The study, published in the Journal of Cleaner Production, examined how artificial intelligence can predict the amount of solar electricity that will be available roughly one day in advance. Such forecasts are essential for utilities and grid operators, which must schedule power plants, manage energy storage and prepare for fluctuations in electricity demand. As solar energy becomes a larger part of the energy mix, even modest forecasting errors can create operational challenges and increase the need for backup generation.

“Solar power generation has expanded rapidly because it is both renewable and widely available,” said Yen-Hsi Chou, a postdoctoral research scholar at NC State and the study’s corresponding author. “But increasing use of solar can also make forecasting supply and demand more challenging because the availability of sunlight isn’t always consistent.” The researchers focused on the relationship between weather conditions and actual solar power production, using historical observations to train and evaluate a series of predictive models.

The team analyzed operational data collected between January 2019 and December 2022 from two California utilities: the Imperial Irrigation District, or IID, and the Los Angeles Department of Water and Power, known as LADWP. The dataset contained more than 20,000 hours of solar generation and weather information. By studying two geographically and operationally distinct regions, the researchers were able to test whether a forecasting strategy that worked well in one location would also perform reliably elsewhere.

The researchers initially compared seven models drawn from two broad categories: statistical forecasting methods and artificial neural networks. Statistical models are designed to identify recurring patterns in historical data, while neural networks can capture complex, nonlinear relationships between variables over time. This distinction is particularly important for solar forecasting, because the effect of weather conditions on electricity generation is rarely simple. A small change in cloud cover, temperature or atmospheric conditions can produce a disproportionate change in power output.

Among the individual models, a bidirectional long short-term memory network, or BiLSTM, delivered the most consistently accurate results. LSTM networks are a type of recurrent neural network designed to process sequences, making them useful for time-dependent problems such as weather and energy forecasting. A BiLSTM examines information in both forward and backward directions within a sequence, allowing it to identify patterns that may depend on relationships across different time steps. The researchers selected this model as the baseline against which their combined approaches were measured.

The first ensemble strategy used weighted averaging. In this approach, forecasts generated by separately trained, location-specific models were combined, but stronger-performing models received greater influence in the final prediction. This method resembles a panel of experts in which more reliable forecasters are given greater weight. For the IID case, weighted averaging produced the strongest results, improving forecast performance by up to approximately 11% during favorable seasons compared with the BiLSTM baseline.

The second strategy, called a multi-input ensemble, supplied individual models with weather information from multiple locations. Rather than relying only on conditions observed near a particular solar generation area, the system could use broader regional information to improve its understanding of incoming weather patterns. This approach was especially effective for LADWP, where it produced improvements of up to about 13%. The result suggests that meteorological information from surrounding areas may help machine-learning systems anticipate changes that have not yet reached the generation site.

The regional contrast was one of the study’s most important findings. Neither ensemble method performed best everywhere, and no single model consistently dominated across all seasons and locations. “There is no universal forecasting strategy that will perform equally well everywhere,” said Anderson De Queiroz, an associate professor at NC State and co-author of the paper. He added that regional characteristics and the availability of meteorological data must be considered when designing tools for real-world grid operations. A model optimized for one utility may therefore require substantial adjustment before it can be deployed elsewhere.

The findings do not suggest that artificial intelligence can eliminate uncertainty from solar generation, but they show that combining models can make forecasts more robust. Better day-ahead predictions could help utilities decide when to charge batteries, schedule conventional generators and coordinate electricity purchases. The researchers emphasized that ensemble methods must be tested and fine-tuned for the specific region in which they will be used. As solar penetration continues to rise, geographically aware forecasting systems could become an important part of keeping power systems reliable while accommodating more renewable energy.

Subject of Research: Not applicable

Article Title: “A Hybrid Machine Learning Framework for Enhanced Day-Ahead Solar Forecasting in Large Scale Systems”

Web References: https://doi.org/10.1016/j.jclepro.2026.148980; Journal of Cleaner Production article

References: Chou, Yen-Hsi; Haldar, Arundhuti; Nisar, Shubh; de Queiroz, Anderson Rodrigo. “A Hybrid Machine Learning Framework for Enhanced Day-Ahead Solar Forecasting in Large Scale Systems.” Journal of Cleaner Production, published July 29, 2026. DOI: 10.1016/j.jclepro.2026.148980

Keywords

Solar forecasting, machine learning, artificial neural networks, BiLSTM, renewable energy, solar power, ensemble models, weather data, smart grids, energy systems

The core challenge in solar radiation forecasting lies in the inherent complexity and variability of solar irradiance. Solar energy availability is influenced by multifaceted factors, including atmospheric conditions, geographic location, temporal fluctuations, and meteorological phenomena. Traditional forecasting models often depend either on purely statistical data-driven methods or deterministic physical models, but each has limitations. Purely statistical models may lack physical interpretability, while deterministic models frequently fail to capture transient and stochastic variations adequately. The synthesis of multivariate signal decomposition with physics-informed time-frequency analysis addresses these challenges by leveraging the strengths of both domains.

Multivariate signal decomposition is a powerful tool in signal processing that enables the breakdown of complex, intertwined data streams into constituent components that reveal hidden patterns and correlations. In the context of solar radiation, this approach facilitates the separation of various influences such as cloud cover fluctuations, aerosol interference, and atmospheric water vapor effects, which often baffle conventional models due to their overlapping temporal and spectral characteristics. By decomposing solar irradiance data into meaningful subcomponents, researchers can better scrutinize and model each pattern independently, leading to more refined and precise predictions.

