Photovoltaic Generation Forecast Research Articles

A Photovoltaic Prediction Model with Integrated Attention Mechanism

Solar energy has become a promising renewable energy source, offering significant opportunities for photovoltaic (PV) systems. Accurate and reliable PV generation forecasts are crucial for efficient grid integration and optimized system planning. However, the complexity of environmental factors, including seasonal and daily patterns, as well as social behaviors and user habits, presents significant challenges. Traditional prediction models often struggle with capturing the complex nonlinear dynamics in multivariate time series, leading to low prediction accuracy. To address this issue, this paper proposes a new PV power prediction method that considers factors such as light, air pressure, wind direction, and social behavior, assigning different weights to them to accurately extract nonlinear feature relationships. The framework integrates long short-term memory (LSTM) and gated recurrent units (GRU) to capture local time features, while bidirectional LSTM (BiLSTM) and an attention mechanism extract global spatiotemporal relationships, effectively capturing key features related to historical output. This improves the accuracy of multi-step predictions. To verify the feasibility of the method for multivariate time series, we conducted experiments using PV power prediction as a scenario and compared the results with LSTM, CNN, BiLSTM, CNN-LSTM and GRU models. The experimental results show that the proposed method outperforms these models, with a mean absolute error (MAE) of 12.133, root mean square error (RMSE) of 14.234, mean absolute percentage error (MAPE) of 2.1%, and a coefficient of determination (R2) of 0.895. These results indicate the effectiveness and potential of the method in PV prediction tasks.

Interval forecasting strategy of photovoltaic generation considering multi-factor self-fluctuation

Nowadays, interval forecasting is crucial for the power dispatching department because it can provide reliable boundaries of photovoltaic (PV) generation. However, few countermeasures are proposed to avoid the multi-factor self-fluctuation involved in the forecasting. To enhance the credibility of PV interval forecasting, this paper mainly conducts the following research around two factors “temperature” and “solar irradiance”: 1. Considering the periodicity of temperature data, an interval forecasting strategy based on auto-regressive integrated moving average model (ARIMA) and Bootstrap algorithms is designed to obtain its fluctuation range. 2. For solar irradiance data, considering it has temporal and non-temporal features, an interval forecasting strategy based on interval optimization and multi-extreme learning machine (ELM) is designed to obtain its fluctuation range; 3. Based on the obtained fluctuation boundaries of temperature and solar irradiance, the ELM is combined with the empirical mode decomposition (EMD) to complete error fitting and apply it to complete the interval forecasting of PV generation. Related case studies are presented in this paper. Finally, on the basis of ensuring the interval coverage requirement, we find the proposed method improves the forecast performance by 7.5 % compared to the existing methods at least, which will provide more reliable data for decision-making and management.

Day-Ahead PV Generation Scheduling in Incentive Program for Accurate Renewable Forecasting

Photovoltaic (PV) power can be a reasonable alternative as a carbon-free power source in a global warming environment. However, when many PV generators are interconnected in power systems, inaccurate forecasting of PV generation leads to unstable power system operation. In order to help system operators maintain a reliable power balance, even when renewable capacity increases excessively, an incentive program has been introduced in Korea. The program is expected to improve the self-forecasting accuracy of distributed generators and enhance the reliability of power system operation by using the predicted output for day-ahead power system planning. In order to maximize the economic benefit of the incentive program, the PV site should offer a strategic schedule. This paper proposes a PV generation scheduling method that considers incentives for accurate renewable energy forecasting. The proposed method adjusts the predicted PV generation to the optimal generation schedule by considering the characteristics of PV energy deviation, energy storage system (ESS) operation, and PV curtailment. It then maximizes incentives by mitigating energy deviations using ESS and PV curtailment in real-time conditions. The PV scheduling problem is formulated as a stochastic mixed-integer linear programming (MILP) problem, considering energy deviation and daily revenue under expected PV operation scenarios. The numerical simulation results are presented to demonstrate the economic impact of the proposed method. The proposed method contributes to mitigating daily energy deviations and enhancing daily revenue.

