which can be prevalent in electrical systems. By managing noise more effectively, the UKF can provide more accurate islanding detection in real-world scenarios where noise is unavoidable.
Computational efficiency: particle filters (PF) are another method used for nonlinear filtering. However, PF often requires a large number of particles for accurate representation, leading to increased computational burden. UKF, in contrast, achieves higher accuracy without needing an extensive number of sampling points, making it computationally more efficient.
Adaptability and flexibility: UKF can adapt to changing system dynamics in real time due to its recursive nature. This adaptability ensures that even if there are changes in system parameters or unexpected events, the UKF remains reliable in its predictions and detections.
Robustness to initial conditions: One common problem with filters like EKF is their sensitivity to initial conditions. An incorrect initial state can lead the filter to converge to an incorrect estimate. UKF, with its unscented transformation, is more robust to such initial condition errors.
In summary, while other filtering techniques and methods provide valuable insights and can be effective in specific scenarios, the unscented Kalman filter presents a well-rounded solution that addresses many of the challenges inherent in islanding detection in modern electric power systems. Figure 9 shows the comparison results of UKF with WT and KF, and from Table 6, we can compare the abovementioned with different attributes.
Its ability to handle nonlinearities without linearization, manage noise effectively, and operate with computational efficiency makes it a prime candidate for the challenges posed by islanding detection.
Additionally, the proposed method uses a single sample frequency for the entire analysis, is less susceptible to existing schemes, and is not affected by variable sampling frequency or time delay. The observed outcomes are reported in Table 7 and evaluated based on detection time, precision, and computational burden. The precision is calculated by accurately identifying the event as an islanding or normal situation, and the computational burden is calculated based on hardware, simulation time, and calculation burden related to the governing formulation. The results from the 240 simulated instances, with 80 instances for islanding and 160 instances for normal conditions including load variation and power imbalances, demonstrate the effectiveness of the proposed method.
3.6. Result Summarisation
The performance of the proposed technique for detecting islands using superimposed voltage components is evaluated through various tests conducted on both an IEEE-399 standard test system and an inverter-based sample test system, as depicted in Figures 2 and 5. The tests encompass a range of islanding and nonislanding scenarios in accordance with the IEEE-1547, UL-1740/1, and IEEE-929 standards. The results demonstrate that the method has a 100% detection accuracy for shunt RLC and balanced load situations. The suggested technique’s effectiveness is further showcased by its ability to detect islanding conditions under diverse operational circumstances, such as variations in actual and reactive power discrepancy, qf, and loading conditions.
The results of the tests indicate that the proposed method is a highly effective solution for detecting islanding in practical systems. The results are accurate as they can accurately distinguish between islanding and normal conditions in various operational conditions. The technique employs the harmonic component of a modal voltage signal and computes THD to detect an islanding state within the time constraints specified by the IEEE-1547 standard. The proposed UKF-based strategy thus appears to be a feasible option for practical application. Furthermore, the nonislanding test conditions demonstrate that the method does not produce false trips and does not fail in nonislanding scenarios.
We acknowledge the concerns about relying solely on the three-phase voltage signal at the point of common coupling (PCC). This singular data source can be influenced by disturbances, limited system views, and possible signal corruption. Additionally, sole reliance reduces redundancy, potentially impacting detection when faced with limited data or ambiguous PCC signals. To mitigate these limitations, we propose the following: multisensor integration, signal integrity checks, and machine learning application.
4. Conclusion
A new passive grid island detection technique based on an unscented Kalman filter (UKF) method that detects voltage signals at the point of common coupling (PCC) has been proposed. This method is designed to distinguish between normal and islanding situations by using a UKF with a suitable variable speed to monitor any changes in the grid. The technique has a low nondetection zone (NDZ) and is cost-effective as it does not require the measurement of current signals. Our analysis highlights several crucial challenges in the domain of power systems:
Voltage instability: system uncertainties can cause voltage fluctuations, which, if misinterpreted by conventional techniques, can lead to incorrect islanding detections. These fluctuations risk damaging sensitive devices and pose fire threats.
Frequency deviations: any deviation in the essential parameter of frequency can desynchronize grid components, potentially triggering expansive blackouts.
Harmonic distortions: uncertainties can enhance grid harmonic content, complicating islanding detection due to equipment overheating and efficiency loss.
Wear and tear: persistent fluctuations intensify strain on components, especially transformers, shortening their lifespans.
Cascading failures: Minor unchecked issues in interconnected systems, like grids, can spiral into larger systemic failures.
Given these challenges, the urgency for a refined detection technique is evident. Our proposed UKF method stands out, promising precision and adaptability and setting a new benchmark over traditional methods in addressing uncertainties.
The proposed UKF technique, with its adaptability and precision, is designed to tackle these uncertainties more effectively than traditional methods. Its ability to handle nonlinearities without linearization, manage noise effectively, and operate with computational efficiency makes it a prime candidate for the challenges posed by islanding detection. The performance of the proposed technique was evaluated using simulations that included various islanding and normal situations such as minor/major power imbalances and balanced/unbalanced load connections/disconnections. The results were compared to other similar tests, and it was found that the UKF-based technique is precise and fast in detecting islanding, with an average detection time of 1-2 cycles of the power frequency. While THD has shown promise in initial tests, it is not without its challenges like disturbance sensitivity and external influences which may result in false alarms. The key advantages of the proposed technique include its speed, precision, and robustness in real-world and noisy situations. These qualities make the UKF-based technique a suitable option for practical applications.
Appendix
A. UKF Process
(1)Set initial estimations for the state vector and its error covariance matrix ( and )(2)Calculate the filter gain instant with (3)Upgrade the estimate with the measurement at instant viawhere belongs to , and based on the abovestated equation, the input parameters for the unscented Kalman filter were initialized (4)Error covariance calculation for the improved estimate with (5)Filter task onwards as(6)Start from II
Data Availability
The datasets generated and/or analyzed during the current study are not publicly available but can be requested from the corresponding author on reasonable request.
Disclosure
A preprint of this manuscript has previously been published [32].
Conflicts of Interest
The authors declare that they have no competing interests.
Acknowledgments
I am deeply grateful to the Department of Electrical Engineering at the School of Engineering at Gautam Buddha University and the Department of Electrical Engineering at the Indian Institute of Technology Roorkee for their invaluable support.