The physics-informed time-frequency feature extraction method enriches this framework by integrating domain-specific knowledge about the physical mechanisms behind solar radiation variability. This approach systematically captures relevant dynamic features across different time scales and frequency bands, reflecting natural oscillations and transient phenomena in solar irradiance. Such integration ensures that the predictive model respects fundamental physical laws, avoids overfitting to noisy data, and maintains robust generalizability across diverse environmental settings and seasonal cycles.

This multifaceted analytical pipeline was validated using expansive datasets collected from diverse climatic regions, encompassing multiple years of high-resolution solar radiation measurements. By applying this dual-layered decomposition and feature extraction strategy, the predictive model delivered remarkable improvements in both short-term and long-term solar irradiation forecasts. Notably, it significantly outperformed traditional models in accurately anticipating sudden fluctuations caused by cloud movement and atmospheric disturbances, a critical factor for the efficiency of photovoltaic power systems.

The implications of these advancements extend far beyond mere academic interest. Accurate solar radiation forecasting is vital for grid operators, energy traders, and policymakers. Grids integrating a high proportion of solar energy require precise irradiance forecasts to manage supply-demand balances, mitigate risks of blackouts, and optimize storage solutions. Moreover, real-time solar predictions enable better scheduling of backup power sources and reduce dependency on carbon-intensive peaking power plants, thereby contributing to the global fight against climate change.

This research further exemplifies a paradigm shift towards hybrid modeling frameworks, where physical insights and statistical rigor coalesce to yield robust and interpretable predictions. The interdisciplinary nature of this work, straddling applied physics, signal processing, and machine learning, underscores the increasing necessity of collaborative innovation to tackle complex environmental challenges. It also sets a precedent for future investigations into other renewable energy resources, such as wind and tidal power, where similarly intricate natural signals govern resource availability.

Attention to detail in the model’s design was paramount. The researchers employed an advanced algorithmic pipeline that systematically cleansed raw irradiance data, eliminating noise stemming from sensor malfunctions or extreme weather events. They crafted decomposition channels sensitive to both slow seasonal trends and rapid diurnal variations, thus capturing the full spectrum of solar radiation dynamics. By coupling these channels with physics-informed feature maps, the model assimilated critical physical variables such as solar zenith angle, atmospheric turbidity, and humidity levels, which profoundly influence irradiance patterns.

A particularly novel aspect of this study was the integration of time-frequency domain analysis, which transcends traditional time series evaluation by enabling simultaneous examination of temporal evolution and spectral characteristics. This method uncovers transient phenomena—like passing clouds or sudden haze—that manifest as ephemeral spikes or oscillations in solar radiation. Capturing these signatures allows the prediction system to adjust dynamically, enhancing its responsiveness and reducing forecast errors that conventional models typically overlook.

Beyond methodological innovation, the research team rigorously evaluated their framework against established benchmarks using metrics including mean absolute error (MAE), root mean square error (RMSE), and correlation coefficients. Their model consistently exhibited superior performance across these quantitative criteria, reporting error reductions upwards of 25% relative to state-of-the-art statistical and physical forecasting models. Such improvements have tangible economic benefits for utilities by improving scheduling accuracy and reducing costs associated with energy imbalances.

Another consequential outcome highlighted in the analysis was the model’s adaptability across geographic regions featuring contrasting climates. The researchers demonstrated that their physics-informed decomposition strategy maintained high predictive accuracy whether applied to temperate urban environments, arid desert locales, or tropical rainforests. This generalizability speaks to the robustness of integrating domain knowledge with advanced signal processing, opening the door for widespread adoption across international solar energy projects.

The broader scientific community has received this research with enthusiasm because it bridges a persistent gap between theoretical understanding and practical application. As solar energy continues to expand its share within the global energy portfolio, advancements in forecasting accuracy directly translate to enhanced grid stability and optimized resource utilization. Furthermore, the model’s transparency and interpretability facilitate easier integration into existing energy management systems, which many earlier complex black-box models failed to achieve effectively.

Looking ahead, the research team intends to refine their approach by incorporating additional geophysical parameters and experimenting with real-time adaptive feature extraction to further improve responsiveness. They also advocate for open-source dissemination of their algorithmic pipeline, aiming to catalyze collaborative improvement and widespread implementation. The vision is a future where predictive solar radiation modeling is so precise and reliable that it seamlessly supports fully renewable, decentralized power networks.

Equally important, this development signals the increasing role of data-driven physics-informed methodologies in environmental science beyond solar energy. Similar hybrid approaches could unlock new understanding and prediction capabilities in fields ranging from climate modeling to ecosystem monitoring, where complex, multivariate time series data are ubiquitous yet challenging to decode with conventional techniques.

In conclusion, the fusion of multivariate signal decomposition with physics-informed time-frequency feature extraction constitutes a paradigm leap in solar radiation prediction, offering a more nuanced, accurate, and physically grounded forecasting framework. This advancement not only promises to accelerate the integration of solar power into modern energy systems but also exemplifies the transformative potential of interdisciplinary research that melds fundamental science with advanced computational techniques. As the world races towards a more sustainable energy future, innovations like these will be indispensable in navigating the complexities of natural variability and harnessing renewable resources to their fullest potential.

Subject of Research: Solar radiation prediction using advanced signal processing and physics-informed feature extraction methods.

Article Title: Solar radiation prediction using multivariate signal decomposition and physics-informed time-frequency feature extraction

Article References:
Mo, X., Xin, J., Jiang, Y. et al. Solar radiation prediction using multivariate signal decomposition and physics-informed time-frequency feature extraction. Commun Eng (2026). https://doi.org/10.1038/s44172-026-00677-6

Image Credits: AI Generated