Gokhan Mert Yagli

Gokhan Mert Yagli

Applied scientist specializing in solar resource assessment and forecasting.

Publications

You can also find my articles on my Google Scholar profile.

Journal Articles

The value of solar forecasts and the cost of their errors: A review. Permalink

Published in Renewable and Sustainable Energy Reviews, 2024

[AI-generated abstract summary] This work reviews the value of solar forecasts and the impact of forecast errors on real applications, such as electricity market bidding, power system operations, and household bill reduction. It finds that while solar forecasts are valuable, their errors have significant costs, especially at the transmission level. The study provides recommendations to minimize solar uncertainty, aiding better integration of photovoltaic (PV) systems and variable generation, which can be adapted by different market operators and regulators. Read more

Non-crossing quantile regression neural network as a calibration tool for ensemble weather forecasts Permalink

Published in Advances in Atmospheric Sciences, 2024

[AI-generated abstract summary] The study tackles under-dispersion and quantile crossing in ensemble numerical weather prediction (NWP) by introducing a non-crossing quantile regression neural network (NCQRNN). This method ensures reliable forecasts without crossing by using a special hidden layer. In a solar irradiance case study, NCQRNN outperformed other models, offering sharp and calibrated forecasts. Its simple design can be broadly applied to various neural networks. Read more

Predictability and forecast skill of solar irradiance over the contiguous United States. Permalink

Published in Renewable and Sustainable Energy Reviews, 2023

[AI-generated abstract summary] This study highlights that current solar forecast verification focuses too much on comparing methods rather than assessing how well the best method performs relative to the best possible outcome. It addresses the gap in understanding predictability and forecast skill of solar irradiance, especially in spatial distribution. By quantifying and mapping these aspects across the U.S., the study clarifies misconceptions about irradiance predictability and revisits the original formulation of the skill score. Read more

Sub-minute probabilistic solar forecasting for real-time stochastic simulations. Permalink

Published in Renewable and Sustainable Energy Reviews, 2022

[AI-generated abstract summary] This paper addresses the limitations of current stochastic simulations for solar energy systems, which often fail to reflect high-frequency fluctuations and changing uncertainties in solar power output. It introduces a cutting-edge probabilistic solar forecasting method that enhances real-time simulations. The method combines lasso-penalized quantile regression with an analog-based preselection algorithm, forecasting irradiance over small areas at sub-minute timescales. By using online training, the model updates its parameters and forecasts every few seconds. Despite its computational demands, it completes each forecasting cycle in under one second. Compared to five benchmarking methods, including analog ensemble (AnEn), the proposed method significantly improves forecast accuracy, achieving up to a 55% skill score. The R code and datasets are publicly available on Github to encourage future adoption. Read more

Ensemble Solar Forecasting and Post-Processing with Neighboring Satellite Pixels. Permalink

Published in Renewable and Sustainable Energy Reviews, 2022

[AI-generated abstract summary] This article addresses the issue of under-dispersed and biased ensemble weather forecasts by generating and post-processing ensemble solar forecasts using spatio-temporal satellite-derived irradiance data. The approach utilizes a dropout neural network with Monte Carlo sampling for ensemble forecasting and employs both parametric and nonparametric post-processing techniques. The framework is validated using four years of data from seven U.S. locations, achieving a 66% improvement in forecast skill compared to a conditional climatology reference. The findings are relevant for power system stakeholders, such as system operators, PV plant owners, and forecast retailers, who benefit from higher quality solar forecasts. Read more

On predictability of solar irradiance. Permalink

Published in Journal of Renewable and Sustainable Energy, 2021

[AI-generated abstract summary] The article critiques the common practice in solar forecasting literature where new methods are claimed to outperform benchmarks without sufficient evidence, casting doubt on actual progress in the field. To assess real advancement, it calls for formal verification and an inquiry into predictability, a complex and debated concept in statistics. The paper suggests using the performance of short-range forecasting methods compared to optimal long-range methods as a proxy for predictability. High predictability would result in significant improvements over climatological forecasts, while low predictability would show minimal gains. The article also introduces a new measure of predictability for solar irradiance. Read more

Operational solar forecasting for grid integration Standards, challenges, and outlook. Permalink

Published in Solar Energy, 2021

[AI-generated abstract summary] This work explores the critical interactions between solar forecasting strategies and grid codes, which significantly influence grid integration. It highlights four overlooked technical aspects that need addressing to develop effective grid-integration standards, such as forecast submission requirements and penalty schemes: (1) assessing forecast accuracy, (2) quantifying predictability, (3) forecast downscaling, and (4) hierarchical forecasting. These challenges are discussed alongside a case study based on industry standards from the National Energy Administration of China, emphasizing the need to balance the interests of both photovoltaic power plant owners and system operators. Read more

