PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES
The decisions regarding power regulation, energy resource planning, and integrating “green” energy into the electrical grid hinge on precise probabilistic forecasts. One of the potential strategies to enhance forecast accuracy is the utilization of ensemble forecasting methods. They represent an app...
Збережено в:
| Дата: | 2024 |
|---|---|
| Автор: | |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
General Energy Institute of the National Academy of Sciences of Ukraine
2024
|
| Теми: | |
| Онлайн доступ: | https://systemre.org/index.php/journal/article/view/870 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | System Research in Energy |
| Завантажити файл: |  |
Репозитарії
System Research in Energy| _version_ | 1871104379866578944 |
|---|---|
| author | Matushkin, Dmytro |
| author_facet | Matushkin, Dmytro |
| author_institution_txt_mv | [
{
"author": "Dmytro Matushkin",
"institution": null
}
] |
| author_sort | Matushkin, Dmytro |
| baseUrl_str | https://systemre.org/index.php/journal/oai |
| collection | OJS |
| datestamp_date | 2026-07-18T12:57:49Z |
| description | The decisions regarding power regulation, energy resource planning, and integrating “green” energy into the electrical grid hinge on precise probabilistic forecasts. One of the potential strategies to enhance forecast accuracy is the utilization of ensemble forecasting methods. They represent an approach where multiple models collaborate to achieve superior results compared to what a single model could produce independently. These methods can be categorized into two main categories: competitive and collaborative ensembles. Competitive ensembles harness the diversity of parameters and data to create a rich pool of base models. This approach may encompass statistical analysis, noise filtering, and anomaly elimination. On the other hand, collaborative ensembles rely on the interaction among models to achieve better outcomes. These methods encompass strategies such as weighted predictions, voting, aggregation, and a combination of model results. The research of ensemble forecasting methods in the context of photovoltaic generation is highly relevant, as solar energy represents a crucial source of renewable energy. Accurate predictions of solar energy production address the challenges related to the efficient utilization of photovoltaic panels and their integration into the overall energy system. This paper investigates conceptual ensemble architectures for photovoltaic energy forecasting. These architectures encompass various methods of aggregating base models within an ensemble, allowing for the consideration of different aspects and peculiarities of solar data, such as solar irradiation intensity, meteorological conditions, geographic factors, and more. These conceptual models are developed based on well-established statistical, machine learning, and artificial intelligence methods. Therefore, this paper provides an overview of ensemble forecasting methods for renewable energy, covering competitive and collaborative ensembles, as well as developing conceptual models for solar energy forecasting. This work aims to elevate the accuracy and efficiency of forecasts in the realm of renewable energy, representing a significant step in the advancement of sustainable and environmentally friendly energy production. |
| doi_str_mv | 10.15407/srenergy2024.04.056 |
| first_indexed | 2026-03-24T02:03:25Z |
| format | Article |
| fulltext |
Системні дослідження в енергетиці. 2024. 4(80) 56
ISSN 2786-7102 (Online), ISSN 2786-7633 (Print)
https://doi.org/10.15407/srenergy2024.04.056
UDC 519.246.8:621.311
Dmytro Matushkin, PhD (Engin.), https://orcid.org/0000-0003-4431-7862
General Energy Institute of NAS of Ukraine, 172, Antonovycha St., Kyiv, 03150, Ukraine
e-mail: dmitry.matushkin@ukr.net
___________________________________________________________________________________
PHOTOVOLTAIC GENERATION FORECASTING MODELS:
CONCEPTUAL ENSEMBLE ARCHITECTURES
Abstract. The decisions regarding power regulation, energy resource planning, and integrating “green”
energy into the electrical grid hinge on precise probabilistic forecasts. One of the potential strategies to
enhance forecast accuracy is the utilization of ensemble forecasting methods. They represent an approach
where multiple models collaborate to achieve superior results compared to what a single model could
produce independently. These methods can be categorized into two main categories: competitive and
collaborative ensembles. Competitive ensembles harness the diversity of parameters and data to create a
rich pool of base models. This approach may encompass statistical analysis, noise filtering, and anomaly
elimination. On the other hand, collaborative ensembles rely on the interaction among models to achieve
better outcomes. These methods encompass strategies such as weighted predictions, voting, aggregation,
and a combination of model results. The research of ensemble forecasting methods in the context of
photovoltaic generation is highly relevant, as solar energy represents a crucial source of renewable energy.
Accurate predictions of solar energy production address the challenges related to the efficient utilization
of photovoltaic panels and their integration into the overall energy system. This paper investigates
conceptual ensemble architectures for photovoltaic energy forecasting. These architectures encompass
various methods of aggregating base models within an ensemble, allowing for the consideration of different
aspects and peculiarities of solar data, such as solar irradiation intensity, meteorological conditions,
geographic factors, and more. These conceptual models are developed based on well-established
statistical, machine learning, and artificial intelligence methods. Therefore, this paper provides an
overview of ensemble forecasting methods for renewable energy, covering competitive and collaborative
ensembles, as well as developing conceptual models for solar energy forecasting. This work aims to elevate
the accuracy and efficiency of forecasts in the realm of renewable energy, representing a significant step
in the advancement of sustainable and environmentally friendly energy production.
