DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING
In energy systems, it is crucial to forecast solar energy generation for optimization of operations and to mitigate the impact of uncertainty. Forecasting solar energy involves predicting solar irradiance, for which historical solar irradiance and weather parameter data are typically required. Howev...
Gespeichert in:
| Datum: | 2023 |
|---|---|
| Hauptverfasser: | , |
| Format: | Artikel |
| Sprache: | Ukrainisch |
| Veröffentlicht: |
Institute of Renewable Energy National Academy of Sciences of Ukraine
2023
|
| Schlagworte: | |
| Online Zugang: | https://ve.org.ua/index.php/journal/article/view/408 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Назва журналу: | Vidnovluvana energetika |
| Завантажити файл: |  |
Institution
Vidnovluvana energetika| _version_ | 1871103657407152128 |
|---|---|
| author | Matushkin, D. Bosak, A. |
| author_facet | Matushkin, D. Bosak, A. |
| author_institution_txt_mv | [
{
"author": "D. Matushkin",
"institution": "National Technical University of Ukraine «Igor Sikorsky Kyiv Polytechnic Institute», Kyiv, Ukraine."
},
{
"author": "A. Bosak",
"institution": "National Technical University of Ukraine «Igor Sikorsky Kyiv Polytechnic Institute», Kyiv, Ukraine."
}
] |
| author_sort | Matushkin, D. |
| baseUrl_str | https://ve.org.ua/index.php/journal/oai |
| collection | OJS |
| datestamp_date | 2026-07-18T06:32:18Z |
| description | In energy systems, it is crucial to forecast solar energy generation for optimization of operations and to mitigate the impact of uncertainty. Forecasting solar energy involves predicting solar irradiance, for which historical solar irradiance and weather parameter data are typically required. However, such data are often unavailable for residential and commercial solar microgrids. This study proposes an hourly forecasting model for next-day Solar Power Production (SPP) that doesn't rely on historical solar irradiance data. The forecasting is performed using deep learning techniques like Long Short-Term Memory (LSTM), and the results are integrated into the MATLAB&Simulink simulation platform. A graphical user interface (GUI) within the MATLAB&Simulink software complex is presented as a simulation platform for hourly SPP forecasting. This platform serves as a useful tool for researchers studying energy management and forecasting as well as planning renewable energy-based energy system operations. The study involves predicting the amount of energy generated by Solar Power Production (SPP). The developed GUI was employed to forecast the SPP output power over test days. The forecasted data were then evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Experimental results demonstrated reasonable accuracy, with the proposed model achieving an RMSE of 0.835 W and an MAE of 0.353 W in certain datasets. |
| doi_str_mv | 10.36296/1819-8058.2023.3(74).32-41 |
| first_indexed | 2025-07-17T11:39:07Z |
| format | Article |
| fulltext |
32
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
УДК 621.31:621.311.243 https://doi.org/10.36296/1819-8058.2023.3(74)32-41
DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING
Received Sept. 13, 2023; accepted Sept. 20, 2023
Available online Sept. 30, 2023
D. Matushkin1, A. Bosak2
Author for correspondence: Dmytro Matushkin,
e-mail: dm.mtkn@lll.kpi.ua
In energy systems, it is crucial to forecast solar energy gener-
ation for optimization of operations and to mitigate the im-
pact of uncertainty. Forecasting solar energy involves predict-
ing solar irradiance, for which historical solar irradiance and weather parameter data are typically required.
However, such data are often unavailable for residential and commercial solar microgrids. This study proposes an
hourly forecasting model for next-day Solar Power Production (SPP) that doesn't rely on historical solar irradiance
data. The forecasting is performed using deep learning techniques like Long Short-Term Memory (LSTM), and the
results are integrated into the MATLAB&Simulink simulation platform. A graphical user interface (GUI) within the
MATLAB&Simulink software complex is presented as a simulation platform for hourly SPP forecasting. This plat-
form serves as a useful tool for researchers studying energy management and forecasting as well as planning
renewable energy-based energy system operations. The study involves predicting the amount of energy generated
by Solar Power Production (SPP). The developed GUI was employed to forecast the SPP output power over test
days. The forecasted data were then evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Error
(MAE). Experimental results demonstrated reasonable accuracy, with the proposed model achieving an RMSE of
0.835 W and an MAE of 0.353 W in certain datasets.
Keywords: short-term forecasting, solar power plant, deep learning, Long Short-Term Memory, graphical user
interface, MATLAB&Simulink simulation.
РОЗРОБКА ГРАФІЧНОГО ІНТЕРФЕЙСУ ДЛЯ КОРОТКОСТРОКОВОГО ПРОГНОЗУВАННЯ
СОНЯЧНОЇ ГЕНЕРАЦІЇ НА ОСНОВІ ГЛИБОКОГО НАВЧАННЯ
Отримано 13 вер. 2023 р.; рекомендовано до публікації 20 вер. 2023 р.
Доступно онлайн 30 вер. 2023 р.
Д. С. Матушкін1, А. В. Босак2
Автор для коресподенції: Матушкін Дмитро,
e-mail: dm.mtkn@lll.kpi.ua
В енергосистемах важливо прогнозувати генерацію со-
нячної енергії для оптимізації роботи та зменшення
впливу невизначеності. Для прогнозування сонячної енергії необхідно прогнозувати сонячне ви-
промінювання, для чого зазвичай потрібні історичні дані про сонячне випромінювання та погодні па-
раметри. Ці дані часто недоступні для домашніх та комерційних сонячних мікромереж. У цьому до-
слідженні пропонується модель погодинного прогнозування генерації SPP на наступний день, яка не
залежить від історичних даних про сонячне випромінювання. Прогнозування виконується за допомо-
гою інструмента глибокого навчання, як-от Long Short-Term Memory (LSTM), а результати подаються
в систему на платформі симуляції MATLAB&Simulink. Як платформу симуляції представлено
графічний інтерфейс користувача (GUI) в програмному комплексі MATLAB&Simulink для прогнозування
погодинної потужності SPP. Ця платформа буде корисним і практичним механізмом для дослідників,
які вивчають енергоменеджмент та для прогнозування й планування режимів роботи енергоси-
стеми з відновлюваними джерелами енергії (ВДЕ). У цьому дослідженні виконано прогноз кількості
1 PhD student.
https://orcid.org/0000-0003-4431-7862
2 Cand. of tech. Sciences., Assoc. Prof.
https://orcid.org/0000-0003-0545-9980
1, 2 National Technical University of Ukraine
«Igor Sikorsky Kyiv Polytechnic Institute», Kyiv,
Ukraine.
1 аспірант.
https://orcid.org/0000-0003-4431-7862
2 канд. техн. наук, доцент.
https://orcid.org/0000-0003-0545-9980
1, 2 НТУУ «Київський політехнічний інститут
ім. Ігоря Сікорського», м. Київ, Україна
33
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
енергії, яка буде згенерована SPP. Розроблений GUI використовувався для прогнозування вихідної пот-
ужності SPP протягом тестових днів. Потім отримані прогнозні дані були оцінені за середньо-квад-
ратичною похибкою (RMSE) і середньою абсолютною похибкою (MAE). Результати експериментів по-
казали достатні показники точності. Запропонована модель досягла RMSE у розмірі 0,835 W і MAE у
розмірі 0,353 W у деяких наборах даних.
