Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Shreyas Rajendra Hole has a total of 30 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    AI-enabled multi-class visual fault diagnosis for large-scale solar photovoltaic systems supporting energy transition and net-zero operations

    by Sudeep Mungara, Vinothkumar Kolluru, Thirunaavukkarasu Murugesan, Saptarshi Maiti, Shreyas Rajendra Hole, Nitin Rakesh

  • Open Access

    AI enabled geospatial intelligence for the energy transition: A comparative CNN study on CCUS, geothermal siting, and NetZero strategies

    by Sudeep Mungara, Vinothkumar Kolluru, Thirunaavukkarasu Murugesan, Saptarshi Maiti, Shreyas Rajendra Hole


  • Open Access

    Design of Evolutionary Algorithm to Restructure Distributed Generation in a Radial Distribution System for power losses

    by Shreyas Rajendra Hole, Pratiksha V Chafle, Girish Bhavekar, Suraj Karpe, Narayana Rao Palepu, Vinothkumar Kolluru, Shreekant Salotagi

    A continuously rising load demand places an increasing pressure and voltage decrease on the current power distribution network. Due to previously unheard-of issues, including a supply-demand gap, growing costs, and global warming, and the power supply sector urgently needs reform. This, in turn, highlights the significance of an intelligent grid. The smart grid includes generating integration at the distribution level as one of its features. If sized and appropriately located, distributed generation (DG) may significantly reduce power losses in the distribution system. For efficient voltage management, power management, and the reduction of power loss, the position and size of the DG are essential in this new power system topology. This article describes applying a genetic algorithm to reduce distribution losses in a feeder by maximizing the size and placement of DG at an existing radial distribution system that represents load with wind production linked to the substation. The performance of a Genetic Algorithm (GA) depends on several factors that must be accurately calibrated. The work in this article is an effort to address this connecting issue. With a voltage-dependent load model, the natural radial distribution system is considered. In this article, the ideal placement and size of the DG are determined by experimenting with different GA operator combinations while maintaining constant values for factors like population, crossover %, and generation. To examine the impact on active and reactive power loss, the best placement and size are implemented for the most minor loss. In the first instance, the available generation is employed, and the Genetic Algorithm (GA) determines the ideal position; in the second instance, both the optimal placement and size are implemented. The test findings show a 56.49 % decrease in loss if existing DG is linked at the ideal location as per GA and a 91.47 % reduction in loss if the location and size of DG are compared to existing DG, respectively. The evolutionary algorithm GA significantly improves the tail-end voltage and power losses.

  • Open Access

    AI—Prediction of Neisseria gonorrhoeae Resistance at the Point of Care from Genomic and Epidemiologic Data

    by Vinothkumar Kolluru, Shreyas Rajendra Hole, Ajeeb Sagar, Advaitha Naidu Chintakunta, Jeevaraj R, Shreekant Salotagi

    Background: Antimicrobial resistance (AMR) in Neisseria gonorrhoeae is an escalating global health challenge, affecting over 82 million individuals each year. The increasing resistance to commonly used antibiotics such as azithromycin, ciprofloxacin, and cefixime hinders timely and effective treatment, primarily due to the delayed detection of resistant strains. Methods: To overcome these limitations, a hybrid machine learning (ML) and deep learning (DL) framework was developed using a dataset comprising 3786 N. gonorrhoeae isolates. The dataset included clinical metadata and phenotypic resistance profiles. The preprocessing steps involved handling 23% data sparsity, imputing 31 skewed columns, and applying resampling and harmonisation techniques sensitive to data skewness. A predictive pipeline was constructed using both clinical variables and genomic unitigs, and a suite of 33 classifiers was evaluated. Results: The CatBoost model emerged as the top-performing ML algorithm, particularly due to its proficiency in handling categorical data, while a three-layered neural network served as the DL baseline. The ML models outperformed genome-wide association study (GWAS) benchmarks, achieving AUC scores of 0.97 (ciprofloxacin), 0.95 (cefixime), and 0.94 (azithromycin), representing a 4–7% improvement. SHAP analysis identified biologically relevant resistance markers, such as penA mosaic alleles and mtrR promoter mutations, validating the interpretability of the model. Conclusions: The study highlights the potential of ML-driven approaches to enhance the real-time prediction of antimicrobial resistance in N. gonorrhoeae. These methods can significantly contribute to antibiotic stewardship programs, although further validation is required in low-resource settings to confirm their generalisability and robustness across diverse populations.


