Do-In Kim

The real-time prediction of nitrogen oxide (NOx) emissions is essential in terms of effective denitrification control for combined cycle power plant (CCPP). However, prediction accuracy and temporal transferability are often challenged by various operational conditions of CCPP including dynamic gas flow in large-scale structure and periodic overhauls. In this regard, this paper presents a robust and precise prediction method applicable to the NOx emission in the CCPP. To overcome these challenges, this study develops a graph attention mechanism that utilizes the complete set of operational monitoring signals, guided by an attention-based signal selection process with learnable input importance. Furthermore, a prediction algorithm is formulated by integrating a parallel GATv2-LSTM framework, which effectively captures operational relationships and temporal dynamics across different time steps. Numerical tests utilizing real plant operation data show that the proposed method outperforms recent deep learning models in terms of prediction accuracy. Moreover, it demonstrates robustness in terms of temporal transferability problems as well as noise. This study offers practical insights for optimizing denitrification systems in CCPP.

As the operational years of power facilities increase, the importance of accurately assessing the remaining lifespan or risk of failures of distribution transformers for the stable power supply has also grown. However, unlike the typical transformer failures, it is not quantified that the failures are also influenced by weather conditions such as temperature, humidity, precipitation, etc. In this regard, we propose the method for investigating the impact of the weather conditions on the failures of distribution transformers to improve the asset management. By employing survival analysis approach, frameworks are employed to quantify the spatial influence of weather conditions on distribution transformer failures, and the weather risk factors are integrated into lifespan indicators to improve maintenance strategies and reliability assessments. Distribution transformer failure cases of over 11 years in South Korea and the corresponding historical weather conditions are configured to enable the survival analysis and derive the meaningful real-world analysis results. Specifically, weather data are quantified and interpolated to configure the spatial data. The notable results underline quantifying significant impact of high summer temperature and humidity in term of hazard ratios. These findings indicate that incorporating historical weather data into asset management indices can significantly enhance the accuracy of transformer lifespan predictions and inform maintenance planning. This study represents the first long-term survival analysis correlating historical weather conditions with transformer failures and provides a guideline for improving power-facility management in the context of environmental changes.

Due to the aging of the power systems, it is one of the important issues to identify the remaining useful life (RUL) of the power facilities including the environmental effects for the reliable power system operation. However, it is hard to identify the effects of spatially distributed environmental effects on each individual facility and manage the individual history of millions of distribution facilities. As a practical solution, we propose a spatiotemporal analysis to derive the comprehensive effects of the lightning strike that is one of the environmental effects on the RUL of the distribution transformers, and provides the guideline for the lightning protection through the decision of lightning-prone areas. We acquired the data related to the distribution transformer failures and the lightning strikes in South Korea, and a spatiotemporally weighted regression analysis is applied to the data for deriving the spatially cumulative effects of a lightning, which is hard to be identified through each field post-analysis. In addition, we propose a method for validating the regression model through the model evaluation using the categorized transformer failures by field post-analysis. All the proposed processes are performed through the real-world data analysis, and practical results are presented for the reinforcement of lightning protection within the Korean power system.

Although the transient stability collapse rarely occurs in daily power system operations, it has a serious ripple effect on the system reliability. Thus, preventing and mitigating the ripple effects from an event or disturbance are important issues for stable operation. In this regard, phasor measurement unit (PMU) facilities enable the acquisition of synchrophasor information and fast transient stability assessments (TSA) before a stability collapse occurs. In addition, the application of a data-driven method to PMU data further enables transient stability prediction (TSP) stage from the near real-time event analysis. However, streaming environments require additional considerations, as always-on prediction must be conducted even before the event occurrences are confirmed, while recognizing events in real-time remains challenges. Consequently, this study proposes a framework that combines a parallel convolutional neural network and long short-term memory (LSTM) structures to automatically perform an event-based TSP. The proposed method uses single event signals to deliver TSP results as streaming data across all the times, eliminating the false alarm rate without requiring the event detection. Finally, the significance of the proposed detection-free TSP method is verified through case studies, addressing anticipated challenges in real-world applications.

This paper presents a distributed synchrophasor data management (DSDM) technique for wide-area monitoring systems. The proposed technique focuses on managing voltage/current phasor data measured from a specific target bus by defining neighboring phasor measurement units (PMUs). We developed an algorithm for searching neighboring PMUs by interpreting the power network topology as a graph structure. By exploiting the data of neighboring PMUs, we constructed a multi-channel structure to recover low-quality entries by using the parallel implementation of long short-term memory networks. A successive output decision-making rule automatically determines the final output, considering the occurrence of defective input, data quality of the measured data, and operational conditions. The proposed DSDM can complement existing centralized management techniques by preserving information on local dynamics. In particular, because of the multi-channel structure, DSDM can provide robust responses to defective data in neighboring PMUs, such as an unintended change in a network topology. We verified the effectiveness of the DSDM with real-world synchrophasor data as well as simulated data from the IEEE 68-bus test system.