Ключові слова: Короткострокове прогнозування, сонячна електростанція, глибоке навчання, довга
короткочасна пам'ять, графічний інтерфейс користувача, симуляція в MATLAB&Simulink.
List of used designations and abbreviations
SES – solar energy systems
ES – Energy System
VRE – variable renewable energy
GUI – graphical user interface
GSR – global solar radiation
SPP – Solar Photovoltaic Power
DL – Deep Learning
ML – Machine Learning
LSTM-RNN – Long Short-Term Memory Recurrent Neural
Networks
Introduction. Photovoltaic generation exhibits a rather
unstable nature due to its dependence on solar energy,
which is influenced by solar radiation and meteorological
parameters such as air temperature, humidity, wind speed,
etc [1, 2]. These destabilizing fluctuations pose significant
challenges for the integration of Solar Energy Systems (SES)
into the Energy System (ES). However, the use of Variable
Renewable Energy (VRE) contributes to reducing the
energy need for balancing and regulating capacity. SES
generation is highly flexible, allowing it to be adjusted to
adapt to changes in energy demand.
The VRE forecasting significantly impacts various
operations for management of the ES, including planning,
dispatch, real-time balancing, and reserve requirements
for the ES. By incorporating forecasts from local
generators, ES operators can predict rapid VRE changes,
enabling them to economically balance consumption and
scheduled generation on a daily and intraday basis.
In particular, demand-side management [3-4] and generation
planning [5-6] become more critical due to research into
Smart Microgrids [6-8]. Effective demand-side management
and grid operation planning are directly related to the
assessment of solar energy. Therefore, the evaluation of solar
energy [8-10] becomes a crucial requirement for both grid
operation planning [9-10] and research into energy
generation planning a day-ahead [11-12].
The volumes of input data and forecast models for different
components can vary. This necessitates the development
of an adaptive forecasting and planning model system.
It is worth noting that this problem is not reliably and
precisely solved. Numerous algorithms and software
complexes are proposed, new software products are being
developed, but there are no widely accepted “industry
standards” for VRE forecasting.
There are two main approaches to solar energy forecasting:
direct and indirect. Direct methods obtain solar energy
directly by forecasting [13]. In indirect methods, solar
irradiation is first forecasted and then converted into solar
energy through mathematical relationships [14-15]. This
method is essential for planners and researchers who lack
solar energy data. In this case, solar irradiation data can be
used for indirect solar energy forecasting.
Graphical user interface (GUI) tools allow for solving solar
energy forecasting and planning problems using
straightforward methods. So far, several GUI tools have
been developed, such as the monthly global solar radiation
forecasting model [16], a comprehensive photovoltaic
simulation model [17], a model for forecasting photovoltaic
power for the day ahead [18], and a model for studying the
efficiency of a solar tower power plant [19].
In this study, a MATLAB GUI was developed based on
existing user interface designs presented in the literature
for forecasting photovoltaic power.
This research can be summarized as follows:
− Solar energy data for the day ahead were forecasted
using 10-minute global solar radiation (GSR) data from
July 1, 2020, to December 31, 2020;
− The developed MATLAB GUI model allows the user to
allow the user to forecast the generated Solar
Photovoltaic Power (SPP) in graphical form.
Various methodologies for generating probabilistic solar
energy forecasts are extensively discussed in the literature.
For instance, the work discusses nearest neighbor methods
[20], vector autoregressive models [21], volatility
estimation methods [22], and ensemble models [23].
Additionally, there are examples of probabilistic solar
energy forecasts using Deep Learning (DL) methods [24].
One of the main advantages of the latter is their ability to
extract simple features from high-dimensional, complex
data [25], making them suitable for forecasting tasks.
Furthermore, DL models are the most suitable approach for
forecasting solar irradiation, especially when dealing with
complex and large datasets. DL models have been
successfully applied in various domains, including image
processing, classification, and forecasting, due to their
34
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
ability to effectively handle complex data without the need
for expert evaluation. For instance, a DL model presented
in [26] is used for short-term wave energy forecasting.
Similarly, DL models are widely employed for various
forecasting tasks such as wind speed [27], PV power [28],
solar irradiation [29, 43], and more. The known issues
associated with conventional neural networks (CNN), such
as gradient vanishing and training complexity, can be easily
addressed using DL networks. For time series forecasting, a
complex neural network is developed in [30], while deep
learning is used for forecasting solar irradiation at 30
locations in Turkey in [31]. Hence, DL models are more
accurate compared to Machine Learning (ML) models and
empirical models in terms of accuracy.
Formulation of the problem. The primary objective of this
article is to address the necessity of reliable SESs by
ensuring the availability of meteorological weather data for
the region where these systems are deployed. Many
countries have developed their own forecasting models,
which serve as valuable tools for energy planning (EP).
However, in Ukraine, there is still a shortage of EP models
based on solar irradiation. This work represents the initial
attempt to create a GSR model for Solar Photovoltaic
Power (SPP) located in the village of Velyka Dymerka in the
Kyiv region of Ukraine. Furthermore, this research
endeavors to provide a user-friendly interface for the
utilization of the developed models.
The central focus of this study is the development of a user
interface within MATLAB & Simulink for the purpose of
forecasting SPP generation. This interface is designed to be
easily accessible and usable by individuals working in the
fields of SPP development and performance evaluation.
Within this article, we propose the utilization of deep Long
Short-Term Memory Recurrent Neural Networks (LSTM-
RNN) to forecast solar irradiation for the upcoming day.
This research represents a preliminary experiment aimed
at establishing a starting point for the creation of intelligent
solar applications.
Deep Learning. Deep learning (DL) is a subset of the broader
family of Machine Learning (ML) methods that employ multiple
layers of processing to learn data with several levels of
abstraction [32]. It uncovers complex structures within large
datasets using the backpropagation algorithm to indicate how
the machine should adjust its internal parameters used for
computation at each layer from the representation at the
previous level [33]. Architectures within DL encompass
Feedforward Neural Networks (FFNN), Recurrent Neural
Networks (RNN), Deep Belief Networks (DBN), and Restricted
Boltzmann Machines (RBM). Among these architectures, FFNN
and RNN are the most widely used. Convolutional neural
networks (CNN), which are a subset of FFNN, excel at processing
images, videos, and audio. Deep Long Short-Term Memory
networks (LSTM), which belong to the RNN family, are well-
suited for sequential data like text, language, and time series. DL
has achieved significant success across various domains. It has
outperformed other ML methods in image recognition [34, 35],
speech recognition [36, 37], natural language understanding
[38], language translation [39, 40], particle accelerator data
analysis [41], brain circuit reconstruction [42], etc. DL has
achieved significant success across various domains. It has
outperformed other ML methods in image recognition [34, 35],
speech recognition [36, 37], natural language understanding
[38], language translation [39, 40], particle accelerator data
analysis [41], brain circuit reconstruction [42], and more.
