DSpace Arşivi :: by Yazar "Korkmaz, Deniz" değerine göre listeleniyor

Yazar "Korkmaz, Deniz" seçeneğine göre listele

Listeleniyor 1 - 20 / 22

10- (589-600) Elektrolüminesans Görüntülerde Arızalı Fotovoltaik Panel Hücrelerin Evrişimli Sinir Ağı ile Otomatik Sınıflandırılması
(2022) AÇIKGÖZ, Hakan; Korkmaz, Deniz
Fotovoltaik (FV) panel hücrelerindeki arızaların tespiti ve sınıflandırılması güneş enerjisi santrallerinin verimli ve güvenilir bir şekilde işletilebilmesi için oldukça önemli bir konu haline gelmiştir. Bu çalışmada, FV panel hücrelerindeki arızaların hızlı ve doğru bir şekilde tespit edilmesi ve sınıflandırılması için etkin bir evrişimli sinir ağı (ESA) modeli önerilmiştir. Önerilen model, daha az parametre ve model boyutuna sahip SqueezeNet ile transfer öğrenme yaklaşımı kullanılarak geliştirilmiştir. Eğitim yakınsamasını iyileştirmek ve sınıflandırma başarımını arttırmak için modelin aktivasyon fonksiyonları değiştirilerek ateşleme modüllerinden atlama bağlantıları oluşturulmuştur. Deneylerde, Elektrolüminesans (EL) görüntülerinden elde edilen bir veri seti kullanılmıştır. Sınıf dağılımının dengesizliğini gidermek ve örnek sayısını arttırmak için veri artırma teknikleri uygulanmıştır. Önerilen yöntemin performansı AlexNet, ShuffleNet, GoogLeNet ve SqueezeNet gibi ön eğitimli ESA mimarileri ile karşılaştırılmıştır. Gerçekleştirilen deneysel çalışmalarda önerilen yöntemin doğruluk, kesinlik, duyarlılık, özgüllük ve F1-skor değerleri sırasıyla %91.29, %84.21, %89.72, %92.04 ve %86.88 olarak elde edilmiştir. Ayrıca önerilen yöntem diğer yöntemlerin doğruluk ölçütündeki değerlerini %0.99 ile %6.29 arasında iyileştirmiştir. Elde edilen tüm sonuçlar analiz edildiğinde önerilen yöntemin FV panel hücrelerindeki arızaların tespitinde üstün bir performansa sahip olduğu gözlemlenmiştir.
Altitude and Attitude Control of a Quadcopter Based on Neuro-Fuzzy Controller
(Springer Science and Business Media Deutschland GmbH, 2022) Korkmaz, Deniz; Açıkgöz, Hakan; Üstündağ, Mehmet
In this paper, a 6-degrees of freedom (DoF) nonlinear dynamic model of the quadcopter is derived and a robust altitude and attitude control is proposed. The motion control is performed with four neuro-fuzzy controllers that ensure rapid and robust performances for nonlinear and uncertain systems. The aim of the designed control scheme is to provide to track the desired yaw, pitch, roll, and altitude trajectories simultaneously. The simulations are realized in the MATLAB/Simulink environment. The obtained results show that the designed control scheme is robust and efficient in both altitude and attitude responses with different uncertain trajectories.
An Automated Diagnosis of Parkinson's Disease from MRI Scans Based on Enhanced Residual Dense Network with Attention Mechanism
(Springer, 2025) Acikgoz, Hakan; Korkmaz, Deniz; Talan, Tarik
The increasing prevalence of neurodegenerative diseases has recently heightened interest in research on early diagnosis of these diseases. Parkinson's disease (PD), among the most prominent of these conditions, is a neurological disorder causing the loss of nerve cells and significantly affecting movement control. Detection of PD in early stages is of critical importance to prevent the progression of the disease and improve treatment processes. The aim of the current study is to develop a deep learning model that can perform accurate classification for early diagnosis of PD from MRI images. In this study, a densely connected feature fusion network with residual learning is designed to diagnose PD patients. The designed network consists of a serial dense block with skip connections and efficient attention mechanisms. In this architecture, squeeze-excitation (SE) blocks with ResNeXt (SE-ResNeXt block) modules are utilized to extract distinctive and high-level features. In the experiments, a publicly available T2-weighted MRI dataset is used, and an offline augmentation process is applied to limited data to increase the generalization ability and classification performance. The proposed method is evaluated and compared with current state-of-the-art deep learning methods. The obtained results show that the proposed model gives higher classification performance with an overall accuracy of 94.44%, precision of 91.67%, sensitivity of 91.67%, specificity of 95.83%, F1-score of 91.67%, and Matthew's correlation coefficient of 87.50% for the PD and healthy control subjects.