This paper presents an event identification process in complementary feature extractions via convolutional neural network (CNN)-based event classification. The CNN is a suitable deep learning technique for addressing the two-dimensional power system data as it directly derives information from a measurement signal database instead of modeling transient phenomena, where the measured synchrophasor data in the power systems are allocated by time and space domains. The dynamic signatures in phasor measurement unit (PMU) signals are analyzed based on the starting point of the subtransient signals, as well as the fluctuation signature in the transient signal. For fast decision and protective operations, the use of narrow band time window is recommended to reduce the acquisition delay, where a wide time window provides high accuracy due to the use of large amounts of data. In this study, two separate data preprocessing methods and multichannel CNN structures are constructed to provide validation, as well as the fast decision in successive event conditions. The decision result includes information pertaining to various event types and locations based on various time delays for the protective operation. Finally, this work verifies the event identification method through a case study and analyzes the effects of successive events in addition to classification accuracy.

This paper presents a data-driven approach for event classification via a regional segmentation of power systems. The data-driven approach is suitable for the complex power systems under transient conditions, as it directly derives the information from the measurement signal database instead of modeling transient phenomena. However, measurement conditions of real-world power system will have unavoidable missing and bad data. Thus, it is necessary for data-driven model to have a robustness and adaptability about varying environment as well as system configurations and measurement conditions. In this work, the clustering-based regional segmentation of power systems is adopted to improve robustness of the data driven model by maintaining the fixed-input-feature format under varieties of measurement conditions. The clustering technique is applied to electrical buses for regional segmentation, and proposed features of phasor measurement unit (PMU) signals are extracted by integrating PMUs in each region based on wavelet analysis. As a result, the regional segmentation achieves improvement of data driven method for event classification with reduced number of calculations and management of bad data. Finally, we verify the event classification algorithm through a case study and analyze the performance of the algorithm for noise and computation time in addition to classification accuracy.

This paper presents the phasor measurement unit (PMU)-based event localization technique. In order to achieve the event localization in real-time manner, proposed method is based on the measurement of the time required of the event propagation at each PMU. When an event propagates through the transmission line network, there exist a time delay to arrive each electrical bus in wide-area. As an initiated time of the event is unknown information in a practical power systems, time difference of arrival (TDoA) is used to estimate the event location instead of time of arrival (ToA). In this regard, the key-point of proposed localization utilizes the comparison of TDoA values which are composed of entire transmission lines to estimate the propagation path. As a result, the directions and time difference in shortest path are developed as features under the assumption of event location. Furthermore, historical event-based network training method is proposed to substitute the system state estimation and calculation of propagation speed. Finally, the comparisons between calculated and measured TDoA features could estimate the event bus, and real-world test shows successful estimation results.

This paper presents a multiscale phasor measurement unit (PMU) data-compression method based on clustering analysis of wide-area power systems. PMU data collected from wide-area power systems involve local characteristics that are significant risk factors when applying dimensionality-reduction-based data compression. Therefore, density-based spatial clustering of applications with noise (DBSCAN) is proposed for the preconditioning of PMU data, except for bad data and the automatic segmentation of correlated local datasets. Clustered PMU datasets of a local area are then compressed using multiscale principal component analysis (MSPCA). When applying MSPCA, each PMU signal is decomposed into frequency sub-bands using wavelet decomposition, approximation matrix, and detail matrices. The detail matrices in high-frequency sub-bands are compressed by using a PCA-based linear-dimensionality reduction process. The effectiveness of DBSCAN for data compression is verified by application of the proposed technique to the real-world PMU voltage and frequency data. In addition, comparisons are made with existing compression techniques in wide-area power systems.

Load forecasting is a key issue for efficient real-time energy management in smart grids. To control the load using demand side management accurately, load forecasting should be predicted in the short term. With the advent of advanced measuring infrastructure, it is possible to measure energy consumption at sampling rates up to every 5 min and analyze the load profile of small-scale energy groups, such as individual buildings. This paper presents applications of deep learning using feature decomposition for improving the accuracy of load forecasting. The load profile is decomposed into a weekly load profile and then decomposed into intrinsic mode functions by variational mode decomposition to capture periodic features. Then, a long short-term memory network model is trained by three-dimensional input data with three-step regularization. Finally, the prediction results of all intrinsic mode functions are combined with advanced measuring infrastructure measured in the previous steps to determine an aggregated output for load forecasting. The results are validated by applications to real-world data from smart buildings, and the performance of the proposed approach is assessed by comparing the predicted results with those of conventional methods, nonlinear autoregressive networks with exogenous inputs, and long short-term memory network-based feature decomposition.