Long Short Term Memory Recurrent Neural Network
(LSTM-RNN). In this research, an LSTM-RNN network was
employed for forecasting solar power generation in a SPP.
The LSTM-RNN method is a subset of DL, developed to
address the issues of vanishing and exploding gradients
inherent in simple RNNs [44]. This is achieved through the
incorporation of memory cells within its gating mechanism
system. Consequently, it generates an additive gradient as
opposed to a vanishing gradient, offering a larger gradient
magnitude to facilitate the training of LSTM-RNN cells [45].
The cells of LSTM-RNN have three key functions: write to
memory, read from memory, and reset memory. LSTM-
RNN is closely tied to recurrent gating units, known as
“forget gates” [46]. These gates mitigate the challenges of
vanishing and exploding gradients during backpropagation
by allowing errors to propagate across multiple layers. In
essence, LSTM-RNN possesses the capability to learn tasks
that require memory of previously learned information,
occurring over a certain time span.
The functioning of LSTM-RNN occurs in four steps. In the
first step (Fig. 1), the “forget gate” algorithm is executed
which determines how much previous information it
should retain. Here, new and previous data presented in
vector form are passed through a sigmoid function. The
sigmoid function confines values between 0 and 1. If the
output is 0, the memory cell forgets the previous data. If
the output is 1, the data is retained.
Fig. 1. Forget Gate Layer
This step is described by the expression (1) [47, 48].
( )1−= ⋅ + ⋅ +t xf t hf t ff σ W x W h b , (1)
where σ denotes the sigmoid activation function., xfW is
the weight matrix for input tx , tx is the memory cell
matrix of the input vector over time t , hfW – weight
matrix for the inputs t-1h , t-1h is matrix of previous states,
and fb is matrix of shift vector.
At the second step (Fig. 2), the “input gate” layer
determines how much of this unit should be added to the
current state and determines which value will be updated.
35
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
Fig. 2. Input Gate Layer
The input gate ti defines which data is stored in the new
candidate state of the cell tC [47, 48]:
( )− −= ⋅ + ⋅ ⋅ + ⋅ + t t t 1 t xc t hc t 1 cC f c i tanh W x W h b (2)
( )−= ⋅ + ⋅ +t xi t hi t 1 ii σ W x W h b (3)
It has two operational functions: sigmoid and hyperbolic
tangent (tanh). Tanh is also an activation function. It is used
to control values processed by the network, transforming
them into a close interval to ensure they always fall within
the range of [ ]-1;1 . Both the vectors of previous and new
data are provided to both functions. The output results are
multiplied together (Fig. 3), and the output is passed to the
cell state, updating the value of the cell state.
Fig. 3. Updating Cell State
The new candidate state of the cell tC and the previous
candidate state of the cell t-1C are used to update the last
state of the cell tC . This step is described by the expression
(4) [47, 48].
−= ⋅ + ⋅t t t 1 t tC f C i C (4)
The final step is the output layer algorithm (Fig. 4).
Fig. 4. Output Gate layer
It determines which part of the cell state will be passed to
the output. Vector values are inputted into the sigmoid
function, and the sigmoid's output is then passed to the
tanh activation function to determine the network's
output. The models consist of layers of neural networks
with activation functions: the sigmoid function and the
tanh function, which are directly associated with the
structure of logical gating mechanisms based on memory
to address the vanishing and exploding gradient problem.
The "Output Gate" controls the output of the cell and
merges it with its activated state using the tanh function to
determine the final output th which is expressed in
expressions (5) and (6) [47, 48].
( )= +t t th o tanh C (5)
( )−= ⋅ + ⋅ +t xo t ho t 1 oo σ W x W h b (6)
When the forward pass is completed at time step t=T ,
which is when the last weight coefficient of the sequence
is reached, the reverse pass known as Back-Propagation
Through Time (BPTT) is initiated. The error gradient is
computed iteratively until t=1 , at this stage, the weights
of the LSTM network are updated iteratively using an
optimization method known as "gradient descent
optimization technique." This training procedure entails
adjusting the weights in a way that minimizes the error
function. In this study, the root mean square error (RMSE)
was utilized, where ( yk ) represents the target generation
and, accordingly, the output of LSTM-RNN for each training
sample ( k ). The cost of the error function is evaluated
based on the network's performance on the data to be
predicted. It is computed after each training iteration. In
the LSTM-RNN model, there are various gradient descent
methods that can be used to work with parameter spaces
for learning tasks. Here, the Adam method is used, which is
a stochastic optimizer that calculates different individual
learning rates for different parameters based on estimated
first and second gradients for each epoch( i ).
Design of the MATLAB-based GUI. A GUI was created to
facilitate the usage of the developed DL model. The
development of a solar irradiance forecasting tool requires
us to first develop the DL model and then create a graphical
interface based on MATLAB, as shown in Fig. 5.
Fig. 5. Steps for designing the MATLAB GUI forecasting tool
36
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
Development of LSTM-RNN model. The development of an
LSTM model involves several steps. In general, there are
five main stages (as shown in Fig. 5): 1) Data collection, 2)
Preprocessing of data, 3) Network creation, 4) Training, and
5) Model performance evaluation.
The first step in model development involves collecting and
preparing the dataset. Solar irradiance and SPP generation
are the main collected data, which serve as input
parameters for training and forecasting for the day ahead.
The data from the dataset were split into two subsets in a
90:10 ratio, where 90% of the data were used for training
the LSTM model, and the forecasted data from the trained
model were validated against the actual 10% of data that
were held out for assessing the prediction performance.
The second step involves data preprocessing, where three
data preprocessing procedures are conducted for more
effective model training. These procedures include: 1)
handling missing data, 2) data normalization, and 3) random
shuffling of data. Missing data are replaced by the mean value
of neighboring values within the same measurement period.
The normalization procedure before inputting the data is
generally a good practice as mixing variables with large and
small amplitudes can confuse the learning algorithm regarding
the importance of each variable and may lead it to discard
variables with smaller amplitudes [49].
Before applying data to the ML algorithm, it is necessary to
preprocess the data to correct or remove outliers and fill in
missing values.
Other steps in data preparation for ML include feature
scaling and encoding. Machine learning algorithms often
perform poorly when input features have vastly different
scales, so we transformed the data to have a zero mean and
unit variance.
During the creation of the LSTM-RNN model, the following
parameters are specified:
– Number of layers;
– Hyperparameters (training parameters);
– Performance evaluation metrics.
We utilized the Matlab Deep Learning Toolbox [50] to develop
the forecasting framework employed in this study. The
framework, is shown in Fig. 6, consists of an input layer, LSTM
hidden layers, a fully connected layer, and an output layer.
Fig. 6. The Deep Learning Framework
The input layer introduces the time series data into the
network. The hidden layers learn dependencies between time
steps in the temporal data. To perform forecasting, the network
concludes with a fully connected layer and an output layer.
