Generation Forecasting Research Articles

Decentralized forecasting of wind energy generation with an adaptive machine learning approach to support ancillary grid services

Abstract. We report on an approach to distributed wind power forecasting, which supports wind energy integration in power grid operation during exceptional and critical situations. Forecasts are generated on-site the wind power plant (WPP) in order to provide blackout-robust data transmission directly from the WPP to the grid operator. An adaptively trained forecasting model uses locally available sensor data to predict the available active power (AAP) signal in a probabilistic fashion. A forecast generated off-site based on numerical weather prediction (NWP) is deposited and combined on-site the WPP with the locally generated forecast. We evaluate the performance of the method in a case study and find that the locally generated forecast significantly improves forecast reliability for a short-term horizon, which is highly relevant for enabling power reserve provision from WPPs.

A Review of Modern Wind Power Generation Forecasting Technologies

The prediction of wind power output is part of the basic work of power grid dispatching and energy distribution. At present, the output power prediction is mainly obtained by fitting and regressing the historical data. The medium- and long-term power prediction results exhibit large deviations due to the uncertainty of wind power generation. In order to meet the demand for accessing large-scale wind power into the electricity grid and to further improve the accuracy of short-term wind power prediction, it is necessary to develop models for accurate and precise short-term wind power prediction based on advanced algorithms for studying the output power of a wind power generation system. This paper summarizes the contribution of the current advanced wind power forecasting technology and delineates the key advantages and disadvantages of various wind power forecasting models. These models have different forecasting capabilities, update the weights of each model in real time, improve the comprehensive forecasting capability of the model, and have good application prospects in wind power generation forecasting. Furthermore, the case studies and examples in the literature for accurately predicting ultra-short-term and short-term wind power generation with uncertainty and randomness are reviewed and analyzed. Finally, we present prospects for future studies that can serve as useful directions for other researchers planning to conduct similar experiments and investigations.

A review of international experience in forecasting renewable energy generation using machine learning methods

In this work, we conduct an analytical review of contemporary international approaches to forecasting the volume of electricity generated by renewable energy sources, as well as to investigate current problems and prospective solutions in this field. The existing forecasting methods were classified following an analysis of published literature on the development of forecasting models, including those based on physical, statistical and machine learning principles. The application practice of these methods was investigated to determine the advantages and disadvantages of each method. In the majority of cases, particularly when carrying out short-term forecasting of renewable electricity generation, machine learning methods outperform physical and statistical methods. An analysis of the current problems in the field of weather data collection systems allowed the major obstacles to a wide application of machine learning algorithms to be determined, which comprise incompleteness and uncertainty of input data, as well as the high computational complexity of such algorithms. An increased efficiency of machine learning models in the task of forecasting renewable energy generation can be achieved using data preprocessing methods, such as normalization, anomaly detection, missing value recovery, augmentation, clustering and correlation analysis. The need to develop data preprocessing methods aimed at optimizing and improving the overall efficiency of machine learning models for forecasting renewable energy generation was justified. Research in this direction, while taking into account the above problems, is highly relevant for the imp lementation of programs for the integration of renewable energy sources into power systems and the development of carbon-free energy.

EMD-based gray combined forecasting model - Application to long-term forecasting of wind power generation

Wind power is the most promising renewable energy source after hydropower because of its mature technology and low price, and has great potential for carbon emission reduction. Long-term forecasts of its power generation can help power companies to develop operational plans, grid configuration and power dispatch, and can also provide a basis for the government to formulate energy and environmental policies. However, due to the characteristics of China's monsoon climate and wind power industry development, wind power generation data are characterized by nonlinear cycles and small data volume, which makes accurate prediction more difficult. To this end, this paper develops a new prediction model and applies it to the long-term prediction of wind power generation in China, and proposes some targeted policy recommendations based on the prediction results to promote the development of China's wind power industry.