Keywords: probabilistic solar forecasting, ensemble model, forecast combination, competitive ensembles,
collaborative ensembles, conceptual models.
1. Introduction
The growing global demand for energy poses challenges such as excessive strain on supply [1],
significant adverse environmental impacts [2] (rising temperatures [3], ozone layer depletion, climate issues,
etc.) [4], and the depletion of energy resources [5]. Over the past decades, new resilient energy systems and
methods for integrating renewable energy sources and energy storage systems into the electrical grid have been
researched and discussed extensively [6, 7].
The impact of new approaches to designing, developing, and managing energy systems focusing on their
resilience throughout their entire life cycle presents a significant challenge [8]. Key terms like “Smart Grids”
and “Nearly Zero-Energy Buildings” are frequently used in the context of various energy system subfields.
Still, it's essential to consider them within the framework of the overall energy system [9, 10]. An example of
this is the concept of intelligent energy systems, commonly referred to as “Smart Energy Systems” [9, 11].
The full potential of managing energy systems using computational resources has not been fully realized,
remaining a relevant topic for further research and development [12]. However, a crucial requirement is the
integration of all subsystems of energy systems on a global scale [10].
The formal definition of a smart energy system, as stated in [11, 13], is that it comprises “new
technologies and infrastructure that create new forms of flexibility, primarily in the energy system conversion
stage” [14]. Various sectors, including electricity and transportation, collaborate to compensate for the lack of
Системні дослідження в енергетиці. 2024. 4(80) 57
flexibility resulting from using renewable energy sources [15–17]. The smartness of energy systems is closely
tied to the ability to predict their behavior in the future, making modeling, simulation, and forecasting essential.
Literature on forecasting in energy systems varies in terms of methods used, as noted in comparative
reviews [3, 15, 18]. These reviews demonstrate that there is no one universal method that outperforms all
others in all aspects. Primary research has focused on point forecasts, for which methods such as Artificial
Neural Networks (ANN) [19, 20], K-Nearest Neighbors (KNN) [21], Support Vector Regression [22], Random
Forests [23], and multi-factor linear regression models [24] are often considered optimal solutions. Traditional
statistical methods like Autoregressive Integrated Moving Average (ARIMA) [25] and Seasonal ARIMA
(SARIMA) are popular due to their simplicity, with the Box-Jenkins methodology often used for model
configuration selection. However, it's essential to consider that ARIMA models assume a linear correlation
structure between forecasts and past data, which may lead to poorer results in the presence of nonlinear
patterns. As real data often contain a combination of linear and nonlinear patterns [26], and solar irradiation
data exhibit intermittency and variability, ARIMA models may perform worse compared to other methods
used in the literature [27, 28].
Recent trends in probabilistic solar forecasting advocate for the use of probabilistic combinations of
individual forecasts as a viable approach to improve accuracy [18, 29, 30]. However, combining probabilistic
forecasts is significantly more complex than combining point forecasts, where simple weighted averaging is
often acceptable. When combining probabilistic forecasts, reliability and accuracy requirements must be met,
and it's crucial to preserve the essential characteristics of the probabilistic forecast in the combined forecasts.
The purpose of this article is to elucidate the essence and content of the concept of “ensemble
forecasting” while investigating the principles and providing conceptual models for ensemble solar
forecasting.
2. Global Ensemble Concept
The concept of ensemble modeling is a powerful approach in machine learning (ML) and statistics that
involves combining predictions or outputs from multiple individual models to create a more accurate and
robust overall forecast. Ensemble modeling is based on the idea that by aggregating the results of several
models, the variability and impact of deviations in predictions can be reduced, leading to improved overall
performance. This approach is widely used in various fields, including data science, predictive analytics, and
ML, to enhance the accuracy and reliability of forecasts and classifications.
An ensemble can be seen as the collaboration of a group of models. The process of modeling a problem
is quite complex, as it involves many entities working together. One of the main challenges is the constant
change in the state of entities. They can become part of the collaboration or leave it, as well as change their
state and adapt to new conditions. Despite various methods that can be used for ensemble modeling, it is
essential to combine them to achieve better results.
One approach to ensemble modeling is the HELENA approach proposed in [31]. Here, ensemble
modeling is defined in terms of roles and their interactions for the shared achievement of a specific goal, which
can be used as a basis for the system development process. In each system, there is a set of different states that
allow us to define the components – ensemble participants and the role they play. The state of both is
determined by the values of their attributes.