On vision transformer for ultra-short-term forecasting of photovoltaic generation using sky images

An accurate photovoltaic (PV) generation forecasting is important for grid scheduling and dispatching. However, ultra-short-term PV generation forecasting is rather challenging because weather conditions may change significantly in a short time period largely due to the dynamics and movement of clouds above a solar PV farm. For monitoring clouds above the solar PV farm, ground-based whole-sky cameras (Sky-Imagers) have been installed. This paper develops a novel cloud image-based ultra-short-term forecasting framework. Within the framework, an integration of the Vision Transformer (ViT) model and the Gated Recurrent Unit (GRU) encoder is designed for the high-dimensional latent feature analysis. A Multi-Layer Perception (MLP) is employed to generate the one-step-ahead PV generation forecasting. Numeric experiments are conducted using real-world solar PV datasets. The results show that the proposed framework and algorithms can achieve higher accuracy compared to several baseline methods for ultra-short-term PV generation forecasting.

Photovoltaic generation forecasting using convolutional and recurrent neural networks

Due to climate change consequences, it is very important to replace fossil energy resources with renewable energy resources. Solar energy is one of the main types of renewable energy resources which is harnessed by Photovoltaic (PV) Cells. It is important to accurately forecast how much electricity these energy resources generate to help operate and maintain the electricity grid. But the generation of electricity by PV is often associated with large uncertainty due to varying features like radiation, wind, humidity, and temperature. Deep learning methods have proved useful for this forecasting problem but the spatial information of features for this type of method has not received the due attention for PV generation forecasting. This study aimed to explore how both spatial and temporal information can be considered via a deep learning approach. In this paper, we propose a PV generation forecaster that considers both spatial and temporal information. A convolutional neural network is used as a pre-processing step to capture spatial information. The convolutional neural network is followed by a gated recurrent unit neural network to model temporal characteristics. The proposed model enriches the forecaster model by feeding more meaningful features into the recurrent neural network rather than raw data. The proposed model can predict a horizon for which there is no available information on irradiance, humidity, or wind. We show experimentally that our method is competitive with the state-of-the-art in terms of time and memory requirement while resulting in better prediction performance. The proposed model is applied to real data collected by the research team, and its performance is compared with some state-of-the-art methods. The results show the advantage of the proposed method.

Data-Driven Minute-Ahead Forecast of PV Generation with Adjacent PV Sector Information

This paper proposes and validates a data-driven minute-ahead forecast model for photovoltaic (PV) generation, which is essential for real-time micro-grid scheduling. Unlike day-ahead PV forecasts that heavily rely on weather forecast information, our proposed model does not require such data as it operates in an ultra-short-term time domain. Instead, the model leverages the generation data of the target PV sector and its adjacent sectors to capture short-term factors that affect electricity generation, such as the movement of clouds. The proposed model employs a long short-term memory (LSTM) network to process the data. By conducting experiments with real PV site data, we demonstrate that the information from adjacent PV sectors improves the accuracy of minute-ahead PV generation forecasts by 3.66% in the mean squared error index and 1.19% in the mean absolute error index compared to the model without adjacent sector information.

An innovative short-term multihorizon photovoltaic power output forecasting method based on variational mode decomposition and a capsule convolutional neural network

Due to its importance regarding the integration, economic dispatch, and operation of PV smart grid systems, infrastructure planning, and budgeting, the accurate forecasting of photovoltaic (PV) power generation has drawn increasing research and industry attention. However, the instability, intermittence, and randomness of solar irradiance impose difficulties on the short-term economic dispatch of a smart integrated power, grid and significantly increase the risks arising from PV generation in a power system, exposing PV generators to potential additional costs. A deep convolutional neural network (CNN)-based method can effectively improve the performance of PV generation point prediction and probabilistic interval prediction by efficiently extracting nonlinear features at each frequency. Nevertheless, existing deep learning (DL) studies have mostly focused on more complex network topologies and data decomposition algorithms, ignoring the importance of simultaneously forecasting the PV power produced over multiple temporal periods. To solve the described challenge, we propose a novel two-stage DL approach for PV generation prediction. The end-to-end trained model used in the proposed PV generation forecasting method is an effort to combine the existing methods and avoid the separation of entire task, such as single time-scale PV generation forecasting, independent point forecasting, and probabilistic interval forecasting. A temporal signal decomposition technique called variational mode decomposition (VMD) is employed in the first stage to construct association mappings from fine-grained features to images. In the second stage, an innovative capsule CNN (ACCNet) is proposed to obtain very short-term multihorizon ahead output power predictions for seven different PV systems based on polycrystalline, monocrystalline, cadmium telluride (CdTe) thin-film, amorphous, copper indium gallium diselenide (CIGS) thin-film, heterojunction with intrinsic thin layer (HIT) hybrid, and concentrated photovoltaic (CPV) technologies. The input parameters for each system include solar radiation (diffuse/global horizontal radiation (DHR/GHR) and radiation diffuse/global tilted (RDT/RGT)) and ambient temperature, while the output parameter is the power output of each PV system. The proposed model is validated with the historical datasets of a PV system downloaded from the Desert Knowledge Precinct in Central Australia (DKASC) homepage. The performance of the developed method is proven in detail based on seven different PV systems over multiple data periods, and the experimental results reveal that the proposed approach displays significant improvements and robustness in point forecasting and probabilistic forecasting tasks. We will release the source code to ensure reproducibility and facilitate future work. Our model is open-sourced at https://github.com/YunDuanFei/ACCNet.