Green hydrogen production ensemble forecasting based on hybrid dynamic optimization algorithm

Solar-powered water electrolysis can produce clean hydrogen for sustainable energy systems. Accurate solar energy generation forecasts are necessary for system operation and planning. Al-Biruni Earth Radius (BER) and Particle Swarm Optimization (PSO) are used in this paper to ensemble forecast solar hydrogen generation. The suggested method optimizes the dynamic hyperparameters of the deep learning model of recurrent neural network (RNN) using the BER metaheuristic search optimization algorithm and PSO algorithm. We used data from the HI-SEAS weather station in Hawaii for 4 months (September through December 2016). We will forecast the level of solar energy production next season in our simulations and compare our results to those of other forecasting approaches. Regarding accuracy, resilience, and computational economy, the results show that the BER-PSO-RNN algorithm has great potential as a useful tool for ensemble forecasting of solar hydrogen generation, which has important ramifications for the planning and execution of such systems. The accuracy of the proposed algorithm is confirmed by two statistical analysis tests, such as Wilcoxon’s rank-sum and one-way analysis of variance (ANOVA). With the use of the proposed BER-PSO-RNN algorithm that excels in processing and forecasting time-series data, we discovered that with the proposed algorithm, the Solar System could produce, on average, 0.622 kg/day of hydrogen during the season in comparison with other algorithms.

Forecasting clean energy power generation in China based on a novel fractional discrete grey model with a dynamic time-delay function

Clean energy plays an essential role in responding to environmental crises. Accurate forecasting of clean energy power generation can provide necessary references for the formulation of energy policy. This paper proposes a novel fractional discrete grey model with a dynamic time delay function (DTDFF-DGM (1,1)) to forecast clean energy power generation. This model introduces the fractional accumulation operator and the dynamic time-delay function into the discrete grey model, which ensures the priority of new information in the original data and improves the model's adaptability to different sample data. The optimal parameters of the model are calculated by the dynamic linkage between the fitting error and the test error, which effectively avoids the overfitting problem. Empirical studies have proved that the model has better prediction accuracy compared to other methods. Finally, the proposed model is employed to forecast clean energy power generation in China. The results show that from 2020 to 2025, the overall growth rate of hydropower, wind power and nuclear power in China will be 9.27%, 119.61% and 36.38%, respectively. Based on the discussion of the forecast results, relevant policy suggestions were made. This paper realizes the transformation from a static model to a dynamic model in methodology and promotes the sustainable development of clean energy power generation in application.

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.

Forecasting China's hydropower generation using a novel seasonal optimized multivariate grey model

Global warming and environmental degradation are essential issues that endanger human survival. The conflict between rising carbon emissions and carbon neutrality goals has prompted an urgent need for China's energy sector to step up efforts to develop clean energy generation. As a significant hydropower country, hydropower generation is China's mainstay of clean energy generation. The work contributing to sustainable hydropower development requires reasonable forecasts of clean energy generation. This paper proposes a seasonal optimized multivariate grey model that optimizes background value and supplements dummy variables to explore hidden factors through an optimization algorithm to the related sequences. The novel model improves the fitting and prediction accuracy through the loop supplementation of dummy variables. The model tests the effect of China's hydropower generation prediction and compares results with other methods. The mean absolute percentage error of the model training and test groups is 3.87 % and 0.83 %. Finally, this paper predicts hydropower generation in China from 2022 to 2025 based on the power generation during China's 13th Five-Year Plan Period.

A VAE-Bayesian deep learning scheme for solar power generation forecasting based on dimensionality reduction

The advancements in distributed generation (DG) technologies such as solar panels have led to a widespread integration of renewable power generation in modern power systems. However, the intermittent nature of renewable energy poses new challenges to the network operational planning with underlying uncertainties. This paper proposes a novel probabilistic scheme for renewable solar power generation forecasting by addressing data and model parameter uncertainties using Bayesian bidirectional long short-term memory (BiLSTM) neural networks, while handling the high dimensionality in weight parameters using variational auto-encoders (VAE). The forecasting performance of the proposed method is evaluated using various deterministic and probabilistic evaluation metrics such as root-mean square error (RMSE), Pinball loss, etc. Furthermore, reconstruction error and computational time are also monitored to evaluate the dimensionality reduction using the VAE component. When compared with benchmark methods, the proposed method leads to significant improvements in weight reduction, i.e., from 76,4224 to 2,022 number of weight parameters, quantifying to 97.35% improvement in weight parameters reduction and 37.93% improvement in computational time for 6 months of solar power generation data.