Classification of Hotspots in Photovoltaic Modules with Deep Learning Methods
(2022) AÇIKGÖZ, Hakan; Korkmaz, Deniz; DANDIL, Çiğdem
Solar energy systems are increasing their capacity in the energy industry day by day by operating with higher efficiency in parallel with technological developments. The functional operation of photovoltaic (PV) module contributes greatly to the optimal performance of these systems. On the other hand, detection and classification of faults occurring in PV modules are of vital importance in the operation and maintenance of solar energy systems. In this study, the classification of hotspots, which is one of the most common faults in Photovoltaic (PV) modules, is carried out by deep learning methods. First, data augmentation is applied to the images in the training dataset to improve the classification performance. Then, pre-trained deep learning models namely AlexNet, GoogLeNet, ShuffleNet, SqueezeNet, ResNet-50, and MobileNet-v2 are compared on the same test dataset. According to the obtained experimental results, AlexNet has the best performance with an accuracy value of 98.65%, while ResNet-50 provides the worst result with 94.59%.
Comparative CFD Simulations of a Soft Robotic Fish for Undulatory Swimming Behaviors
(Mdpi, 2025) Ozmen Koca, Gonca; Ay, Mustafa; Bal, Cafer; Korkmaz, Deniz; Akpolat, Zuhtu Hakan
Studies on autonomous underwater vehicles (AUVs) have gained momentum in recent years, and a special type of AUV, the robotic fish, has become a significant topic, with a superior maneuverability to traditional AUVs. In this paper, a prediction strategy for the hydrodynamic performance of a robotic fish to analyze undulatory swimming behaviors is proposed. The two-dimensional robotic fish model for computational fluid dynamics (CFD) simulations is constructed, and a dynamic network method is applied to orient the generated network based on the wavy motion. For the thrust force of the fin, a body traveling wave is derived. In the simulations, the effects of kinematic parameters such as flapping frequency and speed on swimming efficiency and drag are analyzed, and thrust force production, power expenditure, and overall efficiency of swimming are examined. Later, a deep learning-based prediction model is designed from the obtained parameters, and force predictions are performed. Long short-term memory (LSTM)-, convolutional neural network (CNN)-, and gated recurrent network (GRU)-based time series prediction models are used, and their variations are compared. In these experiments, while the CNN-GRU achieves the higher prediction performance for the root mean square error, with 0.0228, other approaches give a lower performance, between 0.0233 and 0.0359. The proposed method demonstrates a superior performance in CNN and LSTM models and exhibits lower prediction errors.
COVIDiagnosis-Net: Deep Bayes-SqueezeNet based diagnosis of the coronavirus disease 2019 (COVID-19) from X-ray images
(Elsevier, 2020) Uçar, Ferhat; Korkmaz, Deniz
The Coronavirus Disease 2019 (COVID-19) outbreak has a tremendous impact on global health and the daily life of people still living in more than two hundred countries. The crucial action to gain the force in the fight of COVID-19 is to have powerful monitoring of the site forming infected patients. Most of the initial tests rely on detecting the genetic material of the coronavirus, and they have a poor detection rate with the time-consuming operation. In the ongoing process, radiological imaging is also preferred where chest X-rays are highlighted in the diagnosis. Early studies express the patients with an abnormality in chest X-rays pointing to the presence of the COVID-19. On this motivation, there are several studies cover the deep learning-based solutions to detect the COVID-19 using chest X-rays. A part of the existing studies use non-public datasets, others perform on complicated Artificial Intelligent (AI) structures. In our study, we demonstrate an AI-based structure to outperform the existing studies. The SqueezeNet that comes forward with its light network design is tuned for the COVID-19 diagnosis with Bayesian optimization additive. Fine-tuned hyperparameters and augmented dataset make the proposed network perform much better than existing network designs and to obtain a higher COVID-19 diagnosis accuracy.
