Jul 2026· Engineering Research Express· Vol 8, pp. 145322· 0 citations· 48 references
Physics
TL;DR
The spatio-temporal fusion network (STFNet) is proposed, a explainable hybrid model that integrates convolutional neural networks for local feature extraction, unidirectional Long Short-Term Memory networks for temporal modelling, and XGBoost for nonlinear prediction, enhanced by a novel cloud-conditioned bidirectional GHI injection mechanism.
Abstract
As the integration of photovoltaic (PV) systems into modern electrical grids continues to grow, there is an increasing demand for accurate short-term forecasting of both global horizontal irradiance (GHI) and PV power output. However, most existing deep learning models weakly couple GHI and power predictions, degrade under cloudy and partially cloudy conditions, and offer limited interpretability. To address these challenges, this study proposes the spatio-temporal fusion network (STFNet). This explainable hybrid model integrates convolutional neural networks for local feature extraction, unidirectional Long Short-Term Memory networks for temporal modelling, and XGBoost for nonlinear prediction, enhanced by a novel cloud-conditioned bidirectional GHI injection mechanism. The core innovation is a bidirectional gated fusion that couples a dedicated GHI branch and a PV branch in both directions: a GHI→PV blend gate transfers irradiance physics into the power pathway. In contrast, a PV→GHI residual gate allows recent power dynamics to refine the irradiance estimate, with both gates modulated by cloud-type conditioning so that the coupling adapts to prevailing sky conditions. Extensive evaluations on the SHEERM dataset across 15, 30, and 60 min horizons, using mean absolute error, root mean square error (RMSE), and R2 metrics stratified by season and cloud type, demonstrate that STFNet achieves a better overall performance among the eleven models compared, with an R2 of 0.9413 for GHI (RMSE 55.06 W m−2) while maintaining competitive PV performance (R2 0.9069, RMSE 58.49 W m−2). A controlled ablation isolates the bidirectional coupling as the sole driver of the GHI improvement, which is large (∼17% RMSE reduction over the unidirectional gate) and highly reproducible across random seeds; the improvement is consistent across all seasons, horizons, and cloud buckets. LIME and SHAP analyses further reveal that lagged PV power, clear-sky GHI, precipitable water content, and sinusoidal time encodings are the most influential predictors, enhancing both accuracy and physical interpretability for reliable solar forecasting and grid integration.
Accurate short-term solar irradiance forecasting is a critical enabler for Photovoltaic (PV) power integration, and energy management in renewable-dominated power systems. Nevertheless, solar irradiance exhibits strong nonlinearity, and temporal variability, which significantly limit the performance of conventional models. To address these challenges among others, in this study we propose a Dual Reservoir-based Convolutional Neural Network Integrated with Long Short-Term Memory (DR–CNN–LSTM) architecture for high-resolution solar irradiance prediction. The proposed framework introduces two parallel reservoir modules to enrich nonlinear state representations and enhance short- and mid-term temporal memory. The proposed model is validated using multi-year solar irradiance data from multiple stations of the California Irrigation Management Information System (CIMIS) and the Solar Radiation Research Laboratory (SRRL) dataset. Extensive experimentation was conducted across multiple forecasting horizons and benchmarked against state-of-the-art models, including LSTM, CNN–LSTM, CNN–DeepESN and single-reservoir hybrid architectures. Forecasting performance is evaluated using MAE, RMSE, MAPE, and the coefficient of determination ([Formula: see text]). Compared with the CNN–LSTM baseline, the proposed framework achieved average RMSE reductions of 10.7% on CIMIS and 11.5% on SRRL, while MAE was reduced by 14.0% and 15.1%, respectively. Furthermore, compared with the state-of-the-art CNN–DeepESN model, the proposed model achieved RMSE reductions of approximately 5.2% on CIMIS and 5.5% on SRRL. Statistical analyses using Friedman, Nemenyi, and Wilcoxon signed-rank tests confirmed significant improvements over most baseline models. Although the proposed dual-reservoir architecture consistently achieved the best overall forecasting performance, the improvement over the Single Reservoir–CNN–LSTM variant was not statistically significant at [Formula: see text], indicating that the second reservoir provides a moderate but consistent enhancement rather than a statistically dominant gain.
Ibrahem Mohammed Madhat, Omar Alhebashi, Ahmed Azzawi et al.· International Journal of Com...· 0 citations
As the global energy mix shifts toward cleaner sources, the large-scale grid integration of photovoltaic (PV) power poses severe challenges to microgrid frequency stability and security. From a fundamental physical perspective, the solar radiation driving photovoltaic conversion consists of electromagnetic waves on the micrometer scale; as these waves traverse the atmosphere, they undergo intense Rayleigh and Mie scattering caused by cloud dynamics and aerosol attenuation. This atmospheric degradation results in highly nonlinear, transient fluctuations in the effective power reaching the ground. Consequently, computationally intensive full-wave simulation models are impractical for real-time dispatch, while existing purely data-driven deep learning algorithms—lacking physical interpretability—are prone to overfitting and prediction failure under non-stationary meteorological conditions. To bridge this gap between physics and algorithms, this study proposes a novel Physics-Informed Temporal Convolutional Network (PI-TCN) architecture. The framework utilizes Global Horizontal Irradiance (GHI) and Diffuse Horizontal Irradiance (DHI) as inputs to implicitly reconstruct the electromagnetic wave's energy attenuation trajectory, employing 1D causal dilated convolutions to eliminate temporal lag. Furthermore, the model innovatively incorporates non-negative electromagnetic energy boundaries and first-order wave derivatives as penalty functionals during backpropagation, thereby constraining model weights to converge within a physically feasible domain. Benchmarking against a three-year high-resolution dataset from the Desert Knowledge Australia Solar Centre (DKASC) demonstrates that the PI-TCN achieves an exceptionally high coefficient of determination (R2) of 0.9524 and an inference latency of merely 0.08 milliseconds; notably, it attains a Matthews Correlation Coefficient (MCC) of 0.8412 in capturing extreme ramp events. By utilizing the Jacobian matrix of partial derivatives to fully deconstruct the network's "black-box" nature, this research establishes a robust and highly interpretable new paradigm for the convergence of computational electromagnetics and artificial intelligence.