In the context of the task, ensembles are combinations of several models, and the individual forecasts
are combined in a certain way to create the final forecast.
An ensemble consists of a set of individually trained classifiers, such as neural networks or decision
trees, whose predictions are combined when classifying new instances. Each ensemble member is expected to
work collaboratively and complement each other. If the ensemble models used are complementary, the
probability of detecting an error in the prediction increases, as well as the possibility of correcting it with other
models. Instead of many traditional ML algorithms that create a single model, ensemble learning methods
generate multiple models. The ensemble passes a new example to each included model, receives their
predictions, and combines them accordingly.
Fig. 1 illustrates the overall architecture of the ensemble.
Системні дослідження в енергетиці. 2024. 4(80) 58
Fig. 1. The common ensemble architecture
As can be observed in Fig. 1, initially, the training data is divided into multiple datasets, and then several
base models are created, which can be executed sequentially or in parallel. Finally, the models are combined
on the base models [32, 33].
In general, the process of building an ensemble also involves several other steps:
1. Choosing a method by which variety is introduced into the basic models.
2. Choosing a method for combining models.
3. Selecting the type of base model.
Key concepts and components of ensemble modeling include:
Base (underlying) Models: Ensemble modeling starts with selecting base models or individual
predictive models. These base models can be of different types or variations of the same model with different
hyperparameters.
Aggregation: Ensemble methods employ various strategies to combine the predictions of base models.
There are two primary types of aggregation:
● averaging (pooling): Calculating the average (for regression tasks) or majority (for classification
tasks) of predictions made by individual models;
● weighted averaging (voting): Weighting the predictions of each model based on its effectiveness or
expertise, giving more weight to better-performing models.
Pooling is primarily used for combining numerical outputs, while voting is used for combining nominal
outputs [34]. “Voting” is the most popular method, where each base classifier votes for a specific class, and
the class with the most votes is the ensemble's prediction.
The majority voting falls into the category of ensemble combination methods, whereas ensemble selection
methods try to choose the best base models from the available set [35].
Diversity: The effectiveness of an ensemble often depends on the diversity of the base models. Different
models tend to make different types of errors, and when combined, these errors can compensate for each other,
leading to more accurate predictions. Diversity can be achieved through different model architectures, subsets
of training data, or subsets of features.
Ensemble Methods: There are several popular ensemble methods, including:
● bagging (Bootstrap Aggregating): Training multiple instances of the same base model on different
data samples;
● boosting: Boosting algorithms such as AdaBoost and Gradient Boosting combine predictions from
several weak models to create a “stronger” model;
● random forest: It combines bagging with random feature variation to create an ensemble of decision
trees;
● stacking: Combining predictions from multiple models by training a meta-model (the final ensemble
model) that learns to optimally combine their outputs.
Cross-Validation: To ensure that ensemble models generalize well to new data, cross-validation is often
used to assess their performance. Cross-validation helps prevent overfitting, especially when there are many
base models in the ensemble.
Hyperparameter Tuning: Ensemble models may have hyperparameters that require tuning to optimize
their performance. “Grid search” or “random search” is used to find the optimal hyperparameter settings for
both base models and the entire ensemble.
Системні дослідження в енергетиці. 2024. 4(80) 59
The ensemble development process involves two main stages: model training and model combination.
During the training process, each model is trained using the same training examples but with different subsets
of input features.
3. Proposed conceptual architectures of ensemble models
3.1. Architecture with voting-based aggregation
The first architecture developed by the authors and described in [9] is shown in Fig. 2.
Fig. 2. Ensemble architecture based on the voting of basic models [9]
Historical and current data on photovoltaic (PV) panels, weather data along with calendar variables, and
geographical location serve as input data for this forecasting system architecture. These data are used by the
base probabilistic models to create individual probabilistic forecasts. In the end, the forecast quantiles returned
from the base models are used as input data for the ensemble for their proper combination. During the forecast
combination stage, calendar variables and geographical location may or may not be used, differentiating the
estimation parameters. The output of the combination procedure is the probabilistic forecast of PV panel
power.
3.2. Architecture with aggregation based on averaging of base model forecasts
In the ensemble architecture with averaging, the Second-Level model (Forecast Combination in Fig. 2)
is replaced by Equation (1).
( )
( )
1 1 2 2 N N
f
1 2 N
ˆ ˆ ˆa y a y a y
ŷ
a a a
+ + +
= + + + , (1)
where f
ŷ is overall average forecast of Level 1, 1 N
ˆ ˆy , ,y are forecasts, which are given by every single
N model of Level 1, 1 2 N
a , a , , a are the weights assigned to the outputs of Level 1.
Such architecture is depicted in Fig. 3.
Fig. 3. The ensemble architecture based on averaging the base models
The contribution of each base model located at the first level is considered proportionally to its
confidence or performance. In Ensemble Architecture 1, the base models are combined in such a way that the
Системні дослідження в енергетиці. 2024. 4(80) 60
outputs of one level are treated as input data for the next level. Each individual model provides forecasts that
are then aggregated into a training dataset for the second level, where the forecast is evaluated.