The first LSTM block takes the initial state and the first time
step of the training example 1x and computes the first
output 1h and updated cell state 1C . At time step t the
block takes the current network state ( t-1 t-1C , h ) and the
next time step of the training example tx and computes
the output th and updated cell state tC . The final output
data is the solar irradiance forecast 1 Tg ...g .
During the training process, the weights are adjusted to
make the actual outputs of the network closely match the
target (measured) outputs. For each combination of input
variables, various network architectures are explored to
determine the optimal LSTM architecture (i.e., the lowest
mean squared deviation) for each combination of input
variables. Subsequently, different learning algorithms are
employed, involving variations in the number of hidden
layers and activation functions of the hidden/output layers.
Hyperparameter Optimization. Achieving high efficiency
with LSTM-RNN networks involves optimizing numerous
hyperparameters. These parameters affect both perfor-
mance and the time/memory costs of algorithm execution.
The selection of hyperparameters often makes the differen-
ce between average and state-of-the-art performance in ML
algorithms [51]. While there may be some general recom-
mendations for suitable values of these hyperparameters
[52], optimization remains necessary since optimal values
depend on the data type being used, the specific datasets
being compared, evaluation criteria, and other factors. The
hyperparameters we optimized are listed in Table 1.
Learning rate is arguably the most crucial hyperparameter
[53]. The number of hidden layers in the network determines
its depth. We experimented with optimizers such as adam
[54], rmsprop [55], and sgdm [56], finding that adam
performed better than the others. We employed a standard
scaler and employed full gradient descent.
Table 1. Hyperparameter optimization for LSTM model
Hyperparameter Value Search Range
Learning rate 0,0005
Optimization solver adam
LearnRateSchedule piecewise
MiniBatchSize 100
L2Regularization 0,0005
Feature scaling Standard
Number of layers 3
Hidden units/layer 10
Number of epochs 1250
Forecasting Results and performance metrics. The final
step involves evaluating the effectiveness of the developed
LSTM-RNN model. To quantitatively assess the model's
performance and identify any trends in its effectiveness, a
37
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
statistical analysis is conducted. This analysis includes
metrics such as Root Mean Square Error (RMSE) (7) and
Mean Absolute Error (MAE) (8).
Equations of indicators are formulated as follows:
( )
=
= −∑
N 2
i i
i 1
1 ˆRMSE y y
N
; (7)
=
= −∑
N
i i
i 1
1 ˆMAE y y
N
, (8)
where N is the sample size; iy is the actual value; ˆ iy is the
predicted value.
These metrics can be used to characterize the deviation of
forecasted values around measured data, stemming from
spatial aggregation. Lower values of these metrics indicate
higher forecast quality. For predicting the day ahead, the
values of these errors are crucial as they indicate in
absolute terms how much real data can deviate from the
forecasted values [57].
After training and testing the data, the forecasted values
yielded an RMSE of 0.835 W and MAE of 0.353 W,
confirming that the forecast testing provided an effective
outcome (Table 2).
Table 2. Statistical errors of the best performing LSTM model
Epoch RMSE, [kW] MAE, [kW]
1250 0,83486 0,35298
The prediction of SPP generation for the day ahead was
carried out using the trained model with maximum achievable
efficiency. The forecasted data is presented in Fig. 7.
Fig. 7. Measured and predicted SPP generation
Building the MATLAB-based GUI. The tool for
forecasting SPP generation was developed using the
MATLAB Graphical User Interface Development
Environment (GUIDE). GUIDE automatically generates a
code file containing MATLAB functions that control the
GUI's operation. The code helps initialize the GUI and
contains a structure for event handlers of the GUI
components and executable elements when interacting
with the user. The MATLAB editor can be used to add
code to event handlers to perform necessary actions
[58]. The GUI was built based on the best-performing
developed LSTM-RNN model's efficiency.
The main window is designed in such a way that the user
can perform forecasting (“Forecast”), obtain more
information about the tool (“About”), learn about the
description of the studied SPP, and exit the tool (“Exit”), as
shown in Fig. 8 and 9.
Fig. 8. “Main” window
Fig. 9. “Solar Power Plant” window
The “Forecast” window is divided into three sections: input
data, output data, and a set of buttons (Fig. 10).
Fig. 10. “Forecast” window
38
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
The “Input Data” section consists of a text input field, a
dropdown menu, and four buttons. The user can choose a
file with input data by clicking the "Load Data" button. This
will open a window allowing the user to select a file. Then
the user can select a model from the dropdown menu. By
clicking the "Model Param" button, the user can input the
hyperparameters needed for training the LSTM model, and
the "Metrics" button will open a window with calculated
evaluation metrics for each training and testing iteration.
The user can click the "Forecast" button to predict the
output solar power generation, as shown in Fig. 11 and 12,
respectively.
Fig. 11. “Loading dataset” window
Fig. 12. “Selecting Model” window
The prediction results of the SPP generation will be
displayed in a graphical format, as shown in Fig. 13.
At the bottom of the “Forecast” window, there are three
buttons: "Clear", "Main", and "Exit". The functionality
(transition to a specific window) changes according to the
current window. The "Clear" button clears the input data,
graph, and output data table. The "Main" button opens the
main window and closes the current window. Finally, the
"Exit" button closes and terminates the tool. At the bottom
of the window, there are three buttons: “Clear”, “Main”
and “Exit”.
Fig. 13. Forecasting the Dymerka SPP generation
Fig. 14 depicts the "About" window. It provides the user
with information about the tool.
Fig. 14. “About” window
Conclusions. This study developed an LSTM-RNN model for
forecasting solar power generation in the village of Velyka
Dymerka, Kyiv region, Ukraine. In this research, a deep
recurrent neural network called LSTM was utilized, using
only exogenous features to address the task at hand. The
model uses the current solar power plant generation as an
input feature. This approach eliminates the need for
historical solar irradiance data, which is costly to measure.
In comparison to other analogous models available in the
literature, the LSTM-RNN model presented in this study
showcases competitive performance in solar power
generation forecasting. While the specific models in the
literature may vary in architecture and approach, our
LSTM-based model stands out for its ability to forecast
solar energy generation without relying on historical solar
irradiance data. Several studies have explored the use of
traditional regression models, such as linear regression, for
solar energy forecasting. However, these models often
struggle to capture the nonlinear relationships between
environmental factors and solar power generation. In
contrast, our LSTM-RNN model excels in handling such
nonlinearities, providing superior forecasting accuracy.
Additionally, some research has explored the application of
decision tree-based models, such as Random Forests or
Gradient Boosting, for solar energy prediction. While these
models can handle complex interactions, they may require
a substantial amount of historical data and often fall short
when faced with time-series forecasting tasks. The LSTM-
39
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
RNN model, as demonstrated in our study, effectively
addresses this challenge. The achieved RMSE of 0.835 W
and MAE of 0.353 W for most effective model not only
compare favourably with similar deep learning approaches
but also demonstrate competitive performance when
measured against conventional regression-based models.
