Accurate power system state estimation (PSSE) is crucial for effective control and monitoring of power grids. However, existing techniques face numerous challenges including compromised measurements from cyber-attacks and the need for real-time operations. Traditional PSSE methods based on weighted least squares are susceptible to outliers and convergence issues, while convex relaxation techniques like semi-definite programming struggle with computational complexity. Recent deep learning approaches for PSSE are data-driven but unaware of the underlying physical constraints. This paper proposes a novel framework called learned block successive convex approximation (LBSCA) to address PSSE amidst hidden adversarial attacks. LBSCA utilizes an efficient block successive convex approximation algorithm that provably converges to a stationary point and can be learned via algorithm unrolling. The resulting deep neural network achieves real-time, end-to-end PSSE in fractions of a millisecond, even when attacks are unobserved and the network topology is incomplete. Evaluated on the IEEE 118-bus system under realistic noise profiles and stealthy directional and sparse injection attacks, LBSCA achieves lower state estimation errors by up to a factor of three than state-of-the-art deep learning baselines and delivers inference speeds up to four orders of magnitude faster than classical optimization algorithms. Our results demonstrate that model-driven unrolling of robust estimation yields both theoretical rigor and practical efficiency, underscoring its potential in improving real-time power system operation’s efficiency and security.

This paper addresses the increasingly critical challenge of accurate inertia estimation in modern power systems undergoing a significant transition from traditional synchronous generators to renewable energy sources. The high penetration of renewables leads to a reduction in system inertia, causing increased rate of change of frequency (ROCOF) and frequency deviations, which can threaten grid stability. To tackle this issue, we introduce a novel measurement-based hybrid Multi-Head Graph Attention Network and Long Short-Term Memory (GATLSTM) framework for estimating the total inertia constant. Our approach uniquely integrates the spatial awareness of graph attention networks (GATs) to capture the topological dependencies of the power network with the temporal modeling capabilities of Long Short-Term Memory (LSTM) networks to learn from sequential phasor measurement unit (PMU) data. By leveraging PMU measurements of frequency deviation and ROCOF from a dataset based on the IEEE 24-bus system, our experimental results demonstrate that the proposed GATLSTM model significantly outperforms conventional deep learning models such as Graph Convolutional Networks (GCN) particularly under realistic noise conditions. Extensive experiments, sensitivity analysis, and an ambiguity-set evaluation demonstrate the GAT-LSTM's robustness, superior accuracy, and ability to resolve overlapping-signal cases while providing well-calibrated prediction intervals. These results highlight its potential for realtime power system monitoring and enhanced resilience against dynamic system variations and measurement noise.

Day-ahead photovoltaic (PV) forecasting must resolve the coexistence of a smooth diurnal generation envelope and shorter-scale fluctuations while providing uncertainty information suitable for operational decision making. This paper develops a multi-resolution Wavelet–CNN–LSTM framework for 24 h ahead PV power forecasting and evaluates it using a five-year PVGIS-derived dataset for a 334.116 MW Rabigh case study. The forecasting information set is restricted to the preceding 24 h of PV power and meteorological observations, together with deterministic target-day temporal variables. A validation study selects a Haar level-2 decomposition of PV history into $A_2$, $D_2$, and $D_1$ components, which are processed through parallel convolutional branches before recurrent fusion. The resulting network contains 87,032 trainable parameters. On the 2023 temporal evaluation period, the proposed model achieves an RMSE of 17.228 MW, $R^2=0.9641$, and a daylight RMSE of 24.636 MW, corresponding to 25.69% forecast skill relative to 24 h seasonal persistence. A 7-day moving-block bootstrap confirms clear RMSE reductions relative to persistence and XGBoost, while the intervals for LSTM and CNN–LSTM include zero, indicating statistically comparable performance among the strongest recurrent models. Probabilistic forecasts obtained from 100-pass Monte Carlo dropout yield daylight PICP = 0.9279, MPIW = 84.06 MW, and Winkler score = 122.08 for nominal 90% intervals; an independent conformal calibration set requires no additional widening of the neural interval. The results show that multi-resolution representation can improve squared-error performance without sacrificing model compactness, while calibrated uncertainty provides a complementary measure of forecast reliability.