In Ensemble Architecture 2, after averaging the results of the first-level base models, a second regression
model “Model k” is added, which learns the errors made by the previous models. Thus, the primary
contribution of the second-level model is to minimize the errors of the previous models.
3.3. Architecture with combined aggregation
Fig. 4 displays simplified ensemble models.
a) b)
Fig. 4. Ensemble architecture with combined aggregation:
a) simplified competition, b) simplified averaging
In Ensemble 1 (Fig. 4, a), pre-filtered data is provided as input data to the basic models.
Ensemble 2 (Fig. 4, b) provides a weighted average of basic model forecasts as PV generation. The most
suitable weights for each model are determined through trial and error. Weights that yield the lowest
forecasting error are selected.
When considering a generalized ensemble model that can be applied to any PV station, the ensemble
model with the minimum training error should be assigned to the corresponding PVs. Figure 5 depicts the
generalized ensemble model.
Fig. 5. Combined ensemble architecture with combined aggregation
The output power of a solar energy system primarily depends on direct and diffuse solar irradiation, i.e.,
the total solar irradiation received on the tilted PV panel. If the irradiation equals zero (e.g., at night), PV plant
Системні дослідження в енергетиці. 2024. 4(80) 61
generation will be zero. Therefore, when the total irradiation equals zero, the generation is set to zero without
using a forecasting model. This filter for nighttime can help reduce forecasting errors and computational time
during testing. Thus, the accuracy and computational efficiency of the ensemble model can be improved.
Composite, cascade, and hybrid ensembles deserve special attention.
Composite Ensemble (Fig. 6): A composite ensemble is a type of stacking where base models are
organized into several layers, each serving a specific purpose.
Fig. 6. Composite ensemble architecture
The first level includes base models that make individual forecasts. Subsequent levels may include
models that combine the forecasts from the first level and make further predictions (MetaModel). Complex
ensembles can be quite powerful but also more challenging to develop and train.
Cascade: Cascade ensembles are designed to filter cases that are easy to classify at the beginning of the
forecasting process, so that subsequent models can focus on more challenging cases. Each model in the cascade
is trained to be more selective than the previous one, aiming to improve overall efficiency.
Hybrid Ensembles: Hybrid ensembles combine various ensemble techniques or architectures to leverage
their strengths. For example, a hybrid ensemble may use bagging to diversify base models and then use
boosting to enhance prediction accuracy.
4. Conclusions
This study discusses various strategies for combining individual probabilistic forecasts of PV generation
to create more accurate aggregated forecasts than individual predictions.
The article analyzes conceptual approaches to understanding the concepts of an “ensemble” and
“ensemble forecasting”. It is noted that there is no consensus among researchers regarding a specific approach
to forming an ensemble model. The choice of architecture depends on the specific task, data characteristics,
and trade-offs between model complexity and effectiveness. Ensemble methods are a powerful tool in the field
of machine learning and can significantly improve forecasting accuracy compared to individual models.
The synthesis of various approaches to ensemble models has led to the development of conceptual
architectures that utilize well-known aggregation methods and their combinations, which can be
complemented by a weather parameter classification block. The application of these architectures can
contribute to solving current challenges in integrating PV energy into the energy system with high efficiency.
Author Contributions.
The author contributed to the conceptualization, methodology, formal analysis, writing, and final
approval of the manuscript.
Funding.
The research was conducted with the support of the National Academy of Sciences of Ukraine within
the framework of the scientific project "Development of a system of mathematical models for long-term
forecasting of the consumption of major types of fuel and energy resources in the national economy, taking
into account current environmental constraints".
Системні дослідження в енергетиці. 2024. 4(80) 62
References
1. Lytvynchuk, V., Kaplin, M., & Karmazin, O. (2023). Analysis of the impact of distributed generation of renewable
energy sources on the reserve of under frequency load shedding. Vidnovluvana Energetika, 3(70), 19–27 [in
Ukrainian]. https://doi.org/10.36296/1819-8058.2022.3(70)
2. Maliarenko, O., Ivanenko, N., & Sudarykov, O. (2023). Study of the relationship of environmental and energy
efficiency indicators at the country level. System Research in Energy, 4(75), 84–94 [in Ukrainian].
https://doi.org/10.15407/srenergy2023.04.084
3. Stanytsina, V., Zaporozhets, A., & Artemchuk, V. (2024). Demand forecasting mathematical models for residential
electricity consumption considering ambient temperature. In A. Zagorodny, V. Bogdanov, & A. Zaporozhets (Eds.),
Nexus of Sustainability. Studies in Systems, Decision and Control, 559 (pp. 127–145). Springer, Cham.