A bi-level optimization method for voltage control in distribution networks using batteries and smart inverters with high wind and photovoltaic penetrations

In electrical distribution systems, the voltage quality problem considers the most restrictive issue that hinders high photovoltaic (PV) and wind integration. Therefore, this study aims at providing a reliable control method for the voltage problem in distribution networks embedded with high PV and wind sources. The proposed voltage management strategy consists of two controlling stages. The first stage uses battery energy storage systems (BESS), while the second stage employs the smart PV inverters' reactive power injection capability. For BESS operation, a bi-level optimization method based on metaheuristic optimization algorithms (MOA) is developed to regulate the voltage levels by governing the batteries charging/discharging rates. The bi-level optimization problem aims to maximize PV and wind energy exploitation while improving the voltage profile. In this study, a detailed comparison is conducted between three MOA: Social Spider Optimization, Particle Swarm Optimization, and Cuckoo Search Optimization. In addition, considering the uncertainty associated with PV system generation, a PV generation forecast is embedded in the voltage control strategy to reinforce the decision-making process of BESS operation. The efficacy of the proposed method is deployed on a modified IEEE-34 bus test feeder with real-world load and PV and wind power generation data. Experimental results show the effectiveness of the proposed method and indicate that proper coordination between the BESS and smart PV inverters is beneficial for distribution system operations that seamlessly integrate PV and wind energy. Overall, the proposed voltage management strategy can be used to control the voltage to any desired range at different locations.

Improved super-resolution perception convolutional neural network for photovoltaics missing data recovery

Errors in data measurement and recording can induce data missing for photovoltaic (PV) operation. Flawed data, however, can greatly impair the performance of data-driven models such as the accuracy of PV generation forecasting. The super-resolution perception convolutional neural network (SRPCNN) is a powerful tool in data reconstruction from low frequency to high frequency and it can be used for data recovery. The SRPCNN, however, has problems in dealing with null values for PV data recovery. To solve this problem, an improved SRPCNN model, which integrates naïve SPRCNN and linear interpolation, is proposed in this paper. Experiments are carried out under 5 different missing rates. The improved model can handle null values better, and can have a better accuracy improvement than the naïve SPRCNN and other methods. To further improve the data recovery performance, a modified whale optimization algorithm is deployed to optimize the hyperparameters of SPRCNN. To verify the efficiency of the proposed model, extensive numerical experiments on different types of data from an actual PV power plant have been conducted by comparing with benchmarks. Compared with the naïve SRPCNN, the RMSE and MAE of the proposed model are reduced by more than 30% in various cases. Results show that the improved SRPCNN model is efficient in PV data recovery with various missing rates.

