Evaluation of Photovoltaic Generation Forecasting Using Model Output Statistics and Machine Learning

The growth in the adoption of solar photovoltaic (PV) power generation systems has been accelerating around the world, contributing to the debate about the future of policy and regulation in a high distributed energy resources future. Thailand is one of the leaders in solar investment in Southeast Asia. It has recently shifted its policy framework from subsidizing power export through feed-in tariffs toward a policy that is focused on supporting self-consumption of PV electricity. There are three possible forms of self-consumption support scheme: net metering, net billing, and self-consumption only. Net metering and net billing are electricity policies to assign compensation to excess electricity generated from the prosumers? sources, particular in rooftop solar PV system. The design elements on new self-consumption scheme from perspectives of various stakeholders together with the assessment of economic feasibility of rooftop PV system under self-consumption schemes can help to ensure successful implementation of the policy. It is important to analyze the possible impacts of rooftop PV system on distribution network system. This dissertation consists of three main components: economic feasibility analysis, technical analysis, and stakeholders? perspectives analysis. First, this study assesses the economic feasibility of residential and commercial sectors on rooftop solar PV systems under Thailand?s self-consumption scheme, net metering with rolling credit and buyback scheme, and net billing with real-time buyback scheme. These three schemes are compared using 3 indicators: Net Present Values (NPV), Payback Period (PB), and Internal Rate of Return (IRR) to assess the feasibility of rooftop solar PV investment for consumers. To supplement the understanding of selected schemes, this research assesses the technical factors that support or hinder the implementation of solar PV rooftop by reviewing Thailand?s grid code and meter requirement of rooftop solar PV installation. Third, this study investigates the perspectives of stakeholders on the detailed design options of self-consumption schemes for supporting rooftop solar PV system installation. When combined the outcomes of stakeholder?s perspectives, the result also shows that most of stakeholder groups prefer a strong desire to compensate for excess generation from rooftop PV system in the form of net metering. This finding corresponds to the economic feasibility analysis, which indicate that net metering with rolling credit and buyback is economically feasible for both residential and commercial sectors. Finally, this dissertation suggests policy recommendations for each group of stakeholders in order to increase their understanding on how to contribute to a sustainable scheme to scale up rooftop solar PV installation for both residential and commercial customers in Thailand.

Exploring solar and wind energy resources in North Korea with COMS MI geostationary satellite data coupled with numerical weather prediction reanalysis variables

Over the past few decades, there has been a continuous increase in the public interest for solar energy as an alternative and cleaner source of energy. Therefore, it is not surprising that there is a similar interest in developing accurate models to forecast solar photovoltaic (PV) power production. Such models vary depending on the geographical location of PV sites and the seasons considered. The Philippines has yet to have a solar PV output forecasting model adapted to the country's local conditions. This study aims to evaluate the Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX) forecasting model as a tool for forecasting solar PV generation based on the seasonal characteristics of the country and identify which input parameters are significant for each season. This work used solar PV production data as an endogenous variable. Meanwhile, exogenous variables include in-situ solar irradiance data from solar power plants; and cloud cover, wind speed and direction, ambient temperature, precipitation, and relative humidity, which we extracted from ERA5 Reanalysis data. Datasets were divided based on the seasons as defined by the Philippine Atmospheric, Geophysical, and Astronomical Services Administration (PAGASA), namely, hot dry (HD), rainy (R), and cold dry (CD) seasons. Then, we performed a forecast on each season and one full year to assess the performance of the SARIMAX model. The best model per season was done based on the forecasting accuracy measured using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). This work analyzed seasonal and year-round SARIMAX models for Baguio City, Philippines. Results show that the cold dry season got the highest accuracy value in terms of 2.26 MAE and 4.06 RMSE. Meanwhile, the rainy season had the lowest accuracy of 12.91 MAE and 16.16 RMSE. We can infer from our findings that seasonal forecasting is better during hot dry and cold dry seasons. We also found that the year-round forecasting model performs better than the rainy season model. From the significant parameters identified in our best models, our analysis showed that wind direction can be removed from all models; irradiation and relative humidity were significant for all seasons.