https://doi.org/10.1007/978-3-031-66764-0_6
4. Zaporozhets, A., Babak, V., Sverdlova, A., Scherbak, L., & Kuts, Y. (2022). Review of the state of air pollution by
energy objects in Ukraine. System Research in Energy, 2(71), 42–52. https://doi.org/10.15407/srenergy2022.02.042
5. Chou, J.-S., & Tran, D.-S. (2018). Forecasting energy consumption time series using machine learning techniques
based on usage patterns of residential householders. Energy, 165, 709–726.
https://doi.org/10.1016/j.energy.2018.09.144
6. Denysov, V., Kostenko, G., Babak, V., Shulzhenko, S., & Zaporozhets, A. (2023). Accounting the forecasting
stochasticity at the power system modes optimization. In A. Zaporozhets (Ed.), Systems, decision and control in
energy V, 481 (pp. 43–55). Springer, Cham. https://doi.org/10.1007/978-3-031-35088-7_3
7. Kostenko, G., Zaporozhets, A., Zaporozhets, N., & Verpeta, V. (2024). Aspects of integrating renewable distributed
generation into the energy supply system of Ukraine. The Problems of Economy, 2(60), 83–93 [in Ukrainian].
https://doi.org/10.32983/2222-0712-2024-2-83-93
8. Babak, V.P., Babak, S.V., Eremenko, V.S., Kuts, Y.V., Myslovych, M.V., Scherbak, L.M., & Zaporozhets, A.O.
(2021). Models and measures for the diagnosis of electric power equipment. Models and measures in measurements
and monitoring. Studies in Systems, Decision and Control, 360 (pp. 99–126). Springer, Cham.
https://doi.org/10.1007/978-3-030-70783-5_4
9. Lund, H., Østergaard, P. A., Connolly, D., & Mathiesen, B. V. (2017). Smart energy and smart energy systems.
Energy, 137, 556–565. https://doi.org/10.1016/j.energy.2017.05.123
10. Zaporozhets, A., Babak, V., Popov, O., Scherbak, L., & Kuts, Yu. (2023). Methodological support of air pollution
monitoring system. In A. Zaporozhets (Ed.), Systems, decision and control in energy V, 481 (pp. 713–737). Springer,
Cham. https://doi.org/10.1007/978-3-031-35088-7_41
11. Matushkin, D., & Bosak, A. (2023). Concept of an ensemble forecasting system for optimization problems of control
of solar microgrid. Power Engineering: Economics, Technique, Ecology, 3(73), 23–28.
https://doi.org/10.20535/1813-5420.3.2023.289651
12. Kulyk, M., & Zgurovets, O. (2020). Modeling of power systems with wind, solar power plants and energy storage.
In V. Babak, V. Isaienko, & A. Zaporozhets (Eds.), Systems, decision and control in energy I. Studies in Systems,
Decision and Control, 298 (pp. 231–245). Springer, Cham. https://doi.org/10.1007/978-3-030-48583-2_15
13. Lund, H., Mathiesen, B., Connolly, D., & Østergaard, P. (2014). Renewable energy systems—A smart energy
systems approach to the choice and modelling of 100% renewable solutions. Chemical Engineering Transactions,
39, 1–6. https://doi.org/10.3303/CET1439001
14. Shulzhenko, S., Nechaieva, T., & Leshchenko, I. (2024). The application of the optimal unit commitment problem
for the studies of the national power sector development under system risks. In A. Zagorodny, V. Bogdanov, &
A. Zaporozhets (Eds.), Nexus of Sustainability. Studies in Systems, Decision and Control, 559 (pp. 147–164).
Springer, Cham. https://doi.org/10.1007/978-3-031-66764-0_7
15. van der Meer, D., Widén, J., & Munkhammar, J. (2017). Review on probabilistic forecasting of photovoltaic power
production and electricity consumption. Renewable and Sustainable Energy Reviews, 81, 1484–1512.
https://doi.org/10.1016/j.rser.2017.05.212
16. Zgurovets, O., & Kulyk, M. (2021). Comparative analysis and recommendations for the use of frequency regulation
technologies in integrated power systems with a large share of wind power plants. In A. Zaporozhets, &
V. Artemchuk (Eds.), Systems, decision and control in energy II. Studies in Systems, Decision and Control, 346
(pp. 81–99). Springer, Cham. https://doi.org/10.1007/978-3-030-69189-9_5
17. Derii, V., Teslenko, O., Lenchevsky, E., Denisov, V., & Maistrenko, N. (2023). Prospects and energy-economic
indicators of heat energy production through direct use of electricity from renewable sources in modern heat
generators. In A. Zaporozhets (Ed.), Systems, decision and control in energy IV, 454 (pp. 451–463). Springer, Cham.
