Virginia Tech® home

Sujal Bhavsar

Sujal Bhavsar


Sujal obtained a B.E degree in Mechanical Engineering from The Maharaja Sayjirao University of Baroda, India, in 2018. Post his baccalaureate, he joined one of the research groups at the Indian Institute of Technology Kanpur, India, as a project associate for a year. Over the last couple of years, he has been involved with several projects which fall within a broader spectrum of the energy, going from designing a latent energy storage window for the sustainable building to computational modeling of subcomponents of the micro-gas turbine for decentralized power production, and now, to operational aspects of renewable integration into the traditional power grid. During his Ph.D. years, his interest has naturally drifted towards the statistical and computational modeling aspects of the problem, which led him to switch gears and contribute to the emerging field of machine learning. Under the supervision of Dr. Pitchumani at the Advanced Materials and Technologies Laboratory, Virginia Tech, Sujal developed data-driven and machine-learning-based solutions to increasing renewable penetration into the traditional power grid, improving building energy management, and for thermal management of electric vehicle batteries in using probabilistic forecasting. His research interests are in renewable energy integration, applied machine learning, and uncertainty quantification.

Research Projects

Stochastic Economic Dispatch of Wind Power under Uncertainty

Operation of power systems with high penetrations of renewable energy sources requires tools for robust decision making under uncertainty. Stochastic economic dispatch and stochastic unit commitment are effective techniques for planning and operation under uncertainty, whose effectiveness depends on the cardinality and quality of the scenario set. This article proposes a machine learning method of k-means clustering for capturing relevant physical information from a large population of analog scenarios. Scenario samples drawn from the clusters are used in a two-stage stochastic economic dispatch computation. The effectiveness of the proposed approach is assessed on a synthetic 200-bus system with a geographic footprint over Illinois, USA for four months from each season of WIND Toolkit data. The combination of k-means clustering with sampling is shown to reduce the total operational cost by over 43% compared to sampling from populations based on heuristic clustering-based methods. Additionally, the variability in the mean cost is about 56% lower than the variability using Monte Carlo sampling. Moreover, the operational cost with the presented approach is shown to be close to the cost calculated based on a hindsight exact wind profile, signifying a highly accurate quantification of wind uncertainty by the presented k-means clustering based sampling method.

Machine learning building-stock simulator for end-use load forecasting

Building energy models are used to simulate heat and mass transfer and estimate end-use load in buildings. With the proliferation of solar photovoltaics on residential and commercial buildings, increasingly, buildings are expected to provide grid services, for which accurate and computationally efficient building energy simulations and end-use load prediction are imperative. Existing building energy simulation tools, however, have significant computational overhead that make them less practical in real-time deployment for optimization, design, uncertainty quantification and control in building energy management systems. This article presents a data-driven machine learning model based on light gradient boosting method (LightGBM) as a surrogate for a physics-based simulator for residential buildings to predict end-use load. The machine learning based surrogate model accounts for time-series related variables, seasonality and trend component of end-use load, and history of end-use load. The accuracy of the surrogate model is assessed on the prediction of the load profiles of 100 different houses in Cook County, Illinois, USA. The LightGBM surrogate model is shown to reduce the root-mean-squared error by 53% relative to a reference decision tree (DT) based model reported previously in the literature. Moreover, the model predicts the load spikes and high-ramp rate events throughout the year which are often the Achilles heel of other models in the literature. The machine learning based surrogate model is demonstrated to be computationally efficient, with a ten-fold reduction in the computational time compared to a physics-based building energy simulation, and suitable for uncertainty analysis and real-time control of building characteristics in response to uncertainty.

Robust model-predictive thermal control of lithium-ion batteries under drive cycle uncertainty

The exposure of electric vehicle batteries to rapid charging and discharging profiles, particularly under uncertainty in the drive schedules, requires effective strategies to forecast and avoid thermal excursions during operation. In this work, we present a light gradient boosting-based machine learning model to create probabilistic forecasts of the discharge current over a forecast horizon in real time. A surrogate of a physics-based computational model provides the functional relationship for the battery temperature, using which a stochastic model-predictive control strategy is developed to derive the optimal cooling schedule based on the probabilistic forecast. The effectiveness of the stochastic model-predictive control approach is assessed on the US06 driving cycle, in comparison to a constant coolant flow and persistence forecast-based control. It is shown that the total number of temperature excursion instances using the stochastic model-predictive control is reduced by about 69% compared to the constant coolant flow case and by over 51% compared to persistence-forecast-based control. Further, the total coolant usage is reduced by about 73% compared to persistence forecast-based control. The stochastic model-predictive control approach provides more flexibility by allowing a wider range of control and battery pack design parameters to obtain optimal performance with minimum coolant use.

Computationally Efficient Stochastic Unit Commitment and Economic Dispatch Under Wind and Solar Uncertainty

Stochastic unit commitment (UC) and economic dispatch (ED) are the most valuable tools in dealing with uncertainty in renewable forecast for power system operation and planning such that the overall expected production cost is minimized over the planning horizon. However, accurate calculation of the expected production cost requires assessment of a very large number of different scenarios of uncertain renewable resources, such as solar and wind, which is practically infeasible to simulate in real time. This article proposes a hybrid data-driven and physics-based model-predictive paradigm to efficiently solve for stochastic unit commitment and economic dispatch considering uncertainty in wind and solar power forecasts. The novelty of the approach lies in decoupling the production cost estimation from the unit commitment and economic dispatch optimization problems under uncertainty without compromising on the fidelity of the solutions. The presented approach considers, for the first time, solar uncertainty in UC/ED determination and enables efficient and accurate propagation of wind and solar uncertainty to estimate the statistics of the production cost. The expected cost is predicted 62.5% more accurately than the existing state-of-the-art, on unforeseen days during the entire year, and yields, for the first time, the associated physically consistent UC and ED profiles. The solutions are also shown to be flexible in providing adequate daily reserves to address any statistical deviations from probabilistic power forecasts. The computational time associated with the presented method is only about 10 s compared to over 24 h needed for a conventional stochastic UC/ED determination under uncertainty.