This highlights the potential of LSTM-RNNs as a reliable
choice for solar energy forecasting. Furthermore, our
development of a user-friendly GUI tool in MATLAB adds
practical value to the model's applicability. While some
research may focus solely on model development, the
inclusion of a user interface facilitates its adoption and
utilization by solar energy professionals in Ukraine.
Regarding legislative compliance, the proposed model
aligns with the 5% deviation allowance for hourly power
forecasts specified in Section XVII of the Ukrainian Law on
the Electricity Market. This ensures that our forecasting
results meet the regulatory standards, further emphasizing
the practical suitability of our approach in real-world
energy planning and management. In summary, the LSTM-
RNN model presented in this study stands out as a robust
and accurate tool for solar power generation forecasting,
particularly when compared to traditional regression
models and decision tree-based approaches. Its
competitive performance, user-friendly interface, and
compliance with regulatory requirements position it as a
valuable asset in the field of solar energy in Ukraine, with
potential applications in other regions as well. Further
research and validation across diverse geographic locations
and climatic conditions can enhance its versatility and
reliability.
REFERENCES
1. D. Matushkin, A. Bosak and L. Kulakovskyi, “Analysis of
factors for forecasting electric power generation by
solar power plants”, POWER ENGINEERING: economics,
technique, ecology, vol. 4, pp. 64-69, 2020.
https://doi.org/10.20535/1813-5420.4.2020.233597.
2. Bosak, A., Matushkin, D., Davydenko, L., Kulakovskyi, L.,
Bronytskyi, V. (2023). “Short-Term Forecasting of
Photovoltaic Solar Power Generation Based on Time
Series: Application for Ensure the Efficient Operation of
the Integrated Energy System of Ukraine”. In:
Kyrylenko, O., Denysiuk, S., Derevianko, D., Blinov, I.,
Zaitsev, I., Zaporozhets, A. (eds) Power Systems
Research and Operation. Studies in Systems, Decision
and Control, vol 220. Springer, Cham,
https://doi.org/10.1007/978-3-031-17554-1_8.
3. J. Smend, A. Mnatsakanyan and S. Sgouridis, "A Smart
Grid Solution Integrating Distributed Generation and
Internet of Things Sensors for Demand Side
Management and Fault Identification: Case Study,"
2021 IEEE 12th International Symposium on Power
Electronics for Distributed Generation Systems (PEDG),
Chicago, IL, USA, 2021, pp. 1-7,
https://doi.org/10.1109/PEDG51384.2021.9494196.
4. Huang, W., Zhang, N., Kang, C. et al. From demand
response to integrated demand response: review and
prospect of research and application. Prot Control Mod
Power Syst 4, 12 (2019).
https://doi.org/10.1186/s41601-019-0126-4.
5. Helistö, N, Kiviluoma, J, Holttinen, H, Lara, JD, Hodge, B-
M. Including operational aspects in the planning of
power systems with large amounts of variable
generation: A review of modeling approaches. WIREs
Energy Environ. 2019; 8:e341.
https://doi.org/10.1002/wene.341.
6. E. Saeidpour Parizy, S. Choi and H. R. Bahrami, "Grid-
Specific Co-Optimization of Incentive for Generation
Planning in Power Systems With Renewable Energy
Sources," in IEEE Transactions on Sustainable Energy,
vol. 11, no. 2, pp. 947-957, April 2020,
https://doi.org/10.1109/TSTE.2019.2914875.
7. S. Jose and R. L. Itagi, "Data Analytics in Solar Photovoltaics
Power Forecasting for Smart Grid Applications," 2021
International Conference on Intelligent Technologies
(CONIT), Hubli, India, 2021, pp. 1-5,
https://doi.org/10.1109/CONIT51480.2021.9498299.
8. Said, Y., Alanazi, A. AI-based solar energy forecasting for
smart grid integration. Neural Comput & Applic 35,
8625–8634 (2023). https://doi.org/10.1007/s00521-
022-08160-x.
9. W. Cabrera, D. Benhaddou and C. Ordonez, "Solar Power
Prediction for Smart Community Microgrid," 2016 IEEE
International Conference on Smart Computing
(SMARTCOMP), St. Louis, MO, USA, 2016, pp. 1-6,
https://doi.org/10.1109/SMARTCOMP.2016.7501718.
10. Marius Paulescu, Eugenia Paulescu, Viorel Badescu,
"Chapter 9 - Nowcasting solar irradiance for effective solar
power plants operation and smart grid management", in
Ravinesh Deo, Pijush Samui, Sanjiban Sekhar Roy
“Predictive Modelling for Energy Management and Power
Systems Engineering”, Elsevier, 2021, 249-270,
ISBN 9780128177723,
https://doi.org/10.1016/B978-0-12-817772-3.00009-4.
11. Sarker, E, Halder, P, Seyedmahmoudian, M, et al.
Progress on the demand side management in smart grid
and optimization approaches. Int J Energy Res. 2021;
45: 36-64. https://doi.org/10.1002/er.5631.
12. Sajjad Ali, Kalim Ullah, Ghulam Hafeez, Imran Khan,
Fahad R. Albogamy, Syed Irtaza Haider, “Solving day-
ahead scheduling problem with multi-objective energy
optimization for demand side management in smart
grid”, Engineering Science and Technology, an
International Journal, 2022, vol. 36, 101135, ISSN 2215-
0986, https://doi.org/10.1016/j.jestch.2022.101135.
13. Fermín Rodríguez, Fernando Martín, Luis Fontán, Ainhoa
Galarza, “Ensemble of machine learning and spatio-
temporal parameters to forecast very short-term solar
irradiation to compute photovoltaic generators’ output
https://doi.org/10.20535/1813-5420.4.2020.233597
https://doi.org/10.1007/978-3-031-17554-1_8
https://doi.org/10.1109/PEDG51384.2021.9494196
https://doi.org/10.1186/s41601-019-0126-4
https://doi.org/10.1002/wene.341
https://doi.org/10.1109/TSTE.2019.2914875
https://doi.org/10.1109/CONIT51480.2021.9498299
https://doi.org/10.1007/s00521-022-08160-x
https://doi.org/10.1007/s00521-022-08160-x
https://doi.org/10.1109/SMARTCOMP.2016.7501718
https://doi.org/10.1016/B978-0-12-817772-3.00009-4
https://doi.org/10.1002/er.5631
https://doi.org/10.1016/j.jestch.2022.101135
40
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
power”, Energy, vol. 229, 2021, 120647, ISSN 0360-5442,
https://doi.org/10.1016/j.energy.2021.120647.
14. B. H. Mahdi, K. M. Yousif and L. M. Dosky, "Using
Artificial Neural Networks to Predict Solar Radiation for
Duhok City, Iraq," 2020 International Conference on
Computer Science and Software Engineering (CSASE),
Duhok, Iraq, 2020, pp. 1-6,
https://doi.org/10.1109/CSASE48920.2020.9142119.
