Ronnie O. Serfa Juan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ronnie O. Serfa Juan has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
Hi Seok Kim
Department of Electronic Engineering, Cheongju University, Cheongju City, Cheongju Si 28503, South Korea
8 shared publications- Alison R. Gerken
USDA, Agricultural Research Service, Center for Grain and Animal Health Research , Manhattan, KS
4 shared publications - Lean Karlo Santos Tolentino
Electronics Engineering Department, Technological University of the Philippines, Manila, Philippines
4 shared publications - Bubryur Kim
School of Space Engineering Sciences, Kyungpook National University, Daegu, Korea
3 shared publications - Dong-Eun Lee
School of Architecture, Civil, Environmental and Energy Engineering, Kyungpook National University, 80 Daehak-ro, Buk-gu, Daegu, 41566, Republic of Korea
3 shared publications Byoung Hwan Ko
Dept. of Electronic Engg., Cheongju University, Cheongju City, South Korea
2 shared publications
Latest Publications
- Open Access
Spatiotemporal Decision-AI for Species-Specific and Non-Lethal Management of Avian Predators in Aquaculture
Avian predation remains a persistent challenge in aquaculture, as conventional deterrents often lose effectiveness due to bird habituation and limited species specificity. Some methods may also unintentionally affect surrounding ecosystems. To address these limitations, this study presents a spatiotemporal Decision-AI approach for species-specific and non-lethal management of avian predators and nuisance birds by integrating visual intelligence with ecological context. The proposed framework integrates a YOLOv8-based detection model with a metadata extraction and spatiotemporal analysis pipeline that records species identity, population count, time, and location. YOLOv8 serves as the visual detection component, while the methodological contribution of the framework lies in transforming these detections into contextual ecological indicators through temporal occurrence analysis, species-specific weighting, and risk-based Decision-AI. This allows the system to transform raw image detections into meaningful ecological indicators that can guide management decisions. The study was evaluated using a dataset of more than 4500 high-resolution images collected from aquaculture sites across multiple U.S. locations. The proposed model achieved an overall classification accuracy of 97.2%, with precision and recall values of 95.8% and 96.4%, respectively. More importantly, the integration of detection results with temporal and spatial metadata enabled the identification of activity patterns, including peak visitation periods, species dominance trends, and location-specific risk levels. These insights can support targeted and non-lethal intervention strategies, such as optimizing deterrent timing and applying species-aware responses, while minimizing disturbance to non-target wildlife. By linking computer vision with ecological decision support, the proposed study moves beyond passive monitoring toward data-informed, adaptive management. This approach provides a scalable and interpretable pathway for improving aquaculture sustainability while maintaining ecological balance.
Open AccessSpatiotemporal Decision-AI for Species-Specific and Non-Lethal Management of Avian Predators in Aquaculture
Avian predation remains a persistent challenge in aquaculture, as conventional deterrents often lose effectiveness due to bird habituation and limited species specificity. Some methods may also unintentionally affect surrounding ecosystems. To address these limitations, this study presents a spatiotemporal Decision-AI approach for species-specific and non-lethal management of avian predators and nuisance birds by integrating visual intelligence with ecological context. The proposed framework integrates a YOLOv8-based detection model with a metadata extraction and spatiotemporal analysis pipeline that records species identity, population count, time, and location. YOLOv8 serves as the visual detection component, while the methodological contribution of the framework lies in transforming these detections into contextual ecological indicators through temporal occurrence analysis, species-specific weighting, and risk-based Decision-AI. This allows the system to transform raw image detections into meaningful ecological indicators that can guide management decisions. The study was evaluated using a dataset of more than 4500 high-resolution images collected from aquaculture sites across multiple U.S. locations. The proposed model achieved an overall classification accuracy of 97.2%, with precision and recall values of 95.8% and 96.4%, respectively. More importantly, the integration of detection results with temporal and spatial metadata enabled the identification of activity patterns, including peak visitation periods, species dominance trends, and location-specific risk levels. These insights can support targeted and non-lethal intervention strategies, such as optimizing deterrent timing and applying species-aware responses, while minimizing disturbance to non-target wildlife. By linking computer vision with ecological decision support, the proposed study moves beyond passive monitoring toward data-informed, adaptive management. This approach provides a scalable and interpretable pathway for improving aquaculture sustainability while maintaining ecological balance.