Cross-scale fusion network and two-stage decomposition for power forecasting of offshore wind turbines
(Pergamon-Elsevier Science Ltd, 2025) Korkmaz, Deniz; Acikgoz, Hakan; Ustundag, Mehmet
This research presents a hybrid forecasting model for offshore wind power, which is based on a two-stage decomposition process and a densely connected convolutional network. Initially, the offshore wind power data is decomposed into several components utilizing an improved complete ensemble empirical mode decomposition with adaptive noise. The high-frequency component is further divided into multiple components through empirical mode decomposition. Following the decomposition, the dataset is transformed into the HSV color space. The proposed model features a sequential and multi-scale convolutional block architecture, inspired by the clique network approach. Furthermore, a squeeze-and-excitation module is incorporated to enhance the network performance. Comparative experiments are conducted against state-of-the-art deep learning models using data from two offshore wind turbines. The results indicate that the proposed model achieves superior performance metrics for WT3 and WT4, with root mean square error, mean absolute error, and mean absolute percentage error values ranging from 4.4796 to 4.8578, 3.2736 to 3.6543, and 0.2127 to 0.2193 for 1-h ahead forecast; 4.9674 to 5.7693, 3.5980 to 4.2028, and 0.2214 to 0.2295 for 3-h ahead forecast; and 5.8889 to 5.6338, 4.4247 to 4.1148, and 0.3064 to 0.2436 for 5-h ahead forecast, respectively. This pioneering two-stage decomposition and cross-scale CNN outperforms benchmarks by up to 74 % in RMSE. The proposed methodology improves short-term offshore wind power prediction by removing irregularities in the datasets.
An efficient fault classification method in solar photovoltaic modules using transfer learning and multi-scale convolutional neural network
(Elsevier, 2022) Korkmaz, Deniz; Açıkgöz, Hakan
Photovoltaic (PV) power generation is one of the remarkable energy types to provide clean and sustainable energy. Therefore, rapid fault detection and classification of PV modules can help to increase the reliability of the PV systems and reduce operating costs. In this study, an efficient PV fault detection method is proposed to classify different types of PV module anomalies using thermographic images. The proposed method is designed as a multi-scale convolutional neural network (CNN) with three branches based on the transfer learning strategy. The convolutional branches include multi-scale kernels with levels of visual perception and utilize pre-trained knowledge of the transferred network to improve the representation capability of the network. To overcome the imbalanced class distribution of the raw dataset, the oversampling technique is performed with the offline augmentation method, and the network performance is increased. In the experiments, 11 types of PV module faults such as cracking, diode, hot spot, offline module, and other classes are utilized. The average accuracy is obtained as 97.32% for fault detection and 93.51% for 11 anomaly types. The experimental results indicate that the proposed method gives higher classification accuracy and robustness in PV panel faults and outperforms the other deep learning methods and existing studies
Feature fusion-based hand gesture classification with time-domain descriptors and multi-level deep attention network
(Elsevier, 2025) Alcin, Omer Faruk; Korkmaz, Deniz; Acikgoz, Hakan
In conventional human-robot interaction (HRI), it is difficult to provide adaptability by located systems in the human body. Surface Electromyography (sEMG) signals have the potential to meet adaptability in HRI by directly representing movements, and classifying hand gestures with sEMG can be an effective solution to meet the increasing needs of these applications. In this paper, a hybrid and multi-scale convolutional neural network (CNN) model is proposed to obtain an efficient sEMG-based classification approach of human hand gestures. The proposed method includes an effective feature extraction process, including spectral moments, sparseness, irregularity factor, Teager-Kaiser energy, Shannon entropy, Katz fractal dimension, and Higuchi's fractal dimension, and waveform length. The obtained features are then converted to RGB images. The designed network is built on multi-scale convolutional blocks with residual learning and convolutional blocks, including the CBAM to improve the network performance by focusing on channel and spatial features. Furthermore, a pyramid non-pooling local block is utilized at the end of the network to learn more powerful features and their correlations. Five comprehensive publicly available datasets are evaluated in the experiments, and the obtained results are compared with the benchmark CNN models and network variations with different attention mechanisms. In the comparative evaluations, the CBAM achieves a classification accuracy between 84.62 % and 97.56 % while other attention mechanism results give accuracy values between 82.88 % and 97.17 %. The experiments show that the proposed method gives more accurate and robust classification performance compared with other variations and benchmark models.