Short-term wind speed forecasting bias correction in the Hangzhou area of China based on a machine learning model

The golden rule of creating a machine learning algorithm is to handle as much data as possible under consistent conditions. In the field of air quality, these models can be used for short-term air quality forecasting, helping predict high pollution peaks and alerting residents in cases where pollutant concentrations exceed legal or World Health Organization (WHO) limits.For daily air quality predictions and short-term forecasts, the Slovenian Environment Agency uses the CAMx dispersion model, which is frequently upgraded with new input data. Based on the model results and expert opinions, we inform the public about air quality levels and issue alerts when concentrations reach the limit values. The Copernicus Atmosphere Monitoring Service (CAMS) has developed Model Output Statistics (MOS), which downscale air quality forecasts produced by regional ensemble models and incorporate measurement values from observation sites across Europe.In our study, we evaluated and compared CAMS MOS forecasts with forecasts from the Slovenian operational CAMx model and raw CAMS ensemble forecasts at Slovenian observational sites. The evaluation was conducted using time series plots, scatter plots, and Taylor diagrams, alongside statistical metrics such as the fraction of forecasts within a factor of two (FAC2), mean bias (MB), mean gross error (MGE), normalized mean bias (NMB), normalized gross error (NMGE), root mean square error (RMSE), correlation coefficient (r), coefficient of efficiency (COE), and index of agreement (IOA).Our findings indicate that CAMS MOS forecasts are almost always more accurate than raw CAMS or CAMx model predictions, particularly for the MOS day-1 forecast. However, for PM10 and PM2.5, longer-term MOS forecasts (day-2, day-3, and day-4) were less accurate than RAW and/or CAMx predictions based on certain statistical measures. When analysing individual stations, we observed occasional instances where MOS forecasts were less accurate for specific pollutants.CAMS MOS is a useful tool for improving the national alert system, helping to warn residents about poor air quality conditions and enabling timely protective measures. Consequently, our results suggest that the MOS day-1 forecast can serve as a valuable additional tool for issuing daily PM10 forecasts and, for some stations, even O3 daily maxima predictions. However, all MOS forecasts should be used with an awareness of their limitations—for example, the time lag in predicting high PM10 episodes and the systematic underestimation of O3 daily maxima at many Slovenian stations.

The intermittence and fluctuation of photovoltaic power generation seriously affect output power reliability, efficiency, fault detection of photovoltaic power grid, etc. The precise forecasting of photovoltaic power generation is the critical method to solve the above limitations. Current photovoltaic power generation forecasting methods generally usually adopt meteorological data and historical continuous photovoltaic power generation as inputs, but they do not take into account historical periodic photovoltaic power generation as inputs, which makes the existing methods inadequate in learning time correlation. Therefore, to further study the time correlation for improving the prediction accuracy, an LSTM-FC deep learning algorithm composed of long-term short-term memory (LSTM) and fully connected (FC) layers is proposed. The double-branch input of the model enables it not only to consider the impact of meteorological data on power generation but also to consider time continuity and periodic dependence, thereby improving the prediction accuracy to a certain extent. In this paper, meteorological data, historical continuous data, and historical periodic data are used as experimental data, and these three types of data are combined into different input forms to evaluate and compare LSTM-FC with other baseline models, including support vector machines (SVM), gradient boosting decision tree (GBDT), generalized regression neural network (GRNN), feedforward neural network (FFNN), and LSTM. The simulation results show that the accuracy of the models with meteorological data, continuous data, and periodic data as input is higher than that of other input forms, and the accuracy of LSTM-FC is the highest among these models, and its root mean square error (RMSE) is 11.79% higher than that of SVM.

In this work, we evaluate the performance of photovoltaic (PV) power generation forecast using various hybrid deep-learning algorithms, including long and short term memory (LSTM), LSTM with an Autoencoder ( LSTM-Autoencoder ) and LSTM with an attention mechanism (LSTM-Attention). We show that the LSTM-Attention model is significantly more accurate in predicting the hourly power generation of a PV plant with 162kW capacity than the other two reference counterparts. After 100 epochs training, the model achieves a superior Root Mean Square Error (RMSE) below 0.01, Mean Absolute Error (MAE) below 0.005, the Absolute Deviation (AD) below 0.02, and the Mean Absolute Percantage Error (MAPE) is around 35%. Moreover, since the correlation coefficient is up to 92%, this hybrid model not only can be used in solar power generation prediction in PV plants, but potentially can also be extended to other renewable energy sources such as predicting wind power or tide power generation.

Photovoltaics (PV) output power is highly sensitive to many environmental parameters and the power produced by the PV systems is significantly affected by the harsh environments. The annual PV power density of around 2000 kWh/m2 in the Arabian Peninsula is an exploitable wealth of energy source. These countries plan to increase the contribution of power from renewable energy (RE) over the years. Due to its abundance, the focus of RE is on solar energy. Evaluation and analysis of PV performance in terms of predicting the output PV power with less error demands investigation of the effects of relevant environmental parameters on its performance. In this paper, the authors have studied the effects of the relevant environmental parameters, such as irradiance, relative humidity, ambient temperature, wind speed, PV surface temperature and accumulated dust on the output power of the PV panel. Calibration of several sensors for an in-house built PV system was described. Several multiple regression models and artificial neural network (ANN)-based prediction models were trained and tested to forecast the hourly power output of the PV system. The ANN models with all the features and features selected using correlation feature selection (CFS) and relief feature selection (ReliefF) techniques were found to successfully predict PV output power with Root Mean Square Error (RMSE) of 2.1436, 6.1555, and 5.5351, respectively. Two different bias calculation techniques were used to evaluate the instances of biased prediction, which can be utilized to reduce bias to improve accuracy. The ANN model outperforms other regression models, such as a linear regression model, M5P decision tree and gaussian process regression (GPR) model. This will have a noteworthy contribution in scaling the PV deployment in countries like Qatar and increase the share of PV power in the national power production.