- 31 July 2026
- Published by Wiley in Entomologia Experimentalis et Applicata
- Vol. 174 (11) , 1369-1380
A Case Study for the Use of Image‐Based Monitoring of Moth Populations in Stored‐Product Environments Using Smart Sticky Traps
Effective monitoring of stored‐product insect pests is essential for timely intervention and informed decision‐making in integrated pest management (IPM) programs. In practice, however, routine inspection of sticky traps remains labor‐intensive and is often constrained by limited temporal resolution and observer variability. Here, we evaluate an automated, image‐based monitoring system designed to estimate moth populations from sticky traps under realistic storage conditions. The system combines camera‐equipped traps with image preprocessing and a convolutional neural network (CNN) to distinguish moths from non‐moth insects and other trap‐associated artifacts. A dataset of 1739 high‐resolution sticky‐trap images collected under variable lighting, trap orientation, and contamination levels was annotated and used for model training and evaluation. The automated approach achieved an overall classification accuracy of 95.8%, with precision, recall, and F1‐scores consistently exceeding 90%, demonstrating reliable performance even in images containing overlapping insects and debris. Beyond classification accuracy, the system enabled continuous estimation of moth abundance over time, revealing site‐specific and temporal variation in moth activity relevant to IPM decision thresholds. By substantially reducing manual inspection effort while maintaining biologically meaningful population estimates, this approach offers a practical and scalable tool for enhancing stored‐product pest surveillance. The results support the feasibility of integrating automated image analysis into routine monitoring workflows to improve responsiveness and efficiency in postharvest pest management.
- 31 July 2026
- Published by Wiley in Entomologia Experimentalis et Applicata
- Vol. 174 (11) , 1369-1380
A Case Study for the Use of Image‐Based Monitoring of Moth Populations in Stored‐Product Environments Using Smart Sticky Traps
Effective monitoring of stored‐product insect pests is essential for timely intervention and informed decision‐making in integrated pest management (IPM) programs. In practice, however, routine inspection of sticky traps remains labor‐intensive and is often constrained by limited temporal resolution and observer variability. Here, we evaluate an automated, image‐based monitoring system designed to estimate moth populations from sticky traps under realistic storage conditions. The system combines camera‐equipped traps with image preprocessing and a convolutional neural network (CNN) to distinguish moths from non‐moth insects and other trap‐associated artifacts. A dataset of 1739 high‐resolution sticky‐trap images collected under variable lighting, trap orientation, and contamination levels was annotated and used for model training and evaluation. The automated approach achieved an overall classification accuracy of 95.8%, with precision, recall, and F1‐scores consistently exceeding 90%, demonstrating reliable performance even in images containing overlapping insects and debris. Beyond classification accuracy, the system enabled continuous estimation of moth abundance over time, revealing site‐specific and temporal variation in moth activity relevant to IPM decision thresholds. By substantially reducing manual inspection effort while maintaining biologically meaningful population estimates, this approach offers a practical and scalable tool for enhancing stored‐product pest surveillance. The results support the feasibility of integrating automated image analysis into routine monitoring workflows to improve responsiveness and efficiency in postharvest pest management.
- 29 December 2025
- Published by Oxford University Press (OUP) in International Journal of Food Science and Technology
- Vol. 61 (1)
AI-based bread quality assessment using image processing techniques and the developed BQe-CNN
Ensuring consistent bread quality is critical for maintaining industry standards, reducing waste, and sustaining consumer satisfaction. Traditional quality assessment relies on expert visual inspection, which is subjective, time-consuming, and prone to inconsistency. While proprietary machine analysers exist, their high costs limit accessibility for many bakeries and research facilities. This study presents an artificial intelligence (AI)-driven solution that applies advanced image processing to automate and improve bread quality assessment. The Bread Quality Enhanced Convolutional Neural Network (BQe-CNN) analyzes features, such as porosity, texture, air cell structure, and expert-scored colour and crumb metrics. Achieving classification accuracies of 92% for bread colours and 88% for quality levels, the model offers reliable, cost-effective evaluation without requiring expensive equipment or specialised training. By integrating residual connections and attention mechanisms, BQe-CNN captures subtle variations in texture and air-cell distribution, outperforming conventional methods in precision and scalability. However, artisan scoring—valuing nuanced aesthetics central to traditional bread-making—remains challenging for full automation, highlighting opportunities for hybrid or more advanced approaches. Overall, BQe-CNN demonstrates the potential of AI-based image analysis to deliver real-time, standardised quality control. This innovation cuts variability, reduces waste, and provides computational precision with baking expertise—advancing bread quality management.