An improved residual-based convolutional neural network for very short-term wind power forecasting
(ELSEVIER, 2021) Yıldız, Ceyhun; Açıkgöz, Hakan; Korkmaz, Deniz; Budak, Ümit
An accurate forecast of wind power is very important in terms of economic dispatch and the operation of power systems. However, effectively mitigating the risks arising from wind power in power system operations greatly reduces the risk of wind energy producers, exposing them to potential additional costs. Being aware of this challenge, we introduced a two-step novel deep learning method for wind power forecasting. The first stage includes processes of Variational Mode Decomposition (VMD)-based feature extraction and converting these features into images. In the second stage, an improved residual-based deep Convolutional Neural Network (CNN) was utilized to forecast wind power. Meteorological wind speed, wind direction, and wind power data, which are directly related to each other, were employed as a dataset. The combined dataset was procured from a wind farm in Turkey between January 1 and December 31, 2018. The results of the proposed method were compared with the results obtained from the state-of-the-art deep learning architectures namely SqueezeNet, GoogLeNet, ResNet-18, AlexNet, and VGG-16 as well as physical model based on available meteorological forecast data. The proposed method outperformed the other architectures and demonstrated promising results for very short-term wind power forecasting due to its competitive performance.
Long Short-Term Memory Network-based Speed Estimation Model of an Asynchronous Motor
(IEEE (Institute of Electrical and Electronics Engineers), 2021) Açıkgöz‬, ‪Hakan; Korkmaz, Deniz
In this paper, an effective deep rotor speed estimation model of an asynchronous motor is presented. The estimation model is based on the long short-term memory (LSTM) network which is one of the deep learning models. The designed model includes three main steps as the preprocessing, training of the deep speed estimation model, and evaluation of the model with testing. The dataset of the asynchronous motor model is obtained in MATLAB/Simulink environment under variable step speed references. The input parameters of the network are the dq-axis currents (id, iq) and voltages (vd, vq). The output is selected as the rotor speed (wr). The whole data is normalized to increase the estimation performance and then randomly divided into the training and validation. For the testing stage, different test data is also constructed. In the training process, the variation of the network performance is analyzed according to the neuron number increasing and optimum neuron number is achieved. The obtained results show that the proposed model is robust and efficient under the variable step speed references.