https://doi.org/10.1007/978-3-031-22464-5_27
18. Ren, Y., Suganthan, P. S., & Narasimalu, S. (2015). Ensemble methods for wind and solar power forecasting – A
state-of-the-art review. Renewable and Sustainable Energy Reviews, 50, 82–91.
https://doi.org/10.1016/j.rser.2015.04.081
19. Muralitharan, K., Sakthivel, R., & Vishnuvarthan, R. (2017). Neural network based optimization approach for
energy demand prediction in smart grid. Neurocomputing, 273, 199–208.
https://doi.org/10.1016/j.neucom.2017.08.017
https://doi.org/10.36296/1819-8058.2022.3(70).19-27
https://doi.org/10.1007/978-3-031-35088-7_3
https://doi.org/10.32983/2222-0712-2024-2-83-93
https://doi.org/10.1007/978-3-030-69189-9_5
Системні дослідження в енергетиці. 2024. 4(80) 63
20. Alobaidi, M., Chebana, F., & Meguid M. (2018). Robust ensemble learning framework for day-ahead forecasting
of household based energy consumption. Applied Energy, 212, 997–1012.
https://doi.org/10.1016/j.apenergy.2017.12.054
21. Mangalova, E., & Agafonov, E. (2013). Wind power forecasting using the k-nearest neighbors algorithm.
International Journal of Forecasting, 30(2), 402–406. https://doi.org/10.1016/j.ijforecast.2013.07.008
22. Shang, C., & Wei, P. (2018). Enhanced support vector regression based forecast engine to predict solar power output.
Renewable Energy, 127, 269–283. https://doi.org/10.1016/j.renene.2018.04.067
23. Tato, J. H., & Brito, M. C. (2018). Using smart persistence and random forests to predict photovoltaic energy
production. Energies, 12(1), 100, 1–15. https://doi.org/10.3390/en12010100
24. Bracale, A., Carpinelli, G., De Falco, P., & Hong, T. (2019). Short-term industrial reactive power forecasting.
International Journal of Electrical Power & Energy Systems, 107, 177–185.
https://doi.org/10.1016/j.ijepes.2018.11.022
25. Wilson, G. T. (2016). Time series analysis: forecasting and control, 5th edition, by George E.P. Box, G. M. Jenkins,
G. C. Reinsel, & G. M. Ljung, 2015. Published by John Wiley and Sons Inc., Hoboken, New Jersey (p. 712). Journal
of Time Series Analysis, 37(5), 631–633. https://doi.org/10.1111/jtsa.12194
26. Zhang, P. (2003). Time series forecasting using a hybrid ARIMA and neural network model. Neurocomputing, 50,
159–175. https://doi.org/10.1016/S0925-2312(01)00702-0
27. Voyant, C., Notton, G., Kalogirou, S., Nivet, M.-L., Paoli, Ch., Motte, F., & Fouilloy, A. (2017). Machine learning
methods for solar irradiation forecasting: A review. Renewable Energy, 105, 569–582.
https://doi.org/10.1016/j.renene.2016.12.095
28. Guermoui, M., Melgani, F., Gairaa, K., & Mekhalfi, M. (2020). A comprehensive review of hybrid models for solar
irradiation forecasting. Journal of Cleaner Production, 258, 120357. https://doi.org/10.1016/j.jclepro.2020.120357
29. Bracale, A., Carpinelli, G., & De Falco, P. (2016). A probabilistic competitive ensemble method for short-term
photovoltaic power forecasting. IEEE Transactions on Sustainable Energy, 7, 3, 953–961.
https://doi.org/10.1109/TSTE.2016.2610523
30. Wang, Y., Zhang, N., Tan, Y., Hong, T., Kirschen, D. S., & Kang, C. (2018). Combining probabilistic load forecasts.
IEEE Transactions on Smart Grid, 9(4), 3861–3870. https://doi.org/10.1109/TSG.2018.2833869
31. Hennicker, R., & Klarl, A. (2014). Foundations for ensemble modelling – The Helena approach. Lecture Notes in
Computer Science: Specification, Algebra, and Software, 8373, 359–381.
32. Diagne, M., David, M., Lauret, P., Boland, J., & Schmutz, N. (2013). Review of solar irradiance forecasting methods
and a proposition for small-scale insular grids. Renewable and Sustainable Energy Reviews, 27, 65–76.
https://doi.org/10.1016/j.rser.2013.06.042
33. Arora, I., Gambhir, J., & Kaur, T. (2020). Solar irradiance forecasting using decision tree and ensemble models.
Proceedings of the IEEE International Conference on Industrial Informatics, 654–659.