Abstract. In this paper, we evaluate the performance of two global horizontal solar irradiance (GHI) estimates, one derived from Meteosat Second Generation (MSG) and another from the 1-day forecast of the Regional Atmospheric Modeling System (RAMS) mesoscale model. The horizontal resolution of the MSG-GHI is 3 × 5 km2 over Italy, which is the focus area of this study. For this paper, RAMS has the horizontal resolution of 4 km.The performances of the MSG-GHI estimate and RAMS-GHI 1-day forecast are evaluated for 1 year (1 June 2013–31 May 2014) against data of 12 ground-based pyranometers over Italy spanning a range of climatic conditions, i.e. from maritime Mediterranean to Alpine climate.Statistics for hourly GHI and daily integrated GHI are presented for the four seasons and the whole year for all the measurement sites. Different sky conditions are considered in the analysisResults for hourly data show an evident dependence on the sky conditions, with the root mean square error (RMSE) increasing from clear to cloudy conditions. The RMSE is substantially higher for Alpine stations in all the seasons, mainly because of the increase of the cloud coverage for these stations, which is not well represented at the satellite and model resolutions. Considering the yearly statistics computed from hourly data for the RAMS model, the RMSE ranges from 152 W m−2 (31 %) obtained for Cozzo Spadaro, a maritime station, to 287 W m−2 (82 %) for Aosta, an Alpine site. Considering the yearly statistics computed from hourly data for MSG-GHI, the minimum RMSE is for Cozzo Spadaro (71 W m−2, 14 %), while the maximum is for Aosta (181 W m−2, 51 %). The mean bias error (MBE) shows the tendency of RAMS to over-forecast the GHI, while no specific behaviour is found for MSG-GHI.Results for daily integrated GHI show a lower RMSE compared to hourly GHI evaluation for both RAMS-GHI 1-day forecast and MSG-GHI estimate. Considering the yearly evaluation, the RMSE of daily integrated GHI is at least 9 % lower (in percentage units, from 31 to 22 % for RAMS in Cozzo Spadaro) than the RMSE computed for hourly data for each station. A partial compensation of underestimation and overestimation of the GHI contributes to the RMSE reduction. Furthermore, a post-processing technique, namely model output statistics (MOS), is applied to improve the GHI forecast at hourly and daily temporal scales. The application of MOS shows an improvement of RAMS-GHI forecast, which depends on the site considered, while the impact of MOS on MSG-GHI RMSE is small.

Photovoltaic mono and bifacial module/string electrical model parameters identification and validation based on a new differential evolution bee colony optimizer

Forecasts of certain weather elements are improved by linearly relating observed elements to past observations, climatological information, and numerical weather‐prediction‐model output. Model Output Statistics (MOS) is a statistical post‐processing of model output that is capable of forecasting some subgrid scale, synoptically‐forced events and of correcting some systematic, but state dependent, model bias. This tendency of accounting for model error is exploited by feeding MOS forecasts back into the state estimation problem. MOS forecasts and their associated uncertainty are treated as ‘observations’ of the future system state, and a four‐dimensional variational assimilation procedure is employed to improve the original analysis and resulting model forecast. In a simple‐model scenario, it is found that this approach has a small negative impact on the magnitude of forecast errors relative to MOS with a decrease in the number of very good forecasts, but a large positive impact on the variance about the forecast errors: forecast busts are reduced. As a further step, a second round of MOS is performed on the new model forecasts in a manner identical to the original MOS approach. This second application of MOS results in a significant reduction in both the forecast errors and the variance about those errors relative to the first application of MOS, and includes an increase in very good forecasts. Copyright © 2003 Royal Meteorological Society

Research status and application of rooftop photovoltaic Generation Systems

The solar photovoltaic (PV) power forecast is crucial for steady grid operation, scheduling, and grid electricity management. In this work, numerous time series forecast methodologies, including the statistical and artificial intelligence-based methods, are studied and compared fastidiously to forecast PV electricity. Moreover, the impact of different environmental conditions for all of the algorithms is investigated. Hourly solar PV power forecasting is done to confirm the effectiveness of various models. Data used in this paper is of one entire year and is acquired from a 100 MW solar power plant, namely, Quaid-e-Azam Solar Park, Bahawalpur, Pakistan. This paper suggests recurrent neural networks (RNNs) as the best-performing forecasting model for PV power output. Furthermore, the bidirectional long-short-term memory RNN framework delivered high accuracy results in all weather conditions, especially under cloudy weather conditions where root mean square error (RMSE) was found lowest 0.0025, R square stands at 0.99, and coefficient of variation of root mean square error (RMSE) Cv was observed 0.0095%.