Charging scheduling and energy management for mobile chargers in a grid-interactive transportation system

Mobile charging refers to the mobility-on-demand battery packages that can provide flexible charging services and regulation capacities in a grid-interactive transportation system, especially for ridesharing fleets. In this study, we use real-world data to understand the impact of mobile charging in ridesharing operations, and propose a data-driven optimization framework to identify optimal integration scenarios for mobile chargers in urban transportation systems. A mixed-integer linear program is developed to determine how many mobile chargers are required to satisfy the charging demands of shared vehicles, and when and how much to store or discharge energy to the microgrid or to other shared vehicles vs. the time-of-use charging prices regimes during a day. The objective is to minimize the daily operating cost of mobile chargers. In the proposed optimization framework, we first analyze real-world datasets that include information on trips and charging demands of fleets sharing more than one thousand shared vehicles. We then investigate a microgrid power profile that consists of base load analysis and solar photovoltaic generation forecasting. Lastly, we implement our framework in a real-world case study to evaluate model performance. The results show that our approach can reduce the total costs by about 21 % compared to a non-scheduling greedy policy, and thus help stabilize the power profile by vehicle-to-grid implementation.

Multivariate Strategy Using Artificial Neural Networks for Seasonal Photovoltaic Generation Forecasting

Electric power systems have experienced the rapid insertion of distributed renewable generating sources and, as a result, are facing planning and operational challenges as new grid connections are made. The complexity of this management and the degree of uncertainty increase significantly and need to be better estimated. Considering the high volatility of photovoltaic generation and its impacts on agents in the electricity sector, this work proposes a multivariate strategy based on design of experiments (DOE), principal component analysis (PCA), artificial neural networks (ANN) that combines the resulting outputs using Mixture DOE (MDOE) for photovoltaic generation prediction a day ahead. The approach separates the data into seasons of the year and considers multiple climatic variables for each period. Here, the dimensionality reduction of climate variables is performed through PCA. Through DOE, the possibilities of combining prediction parameters, such as those of ANN, were reduced, without compromising the statistical reliability of the results. Thus, 17 generation plants distributed in the Brazilian territory were tested. The one-day-ahead PV generation forecast has been considered for each generation plant in each season of the year, reaching mean percentage errors of 10.45% for summer, 9.29% for autumn, 9.11% for winter and 6.75% for spring. The versatility of the proposed approach allows the choice of parameters in a systematic way and reduces the computational cost, since there is a reduction in dimensionality and in the number of experimental simulations.

Assessing the Impact of Features on Probabilistic Modeling of Photovoltaic Power Generation

Photovoltaic power generation has high variability and uncertainty because it is affected by uncertain factors such as weather conditions. Therefore, probabilistic forecasting is useful for optimal operation and risk hedging in power systems with large amounts of photovoltaic power generation. However, deterministic forecasting is the mainstay of photovoltaic generation forecasting; there are few studies on probabilistic forecasting and feature selection from weather or time-oriented features in such forecasting. In this study, prediction intervals were generated by the lower upper bound estimation (LUBE) using neural networks with two outputs to make probabilistic modeling for predictions. The objective was to improve prediction interval coverage probability (PICP), mean prediction interval width (MPIW), continuous ranked probability score (CRPS), and loss, which is the integration of PICP and MPIW, by removing unnecessary features through feature selection. When features with high gain were selected by random forest (RF), in the modeling of 14.7 kW PV systems, loss improved by 1.57 kW, CRPS by 0.03 kW, PICP by 0.057 kW, and MPIW by 0.12 kW on average over two weeks compared to the case where all features were used without feature selection. Therefore, the low gain features from RF act as noise and reduce the modeling accuracy.

A Photovoltaic Power Predicting Model Using the Differential Evolution Algorithm and Multi-Task Learning

The rapid growth of photovoltaic installed capacity exacerbates the power management challenges faced by photovoltaic power stations, emphasizing the importance of accurate and stable photovoltaic generation forecasting. As a result, researchers conducted research and developed several photovoltaic power prediction models. However, many prediction models focus exclusively on the algorithm structure in order to improve model accuracy, oblivious to how the dataset is constructed and divided for the prediction model. This paper proposes a comprehensive model to address this gap. To be more precise, the differential evolution algorithm is constantly looking for optimal values between different populations and determining the best way to construct datasets for prediction tasks. Multi-task learning enables the transfer of knowledge between related tasks via parameter sharing layers, referring to the accuracy and stability of prediction models. Overall, the proposed model achieves high prediction accuracy and stability. The prediction error of the proposed model is less than 450W in RMSE, NRMSE is less than 2.5%, and R-Square is greater than 99% in multiple prediction tasks. Additionally, when compared to other single-task prediction models with an R-Square greater than 96%, the proposed model further reduces the root mean squared error by an average of 28% and the standard deviation of root mean squared error by 54%.

