This paper presents a deep residual network for improving time-series forecasting models, indispensable to reliable and economic power grid operations especially with high shares of renewable energy sources. Motivated by the potential performance degradation due to the overfitting of the prevailing stacked bidirectional long short-term memory (Bi-LSTM) layers associated with its linear stacking, we propose a concatenated residual learning by connecting the multilevel residual network (MRN) and DenseNet. This method further integrates long and short Bi-LSTM networks, ReLU and SeLU for its activating function. Rigorous studies present superior prediction accuracy and parameter efficiency for the widely used temperature dataset as well as actual wind power dataset. The peak value forecasting and generalization capability along with the credible confidence range demonstrate that the proposed model offers essential features of a time-series forecasting, enabling a general forecasting framework in grid operations. The source code of this paper can be found in https://github.com/MinseungKo/DRNet.git.

This study prognoses the remaining useful life of a turbofan engine using a deep learning model, which is essential for the health management of an engine. The proposed deep learning model affords a significantly improved accuracy by organizing networks with a one-dimensional convolutional neural network, long short-term memory, and bidirectional long short-term memory. In particular, this paper investigates two practical and crucial issues in applying the deep learning model for system prognosis. The first is the requirement of numerous sensors for different components, i.e., the curse of dimensionality. Second, the deep neural network cannot identify the problematic component of the turbofan engine due to its “black box” property. This study thus employs dimensionality reduction and Shapley additive explanation (SHAP) techniques. Dimensionality reduction in the model reduces the complexity and prevents overfitting, while maintaining high accuracy. SHAP analyzes and visualizes the black box to identify the sensors. The experimental results demonstrate the high accuracy and efficiency of the proposed model with dimensionality reduction and show that SHAP enhances the explainability in a conventional deep learning model for system prognosis.

4차산업혁명 기술 중 가상·증강현실 기술은 최근 급격한 요소기술의 발전과 함께 다양한 분야의 어플리케이션 개발이 활발히 이루어지는 중이며, 특히 위험성이 크고 비용이 많이 소모되어 반복적인 숙달이 제한되는 군 교육훈련 분야에 적용 시 상당한 비용 절감과 전투력 향상 효과가 기대된다. 본 논문에서는 가상ㆍ증강현실 교육훈련 체계 개발을 위한 주요 요소 기술들인 디스플레이 기술, 객체 인식 및 추적 기술, 영상합성 기술, 상호작용 기술의 발전 동향에 대해 살펴보고, 가상·증강현실 기술이 국방 교육훈련 체계에 적용되기 위해 연구되고 있는 사례들을 소개한다. 마지막으로 군의 교육훈련 분야에 적용하기 위한 가상·증강현실 기술의 활용과 개발 방향을 제시한다.

Deploying deep learning for state-of-charge (SOC) estimation on resource-constrained battery management systems (BMS) is critical for platforms under stringent Size, Weight, and Power (SWaP) constraints, yet existing data-driven approaches optimize accuracy alone and address model complexity and latency only after architecture selection, so the resulting designs often fail to meet edge deployment budgets. We propose a multi-objective neural architecture search (NAS) framework that integrates NSGA-II and Random Search via a Combined Pareto Front, evaluating CNN, LSTM, GRU, and Transformer families in a unified search space at an identical 300-trial budget per algorithm. Pareto-optimal candidates pass through a three-dimensional Edge-Realistic Ideal Zone grounded in deployment envelopes: Val MAE < 0.025 (±5% SOC band per automotive BMS practice), FLOPs < 2.0 M (ARM Cortex-M7 real-time throughput), and parameter count < 30 K (ARM Cortex-M3 SRAM). On the NASA (long-term degradation) and CALCE (dynamic stress test) datasets with five-run repeated training, the preferred architecture depends on analysis depth: NASA's Ideal Zone admits exclusively GRU candidates (5/5) while CALCE admits a mixed CNN/GRU/LSTM composition, yet once five-run robustness validation is applied, both converge to GRU as the sole val_safe main candidate — Trial 208 for NASA (16,161 params, 1.49 MFLOPs, val MAE 0.0119 ± 0.0015) and Trial 136 for CALCE (2,737 params, 0.058 MFLOPs, val MAE 0.0180 ± 0.0016). NSGA-II accounts for 80% of Ideal Zone candidates across both datasets and 100% on CALCE, concentrating trials where deployment constraints bind. By embedding hardware constraints directly in the search loop, the framework supports SOC estimation across SWaP-constrained platforms — unmanned vehicles, electric vehicles, and stationary energy storage.