Reduced Order Scenario Generation for Stochastic Analysis of Solar Power Forecast

With increased reliance on solar-based energy generation in modern power systems, the problem of managing uncertainty in power system operation becomes crucial. This work presents a method to generate statistically accurate scenarios from probabilistic forecasts and a method based on unsupervised machine learning to reduce the cardinality of the scenarios set and speed up the computations, while preserving the statistical properties of the original set. Through a systematic parametric study, an optimum clustering-based machine learning method and its associated parameters are derived. This approach yields statistically equivalent characteristics as a full set with a substantially reduced cardinality (from 7000 to 20). The reduced set of scenarios also preserves the temporal correlation, which is imperative in time-series data and complies with the nonparametric distribution of power obtained from a probabilistic forecast at any particular time. Applying the proposed algorithm to the RTS-GMLC and California ISO yearly solar production data, it is shown that the uncertainty in the estimation of the statistical moments is reduced to less than 2–4% of the daily peak power value.

Machine-Learning Based Identification of Potential Adopters of Rooftop Solar PV

With the proliferation of rooftop solar photovoltaic installations, there is a need to proactively predict consumer potential for solar photovoltaic adoption, for improved electric utility planning and operation. Traditional analytical modeling approaches are limited to a few survey features and a larger part of survey would remain untouched by the decision model.  The present method is based on a data-driven modeling approach that utilizes a large set of consumer profile features that are strategically pruned in a machine learning framework to train a model for predicting potential solar adoption. The approach utilizes the Gradient Boosting Decision Tree model through a Light Gradient Boosting framework. Model training using focal-loss based supervision is used to overcome the difficulty in identifying the potential adopters that is inherent in conventional data-driven models. A Bayesian optimization approach is used to systematically arrive at the hyperparameters of the proposed model.  In addition, to overcome possible data sparsity in a limited survey sample, a Generative Adverserial Network has been adopted to create synthetic user samples and its effectiveness on model training is assessed. Validation of the proposed approach on a survey data collected by National Rural Electric Cooperative Association in Virginia in 2018 demonstrates the excellent predictive capability of the machine learning based approach to modeling solar adoption reliably.

A Reforecasting Based Reserve Estimation Under Uncertainty

The install capacity of solar-based energy sources has been increasing in a traditional power system to accelerate the production of carbon-free electricity. However, the reliability of the conventional control reserve practice is challenged under the increasing penetration of renewable sources like solar because of its unpredictable stochastic nature. The present article proposes a dynamic reserve estimation method that incorporates machine learning based re-forecasting to provide the day-ahead estimate of the mean and spread of uncertainty around the base-forecast, to estimate the up and down reserve relative to the base-forecast. The present method takes various endogeneous and exogeneous, mostly the calendar variables, as an input to estimate the day-ahead reserve. The proposed approach relies on the combination of re-forecasting and dynamic reserve estimation techniques. The proposed model showed consistent performance enhancement by reducing the amount of reserve allocation during the period of the larger peak-power production, which would otherwise be required with traditional reserve sizing methods. On average, an 80% reduction in the amount of energy needed for a one day reserve is found. Apart from a reduction in the amount of energy, the uncovered energy demand and overproduction are also seen to be significantly less. The validation of the proposed approach is demonstrated with a case study on CAISO solar production data.


  1. S. Bhavsar, R. Pitchumani, J. Maack, I. Satkauskas, M. Reynolds, W. Jones, “Stochastic Economic Dispatch of Wind Power under Uncertainty using Clustering-based Extreme Scenarios,” Electric Power Systems Research, 229, 110158, 2024.
  2. S. Bhavsar, R. Pitchumani, M. Reynolds, N. Merket and J. Reyna, “Machine Learning Surrogate of Physics-Based Building-Stock Simulator for End-Use Load Forecasting,” Energy and Buildings296, 113395, 2023.
  3. S. Bhavsar, R. Pitchumani, M.A. Ortega-Vazquez and N. Costilla-Enriquez, “A Hybrid Data-driven and Model-based Approach for Computationally Efficient Stochastic Unit Commitment and Economic Dispatch Under Wind and Solar Uncertainty,” International Journal of Electrical Power & Energy Systems, 151, 109144, 2023.
  4. S. Bhavsar, K. Kant and R. Pitchumani, “Robust Model-Predictive Thermal Control of Lithium-Ion Batteries under Drive Cycle Uncertainty,” Journal of Power Sources, 557, 232496, 2023.
  5. S. Bhavsar, R. Pitchumani and M. A. Ortega-Vazquez, “A Reforecasting-Based Dynamic Reserve Estimation for Variable Renewable Generation and Demand Uncertainty,” Electric Power Systems Research, 211, 108157, 2022.
  6. S. Bhavsar, R. Pitchumani and M. A. Ortega-Vazquez, “Reduced-order Scenario Generation for Stochastic Analysis of Solar Power Forecasts,” Applied Energy, 293, 116964, 2021.
  7. S. Bhavsar and R. Pitchumani, “A Novel Machine Learning Based Identification of Potential Adopter of Rooftop Solar Photovoltaics,” Applied Energy, 286, 116503, 2021.


Department of Energy