Short-term PV power forecast using hybrid deep learning model and Variational Mode Decomposition

In recent decades, the dramatic transformation from conventional energy to renewables, such as photovoltaic (PV), has been extensively occurred to address the increasing electricity demand and environmental issues. Nevertheless, the operation of PV systems is adversely affected by the intermittence of meteorological factors, such as solar irradiance. Therefore, accurate PV power generation forecast has played a vital role to stimulate the integration of renewable energy sources into the power grid. This study proposes a novel scheme for short-term PV power forecast based on Transformer Neural Network (TransNN) integrated with Convolutional Neural Network (CNN). In addition, Variational Mode Decomposition (VMD) is adopted in the data pre-processing stage to enhance the forecast accuracy. The performance of the proposed model is validated utilizing two real-world datasets, and five benchmark models are dedicated to forecast outcome comparison. The predicted results indicated the superiority of the proposed model compared to the referenced models by achieving the Mean Absolute Error (MAE) values under 1 kW in all cases.

Day-Ahead Wind Power Forecasting Using Numerical Weather Prediction

New power plants are included in the grid continuously to meet the increasing need for electrical energy. More than half of the power plants that have been newly added to the grid since 2015 are renewable energy power plants. As of 2022, the total installed renewable energy power increased by 11% compared to the previous year and reached 3146 GW. This increase will likely accelerate with the widespread use of electric vehicles. Wind energy is one of the renewable energy types with the highest installed capacity among renewable energy sources. Due to the nature of the wind, fluctuations in production cause adverse effects on the management of the electricity grid. Therefore, the forecast of renewable energy generation is very important for managing the electricity grid. This paper presents a wind energy forecast for the next 24 hours using data from a wind farm in Turkey. The forecast model consists of two stages. In the first stage, numerical weather predictions (NWP) have improved by using Artificial Neural Networks (ANN) and Artificial Bee Colony (ABC) methods. In the second stage, estimation is carried out by the ANFIS method. Improving the NWP using past NWP data and actual wind speed data increases the power forecast accuracy. The forecast results are compared with the forecasts made by the wind power plant (WPP). The proposed model has forecasted wind power with NRMSE 16.73%, and it has given more successful results than the forecast made by the power plant.

Predicting Power Generation from a Combined Cycle Power Plant Using Transformer Encoders with DNN

With the development of the Smart Grid, accurate prediction of power generation is becoming an increasingly crucial task. The primary goal of this research is to create an efficient and reliable forecasting model to estimate the full-load power generation of a combined-cycle power plant (CCPP). The dataset used in this research is a subset of the publicly available UCI Machine Learning Repository. It contains 9568 items of data collected from a CCPP during its full load operation over a span of six years. To enhance the accuracy of power generation forecasting, a novel forecasting method based on Transformer encoders with deep neural networks (DNN) was proposed. The proposed model exploits the ability of the Transformer encoder to extract valuable information. Furthermore, bottleneck DNN blocks and residual connections are used in the DNN component. In this study, a series of experiments were conducted, and the performance of the proposed model was evaluated against other state-of-the-art machine learning models based on the CCPP dataset. The experimental results illustrated that using Transformer encoders along with DNN can considerably improve the accuracy of predicting CCPPs power generation (RMSE = 3.5370, MAE = 2.4033, MAPE = 0.5307%, and R2 = 0.9555).

Scenario-Based Uncertainty Modeling for Power Management in Islanded Microgrid Using the Mixed-Integer Distributed Ant Colony Optimization

Reliable droop-controlled islanded microgrids are necessary to expand coverage and maximize renewables potential. Nonetheless, due to uncertainties surrounding renewable generation and load forecast, substantial power mismatch is expected at off-peak hours. Existing energy management systems such as storage and demand response are not equipped to handle a large power mismatch. Hence, utilizing dump loads to consume excess power is a promising solution to keep frequency and voltage within permissible limits during low-load hours. Considering the uncertainty in wind generation and demand forecast during off-peak hours, the dump load allocation problem was modeled within a scenario-based stochastic framework. The multi-objective optimization with uncertainty was formulated to minimize total microgrid cost, maximum voltage error, frequency deviation, and total energy loss. The mixed-integer distributed ant colony optimization was utilized in a massive parallelization framework for the first time in microgrids to solve the decomposed deterministic problem of the most probable scenarios. Moreover, a flexible and robust load-flow method called general backward/forward sweep was used to obtain the load-flow solution. The optimization problem was applied to the IEEE 69-bus and 118-bus systems. Furthermore, a cost benefit analysis was provided to highlight the proposed method’s advantage over battery-based power management solutions. Lastly, the obtained results further demonstrate the fundamental role of dump load as power management solution while minimizing costs and energy losses.

