15 March 2022 in IEEE Transactions on Sustainable Energy
The wind turbines (WTs) present in a wind farm (WF) are subjected to varying wind speeds owing to the wake effect. Additionally, the response of a WF is also affected by the low voltage ride-through (LVRT) characteristic imbibed in each doubly-fed induction generator (DFIG) of WF. Consequently, for deriving an equivalent of a large WF, consideration of wake effect and LVRT are crucial. In this paper, a three-step WF aggregation is proposed. Firstly, various wind scenarios from historical wind data are taken, and the Weibull-based probabilistic and wake effect models are developed. Secondly, WTs’ clusters based on their LVRT activation status are obtained. In the final step, the WTs are aggregated based on the pre-fault wind speed, and a multi-machine dynamic equivalent model is developed. Also, a new collector network transformation method is proposed. Various low voltage cases are simulated on a 40 MW test WF at different wind conditions. A comparison of the responses from the proposed equivalent model and the detailed model of the test WF indicates the efficacy of the proposed multi-machine model. A comparison of the errors in responses confirms the superiority of the proposed model over the previous relevant works.
30 June 2023 in IEEE Transactions on Instrumentation and Measurement
Accurate identification of various loads for Non-Intrusive Load Monitoring (NILM) plays a crucial role in a variety of power system applications and strategies. In this paper, a novel approach is proposed for performing load identification which consists of three main parts: features extraction considering decomposition-transformation strategy, appropriate feature screening, and employing a suitable classification strategy for performance improvement. The proposed approach firstly decomposes a load current into multiple modes by utilizing an adaptive signal decomposition technique, namely, Variational Mode Decomposition (VMD), followed by the determination of the auto-correlograms of the modes. Multiple features are extracted from the obtained modes and their respective auto-correlograms. Post feature extraction, combinations of multiple feature ranking (FR) techniques via union and intersection strategies are explored in contrast to employing only a single FR technique as prevalent in existing literature in NILM. Finally, considering the selected features by above stated FR combinations, a heterogeneous classifier strategy is adopted for performing classification that outperforms a single classifier as well as random forest classifiers on account of its heterogeneous structure. For further performance improvement, a surrogate based optimization technique is utilized for determining the appropriate hyper-parameter values in contrast to manual tuning which is prominent in existing works. Experimental appliance data obtained from publicly available data-sets and execution via the software-in-loop mode have been considered to put forward the efficacy of the proposed approach.
16 June 2023 in IEEE Transactions on Smart Grid
Non-Intrusive Load Monitoring (NILM) aids in estimating the individual energy consumed by different appliances using aggregate signals measured at point of common connection. This paper explores various unexplored aspects concerning load classification in NILM. Rich feature sets are extracted from load signatures using auto-correlogram, Variational Mode Decomposition (VMD), and Wavelet Packet Tree (WPT) decomposition. Typically, previous works employ a single Feature Selection (FS) technique to earmark the relevant features for improved performance of NILM. However, each FS technique behaves uniquely, and the effects of using different FS techniques or their combinations have yet to be explored for NILM. Hence, Combinatorial Fusion Analysis (CFA), which surpasses the performance of a single FS technique by considering an optimal combination of multiple FS techniques, is used in this paper. The final selected features are fused and provided to a classifier. Rather than manually tuning or performing an exhaustive search for setting the classifier’s hyper-parameters, as prevalent in previous works, the present work utilizes Bayesian Optimization (BO) comprising surrogate and acquisition functions. The proposed work’s efficacy is validated on publicly available data-sets in a host computer and an embedded device, namely, R-Pi, via comparison with previous works.
29 October 2020 in IEEE Transactions on Energy Conversion
In this article, an online approach for joint parameter estimation of the exciter and Synchronous Generator (SG) is proposed. It only utilizes the Phasor Measurement Unit (PMU) based measurements, which can be obtained at the SG's terminal and does not require field measurements (as in brushless exciters). Compared to existing joint estimation approaches, the proposed work includes multiple aspects to put forward a pragmatic estimation framework. Firstly, the unobservable erroneous initial states are duly considered by utilizing the proposed Augmented Parameter Vector (APV). It is an essential feature in practical scenarios as an accurate representation of a dynamic system entails knowledge of true initial states and parameters. Secondly, suitable models of SG and exciter necessitate the inclusion of saturation characteristics, which is also considered in present work. Further, the estimation procedure is independent of multiple pre-tuned parameters, as such dependence may lead to tedious implementation and inaccurate estimates. Additionally, Wavelet denoising is used to improve the parameter estimation accuracy by pre-processing the noisy measurements prior to estimation. Real-Time Digital Simulator (RTDS) based simulation results, considering noise and measurement rate analogous to commercial synchrophasors, demonstrate the proposed approach's improved accuracy.
23 December 2021 in IEEE Transactions on Power Systems
With the power system operating close to its stability margin, the increased penetration of renewable resources and loads with diverse characteristics, appropriate stability studies, and control decisions have become crucial. Due to the importance of accurate representation of aggregate loads in such studies, research interests for load modeling have increased. This paper puts forward a novel measurement-based load modeling approach that adapts itself based on the complex dynamics exhibited by the installed aggregate loads. The proposed approach primarily utilizes the collected measurements to gain insight regarding the dynamic trend in the measurements and subsequently determines the optimal order of an equivalent system or load model capable of effectively representing the aggregate load dynamics. An advanced signal pre-processing technique is employed, which effectively suppresses the noise and also preserves the transients present in the measurements. Afterward, an adaptive dynamic load model (ADLM) accurately represents the installed aggregate load. Lastly, the Refined Instrumental Variable (RIV) approach estimates the equivalent dynamic load model parameters. The results on a 2 bus and the IEEE 118 bus networks in DIgSILENT Powerfactory and a 7 bus network in Real-Time Digital Simulator (RTDS) reveal the efficacy of the proposed approach.
1 August 2023 in IEEE Transactions on Smart Grid
The present work proposes to utilize a data-driven technique, namely, phase space reconstruction (PSR), followed by an exploration of multiple delays, which, in turn, enables the generation of multiple distinct image sets for each of the load signatures without involving data-normalization. Considering the multiple generated images, individual convolutional neural networks (CNNs) for each of the images are considered, resulting in variegated predictions which are effectively combined within a majority voting framework in order to limit the erroneous predictions of the individual classifiers. The proposed strategy leads to significant improvement in classification accuracy compared to the relevant existing literature.