F. Liu, C. H. Zhao, Z. P. Xie· Advanced Electromagnetics· 0 citations
With the global shift toward green energy, solar photovoltaic (PV) power has expanded rapidly. However, the unpredictable nature of PV generation, caused by changing weather conditions like irradiance and temperature, challenges grid stability and power scheduling. Therefore, developing smart forecasting models for high-precision PV power prediction is essential for modern grid management. This paper introduces an optimized forecasting framework using Long Short-Term Memory (LSTM) networks. By integrating historical power generation data with localized meteorological factors, a multivariate predictive model was developed and validated using empirical data from a 100 MW PV plant. Based on Pearson correlation analysis, four key features—temperature, direct normal irradiance (DNI), relative humidity, and cloud cover—were chosen as the main drivers of PV output. A multivariate LSTM model was then trained and carefully tested using time series cross-validation. Results show the multi-feature architecture consistently outperforms and surpasses single-feature benchmarks. Specifically, the model achieved a peak R2 of 0.9864 and minimum MAE of 1.4057 kW. Across four validation sets, R2 remained stable (0.9755–0.9836), with most errors tightly bounded within ±5 kW. These findings confirm the model’s excellent accuracy and its value for improving power system dispatch and resource planning.
Accurate short-term wind power forecasting (STWPF) is critical to maintaining grid stability and improving renewable integration. Physical, statistical, and deep-learning methods are widely used and have produced promising results. However, the above methods often struggle to balance prediction accuracy with computational efficiency and to capture spatiotemporal dependencies while keeping hyperparameter tuning manageable. To address these limitations, we propose a novel cloud–edge collaborative intelligence framework which enables synergy between large and small models for STWPF. The framework consists of two components: a LightGBM module for efficient feature selection, and a cloud module with a temporal 1D Convolutional Neural Network (1D-CNN) for local temporal pattern extraction and cross-channel interactions, followed by a Bidirectional Long Short-Term Memory (BiLSTM) for bidirectional temporal modeling. Moreover, we use a resource aware tuning strategy that speeds up tuning without loss of accuracy. Through extensive experiments on real-world datasets, our method outperforms state-of-the-art baselines, highlighting the practical value of our framework.
Zhiqiang Jiang, Changfu You, Dong Ma et al.· Journal of Cloud Computing· 0 citations
Accurate multi-site photovoltaic (PV) power forecasting plays a pivotal role in enhancing smart grid operation and facilitating large-scale renewable energy integration. However, existing forecasting approaches often struggle to simultaneously capture the dynamic spatial interactions among geographically distributed PV plants and the diverse temporal dependencies embedded in PV power generation data. To overcome these limitations, this paper proposes a Bidirectional Multi-Scale Dynamic Graph Convolutional Network (BiMSDGCN). The proposed framework integrates two key components. First, a Dynamic Graph Convolutional Network (DGCN) is designed to adaptively learn a time-dependent adjacency matrix to capture evolving spatial correlations among PV sites. Second, a Bidirectional Multi-Scale Attention (BiMS) is developed to model both cross-scale and intra-scale dependencies, enabling more comprehensive spatiotemporal feature extraction and information fusion. Experimental results on various publicly available PV power datasets show that BiMS-DGCN consistently outperforms several state-of-the-art forecasting methods. Ablation studies further verify that both DGCN and BiMS play critical roles in modeling complex spatiotemporal dependencies, thereby improving the accuracy of multi-site photovoltaic power forecasting.
Liu Xinqi, Lilin Peng, Linong Du et al.· International journal of pat...· 0 citations
Accurate photovoltaic (PV) power forecasting is essential for reliable microgrid operation, efficient energy dispatch, and improved utilization of renewable energy resources. Existing forecasting methods often have limited capacity to represent the nonlinear relationships between meteorological conditions and PV power output. They also tend to underrepresent the temporal dynamics of PV generation and the physical principles governing photovoltaic energy conversion. To address these limitations, this study proposes a hybrid forecasting framework, CNN-LSTM-PINNs, that integrates Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Physics-Informed Neural Networks (PINNs). In the proposed framework, CNNs extract spatial dependencies among multivariate meteorological variables, LSTM networks capture temporal dependencies in PV generation, and PINNs incorporate soft physical constraints derived from photovoltaic energy conversion mechanisms. The proposed model is evaluated using publicly available datasets from three large-scale PV power stations in China, with observations recorded at 15-min intervals. The empirical results show that CNN-LSTM-PINNs outperform the conventional CNN-LSTM benchmark across the primary station-level datasets. Relative to the benchmark model, the proposed framework reduces Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) and improves the coefficient of determination (R2). These results indicate that embedding physical constraints into data-driven forecasting models can improve PV power prediction accuracy. The model also shows stronger robustness and generalization performance under heterogeneous operating conditions, although its effectiveness is contingent on relatively stable data distributions. Feature-importance analysis further indicates that global horizontal irradiance (GHI) and irradiance-derived variables are the most informative predictors of PV power output. Overall, this study provides a physics-informed hybrid modeling approach for high-resolution PV power forecasting in microgrid applications.
Jiabo Gou, Xiaoqiao Liao, Sheng Li et al.· Energies· 0 citations