15. Z. Zarkov, L. Stoyanov, I. Draganovska and V. Lazarov,
"The Comparison of different approaches for solar
radiation forecasting using Artificial Neural Network,"
2019 11th Electrical Engineering Faculty Conference
(BulEF), Varna, Bulgaria, 2019, pp. 1-6,
https://doi.org/10.1109/BulEF48056.2019.9030781.
16. Assi, A. H., Shamisi, M. H. A., & Hejase, H. A. N. (2012).
MATLAB tool for predicting the global solar radiation in
UAE. In 2012 International Conference on Renewable
Energies for Developing Countries, REDEC 2012
[6416698] (2012 International Conference on
Renewable Energies for Developing Countries, REDEC
2012). https://doi.org/10.1109/REDEC.2012.6416698.
17. M B Eteiba, El E T Shenawy, J H Shazly and A Z Hafez.
Article: A Photovoltaic (Cell, Module, Array) Simulation
and Monitoring Model using MATLAB/GUI Interface.
International Journal of Computer Applications
69(6):14-28, May 2013,
https://doi.org/10.5120/11845-7579.
18. S. Ç. BEKTAŞ, R. ÇAKMAK and İ. H. ALTAŞ, "Design of a
MATLAB GUI for Day Ahead Forecasting of PV Panel
Power," 2019 Innovations in Intelligent Systems and
Applications Conference (ASYU), Izmir, Turkey, 2019, pp.
1-5, https://doi.org/10.1109/ASYU48272.2019.8946349.
19. Jawad, I. N., Rishack, Q. A., & Sultan, H. S. (2021).
Matlab graphical user interface (GUI) code for solar
tower power plant performance calculations. Basrah
Journal for Engineering Sciences, 21(1), 8-14,
https://dx.doi.org/10.33971/bjes.21.1.2.
20. Zhang, Yao & Jianxue, Wang. (2015). GEFCom2014
Probabilistic Solar Power Forecasting based on k-
Nearest Neighbor and Kernel Density Estimator.
https://doi.org/10.1109/PESGM.2015.7285696.
21. Bessa, Ricardo J., Artur Trindade, Cátia S. P. Silva and
Vladimiro Miranda. “Probabilistic solar power
forecasting in smart grids using distributed
information.” International Journal of Electrical Power
& Energy Systems 72 (2015): 16-23.
https://doi.org/10.1016/J.IJEPES.2015.02.006.
22. Fliess, Michel & Join, Cédric & Voyant, Cyril. (2018).
Prediction bands for solar energy: New short-term time
series forecasting techniques. Solar Energy. 166.
https://doi.org/10.1016/j.solener.2018.03.049.
23. Alessandrini, Stefano & Delle Monache, Luca & Sperati,
Simone & Cervone, Guido. (2015). An analog ensemble
for short-term probabilistic solar power forecast.
Applied Energy. 157. 95-110.
https://doi.org/10.1016/j.apenergy.2015.08.011.
24. Gensler, A., Henze, J., Sick, B., & Raabe, N. (2016). Deep
Learning for solar power forecasting — An approach
using AutoEncoder and LSTM Neural Networks. 2016
IEEE International Conference on Systems, Man, and
Cybernetics (SMC), 002858-002865.
https://doi.org/10.1109/SMC.2016.7844673.
25. Kumari, Pratima & Toshniwal, Durga. (2020). Real-time
estimation of COVID-19 cases using machine learning
and mathematical models - The case of India. 369-374.
https://doi.org/10.1109/ICIIS51140.2020.9342735.
26. Li, Liang & Yuan, Zhi-Ming & Yan, Gao. (2018).
Maximization of energy absorption for a wave energy
converter using the deep machine learning. Energy.
165. 340-349.
https://doi.org/10.1016/j.energy.2018.09.093.
27. Hu, Qinghua & Zhang, Rujia & Zhou, Yucan. (2016).
Transfer learning for short-term wind speed prediction
with deep neural networks. Renewable Energy. 85.
https://doi.org/10.1016/j.renene.2015.06.034.
28. Mishra, Manohar & Dash, Pandit & Nayak, Janmenjoy &
Naik, Bighnaraj & Swain, Subrat. (2020). Deep Learning and
Wavelet Transform integrated approach for Short-term
solar PV power prediction. Measurement. 166. 108250.
https://doi.org/10.1016/j.measurement.2020.108250.
29. Qing, Xiangyun & Niu, Yugang. (2018). Hourly day-
ahead solar irradiance prediction using weather
forecasts by LSTM. Energy. 148.
https://doi.org/10.1016/j.energy.2018.01.177.
30. Kuremoto, Takashi & Kimura, Shinsuke & Kobayashi,
Kunikazu & Obayashi, Masanao. (2014). Time series
forecasting using a deep belief network with restricted
Boltzmann machines. Neurocomputing. 137. 47-56.
https://doi.org/10.1016/j.neucom.2013.03.047.
31. Kaba, Kazım & Sarigul, Mehmet & Avci, Mutlu &
Kandırmaz, H.. (2018). Estimation Of Daily Global Solar
Radiation Using Deep Learning Model. Energy. 162. 126-
135. https://doi.org/10.1016/j.energy.2018.07.202.
32. Bengio, Y. (2009). Learning Deep Architectures for AI.
Foundations. 2. 1-55.
https://doi.org/10.1561/2200000006.
33. LeCun, Yann & Bengio, Y. & Hinton, Geoffrey. (2015).
Deep Learning. Nature. 521. 436-44.
https://doi.org/10.1038/nature14539.
34. Krizhevsky, A., Sutskever, I. and Hinton, G.E. (2017)
Imagenet Classification with Deep Convolutional Neural
Networks. Communications of the ACM, 60, 84-90.
https://doi.org/10.1145/3065386.
35. Farabet, Clement & Couprie, Camille & Najman, Laurent
& Lecun, Yann. (2013). Learning Hierarchical Features
for Scene Labeling. IEEE transactions on pattern analysis
and machine intelligence. 35. 1915-1929.
https://doi.org/10.1109/TPAMI.2012.231.
https://doi.org/10.1016/j.energy.2021.120647
https://doi.org/10.1109/CSASE48920.2020.9142119
https://doi.org/10.1109/BulEF48056.2019.9030781
https://doi.org/10.1109/REDEC.2012.6416698
https://doi.org/10.5120/11845-7579
https://doi.org/10.1109/ASYU48272.2019.8946349
https://dx.doi.org/10.33971/bjes.21.1.2
https://doi.org/10.1109/PESGM.2015.7285696
https://doi.org/10.1016/J.IJEPES.2015.02.006
https://doi.org/10.1016/j.solener.2018.03.049
https://doi.org/10.1016/j.apenergy.2015.08.011
https://doi.org/10.1109/SMC.2016.7844673
https://doi.org/10.1109/ICIIS51140.2020.9342735
https://doi.org/10.1016/j.energy.2018.09.093
https://doi.org/10.1016/j.renene.2015.06.034
https://doi.org/10.1016/j.measurement.2020.108250
https://doi.org/10.1016/j.energy.2018.01.177
https://doi.org/10.1016/j.neucom.2013.03.047
https://doi.org/10.1016/j.energy.2018.07.202
https://doi.org/10.1561/2200000006
https://doi.org/10.1038/nature14539
https://doi.org/10.1145/3065386
https://doi.org/10.1109/TPAMI.2012.231
41
Відновлювана енергетика. №3/2023 | Комплексні проблеми енергетичних систем на основі НВДЕ
36. Hinton, Geoffrey & Deng, li & Yu, Dong & Dahl, George
& Mohamed, Abdel-rahman & Jaitly, Navdeep & Senior,
Andrew & Vanhoucke, Vincent & Nguyen, Phuongtrang
& Sainath, Tara & Kingsbury, Brian. (2012). Deep Neural
Networks for Acoustic Modeling in Speech Recognition:
The Shared Views of Four Research Groups. Signal
Processing Magazine, IEEE. 29. 82-97.
https://doi.org/10.1109/MSP.2012.2205597.