MEKANSAL PİRAMİT HAVUZLAMA TABANLI EVRİŞİMLİ SİNİR AĞI İLE OTOMATİK DRONE SINIFLANDIRMA
(2022) Korkmaz, Deniz; AÇIKGÖZ, Hakan
Hava sahalarının önemli olduğu bölgelerde dronları tespit etmek zorlu bir konu haline gelmiştir. Bu insansız hava araçlarının kontrolsüz uçuşları ve konuşlanmaları da istenmeyen bölgelerde çeşitli güvenlik sorunlarına sebep olur. Bu çalışmada, dronları kuşlardan ayırarak etkili bir şekilde sınıflandırabilmek için bir evrişimli sinir ağı (ESA) modeli önerilmiştir. Önerilen model, ön eğitimli AlexNet ile mekansal piramit havuzlama (MPH) yapısı kullanılarak tasarlanmıştır. Böylece, ağın evrişimsel katmanlarından gelen yerel öznitelikler birleştirerek ağın nesne özelliklerini daha kapsamlı bir şekilde öğrenmesi sağlanmış ve önerilen modelin sınıflandırma performansı artırılmıştır. Ayrıca, eğitim görüntülerinde çevrimdışı veri artırma tekniği uygulanarak örnek sayısı artırılmıştır. Önerilen yöntemin performansı AlexNet, ShuffleNet, GoogLeNet ve DarkNet gibi sıklıkla kullanılan ön eğitimli ESA mimarileri ile karşılaştırılmıştır. Gerçekleştirilen deneysel çalışmalarda önerilen yöntemin doğruluk, kesinlik, duyarlılık, özgüllük ve F1-skor değerleri sırasıyla %98.89, %97.83, %100, %97.78 ve %98.90 olarak elde edilmiştir. Çalışmada elde edilen tüm sonuçlar incelendiğinde, önerilen yöntemin farklı ortamlara ait drone görüntülerini kuşlardan ayırarak başarımı yüksek bir şekilde sınıflayabildiğini ortaya koymaktadır.
MSRConvNet: Classification of railway track defects using multi-scale residual convolutional neural network
(Pergamon-Elsevier Science Ltd, 2023) Acikgoz, Hakan; Korkmaz, Deniz
The development of an automated rail line defect classification system is of great benefit, as railway tracks must be periodically monitored and inspected to guarantee the safety of rail transportation. In this paper, an effective multi-scale residual convolutional network (MSRConvNet) model is proposed to classify the different types of railway track defects. The skip connections with residual learning blocks are used to increase the effectiveness of the network. The multi-scale convolutions are connected with parallel and two skip connections in the structure to distribute detailed feature maps with each other. Therefore, different scale feature maps can be extracted. The data augmentation method is performed to ensure a balanced class distribution and to eliminate the negative effect of the imbalanced dataset. The proposed model is compared with both benchmark deep learning models and the different variations of the designed network. The results verify that the proposed model can reach superior classification fulfillment, and the MSRConvNet provides an overall accuracy of 99.83%, precision of 99.83%, sensitivity of 99.83%, specificity of 99.94%, F1-score of 99.83%, and Matthew's correlation coefficient of 99.78% for four defect classes.
A novel ship classification network with cascade deep features for line-of-sight sea data
(2021) Ucar, Ferhat; Korkmaz, Deniz
In ship classifcation, selecting distinctive features and designing a proper classifer are two key points of the process. As a lack of most of the studies, these two essential points are considered separately. In this study, our proposal includes joint feature extraction, selection, and classifer design framework to build a novel deep cascade network for ship classifcation. We propose a transfer learning-based deep feature extraction using cascade Convolutional Neural Network architecture to convert the input image to multi-dimensional feature maps. The distributions of the MUTual Information (MUTInf) based feature selection algorithm compose a distinctive feature set originated for a public ship imagery dataset. The dataset consists of fve specifc classes of ships most existed in the maritime domain. A quadratic kernel-based non-linear Support Vector Machine is the designed classifer. Extensive experiments on the benchmark dataset indicate that the proposed framework can integrate the optimal feature set and a well-designed classifer to increase the performance of the classifcation process in ship imagery. In the experiments, the proposed method achieves an overall accuracy of 95.06%. The ship classes are also performed high classifcation performances into cargo, military, carrier, cruise, and tanker with an accuracy of 88.26%, 98.38%, 98.38%, 98.78%, and 91.50%, respectively. In addition, MUTInf feature selection reduces the features at a rate of 50.04%. These results show that the proposed method provides the highest performance value with less number of elements and outperforms state-of-the-art methods
A Novel Short-Term Photovoltaic Power Forecasting Approach based on Deep Convolutional Neural Network
(2021) Korkmaz, Deniz; Açıkgöz, Hakan; Yıldız, Ceyhun