- 29 December 2025
- Published by Oxford University Press (OUP) in International Journal of Food Science and Technology
- Vol. 61 (1)
AI-based bread quality assessment using image processing techniques and the developed BQe-CNN
Ensuring consistent bread quality is critical for maintaining industry standards, reducing waste, and sustaining consumer satisfaction. Traditional quality assessment relies on expert visual inspection, which is subjective, time-consuming, and prone to inconsistency. While proprietary machine analysers exist, their high costs limit accessibility for many bakeries and research facilities. This study presents an artificial intelligence (AI)-driven solution that applies advanced image processing to automate and improve bread quality assessment. The Bread Quality Enhanced Convolutional Neural Network (BQe-CNN) analyzes features, such as porosity, texture, air cell structure, and expert-scored colour and crumb metrics. Achieving classification accuracies of 92% for bread colours and 88% for quality levels, the model offers reliable, cost-effective evaluation without requiring expensive equipment or specialised training. By integrating residual connections and attention mechanisms, BQe-CNN captures subtle variations in texture and air-cell distribution, outperforming conventional methods in precision and scalability. However, artisan scoring—valuing nuanced aesthetics central to traditional bread-making—remains challenging for full automation, highlighting opportunities for hybrid or more advanced approaches. Overall, BQe-CNN demonstrates the potential of AI-based image analysis to deliver real-time, standardised quality control. This innovation cuts variability, reduces waste, and provides computational precision with baking expertise—advancing bread quality management.
- Open Access
- 12 December 2025
- Published by Springer Nature in Scientific Data
- Vol. 13 (1)
Numerical Signature Dataset of Curculionidae and Tenebrionidae Beetle Fragments for ML Identification
This data descriptor presents a curated dataset of numerical signature descriptors derived from fragment images of six economically significant stored-product beetle species from the families Curculionidae ( Sitophilus zeamais , Sitophilus oryzae , Sitophilus granarius ) and Tenebrionidae ( Tribolium castaneum , Tribolium confusum , Latheticus oryzae ). Anatomical fragments—including antennae, elytra, thorax, snout (Curculionidae), and head aspect ratio (Tenebrionidae)—were imaged using digital microscopy and processed with standardized image acquisition and segmentation techniques. From each image, four statistical descriptors—skewness, kurtosis, entropy, and standard deviation—were extracted, which form compact numerical signatures that capture fragment-level texture and morphological variation. These descriptors are designed to support artificial intelligence and machine learning workflows for automated classification in entomological diagnostics and post-harvest pest detection. The dataset includes 3,423 fragment images, each linked to a numerical signature vector and labeled by species, anatomical region, and metadata. This dataset adheres to Findable, Accessible, Interoperable, Reusable (FAIR) principles and is intended for open reuse in entomological AI research and machine learning-driven insect fragment identification workflows.
- 12 December 2025
- Published by Springer Nature in Scientific Data
- Vol. 13 (1)
Open AccessNumerical Signature Dataset of Curculionidae and Tenebrionidae Beetle Fragments for ML Identification
This data descriptor presents a curated dataset of numerical signature descriptors derived from fragment images of six economically significant stored-product beetle species from the families Curculionidae ( Sitophilus zeamais , Sitophilus oryzae , Sitophilus granarius ) and Tenebrionidae ( Tribolium castaneum , Tribolium confusum , Latheticus oryzae ). Anatomical fragments—including antennae, elytra, thorax, snout (Curculionidae), and head aspect ratio (Tenebrionidae)—were imaged using digital microscopy and processed with standardized image acquisition and segmentation techniques. From each image, four statistical descriptors—skewness, kurtosis, entropy, and standard deviation—were extracted, which form compact numerical signatures that capture fragment-level texture and morphological variation. These descriptors are designed to support artificial intelligence and machine learning workflows for automated classification in entomological diagnostics and post-harvest pest detection. The dataset includes 3,423 fragment images, each linked to a numerical signature vector and labeled by species, anatomical region, and metadata. This dataset adheres to Findable, Accessible, Interoperable, Reusable (FAIR) principles and is intended for open reuse in entomological AI research and machine learning-driven insect fragment identification workflows.