Downscaling in Weather Forecasting The first downscaling methods had been invented already in the late 1940s (Klein 1948) and became operational in the early days of numerical weather prediction at the end of the 1950s. Back then, operational numerical weather prediction models were by far too coarse to predict local weather, and furthermore they did not forecast all variables of interest but only a few such as pressure and temperature. At that time a considerable network of observed weather time series was available already. Klein et al. (1959) employed this data to infer statistical relationships between the observed large-scale circulation – for those variables that were simulated by the models – and the observed local-scale weather variables of interest. The statistical model was then applied to downscale the actual numerical forecast of the large-scale circulation to a forecast of the local weather. The key assumption of this approach is that the large-scale predictor has been perfectly forecasted by the numerical model, hence the approach itself has been coined perfect prognosis (PP). After some years, a considerable database of past forecasts had been archived. Analyses of this data revealed that numerical forecasts even of the largescale weather were of course not perfect but showed systematic deviations compared to observations. Yet this database also became key to mitigate this problem: Glahn and Lowry (1972) developed a new approach that – during calibration – did not take the predictors from observations but from the archived numerical forecasts. For a new weather prediction, the inferred statistical link is then applied to the new numerical forecast. As this approach is basically a post-processing of numerical model data, it has been coined model output statistics (MOS). The key advantage of MOS is that it contains by construction a bias correction of the numerical model. Current weather prediction systems employ complex MOS approaches with several predictors that are continually recalibrated to provide the highest predictive skill. In parallel, numerical approaches were developed to improve the resolution and accuracy of forecasts over a target region. The first limited-area model was developed at the US National Meteorological Center (Howcroft 1966, Gerrity and McPherson 1969) and became operational in 1971. This model covered the US, Canada and the Arctic Ocean at a horizontal resolution of 190.5km at 60◦N and was driven at the lateral boundaries with input from a Northern Hemisphere numerical weather prediction model.

Tropical cyclones (TCs) are destructive meteorological phenomena, necessitating accurate predictions of TC track and intensity to reduce risks to human life. This study evaluates three Transformer-based models – vanilla Transformer (Transformer), inverted Transformer (iTransformer), and temporal-variate Transformer (TVFormer) – which are trained, validated, and tested on best track data from 1980 to 2021 from the China Meteorological Administration for TC prediction, integrating temporal, variate, and hybrid token embeddings to analyze temporal and variate correlations. Comparative analysis with four recurrent neural network (RNN) models demonstrates the superiority of the refined Transformer models over RNNs: iTransformer reduces mean absolute error (MAE) and root mean square error (RMSE) by 29.55% and 25.80% (latitude), 50.31% and 46.18% (longitude), 8.71% and 9.98% (pressure), and 8.68% and 9.45% (wind speed), while TVFormer achieves MAE and RMSE reductions of 13.98% and 13.84% (latitude), 39.11% and 38.02% (longitude), 13.69% and 14.02% (pressure), and 12.84% and 12.94% (wind speed) on average. Among Transformer variants, iTransformer excels in track prediction, outperforming Transformer with 21.74% lower MAE and 18.26% lower RMSE for latitude, and 32.73% lower MAE and 24.01% lower RMSE for longitude. TVFormer dominates intensity prediction, reducing pressure errors by 4.42% (MAE) and 3.92% (RMSE) and wind speed errors by 19.21% (MAE) and 14.79% (RMSE) compared to Transformer, while outperforming iTransformer with 4.59% lower MAE and 3.68% lower RMSE for pressure and 3.83% lower MAE and 3.18% lower RMSE for wind speed. Notably, TVFormer also enhances track prediction, with 7.10% reduction in MAE and 7.02% reduction in RMSE for latitude, and 22.84% reduction in MAE and 17.09% reduction in RMSE for longitude compared to Transformer. These results highlight the superiority of iTransformer in track prediction and the efficacy of TVFormer in intensity prediction, thanks to their ability to exploit temporal and variate dependencies, offering potential for TC disaster preparedness systems.