In this study, a novel photovoltaic power forecasting system that utilizes a deep Convolutional Neural Network (CNN) structure and an input signal decomposition algorithm is proposed. The proposed CNN architecture extracts deep features to forecast short-term power using transfer learning-based AlexNet. The historical power, solar radiation, wind speed, and temperature data are selected as the input. The signal decomposition algorithm called Empirical Mode Decomposition (EMD) is utilized to decompose the historical power signal into sub-components. In order to extract deep features, all input parameters are converted to 2D feature maps and feed to the input of the CNN. The experiments are realized on a grid-tied Photovoltaic Power Plant (PVPP) that has 1000 kW installed capacity located in Turkey. The experiments are performed under four weather conditions as partial cloudy, cloudy-rainy, heavy-rainy, and sunny days to show the effectiveness of the proposed method. The obtained results are compared with the benchmark regression algorithms. When the results are analyzed, the proposed method gives the highest Correlation Coefficient (R) and the lowest Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and SMAPE values under all horizons and weather conditions. For 1-h to 5-h ahead, the average R values of the proposed method are obtained as 97.28%, 95.77%, 94.49%, 93.61%, and 92.62%, respectively. The average RMSE values are observed as 4.90%, 6.30%, 7.50%, 8.00%, and 9.17% for 1-h to 5-h ahead. The experimental results confirm that the proposed method outperforms the conventional regression algorithms and reveals effective results with its competitive performance.
Photovoltaic cell defect classification based on integration of residual-inception network and spatial pyramid pooling in electroluminescence images
(Pergamon-Elsevier Science Ltd, 2023) Acikgoz, Hakan; Korkmaz, Deniz; Budak, Umit
Electroluminescence (EL) imaging provides high spatial resolution and better identifies micro-defects for in-spection of photovoltaic (PV) modules. However, the analysis of EL images could be typically a challenging process due to complex defect patterns and inhomogeneous background structure. In this study, a deep con-volutional neural network (CNN) model using residual connections and spatial pyramid pooling (SPP) is pro-posed for the efficient classification of PV cell defects. The proposed CNN model is built on the Inception-v3 network. In this way, feature maps in inception modules are shared to reuse in deeper layers and the repre-sentation ability of features is enriched with the pooling process of the SPP in different sizes. Due to the imbalanced class distribution, offline data augmentation strategies are applied and network performance is further improved. The proposed method is evaluated on a publicly available dataset of 8 classes, of which 7 classes are defective and one class is defect-free images. In the comparative evaluation, while other approaches give accuracy values between 76.49% and 89.17%, this value is increased to 93.59% with the proposed method. The experimental results show that the proposed method exhibits more accurate and robust classification per-formance compared with other model combinations and CNN models.
PVEL-ViT: Adaptive token selective vision transformer for photovoltaic cell defect classification in electroluminescence imaging
(Elsevier, 2026) Acikgoz, Hakan; Korkmaz, Deniz; Bal, Cafer; Coteli, Resul; Dandil, Besir
The rapid expansion of photovoltaic (PV) technology has increased the need for reliable and automated defect detection to ensure the long-term efficiency and durability of PV modules. Electroluminescence (EL) imaging provides detailed visualization of internal defects and is a key modality for data-driven classification systems. This study proposes a vision transformer (ViT)-based framework, named PVEL-ViT, for accurate classification of PV cell defects from EL images. The architecture is constructed on a MetaFormer-based backbone and integrates an embedded multi-scale feature fusion module with an adaptive token selective attention mechanism. The designed PVEL-ViT captures fine-grained local defect patterns and global context while dynamically emphasizing defect-relevant tokens and suppressing redundant background information, thereby improving discriminative feature learning and robustness. The proposed framework is evaluated on an EL imaging dataset and compared against recent convolutional and transformer-based models. Experiments show that EfficientNetV2-m and EfficientViT-b2 achieve accuracy rates of 0.9386 and 0.9359, respectively, whereas PVEL-ViT attains a higher accuracy of 0.9584, corresponding to accuracy gains of 2.11% and 2.40% over these models. The obtained results indicate that PVEL-ViT provides a robust and scalable defect classification performance on EL imagery, enhancing the reliability of automated PV inspection pipelines and supporting monitoring in PV modules.