- 22 October 2025
- Published by Wiley in Entomologia Experimentalis et Applicata
- Vol. 173 (12) , 1285-1305
AI ‐Based Image Profiling and Detection for the Beetle Byte Quintet Using Vision Transformer ( ViT ) in Advanced Stored Product Infestation Monitoring
Managing beetles that infest stored products is crucial for reducing losses in harvest supply chains and improving food security and safety. Successful pest management programs require effective and timely monitoring systems; however, traditional detection methods are often time‐consuming, labor‐intensive, and reliant on taxonomic expertise. Automated approaches using computer vision offer improved efficiency, yet consistently struggle to distinguish between species due to the small size and morphological similarities among beetles. This study introduces an AI‐based image profiling framework for detecting the “Beetle Byte Quintet”—a group of five morphologically similar and economically important stored‐product beetle species: the maize weevil ( Sitophilus zeamais ), red flour beetle ( Tribolium castaneum ), rusty grain beetle ( Cryptolestes ferrugineus ), sawtoothed grain beetle ( Oryzaephilus surinamensis ), and lesser grain borer ( Rhyzopertha dominica ). Leveraging the capabilities of Vision Transformers (ViT), a model was trained on a manually curated, high‐resolution image dataset annotated to highlight distinctive morphological traits such as body shape, coloration, and exoskeleton patterns. Through this profiling approach, the ViT model achieved a classification accuracy of 99.34% during training and 96.57% on unseen test data, demonstrating strong generalizability and outperforming traditional CNN‐based classifiers. The integration of ViT enables interpretable attention maps, offering entomological insight and validation while supporting real‐time detection applications. This AI‐enhanced system presents a scalable solution adaptable to large‐scale storage facilities, contributing to precise, targeted pest control strategies that reduce economic losses and promote sustainable food security practices.
- 22 October 2025
- Published by Wiley in Entomologia Experimentalis et Applicata
- Vol. 173 (12) , 1285-1305
AI ‐Based Image Profiling and Detection for the Beetle Byte Quintet Using Vision Transformer ( ViT ) in Advanced Stored Product Infestation Monitoring
Managing beetles that infest stored products is crucial for reducing losses in harvest supply chains and improving food security and safety. Successful pest management programs require effective and timely monitoring systems; however, traditional detection methods are often time‐consuming, labor‐intensive, and reliant on taxonomic expertise. Automated approaches using computer vision offer improved efficiency, yet consistently struggle to distinguish between species due to the small size and morphological similarities among beetles. This study introduces an AI‐based image profiling framework for detecting the “Beetle Byte Quintet”—a group of five morphologically similar and economically important stored‐product beetle species: the maize weevil ( Sitophilus zeamais ), red flour beetle ( Tribolium castaneum ), rusty grain beetle ( Cryptolestes ferrugineus ), sawtoothed grain beetle ( Oryzaephilus surinamensis ), and lesser grain borer ( Rhyzopertha dominica ). Leveraging the capabilities of Vision Transformers (ViT), a model was trained on a manually curated, high‐resolution image dataset annotated to highlight distinctive morphological traits such as body shape, coloration, and exoskeleton patterns. Through this profiling approach, the ViT model achieved a classification accuracy of 99.34% during training and 96.57% on unseen test data, demonstrating strong generalizability and outperforming traditional CNN‐based classifiers. The integration of ViT enables interpretable attention maps, offering entomological insight and validation while supporting real‐time detection applications. This AI‐enhanced system presents a scalable solution adaptable to large‐scale storage facilities, contributing to precise, targeted pest control strategies that reduce economic losses and promote sustainable food security practices.
- Open Access
- 1 November 2024
- Published by Qeios Ltd
Review of: "Academic Performance Prediction Based on Convolutional Neural Networks and IRT Parameters as RGB Images"
The manuscript provides valuable insights, but some sections could benefit from greater clarity. For instance, simplifying complex terms and rephrasing lengthy sentences could improve readability, making the research more accessible to a broader audi...
- 1 November 2024
- Published by Qeios Ltd
Open AccessReview of: "Academic Performance Prediction Based on Convolutional Neural Networks and IRT Parameters as RGB Images"
The manuscript provides valuable insights, but some sections could benefit from greater clarity. For instance, simplifying complex terms and rephrasing lengthy sentences could improve readability, making the research more accessible to a broader audi...