37. Sainath, Tara & Mohamed, Abdel-rahman & Kingsbury,
Brian & Ramabhadran, Bhuvana. (2013). Deep
convolutional neural networks for LVCSR. Acoustics,
Speech, and Signal Processing, 1988. ICASSP-88., 1988
International Conference on. 8614-8618.
https://doi.org/10.1109/ICASSP.2013.6639347.
38. Collobert, Ronan & Weston, Jason & Bottou, Leon &
Karlen, Michael & Kavukcuoglu, Koray & Kuksa, Pavel.
(2011). Natural Language Processing (Almost) from
Scratch. Journal of Machine Learning Research. 12.
2493-2537.
39. Sutskever, Ilya & Vinyals, Oriol & Le, Quoc. (2014).
Sequence to Sequence Learning with Neural
Networks. 10.
40. Jean, Sébastien & Cho, Kyunghyun & Memisevic, Roland
& Bengio, Y.. (2014). On Using Very Large Target
Vocabulary for Neural Machine Translation. 1.
https://doi.org/10.3115/v1/P15-1001.
41. Ciodaro, T & Dd, Mr D & Seixas, Joao & Oliveira
Damazio, Denis. (2012). Online particle detection with
Neural Networks based on topological calorimetry
information. Journal of Physics: Conference Series. 368.
https://doi.org/10.1088/1742-6596/368/1/012030.
42. Helmstaedter, M., Briggman, K. L., Turaga, S. C., Jain, V.,
Seung, H. S., & Denk, W. (2013). Connectomic
reconstruction of the inner plexiform layer in the mouse
retina. Nature, 500(7461), 168-174.
https://doi.org/10.1038/nature12346.
43. Lago, Jesus & Brabandere, Karel & De Ridder, Fjo & De
Schutter, Bart. (2018). Short-term forecasting of solar
irradiance without local telemetry: A generalized model
using satellite data. Solar Energy. 173. 566-577.
https://doi.org/10.1016/j.solener.2018.07.050.
44. Goswami, B., & Bhandari, G. (2011). Automatically
adjusting cloud movement prediction model from satellite
infrared images. 2011 Annual IEEE India Conference, 1-4.
https://doi.org/10.1109/INDCON.2011.6139604.
45. Ali, Ahmed & Obiora, Chibuzor & Ali, A & Hasan, Ali.
(2020). Forecasting Hourly Solar Irradiance Using Long
Short-Term Memory (LSTM) Network.
https://doi.org/10.1109/IREC48820.2020.9310449.
46. F. A. Gers, N. N. Schraudolph, and J. Schmidhuber,
“Learning precise timing with LSTM recurrent
networks,” Journal of machine learning research, vol. 3,
no. Aug, pp. 115–143, 2002.J. Schmidhuber, “Deep
learning in neural networks: An overview,” Neural
Networks, vol. 61, pp. 85-117, 2015.
47. S U Sabareesh, K S N Aravind, Kandru Bhargav
Chowdary, S Syama, and Kirthika Devi V S. 2023. LSTM
Based 24 hours Ahead Forecasting of Solar PV System
for Standalone Household System. Procedia Comput.
Sci. 218, C (2023), 1304–1313.
https://doi.org/10.1016/j.procs.2023.01.109.
48. Assaf, Abbas & Haron, Habibollah & Hamed, Haza Nuzly
Abdull & Ghaleb, Fuad & Qasem, Sultan & Albarrak,
Abdullah. (2023). A Review on Neural Network Based
Models for Short Term Solar Irradiance Forecasting.
Applied Sciences. 13. 8332.
https://doi.org/10.3390/app13148332.
49. Tymvios, Filippos & Michaelides, Silas & Skouteli, Chara.
(2008). Estimation of Surface Solar Radiation with
Artificial Neural Networks. Modeling Solar Radiation at
the Earth's Surface: Recent Advances. 221-256.
https://doi.org/10.1007/978-3-540-77455-6_9.
50. Deep Learning Toolbox. Available online:
https://mathworks.com/products/deep-learning.html
(accessed on 28 August 2023).
51. Hutter, Frank & Lücke, Jörg & Schmidt-Thieme, Lars.
(2015). Beyond Manual Tuning of Hyperparameters. KI
- Künstliche Intelligenz. 29.
https://doi.org/10.1007/s13218-015-0381-0.
52. Ng, A. Machine Learning Yearning: Technical Strategy
for AI Engineers in the Era of Deep Learning. Available
online: https://deeplearning.ai/machine-learning-year-
ning (accessed on 28 August 2023).
53. Goodfellow, I.; Bengio, Y.; Courville, A. Deep Learning;
MIT Press: Cambridge, MA, USA, 2017.
54. Kingma, D.P.; Ba, J.L. Adam: A method for stochas-
tic optimization. arXiv 2014, arXiv:1412.6980.
55. Hinton, G. Neural Networks for Machine Learn-
ing. Available online:
https://www.coursera.org/learn/neuralnetworks/hom
e/welcome (accessed on 28 August 2023).
56. Qian, Ning. (1999). On the momentum term in gradient
descent learning algorithms. Neural networks: the
official journal of the International Neural Network
Society. 12. 145-151. https://doi.org/10.1016/S0893-
6080(98)00116-6.
57. Bosak, A., Matushkin, D., Dubovyk, V., Homon, S., &
Kulakovskyi, L. (2021). Determination of the Concepts
of Building a Solar Power Forecasting Model. Scientific
Horizons, 24(10), 9-16.
https://doi.org/10.48077/scihor.24(10).2021.9-16.