A Ship Detector Design Based on Deep Convolutional Neural Networks for Satellite Images
(Sakarya Üniversitesi, 2020) Uçar, Ferhat; Korkmaz, Deniz
Ship detection and classification systems from satellite images are challenging tasks with their requirements of feature extracting, advanced pre-processing, a variety of parameters obtained from satellites and other types of images, and analyzing of images. The dissimilarity of results, enhanced dataset requirement, the intricacy of the problem domain, general use of Synthetic Aperture Radar (SAR) images and problems on generalizability are some topics of the issues related to ship detection. In this study, we propose a Deep Convolutional Neural Network (DCNN) model for detecting the ships using the satellite images as inputs. Our model has acquired an adequate accuracy value by just using a pre-processed satellite image with a deep learning model built from scratch. The designed CNN model is constructed with a plain and easy to implement form in particular to the preferred satellite image set. Visual and graphical results show that the proposed model provides an efficient detection process with an accuracy of 99.60%.
Short-term offshore wind speed forecasting approach based on multi-stage decomposition and deep residual network with self-attention
(Pergamon-Elsevier Science Ltd, 2025) Acikgoz, Hakan; Korkmaz, Deniz
Wind energy is one of the widely used renewable energy systems. Wind speed forecasting is used to produce of wind energy and to ensure the sustainability of the power system. However, offshore wind speed forecasting is a challenging task with complex variables and highly nonlinear temporal dynamics of the ocean. This paper proposes a hybrid and robust offshore wind speed forecasting approach based on multi-stage decomposition, deep convolutional neural network (CNN), and extreme learning machine (ELM). Unlike conventional preprocessing for forecasting of renewable energy problems, the proposed approach combines two efficient decomposition methods as complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and ensemble empirical mode decomposition (EEMD). This method can decompose high-frequency and lowfrequency components of the wind speed. While high-frequency components are decomposed with the EEMD, low-frequency components are directly sent to the ELM model. The obtained mode functions from the EEMD are then fed to the designed network for forecasting. The CNN model is constructed with the deep residual network and self-attention (SA) mechanism to improve the network performance. In the comparative evaluations, while other approaches give lower forecasting performance between 0.8233 and 2.1885 for the root mean square error (RMSE), the proposed method presents the lowest RMSE value as 0.5400. The experimental results show that the proposed method exhibits more accurate and robust forecasting performance compared with other model combinations and deep learning models.
SolarNet: A hybrid reliable model based on convolutional neural network and variational mode decomposition for hourly photovoltaic power forecasting
(Elsevier, 2021) Korkmaz, Deniz
Photovoltaic (PV) power generation has high uncertainties due to the randomness and imbalance nature of solar energy and meteorological parameters. Hence, accurate PV power forecasts are essential in the operation of PV power plants (PVPP) for short-term dispatches and power generation schedules. In this study, a novel convolutional neural network (CNN) model, namely SolarNet, is proposed for short-term PV output power forecasting under different weather conditions and seasons. The proposed CNN model is designed as a parallel pooling structure to increase the forecasting performance. This structure consists of max-pooling and average-pooling blocks. The input parameters are the measured historical solar radiation, temperature, humidity, and active power data. The power data is decomposed into sub-components with the variational mode decomposition method and a data preprocessing and reconstruction process is utilized to obtain deep input feature maps. After input parameters are converted to hue-saturation-value (HSV) color space, the subsets feed to the input of the network. The experimental studies are performed with a case study using a 23.40 kW PVPP dataset from the Desert Knowledge Australia Solar Centre. The design CNN model is also compared with benchmark deep learning methods. In the experiments, the average correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE) of the proposed method for 1-h different weather conditions are achieved as 0.9871, 0.3090, and 0.1750, respectively. The experimental results show that the proposed deep forecasting method has higher accuracy and stability in short-term PV power forecasting and outperforms the other deep learning methods.

Yazar "Korkmaz, Deniz" seçeneğine göre listele

Sayfa Başına Sonuç

Sıralama seçenekleri