Short-term photovoltaic (PV) power forecasting is pivotal for grid stability and high renewable-energy integration, yet existing hybrid deep-learning models face three unresolved challenges: they fail to balance accuracy, computational efficiency, and interpretability; cannot mitigate iTransformer’s inherent weakness in local feature capture (critical for transient events like minute-level cloud shading); and rely on linear concatenation that mismatches the nonlinear correlations between global multivariate trends and local fluctuations in PV sequences. To address these gaps, this study proposes a novel lightweight hybrid framework—DSC-Chebyshev KAN-iTransformer—for 15-min short-term PV power forecasting. The core novelty lies in the synergistic integration of Depthwise Separable Convolution (DSC) for low-redundancy local temporal pattern extraction, Chebyshev Kolmogorov–Arnold Network (Chebyshev KAN) for adaptive nonlinear fusion and global nonlinear modeling, and iTransformer for efficient capture of cross-variable global dependencies. This design not only compensates for iTransformer’s local feature deficiency but also resolves the linear fusion mismatch issue of traditional hybrid models. Experimental results on real-world PV datasets demonstrate that the proposed model achieves an $R^2$ of 0.996, with root mean square error (RMSE) and mean absolute error (MAE) reduced by 19.6–62.1% compared to state-of-the-art baselines (including iTransformer, BiLSTM, and DSC-CBAM-BiLSTM), while maintaining lightweight characteristics (2.04M parameters, 3.90 GFLOPs) for urban edge deployment. Moreover, Chebyshev polynomial weight visualization enables quantitative interpretation of variable contributions (e.g., solar irradiance dominates via low-order polynomials), enhancing model transparency for engineering applications. This research provides a lightweight, accurate, and interpretable forecasting solution, offering policymakers a data-driven tool to optimize urban PV-infrastructure integration and improve grid resilience amid the global energy transition.

Significant Wave Height (SWH) is a critical parameter for ocean observation. SWH retrieval using GNSS-R data faces challenges including difficult feature selection, insufficient temporal dependency modeling, and limitations due to single-modality data. This paper proposes a novel Multimodal Transformer Network (MultTransNet) to enhance the accuracy of GNSS-R SWH retrieval. To optimize the feature set, we designed an XGBoost-based iterative feature selection module that effectively eliminates redundant features. To capture complex temporal dependencies and global context, the model employs a Transformer encoder utilizing its self-attention mechanism. Furthermore, to overcome the constraints of single-modality data, we innovatively fused 2D DDM image data with 1D auxiliary parameters, enabling multi-source information integration. Simulation results show that the Transformer architecture reduces Root Mean Square Error (RMSE) by 8.91% and increases Correlation Coefficient (CC) by 4.05% compared to a conventional Deep Neural Network (DNN) model. More significantly, the proposed multimodal algorithm further improves retrieval accuracy by 27.05% (RMSE reduction) and 7.21% (CC increase) compared to its single-modality Transformer counterpart, demonstrating superior performance, especially in complex sea-state conditions.