Feasible Dispatch Limits of PV Generation With Uncertain Interconnection of EVs in the Unbalanced Distribution Network

This paper presents a framework to determine the feasible dispatch limits of solar photovoltaic (PV) generation in the unbalanced distribution network considering the interconnection of electric vehicles (EVs) and associated uncertainties. The proposed framework determines the lower and upper dispatch limits of PV generation considering the worst-case realization of a) the minimum and maximum storage capacity of EVs, b) the minimum and maximum power dispatch of EVs, c) the lower and upper bounds for the arrival and departure times, and d) the available energy at arrival and departure times. The unbalanced operation of the distribution network as well as the uncertainty in the maximum PV generation and demand forecasts were considered. The problem formulation and solution approach are validated using the modified IEEE 34-bus and IEEE 123-bus distribution systems. The impacts of the budget of uncertainty and the vehicle-to-grid operation mode of EV clusters were addressed in the case studies. It is shown that integrating EVs with charging capability will increase the lower dispatch limit or increase the upper dispatch limit of the PV generation. Moreover, the increase in the budget of uncertainty will reduce the difference between the upper and lower dispatch limits by increasing the lower dispatch or decreasing the upper dispatch limit of the PV generation. Finally, it is shown that the vehicle-to-grid capability will reduce the total lower dispatch limit or increase the total upper dispatch limit of the PV generation in the operation horizon.

Hour-ahead photovoltaic generation forecasting method based on machine learning and multi objective optimization algorithm

As the penetration rate of solar energy in the grid continues to enhance, solar power photovoltaic generation forecasts have become an indispensable aspect of mechanism mobilization and maintenance of the stability of the power system. In this regard, many researchers have done a lot of study, and put forward some predictive models. However, many individual prediction systems only consider the prediction accuracy rate without further considering the prediction utility and stability. To fill this gap, a comprehensive system is designed in this paper, which is on the basis of automatic optimization of variational mode decomposition mechanism, and the weight of system is determined by multi objective intelligent optimization algorithm. In particular, it can be proved theoretically that the developed predictive system can achieve the pareto optimal solution. And the designed system is shown to be very effective in forecasting the 2021 photovoltaic power data obtained from Belgium. The empirical study reports that the combination of variational mode decomposition strategy based on genetic algorithm and multi objective grasshopper optimization algorithm is found to be the satisfactory strategy to optimize the predictive system compared with other common mechanism. And the results of several numerical studies show that the designed predictive system achieves the superior performance as compared to the control systems, and in multi-step forecasting, the designed system has better stability than the comparison systems.

Improving Forecast Reliability for Geographically Distributed Photovoltaic Generations

Photovoltaic (PV) generation is potentially uncertain. Probabilistic PV generation forecasting methods have been proposed with prediction intervals (PIs) to evaluate the uncertainty quantitively. However, few studies have applied PIs to geographically distributed PVs in a specific area. In this study, a two-step probabilistic forecast scheme is proposed for geographically distributed PV generation forecasting. Each step of the proposed scheme adopts ensemble forecasting based on three different machine-learning methods. When individual PV generation is forecasted, the proposed scheme utilizes surrounding PVs’ past data to train the ensemble forecasting model. In this case study, the proposed scheme was compared with conventional non-multistep forecasting. The proposed scheme improved the reliability of the PIs and deterministic PV forecasting results through 30 days of continuous operation with real data in Japan.

Solar Irradiance Nowcasting for Virtual Power Plants Using Multimodal Long Short-Term Memory Networks

The rapid penetration of photovoltaic generation reduces power grid inertia and increases the need for intelligent energy resources that can cope in real time with the imbalance between power generation and consumption. Virtual power plants are a technology for coordinating such resources and monetizing them, for example on electricity markets with real-time pricing or on frequency reserves markets. Accurate short-term photovoltaic generation forecasts are essential for such virtual power plants. Although significant research has been done on medium- and long-term photovoltaic generation forecasting, the short-term forecasting problem requires special attention to sudden fluctuations due to the high variability of cloud cover and related weather events. Solar irradiance nowcasting aims to resolve this variability by providing reliable short-term forecasts of the expected power generation capacity. Sky images captured in proximity to the photovoltaic panels are used to determine cloud behavior and solar intensity. This is a computationally challenging task for conventional computer vision techniques and only a handful of Artificial Intelligence (AI) methods have been proposed. In this paper, a novel multimodal approach is proposed based on two Long Short-Term Memory Networks (LSTM) that receives a temporal image modality of a stream of sky images, a temporal numerical modality of a time-series of past solar irradiance readings and cloud cover readings as inputs for irradiance nowcasting. The proposed nowcasting pipeline consists of a preprocessing module and an irradiance augmentation module that implements methods for cloud detection, Sun localization and mask generation. The complete approach was empirically evaluated on a real-world solar irradiance case study across the four seasons of the northern hemisphere, resulting in a mean improvement of 39% for multimodality.