https://doi.org/10.1109/ICIRCA48905.2020.9182876
34. Zhou, Z.-H. (2012). Ensemble methods: Foundations and algorithms. CRC Press (p. 350).
https://doi.org/10.1201/b12207
35. Way, M., Scargle, J. D., Ali, K. M., & Srivastava, A. N. (2012). Advances in machine learning and data mining for
astronomy. CRC Press (p. 325). https://doi.org/10.1201/b11822
МОДЕЛІ ПРОГНОЗУВАННЯ ФОТОЕЛЕКТРИЧНОЇ
ГЕНЕРАЦІЇ: КОНЦЕПТУАЛЬНІ АНСАМБЛЕВІ АРХІТЕКТУРИ
Дмитро Матушкін, д-р філософії (PhD), https://orcid.org/0000-0003-4431-7862
Інститут загальної енергетики НАН України, вул. Антоновича, 172, м. Київ, 03150, Україна
e-mail: dmitry.matushkin@ukr.net
Анотація. Рішення щодо регулювання потужності, планування енергетичних ресурсів та
інтеграції «зеленої» енергії в електричну мережу залежать від точних імовірнісних прогнозів.
Однією з потенційних стратегій підвищення точності прогнозів є використання методів
ансамблевого прогнозування. Це підхід, при якому кілька моделей працюють спільно, досягаючи
кращих результатів порівняно з тим, що може забезпечити окрема модель. Ці методи
поділяються на дві основні категорії: конкурентні та співпрацюючі ансамблі. Конкурентні
ансамблі використовують різноманітність параметрів і даних для створення багатої групи
базових моделей. Такий підхід може включати статистичний аналіз, фільтрацію шумів та
усунення аномалій. З іншого боку, співпрацюючі ансамблі базуються на взаємодії між моделями для
досягнення кращих результатів. До цих методів належать стратегії, такі як зважені прогнози,
голосування, агрегування та комбінування результатів моделей. Дослідження методів
ансамблевого прогнозування у контексті фотоелектричної генерації є надзвичайно актуальним,
оскільки сонячна енергія представляє важливе джерело відновлюваної енергії. Точні прогнози
Системні дослідження в енергетиці. 2024. 4(80) 64
виробництва сонячної енергії вирішують проблеми, пов’язані з ефективним використанням
фотоелектричних панелей та їх інтеграцією в загальну енергетичну систему. У даній статті
досліджуються концептуальні архітектури ансамблів для прогнозування фотоелектричної енергії.
Ці архітектури охоплюють різні методи агрегування базових моделей у межах ансамблю, що
дозволяє враховувати різні аспекти та особливості сонячних даних, такі як інтенсивність
сонячного випромінювання, метеорологічні умови, географічні фактори тощо. Ці концептуальні
моделі розроблені на основі перевірених методів статистики, машинного навчання та штучного
інтелекту. Таким чином, дана робота надає огляд методів ансамблевого прогнозування для
відновлюваної енергії, охоплюючи конкурентні та співпрацюючі ансамблі, а також розробку
концептуальних моделей для прогнозування сонячної енергії. Це дослідження має на меті
підвищити точність і ефективність прогнозів у сфері відновлюваної енергетики, представляє
важливий крок у розвитку стійкого та екологічно чистого виробництва енергії.
Ключові слова: ймовірнісний сонячний прогноз, ансамблева модель, комбінація прогнозів,
конкурентні ансамблі, співпрацюючі ансамблі, концептуальні моделі.
Надійшла до редколегії: 08.10.2024
|
| id | systemreorg-article-870 |
| institution | System Research in Energy |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2026-07-19T01:23:10Z |
| publishDate | 2024 |
| publisher | General Energy Institute of the National Academy of Sciences of Ukraine |
| record_format | ojs |
| resource_txt_mv | systemreorg/ff/0da831ad57478127c44c359c38a023ff.pdf |
| spelling | systemreorg-article-8702026-07-18T12:57:49Z PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES Моделі прогнозування фотоелектричної генерації: концептуальні ансамблеві архітектури Matushkin, Dmytro probabilistic solar forecasting, ensemble model, forecast combination, competitive ensembles, collaborative ensembles, conceptual models. ймовірнісний сонячний прогноз, ансамблева модель, комбінація прогнозів, конкурентні ансамблі, співпрацюючі ансамблі, концептуальні моделі. The decisions regarding power regulation, energy resource planning, and integrating “green” energy into the electrical grid hinge on precise probabilistic forecasts. One of the potential strategies to enhance forecast accuracy is the utilization of ensemble forecasting methods. They represent an approach where multiple models collaborate to achieve superior results compared to what a single model could produce independently. These methods can be categorized into two main categories: competitive and collaborative ensembles. Competitive ensembles harness the diversity of parameters and data to create a rich pool of base models. This approach may encompass statistical analysis, noise filtering, and anomaly elimination. On the other hand, collaborative ensembles rely on the interaction among models to achieve better outcomes. These methods encompass strategies such as weighted predictions, voting, aggregation, and a combination of model results. The research of ensemble forecasting methods in the context of photovoltaic generation is highly relevant, as solar energy represents a crucial source of renewable energy. Accurate predictions of solar energy production address the challenges related to the efficient utilization of photovoltaic panels and their integration into the