- Open Access
An Automated Image-Based Multivariant Concrete Defect Recognition Using a Convolutional Neural Network with an Integrated Pooling Module
Buildings and infrastructure in congested metropolitan areas are continuously deteriorating. Various structural flaws such as surface cracks, spalling, delamination, and other defects are found, and keep on progressing. Traditionally, the assessment and inspection is conducted by humans; however, due to human physiology, the assessment limits the accuracy of image evaluation, making it more subjective rather than objective. Thus, in this study, a multivariant defect recognition technique was developed to efficiently assess the various structural health issues of concrete. The image dataset used was comprised of 3650 different types of concrete defects, including surface cracks, delamination, spalling, and non-crack concretes. The proposed scheme of this paper is the development of an automated image-based concrete condition recognition technique to categorize, not only non-defective concrete into defective concrete, but also multivariant defects such as surface cracks, delamination, and spalling. The developed convolution-based model multivariant defect recognition neural network can recognize different types of defects on concretes. The trained model observed a 98.8% defect detection accuracy. In addition, the proposed system can promote the development of various defect detection and recognition methods, which can accelerate the evaluation of the conditions of existing structures.
Open AccessAn Automated Image-Based Multivariant Concrete Defect Recognition Using a Convolutional Neural Network with an Integrated Pooling Module
Buildings and infrastructure in congested metropolitan areas are continuously deteriorating. Various structural flaws such as surface cracks, spalling, delamination, and other defects are found, and keep on progressing. Traditionally, the assessment and inspection is conducted by humans; however, due to human physiology, the assessment limits the accuracy of image evaluation, making it more subjective rather than objective. Thus, in this study, a multivariant defect recognition technique was developed to efficiently assess the various structural health issues of concrete. The image dataset used was comprised of 3650 different types of concrete defects, including surface cracks, delamination, spalling, and non-crack concretes. The proposed scheme of this paper is the development of an automated image-based concrete condition recognition technique to categorize, not only non-defective concrete into defective concrete, but also multivariant defects such as surface cracks, delamination, and spalling. The developed convolution-based model multivariant defect recognition neural network can recognize different types of defects on concretes. The trained model observed a 98.8% defect detection accuracy. In addition, the proposed system can promote the development of various defect detection and recognition methods, which can accelerate the evaluation of the conditions of existing structures.
- Open Access
Multivariate Analysis of Concrete Image Using Thermography and Edge Detection
With the growing demand for structural health monitoring system applications, data imaging is an ideal method for performing regular routine maintenance inspections. Image analysis can provide invaluable information about the health conditions of a structure’s existing infrastructure by recording and analyzing exterior damages. Therefore, it is desirable to have an automated approach that reports defects on images reliably and robustly. This paper presents a multivariate analysis approach for images, specifically for assessing substantial damage (such as cracks). The image analysis provides graph representations that are related to the image, such as the histogram. In addition, image-processing techniques such as grayscale are also implemented, which enhance the object’s information present in the image. In addition, this study uses image segmentation and a neural network, for transforming an image to analyze it more easily and as a classifier, respectively. Initially, each concrete structure image is preprocessed to highlight the crack. A neural network is used to calculate and categorize the visual characteristics of each region, and it shows an accuracy for classification of 98%. Experimental results show that thermal image extraction yields better histogram and cumulative distribution function features. The system can promote the development of various thermal image applications, such as nonphysical visual recognition and fault detection analysis.
Open AccessMultivariate Analysis of Concrete Image Using Thermography and Edge Detection
With the growing demand for structural health monitoring system applications, data imaging is an ideal method for performing regular routine maintenance inspections. Image analysis can provide invaluable information about the health conditions of a structure’s existing infrastructure by recording and analyzing exterior damages. Therefore, it is desirable to have an automated approach that reports defects on images reliably and robustly. This paper presents a multivariate analysis approach for images, specifically for assessing substantial damage (such as cracks). The image analysis provides graph representations that are related to the image, such as the histogram. In addition, image-processing techniques such as grayscale are also implemented, which enhance the object’s information present in the image. In addition, this study uses image segmentation and a neural network, for transforming an image to analyze it more easily and as a classifier, respectively. Initially, each concrete structure image is preprocessed to highlight the crack. A neural network is used to calculate and categorize the visual characteristics of each region, and it shows an accuracy for classification of 98%. Experimental results show that thermal image extraction yields better histogram and cumulative distribution function features. The system can promote the development of various thermal image applications, such as nonphysical visual recognition and fault detection analysis.