Novel forecast-based dispatch strategy optimization for PV hybrid systems in real time. Permalink

Published in Energy, 2021

[AI-generated abstract summary] This paper presents a new method for optimizing the real-time scheduling of off-grid systems composed of solar panels, batteries, and diesel generators. By using forecasted load and irradiance values, the approach determines the optimal power generation and energy source operation to minimize costs while meeting system constraints. A real-time simulator is used for accurate validation of results and analysis of grid quality parameters. Compared to two benchmark algorithms, the proposed method achieves 5% cost savings while maintaining grid quality, making it suitable for enhancing the performance of real off-grid systems. Read more

Reconciling solar forecasts: Probabilistic forecasting with homoscedastic Gaussian errors on a geographical hierarchy. Permalink

Published in Solar Energy, 2020

[AI-generated abstract summary] This paper explores hierarchical forecasting and reconciliation in solar engineering, focusing on their effects on probabilistic forecasting. Building on previous works, it examines how reconciliation impacts parametric predictive distributions for solar forecasts. Using two datasets—distributed solar power in California and satellite-derived irradiance in Arizona—four minimum-trace-based reconciliation techniques are applied to day-ahead and hour-ahead forecasts. The results show that reconciliation not only enhances the accuracy of point forecasts but also improves the quality of predictive distributions in terms of sharpness, calibration, and skill score. These improvements are general and independent of data, hierarchy structure, or forecasting model. Read more

Ensemble solar forecasting using data-driven models with probabilistic post-processing through GAMLSS. Permalink

Published in Solar Energy, 2020

[AI-generated abstract summary] This study addresses the challenge of forecast performance variability in data-driven models due to local weather conditions, proposing ensemble forecasting to combine multiple models for better accuracy. However, raw ensemble forecasts often suffer from underdispersion, which affects calibration. To improve forecast accuracy, the study calibrates hourly ensemble clear-sky index forecasts from 20 models using parametric and nonparametric post-processing techniques. Using four years of data from seven sites, the results show that post-processed forecasts significantly outperform raw forecasts. The proposed parametric technique, generalized additive models for location, scale, and shape, notably improves forecast accuracy, reducing the continuous ranked probability score (CRPS) by 38–58% compared to a climatology reference. Read more

Solar irradiance resource and forecasting: a comprehensive review. Permalink

Published in IET Renewable Power Generation, 2020

[AI-generated abstract summary] This study reviews the increasing integration of photovoltaic (PV) energy into power grids and the need for accurate solar irradiance forecasting to ensure reliable power supply and grid stability. It examines various forecasting methods, including numerical weather prediction, satellite-based, cloud-image, data-driven, and sensor-network approaches. The study also provides an extensive review of sensor networks used to determine solar irradiance and discusses available error metrics and datasets for validating these forecasts across different time horizons. Accurate forecasting is essential for large-scale PV plant deployment and efficient management of power demand and supply. Read more

Can we justify producing univariate machine-learning forecasts with satellite-derived solar irradiance? Permalink

Published in Applied Energy, 2020

[AI-generated abstract summary] This study evaluates the quality of solar forecasts generated using ground-based and bias-corrected satellite-derived irradiance data, focusing on error decomposition. Five machine-learning models generate hourly forecasts using data from 15 research stations in Europe, South America, and Africa, as well as satellite-derived data for the same locations. Ensemble forecasts are also assessed. Instead of using traditional accuracy measures, the study employs the Murphy–Winkler forecast verification framework to analyze calibration, conditional bias, and discrimination. Results show that forecasts based on bias-corrected satellite-derived data are of comparable quality to those using ground-based data. Read more

Automatic hourly solar forecasting using machine learning models. Permalink

Published in Renewable and Sustainable Energy Reviews, 2019

[AI-generated abstract summary] This article evaluates the performance of 68 machine learning algorithms for hourly solar forecasting across different sky conditions, locations, and climate zones in the U.S. Despite the prevalence of machine learning in solar forecasting, the notion of a universally superior model is a myth, as models can only be judged accurately after empirical evaluation. The study avoids hybrid models and uses off-the-shelf algorithms with automatic tuning for fairness. Tree-based methods consistently perform well in long-term results, but no single model excels daily across all conditions. The article suggests preferred models for specific sky and climate conditions. Read more

Reconciling solar forecasts: Sequential reconciliation. Permalink

Published in Solar Energy, 2019

[AI-generated abstract summary] This article extends previous works on forecast reconciliation for photovoltaic (PV) power generation by addressing both geographical and temporal hierarchies simultaneously. Building on earlier studies that handled these hierarchies separately, this work applies sequential reconciliation for day-ahead forecasting of 318 PV systems in California. The results demonstrate that using sequential reconciliation ensures consistent forecasts across both geographical and temporal hierarchies, and also improves forecast accuracy compared to single-hierarchy approaches. Read more