overall energy system. This paper investigates conceptual ensemble architectures for photovoltaic energy forecasting. These architectures encompass various methods of aggregating base models within an ensemble, allowing for the consideration of different aspects and peculiarities of solar data, such as solar irradiation intensity, meteorological conditions, geographic factors, and more. These conceptual models are developed based on well-established statistical, machine learning, and artificial intelligence methods. Therefore, this paper provides an overview of ensemble forecasting methods for renewable energy, covering competitive and collaborative ensembles, as well as developing conceptual models for solar energy forecasting. This work aims to elevate the accuracy and efficiency of forecasts in the realm of renewable energy, representing a significant step in the advancement of sustainable and environmentally friendly energy production. Рішення щодо регулювання потужності, планування енергетичних ресурсів та інтеграції «зеленої» енергії в електричну мережу залежать від точних імовірнісних прогнозів. Однією з потенційних стратегій підвищення точності прогнозів є використання методів ансамблевого прогнозування. Це підхід, при якому кілька моделей працюють спільно, досягаючи кращих результатів порівняно з тим, що може забезпечити окрема модель. Ці методи поділяються на дві основні категорії: конкурентні та співпрацюючі ансамблі. Конкурентні ансамблі використовують різноманітність параметрів і даних для створення багатої групи базових моделей. Такий підхід може включати статистичний аналіз, фільтрацію шумів та усунення аномалій. З іншого боку, співпрацюючі ансамблі базуються на взаємодії між моделями для досягнення кращих результатів. До цих методів належать стратегії, такі як зважені прогнози, голосування, агрегування та комбінування результатів моделей. Дослідження методів ансамблевого прогнозування у контексті фотоелектричної генерації є надзвичайно актуальним, оскільки сонячна енергія представляє важливе джерело відновлюваної енергії. Точні прогнози виробництва сонячної енергії вирішують проблеми, пов’язані з ефективним використанням фотоелектричних панелей та їх інтеграцією в загальну енергетичну систему. У даній статті досліджуються концептуальні архітектури ансамблів для прогнозування фотоелектричної енергії. Ці архітектури охоплюють різні методи агрегування базових моделей у межах ансамблю, що дозволяє враховувати різні аспекти та особливості сонячних даних, такі як інтенсивність сонячного випромінювання, метеорологічні умови, географічні фактори тощо. Ці концептуальні моделі розроблені на основі перевірених методів статистики, машинного навчання та штучного інтелекту. Таким чином, дана робота надає огляд методів ансамблевого прогнозування для відновлюваної енергії, охоплюючи конкурентні та співпрацюючі ансамблі, а також розробку концептуальних моделей для прогнозування сонячної енергії. Це дослідження має на меті підвищити точність і ефективність прогнозів у сфері відновлюваної енергетики, представляє важливий крок у розвитку стійкого та екологічно чистого виробництва енергії. General Energy Institute of the National Academy of Sciences of Ukraine 2024-11-28 Article Article application/pdf https://systemre.org/index.php/journal/article/view/870 10.15407/srenergy2024.04.056 System Research in Energy; No. 4 (80) (2024): System Research in Energy; 56-64 Системні дослідження в енергетиці; № 4 (80) (2024): Системні дослідження в енергетиці; 56-64 2786-7102 2786-7633 en https://systemre.org/index.php/journal/article/view/870/777 Copyright (c) 2024 Dmytro Matushkin https://creativecommons.org/publicdomain/zero/1.0 |
| spellingShingle | probabilistic solar forecasting ensemble model forecast combination competitive ensembles collaborative ensembles conceptual models. Matushkin, Dmytro PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES |
| title | PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES |
| title_alt | Моделі прогнозування фотоелектричної генерації: концептуальні ансамблеві архітектури |
| title_full | PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES |
| title_fullStr | PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES |
| title_full_unstemmed | PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES |
| title_short | PHOTOVOLTAIC GENERATION FORECASTING MODELS: CONCEPTUAL ENSEMBLE ARCHITECTURES |
| title_sort | photovoltaic generation forecasting models: conceptual ensemble architectures |
| topic | probabilistic solar forecasting ensemble model forecast combination competitive ensembles collaborative ensembles conceptual models. |
| topic_facet | probabilistic solar forecasting ensemble model forecast combination competitive ensembles collaborative ensembles conceptual models. ймовірнісний сонячний прогноз ансамблева модель комбінація прогнозів конкурентні ансамблі співпрацюючі ансамблі концептуальні моделі. |
| url | https://systemre.org/index.php/journal/article/view/870 |
| work_keys_str_mv | AT matushkindmytro photovoltaicgenerationforecastingmodelsconceptualensemblearchitectures AT matushkindmytro modelíprognozuvannâfotoelektričnoígeneracííkonceptualʹníansamblevíarhítekturi |