Accurate forecasting of photovoltaic (PV) power is essential for grid scheduling and energy management. In recent years, deep learning technology has made significant progress in time-series forecasting, offering new solutions for PV power forecasting. This study provides a systematic review of deep learning models for PV power forecasting, concentrating on comparisons of the features, advantages, and limitations of different model architectures. First, we analyze the commonly used datasets for PV power forecasting. Additionally, we provide an overview of mainstream deep learning model architectures, including multilayer perceptron (MLP), recurrent neural networks (RNN), convolutional neural networks (CNN), and graph neural networks (GNN), and explain their fundamental principles and technical features. Moreover, we systematically organize the research progress of deep learning models based on different architectures for PV power forecasting. This study indicates that different deep learning model architectures have their own advantages in PV power forecasting. MLP models have strong nonlinear fitting capabilities, RNN models can capture long-term dependencies, CNN models can automatically extract local features, and GNN models have unique advantages for modeling spatiotemporal characteristics. This manuscript provides a comprehensive research survey for PV power forecasting using deep learning models, helping researchers and practitioners to gain a deeper understanding of the current applications, challenges, and opportunities of deep learning technology in this area.

This paper addresses the multi-source uncertainties faced by horizontal pumped storage-wind-solar (HWS) hybrid systems in the day-ahead market by proposing a hybrid stochastic-robust optimization model for bidding and scheduling. The model employs a scenario-based method to capture the randomness of wind and solar power output, utilizes Information Gap Decision Theory (IGDT) to handle the epistemic uncertainty in runoff inflow forecasting, and constructs a price-acceptance probability function based on historical statistics to characterize the market mechanism. By maximizing the system’s tolerable uncertainty immunity gap, the model co-optimizes generation schedules, pumped-storage operation, and market bids while ensuring that revenue under the worst-case inflow scenario does not fall below a predefined threshold. Simulation results based on an actual project in Hubei Province demonstrate that the proposed method effectively balances revenue and risk, showing significant advantages in both revenue stability and robustness compared to the system before retrofitting. This study provides practical decision-making support for hybrid systems with horizontal pumped storage participating in electricity markets.

Within a single flood event, this study evaluated the empirical relationship between underwater hyperspectral observations and layer-scale suspended sediment concentration (SSC) and calibrated its predictive uncertainty. The dataset comprised 101 paired observations from 25 complete four-position verticals (n = 100) plus one isolated record. Sample-level leave-one-out cross-validation (LOOCV) provided the primary estimate of event-internal layer-level interpolation, while leave-one-vertical-out validation (LOVO) assessed sensitivity to an unseen vertical. Under LOOCV, random-forest models reached R² ≈ 0.78–0.79 and RMSE ≈ 0.087 kg/m³; on the complete-vertical subset, spectra-only RF_log1p decreased to R² = 0.595 and RMSE = 0.120 kg/m³ under LOVO. Discharge and interpolated stage added event-specific context (Q/stage-only LOVO R² = 0.699), but cross-date transfer remained poor. Raw quantile-regression-forest intervals under-covered the nominal 90% level; conformal calibration restored coverage at the cost of wider intervals (grouped PICP90 0.721 → 0.927; MPIW90 0.234 → 0.617 kg/m³). The four positions showed a weak within-vertical effect (Friedman χ² = 11.93, p = 0.0076; Kendall’s W = 0.159), with no pairwise contrast significant after Holm correction. The results support calibrated event-internal interpolation at sampled layers, but do not resolve a monotonic vertical profile or support transfer across dates, stages, stations, or events without recalibration.