  • Design of an iterative model incorporating transformer-based self-attention and proximal policy optimization for solar power systems

    by S.R. Hole, U.T. Kute, S. Upadhye, J.S. Morbale, S.P. Tondare, D.J. Saini

    Solar power integration into modern energy systems relies on accurate and adaptive control models, which are variable and uncertain. Traditional models such as ecurrent neural networkd and long short-term memory (LSTM) networks have problems with scalability, prediction accuracy, and computational efficiency. We present an enhanced control framework that combines three promising methods: Transformer-based models with self-attention mechanisms to understand predispositions, proximal policy optimisation for adaptive learning, and BNNs to update the probabilistic forecast. Transformer-based models use self-attention techniques to capture long-range relationships in time series, improving prediction accuracy and scalability. This model parallelizes inputs to reduce training time, making it better for large-scale applications. Empirically, this approach achieves 95 % prediction accuracy on validation data and 20 % faster training time than LSTM models. This works well with proximal policy optimization, an advanced reinforcement learning technique that allows real-time control policy adjustments based on solar power system data. Since maximum power point tracking and proportional-integral-derivative controllers optimize control actions, this actor-critic structure ensures learning stability and dependability. The model includes VARMAX for pre-emption. The system improved global efficiency by 15 % and updated the control strategy in 100 ms. BNNs are neural networks with uncertainty quantification for solar power generation variability management. BNNs provide probabilistic mean and variance predictions with uncertainty intervals, improving resilience and decision-making. The article found that the model covered the uncertainty interval at 90 %, while the typical neural network had a 10 % reduction in breadth. Using such methods, a solar power system control model would be robust, adaptive, and efficient. This comprehensive solution improves real-time accuracy, adaptiveness, and uncertainty handling, enhancing connected-solar load deployment performance and dependability. Интеграция солнечной энергии в современные энергетические системы опирается на точные и адаптивные модели управления, которые изменчивы и неопределенны. Традиционные модели, такие как сети RNN и LSTM, имеют проблемы с масштабируемостью, точностью прогнозирования и экономичностью вычислений. Мы представляем улучшенную структуру управления, которая объединяет три многообещающих метода: модели на основе трансформера с механизмами внутреннего внимания для понимания предрасположенностей, оптимизацию проксимальной политики для адаптивного обучения и байесовские нейронные сети для обновления вероятностного прогноза. Трансформируемые модели, применяющие методы самовнимания для использования долгосрочных временных отношений, повышают точность прогнозирования и масштабируемость. Эта модель распараллеливает входные данные для сокращения времени обучения, что делает ее лучшей для крупномасштабных приложений. Эмпирически этот подход достигает 95 % точности прогнозирования на данных проверки и приводит к уменьшению времени обучения на 20 %, чем модели LSTM. Это хорошо работает с PPO,передовойусиленной методикой обучения, которая позволяет корректировать политику управления в реальном времени на основе данных солнечной энергосистемы. Поскольку контроллеры MPPTиPIDоптимизируют управляющие действия, эта структура субъект–критик обеспечивает стабильность и надежность обучения. Модель включает VARMAx для упреждения. Система повысила глобальную эффективность на 15 % и обновила стратегию управления за 100 мс. BNN—это нейронные сети с количественной оценкой неопределенности для управления изменчивостью генерации солнечной энергии. BNN предоставляют вероятностные прогнозы среднего и дисперсии с интервалами неопределенности, повышая устойчивость и принятие решений. Установлено, что модель охватывает интервал неопределенности на 90 %, в то время как типичная нейронная сеть имеет 10 %-ный охват. При использовании таких методов модель управления солнечной энергосистемой будет надежной, адаптивной и эффективной. Это комплексное решение повышает точность, адаптивность и обработку неопределенности в...

  • Open Access

    Efficient Quantum-Enhanced Ensemble Fault Detection for Solar Energy Integration using an Iterative Game-Theoretic Approach with Adaptive Neuro-Fuzzy Inference and Energy Storage

    by Shreyas Hole, Jayavrinda Vrindavanam, Zahid Akhtar

    The rising energy requirements across the globe coupled with the need for sustainable development makes it imperative to harness renewable sources such as solar energy. That being said, changes in the weather and instability in the systems can enhance the reliability and efficiency of solar power systems which in the case of the current systems poses quite a challenge. The current solar power systems cannot adapt to such changes therefore causing energy wastages as well as problems with the grid. The new model presented here, the Ensemble Fault Detection Model for Solar Deployments (EFDMSD), uses technology to resolve a number of issues. This model has been built to harvest solar energy with greater speed and accuracy by means of Quantum Machine Learning (QML). A system’s changes can be addressed appropriately in real-time, thanks to the model’s Adaptive Neuro-Fuzzy Inference Control’s (ANFIS) control system. In other words, Game Theory is applied to explain energy shortage scenarios better and manage supply for peak periods. Energy Storage Systems are also present within the mix which allows for excess solar energy to be accumulated making the supply of energy more secure. These techniques, as a result of using artificial intelligence, are able to enhance energy production, stabilize grids, and optimize performance for many hardware configurations. With the implementation of the EFDMSD model, the availability and reliability of energy through the use of simpler solar systems is enhanced. Hence, there would be changes in the economic impacts while reducing the grid impact thus leading the country towards clean energy.