A Regional Day-Ahead Rooftop Photovoltaic Generation Forecasting Model Considering Unauthorized Photovoltaic Installation

Rooftop photovoltaic (PV) systems are usually behind the meter and invisible to utilities and retailers and, thus, their power generation is not monitored. If a number of rooftop PV systems are installed, it transforms the net load pattern in power systems. Moreover, not only generation but also PV capacity information is invisible due to unauthorized PV installations, causing inaccuracies in regional PV generation forecasting. This study proposes a regional rooftop PV generation forecasting methodology by adding unauthorized PV capacity estimation. PV capacity estimation consists of two steps: detection of unauthorized PV generation and estimation capacity of detected PV. Finally, regional rooftop PV generation is predicted by considering unauthorized PV capacity through the support vector regression (SVR) and upscaling method. The results from a case study show that compared with estimation without unauthorized PV capacity, the proposed methodology reduces the normalized root mean square error (nRMSE) by 5.41% and the normalized mean absolute error (nMAE) by 2.95%, It can be concluded that regional rooftop PV generation forecasting accuracy is improved.

Methods for Integrating Extraterrestrial Radiation into Neural Network Models for Day-Ahead PV Generation Forecasting

Variability, intermittency, and limited controllability are inherent characteristics of photovoltaic (PV) generation that result in inaccurate solutions to scheduling problems and the instability of the power grid. As the penetration level of PV generation increases, it becomes more important to mitigate these problems by improving forecasting accuracy. One of the alternatives to improving forecasting performance is to include a seasonal component. Thus, this study proposes using information on extraterrestrial radiation (ETR), which is the solar radiation outside of the atmosphere, in neural network models for day-ahead PV generation forecasting. Specifically, five methods for integrating the ETR into the neural network models are presented: (1) division preprocessing, (2) multiplication preprocessing, (3) replacement of existing input, (4) inclusion as additional input, and (5) inclusion as an intermediate target. The methods were tested using two datasets in Australia using four neural network models: Multilayer perceptron and three recurrent neural network(RNN)-based models including vanilla RNN, long short-term memory, and gated recurrent unit. It was found that, among the integration methods, including the ETR as the intermediate target improved the mean squared error by 4.1% on average, and by 12.28% at most in RNN-based models. These results verify that the integration of ETR into the PV forecasting models based on neural networks can improve the forecasting performance.

A Compressive Receding Horizon Approach for Smart Home Energy Management

Current increases in the demand for electricity require sustainable energy management measures and have promoted the adoption of clean and renewable sources, particularly at the residential building level. Active demand management is usually carried out through load shifting based on specific techniques, such as optimisation, heuristics, model-based predictive control and machine learning methodologies. This work addresses the problem of residential load scheduling via optimisation techniques. A compressive receding horizon strategy is proposed for week-ahead load shifting, and the selection is driven by traditional receding horizon and day-ahead allocation strategy misalignment, with weekly household appliance usage patterns. The proposed approach is compared with receding horizon and day-ahead scheduling techniques over 30 different weeks for a prototypical smart home with non-controllable demand, which is representative of a four-resident family and includes micro power generation and battery storage. The simulation results confirm the validity of the proposed strategy in the context of household appliance scheduling problems and show competitive electricity costs and resident discomfort performance compared to state-of-the-art approaches. Furthermore, the proposed compressive receding horizon strategy fully exploits weather and photovoltaic generation forecasts to promote self-consumption and grid demand stress reduction while providing environmental gains and financial benefits to the utility service and consumers, particularly in the case of simultaneously scheduling a huge number of households.