Date of Award
2026
Degree Type
Dissertation
Degree Name
Doctor of Philosophy in Electrical Engineering
Department
Electrical, Computer, and Biomedical Engineering
First Advisor
Kaushallya Adhikari
Abstract
The state of charge (SOC) of a battery indicates the remaining charge in the battery relative to its nominal maximum charge capacity. The accurate estimation of SOC is of utmost importance for efficient energy management as it indicates the level of energy stored in the battery.
Bayesian and machine learning (ML) approaches represent two distinct estimation strategies for battery state estimation. Bayesian filters, such as the Kalman filter (KF) and particle filter (PF), operate sequentially by incorporating physical models of battery dynamics. In contrast, ML-based methods are data-driven, and their performance largely depends on the quantity and quality of training data. Both approaches have gained significant attention for SOC estimation in recent years.
This dissertation employs Bayesian and ML-based estimation techniques to address three key challenges in SOC estimation for photovoltaic battery energy storage systems (PV-BESS). First, most existing SOC estimators are developed for electric vehicle (EV) applications, which operate under charging and discharging profiles that differ significantly from those of PV-BESS systems. To address this challenge, this work develops a novel KF-based estimation framework by introducing a specialized battery modeling approach for PV-BESS that incorporates weather-related parameters, including temperature, solar irradiance, and humidity. The proposed framework demonstrates improved estimation performance for PV operating conditions. In addition, a separate study using ML-based SOC estimation investigates the influence of various input features, including weather parameters. The results indicate that incorporating weather-related data improves estimation accuracy, thereby providing cross-validation of the proposed PV-BESS modeling framework from a data-driven perspective.
Second, this dissertation proposes a simple yet effective noise initialization technique for the KF estimator. The proposed approach eliminates the need for speculative noise tuning, which may otherwise rely on manual parameter selection and lead to inconsistent estimation performance under varying operating conditions.
Third, this work investigates the impact of missing measurements on battery state estimation from both Bayesian and ML perspectives. The effect of missing measurements is analyzed for multiple Bayesian estimators, including KF and PF. An analytical framework is developed to derive a closed-form expression that quantifies the impact of consecutive missing measurements on KF-based SOC estimation, enabling theoretical insight into error propagation. Furthermore, a novel physics-informed (PI) data imputation technique is proposed to reconstruct missing voltage measurements, which constitutes one of the most influential features for SOC estimation. The proposed PI-based imputation method is integrated with ML-based SOC estimators, resulting in improved estimation accuracy and robustness compared with ML estimators with conventional data imputation process operating under missing measurements.
In addition to these original contributions, this dissertation includes a comprehensive and systematic review of SOC estimation methods applied to BESS used in stationary applications. This survey synthesizes existing techniques, highlights current challenges and limitations, and identifies opportunities for future research in the context of stationary BESS, thereby providing a complete reference framework for both model-based and data-driven SOC estimation approaches.
Chapters 1 through 7 of this dissertation focus on the core contributions of the work, which are the development and application of Bayesian and ML methods for SOC estimation in PV battery systems. Chapters 8 and 9 extend the methodological foundation of this dissertation by exploring ML and EKF techniques in the context of sparse array signal processing. Although these applications are outside the primary focus of SOC estimation, they provide complementary insights into the behavior, implementation, and performance of ML and EKF-based approaches, thereby strengthening the theoretical and practical understanding that underpins the core contributions of this work.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Rahman, Showrov, "UNIFIED BAYESIAN AND MACHINE LEARNING-BASED STATE OF CHARGE ESTIMATION IN A PV BATTERY SYSTEM" (2026). Open Access Dissertations. Paper 4543.
https://digitalcommons.uri.edu/oa_diss/4543