  • Hybrid Approach of TabNet and Transformer-XGBoost for Predicting Traffic Flow in Smart Cities

    by Shreyas Rajendra Hole, Vinothkumar Kolluru, Shreekant Salotagi, Yagnesh Challagundla, Sudeep Mungara, Mude Nagarjuna Naik

    A smart city's transportation system can be en-hanced if there is an accurate prediction of the traffic flow which develops in an area as time goes on. This article investigates the relative accuracy of predicting traffic flow at target intersections in a city, taking advantage of two models: Tab Net and a hybrid Transformer-XGBoost. Several metrics including Mean Squared Error (MSE) and Coefficient of determination (R2 score) among others were employed to evaluate the models employed. A sequential deep learning model known as Tab Net triumphing in the competition with MSE, RMSE, and MAE being 34.48, 5.87, and 3.73 respectively, and R2 score as high as 0.917, although performing well while enhancing identification of features such as year and junction. On the other hand, while reporting similar R2 score as low as 0.184, the Transformer-XGBoost hybrid model had structuring limitations where complex interdependencies and time variations were involved recording MSE, RMSE and MAE as 338.52, 18.40 and 13.24 respectively. The graphs of residuals and QQ plots served to uncover that the hybrid model failed making a right prediction during high traffic volume days, as general tabnet results shown in the final graphs are more stable. Following investigation of the models in this study, the researchers suggest implementing TabNet as a better option for traffic management explaining its high accuracy and providing valuable guidance for cities. This would result in improvement of the traffic flow prediction, which is highly needed for effective urban mobility solutions.


  • A Design of Hybrid Model and Bayesian Neural Networks for Smart Grid Stability Prediction

    by Shreyas Rajendra Hole, Vinothkumar Kolluru, Shreekant Salotagi, Yagnesh Challagundla, Sudeep Mungara, Sriramkumar R

    The latest evolution in power systems, is ‘smart grids'that offers real time monitoring, control features as well as effective management of renewable energy sources. Nonetheless, with the increase in the system complexity, the chances for insta- bility also increase and thus it is crucial to forecast the stability accurately for the safety of the grid. Weather it is in terms of Gradient Boosting Tree Algorithm or any other traditional algorithm, these machine learning models are successful in achieving the objectives of the task but they lack a forthcoming metric to assist in high stake situations, uncertainty quantification. The paper is devoted to the analysis of the reliability of smart grid systems based on Bayesian Neural Networks (BNN) taking into account both the reliability of predictive modeling and forecasting uncertainty. The smart grid stability augmented dataset was used to benchmark the efficacy of using BNNs or BNNs combined hybrid models, one with XGBoost, for feature extraction and the other with LightGBM. The present study shows that the BNN used with the above mentioned smart grid system achieved an accuracy of 92% with precision of 0.89, recall of 0.91 and F1 of 0.90. These metrics are appending with the help of the additional features of BNN where they are capable of addressing the key questions around uncertainty and thus the metrics confirming the predictions are quite strong in terms of real-time grid stability evaluation. This article encapsulates a detailed study of BNN architecture, the role of uncertainty in machine learning models and the natural application of such models in the design of dependable and fault-tolerant smart grids.


  • Hybrid PCA-Based Machine Learning Models for Predictive Analytics in Urban Health Monitoring Systems

    by Shreyas Rajendra Hole, Girish S Bhavekar, Arvind Kumar Prajapati, Vinothkumar Kolluru, Suraj Rajesh Karpe, Jayavrinda Vrindavanam

    In this paper, a hybrid PCA-based approach is proposed for predicting the respiratory imbalance in an urban health monitoring system. To do this, three models namely Random Forest, XGBoost, and Multi-Layer Perceptron (MLP) were integrated into one system to improve the probability of prediction while maintaining the trade-off with the dimension reduction - a common variable in big data tasks (which, this effort is focused on). Several variables regarding the environment and health of the patients were used to form a dataset that was later transformed with the help of PCA meaning the results would get better because models with better dimensionality tend to perform better. The Hybrid PCA + MLP model produced the best accuracy which 99% and AUC - ROC which 0.90, there were also 99% produced by XGBoost but AUC - ROC was 0.83. Random forest produced “slightly” less accurate predictions of 95% with an AUC - ROC of 0.72 however; this proved the best method in terms of the computational burden since the training time recorded was about 0.55 seconds, which was lower than that of XGBoost 0.45 seconds and MLP 0.94 seconds. These findings show that hybrid PCA based models do not only improve the accuracy of predictions made but also reduce the amount of computational resources needed, which is ideal for tasks that occur in real time whereby smart urban health monitoring is a part of a bigger CPS system.