A novel data-driven seasonal multivariable grey model for seasonal time series forecasting

Rapid growth in solar power generation, accompanied by random fluctuations, has important implications for the security, stability, and productivity of the grid. Therefore, accurate predictions of solar power generation provide an essential benchmark for relevant institutions to conduct grid planning and power dispatch. Thereby, a novel data-driven seasonal multivariable grey model is proposed for simulating seasonal fluctuations and non-linear trends in solar power generation. To be specific, the novel model contains the following refinements: first, the seasonal factors and time-power item are introduced to modify the insensitivity of traditional multivariable grey models to seasonal fluctuations and non-linear trends. Second, to enhance the generalizability and adaptability of the model, the parameter associated with the time-power item is determined by using the genetic algorithm. Third, on the premise of proving the model’s requirements for period length and sample size, the derivation of the time response function is expounded. For illustration and verification purposes, comprehensive experimental studies are conducted with quarterly and monthly time series samples. In addition, forecasts of the novel model are compared with five prevailing models, including grey prediction models, traditional econometric technology, and artificial intelligence. Case studies indicate that this novel model exhibits improved generalizability, stability, and reliability in the face of quarterly or monthly time series data, confirming it as an auspicious and powerful tool for future solar power generation forecasting.

Forecasting China's electricity generation using a novel structural adaptive discrete grey Bernoulli model

The generation forecast helps to develop the scheduling plan of power sources, build a power balance system and ensure the adequacy of power supply to the power system at the macro level. Therefore, a novel structurally adaptive discrete grey Bernoulli model, called DHGBM(1,1), is proposed in this paper for the prediction of power generation. Firstly, the Hausdorff fractional order cumulative generation operation makes it possible to achieve new information prioritisation. As for the structure of the model, we have extended the nonlinear grey Bernoulli model with the introduction of a nonlinear dynamic structure term, which allows the model to simulate variable generation data and improves the adaptability of the model. In a comparative test of the algorithms, the well-performing differential evolution algorithm was used for hyperparameter optimisation of the model to obtain better predictive performance. Monte Carlo simulations and probability density analysis verified the excellent robustness of the model. In addition, through the simulation tests, we obtained a reasonable range of parameters for the structural terms, thus limiting the complexity of the model, and the phenomenon of model overfitting was effectively prevented. Finally, five actual cases of China's hydro, China's thermal, China's nuclear, China's wind power generation and China's electricity exports are predicted.

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.

Differential Evolution and Fuzzy-Logic-Based Predictive Algorithm for V2G Charging Stations

This work proposes a differential evolution algorithm to control a vehicle-to-grid (V2G) system based on photovoltaic generation and energy cost curves, and constraints associated with the power converters’ operation, battery charging strategy, and initial budgets. The algorithm is designed to trade off the batteries’ state of charge and the profits gained from selling energy to the grid. To achieve this balance, a fuzzy controller is employed and acts based on forecasts of the photovoltaic generation and the cost of electricity, within prediction windows of 120 min, adapting the batteries’ rate of charging or discharging. Simulation results show that for different curves and different initial budgets, the target state of charge is reached at the end of the time horizon. By evaluating the proposed scheme under different scenarios, the algorithm’s performance is proven to be suitable for future practical deployment.

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.

AI 기반 재생에너지 발전 예측 기술 특허 동향

In order to utilize renewable power generation that contains uncertainty due to natural intermittence, it is essential to develop renewable power generation forecasting technologies. This paper focuses on identifying the technology development of artificial intelligence (AI)-based renewable power generation by the analysis of patents. In this study, AI-based renewable power generation forecasting technology is comprehensively defined, and related patent applications are systematically identified to analyze country, period, and technological trends. The results indicate that 1) renewable power generation forecasting has reached a technological maturity in Japan, the United States, and Europe, 2) to overcome this, AI technology has been being grafted onto renewable power generation forecasting, 3) the use of meteorological data was described as a major right, and AI technology was described as an applicable additional item due to the characteristics of AI technology patents.

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.