- Open Access
- 9 September 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (18) , 8388
Importance of Image Enhancement and CDF for Fault Assessment of Photovoltaic Module Using IR Thermal Image
Infrared thermography is the science of measuring the infrared energy emitted by an object, translating it to apparent temperature variance, and displaying the result as an infrared image. Significantly, acquiring thermal images delivers distinctive levels of temperature differences in solar panels that correspond to their health status, which is beneficial for the early detection of defects. The proposed algorithm aims to analyze the thermal solar panel images. The acquired thermal solar panel images were segmented into solar cell sizes to provide more detailed information by region or cell area instead of the entire solar panel. This paper uses both the image histogram information and its corresponding cumulative distribution function (CDF), useful for image analysis. The acquired thermal solar panel images are enhanced using grayscale, histogram equalization, and adaptive histogram equalization to represent a domain that is easier to analyze. The experimental results reveal that the extraction results of thermal images provide better histogram and CDF features. Furthermore, the proposed scheme includes the convolutional neural network (CNN) for classifying the enhanced images, which shows that a 97% accuracy of classification was achieved. The proposed scheme could promote different thermal image applications—for example, non-physical visual recognition and fault detection analysis.
- 9 September 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (18) , 8388
Open AccessImportance of Image Enhancement and CDF for Fault Assessment of Photovoltaic Module Using IR Thermal Image
Infrared thermography is the science of measuring the infrared energy emitted by an object, translating it to apparent temperature variance, and displaying the result as an infrared image. Significantly, acquiring thermal images delivers distinctive levels of temperature differences in solar panels that correspond to their health status, which is beneficial for the early detection of defects. The proposed algorithm aims to analyze the thermal solar panel images. The acquired thermal solar panel images were segmented into solar cell sizes to provide more detailed information by region or cell area instead of the entire solar panel. This paper uses both the image histogram information and its corresponding cumulative distribution function (CDF), useful for image analysis. The acquired thermal solar panel images are enhanced using grayscale, histogram equalization, and adaptive histogram equalization to represent a domain that is easier to analyze. The experimental results reveal that the extraction results of thermal images provide better histogram and CDF features. Furthermore, the proposed scheme includes the convolutional neural network (CNN) for classifying the enhanced images, which shows that a 97% accuracy of classification was achieved. The proposed scheme could promote different thermal image applications—for example, non-physical visual recognition and fault detection analysis.
- 21 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 294-299
Implementation of Generalized Regression Neural Network (GRNN) for Solar Panel Power Estimation
An efficient algorithm to characterize the current-voltage (IV) curve of photovoltaic (PV) modules under different operating conditions is necessary for solar power estimation and some operational stability. In this paper, the developed algorithm utilizes a generalized regression neural network (GRNN) as a power estimator for solar panels. The proposed model uses the seven input variables namely, the IV characteristic curve, weather condition, and temperature parameters from the testbed solar panel modules of Cheongju University dated August 2019 to July 2020. The dataset is divided into three sections as training, validation, and testing set to 60%, 20%, and 20 %, respectively. The simulation provides comparative results for the actual and predicted power. Moreover, GRNN shows a better predicted power output compared to a feed-forward neural network (FFNN). The regression value results in a much nearer to 1 that fits the dataset. The correlation coefficient is 0.9961 which identifies the provided dataset is on the line of best fit.
- 21 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 294-299
Implementation of Generalized Regression Neural Network (GRNN) for Solar Panel Power Estimation
An efficient algorithm to characterize the current-voltage (IV) curve of photovoltaic (PV) modules under different operating conditions is necessary for solar power estimation and some operational stability. In this paper, the developed algorithm utilizes a generalized regression neural network (GRNN) as a power estimator for solar panels. The proposed model uses the seven input variables namely, the IV characteristic curve, weather condition, and temperature parameters from the testbed solar panel modules of Cheongju University dated August 2019 to July 2020. The dataset is divided into three sections as training, validation, and testing set to 60%, 20%, and 20 %, respectively. The simulation provides comparative results for the actual and predicted power. Moreover, GRNN shows a better predicted power output compared to a feed-forward neural network (FFNN). The regression value results in a much nearer to 1 that fits the dataset. The correlation coefficient is 0.9961 which identifies the provided dataset is on the line of best fit.