58. MATLAB Creating Graphical User Interfaces, revision of
March 2015 for MATLAB 8.5 (Release 2015a), The
MathWorks Inc., Natick, MA, USA.
https://doi.org/10.1109/MSP.2012.2205597
https://doi.org/10.1109/ICASSP.2013.6639347
https://doi.org/10.3115/v1/P15-1001
https://doi.org/10.1088/1742-6596/368/1/012030
https://doi.org/10.1038/nature12346
https://doi.org/10.1016/j.solener.2018.07.050
https://doi.org/10.1109/INDCON.2011.6139604
https://doi.org/10.1109/IREC48820.2020.9310449
https://doi.org/10.1016/j.procs.2023.01.109
https://doi.org/10.3390/app13148332
https://doi.org/10.1007/978-3-540-77455-6_9
https://doi.org/10.1007/s13218-015-0381-0
https://doi.org/10.1016/S0893-6080(98)00116-6
https://doi.org/10.1016/S0893-6080(98)00116-6
https://doi.org/10.48077/scihor.24(10).2021.9-16
|
| id | veorgua-article-408 |
| institution | Vidnovluvana energetika |
| keywords_txt_mv | keywords |
| language | Ukrainian |
| last_indexed | 2026-07-19T01:11:41Z |
| publishDate | 2023 |
| publisher | Institute of Renewable Energy National Academy of Sciences of Ukraine |
| record_format | ojs |
| resource_txt_mv | veorgua/7d/b7ae3fa58028efa7060e590431ba947d.pdf |
| spelling | veorgua-article-4082026-07-18T06:32:18Z DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING РОЗРОБКА ГРАФІЧНОГО ІНТЕРФЕЙСУ ДЛЯ КОРОТКОСТРОКОВОГО ПРОГНОЗУВАННЯ СОНЯЧНОЇ ГЕНЕРАЦІЇ НА ОСНОВІ ГЛИБОКОГО НАВЧАННЯ Matushkin, D. Bosak, A. short-term forecasting, solar power plant, deep learning, Long Short-Term Memory, graphical user inter-face, MATLAB&Simulink simulation. Короткострокове прогнозування, сонячна електростанція, глибоке навчання, довга короткочас-на пам'ять, графічний інтерфейс користувача, симуляція в MATLAB&Simulink. In energy systems, it is crucial to forecast solar energy generation for optimization of operations and to mitigate the impact of uncertainty. Forecasting solar energy involves predicting solar irradiance, for which historical solar irradiance and weather parameter data are typically required. However, such data are often unavailable for residential and commercial solar microgrids. This study proposes an hourly forecasting model for next-day Solar Power Production (SPP) that doesn't rely on historical solar irradiance data. The forecasting is performed using deep learning techniques like Long Short-Term Memory (LSTM), and the results are integrated into the MATLAB&Simulink simulation platform. A graphical user interface (GUI) within the MATLAB&Simulink software complex is presented as a simulation platform for hourly SPP forecasting. This platform serves as a useful tool for researchers studying energy management and forecasting as well as planning renewable energy-based energy system operations. The study involves predicting the amount of energy generated by Solar Power Production (SPP). The developed GUI was employed to forecast the SPP output power over test days. The forecasted data were then evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Experimental results demonstrated reasonable accuracy, with the proposed model achieving an RMSE of 0.835 W and an MAE of 0.353 W in certain datasets. В енергосистемах важливо прогнозувати генерацію сонячної енергії для оптимізації роботи та зменшення впливу невизначеності. Для прогнозування сонячної енергії необхідно прогнозувати сонячне випромінювання, для чого зазвичай потрібні історичні дані про сонячне випромінювання та погодні параметри. Ці дані часто недоступні для домашніх та комерційних сонячних мікромереж. У цьому дослідженні пропонується модель погодинного прогнозування генерації SPP на наступний день, яка не залежить від історичних даних про сонячне випромінювання. Прогнозування виконується за допомогою інструмента глибокого навчання, як-от Long Short-Term Memory (LSTM), а результати подаються в систему на платформі симуляції MATLAB&Simulink. Як платформу симуляції представлено графічний інтерфейс користувача (GUI) в програмному комплексі MATLAB&Simulink для прогнозування погодинної потужності SPP. Ця платформа буде корисним і практичним механізмом для дослідників, які вивчають енергоменеджмент та для прогнозування й планування режимів роботи енергосистеми з відновлюваними джерелами енергії (ВДЕ). У цьому дослідженні виконано прогноз кількості енергії, яка буде згенерована SPP. Розроблений GUI використовувався для прогнозування вихідної потужності SPP протягом тестових днів. Потім отримані прогнозні дані були оцінені за середньо-квадратичною похибкою (RMSE) і середньою абсолютною похибкою (MAE). Результати експериментів показали достатні показники точності. Запропонована модель досягла RMSE у розмірі 0,835 W і MAE у розмірі 0,353 W у деяких наборах даних. Institute of Renewable Energy National Academy of Sciences of Ukraine 2023-10-19 Article Article application/pdf https://ve.org.ua/index.php/journal/article/view/408 10.36296/1819-8058.2023.3(74).32-41 Vidnovluvana energetika ; No. 3(74) (2023): Scientific and applied Journal renewable energy ; 32-41 Возобновляемая энергетика; ##issue.no## 3(74) (2023): Scientific and applied Journal renewable energy ; 32-41 Відновлювана енергетика; № 3(74) (2023): Науково-прикладний журнал Відновлювана енергетика; 32-41 2664-8172 1819-8058 10.36296/1819-8058.2023.3(74) uk https://ve.org.ua/index.php/journal/article/view/408/319 Copyright (c) 2023 D. Matushkin, A. Bosak https://creativecommons.org/licenses/by-nc-nd/4.0 |
| spellingShingle | short-term forecasting solar power plant deep learning Long Short-Term Memory graphical user inter-face MATLAB&Simulink simulation. Matushkin, D. Bosak, A. DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING |
| title | DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING |
| title_alt | РОЗРОБКА ГРАФІЧНОГО ІНТЕРФЕЙСУ ДЛЯ КОРОТКОСТРОКОВОГО ПРОГНОЗУВАННЯ СОНЯЧНОЇ ГЕНЕРАЦІЇ НА ОСНОВІ ГЛИБОКОГО НАВЧАННЯ |
| title_full | DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING |
| title_fullStr | DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING |
| title_full_unstemmed | DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING |
| title_short | DESIGN OF A MATLAB GUI FOR SHORT-TERM SOLAR FORECASTING BASED ON DEEP LEARNING |
| title_sort | design of a matlab gui for short-term solar forecasting based on deep learning |
| topic | short-term forecasting solar power plant deep learning Long Short-Term Memory graphical user inter-face MATLAB&Simulink simulation. |
| topic_facet | short-term forecasting solar power plant deep learning Long Short-Term Memory graphical user inter-face MATLAB&Simulink simulation. Короткострокове прогнозування сонячна електростанція глибоке навчання довга короткочас-на пам'ять графічний інтерфейс користувача симуляція в MATLAB&Simulink. |
| url | https://ve.org.ua/index.php/journal/article/view/408 |
| work_keys_str_mv | AT matushkind designofamatlabguiforshorttermsolarforecastingbasedondeeplearning AT bosaka designofamatlabguiforshorttermsolarforecastingbasedondeeplearning AT matushkind rozrobkagrafíčnogoínterfejsudlâkorotkostrokovogoprognozuvannâsonâčnoígeneracíínaosnovíglibokogonavčannâ AT bosaka rozrobkagrafíčnogoínterfejsudlâkorotkostrokovogoprognozuvannâsonâčnoígeneracíínaosnovíglibokogonavčannâ |