تخمين داده محور پارامترهاي نشان دهنده وضعيت شارژ و سلامت بسته باتري

In recent years, battery-based energy storage systems have become increasingly important in applications such as electric vehicles, microgrids, an‎d renewable energy storage systems. The safe, stable, an‎d reliable operation of these systems strongly depends on accurate monitoring an‎d management of battery conditions. In particular, accurate estimation of the state of charge an‎d state of health is essential for improving efficiency, enhancing safety, preventing overcharge an‎d over discharge, an‎d extending the lifetime of battery packs. However, reliable estimation of these parameters at the pack level remains a significant challenge due to cell inconsistencies, thermal variations, aging effects, an‎d the complexity of series–parallel battery configurations. This dissertation proposes a data-driven framework based on machine learning techniques for estimating the state of charge an‎d state of health of lithium-ion battery packs with series–parallel structures. To provide the required experimental data, a battery testing an‎d data acquisition platform was designed an‎d implemented. The developed setup enables the recording of parameters such as cell terminal voltage, pack current, an‎d cell surface temperature during repeated charge an‎d discharge cycles. In addition, the thermal behavior of the cells was investigated through the analysis of both surface an‎d core temperature variations under different operating conditions. Subsequently, a supervised machine learning model was developed to estimate the core temperature of battery cells using measured surface temperature data. The estimated core temperature was then incorporated as one of the key inputs to the proposed algorithms for both state of charge an‎d state of health estimation. The results demonstrated that incorporating core temperature information improves estimation accuracy compared with approaches that rely solely on surface temperature measurements. For battery pack state of charge estimation, a representative-cell-based strategy was proposed. In this approach, a criterion was defined to eva‎luate the similarity among cells, an‎d one representative cell from each parallel branch was selec‎ted for the estimation process. As a result, the proposed estimation algorithm relies only on measurements obtained from the selec‎ted representative cells. This strategy significantly reduces the number of required sensors an‎d computational complexity while preserving satisfactory estimation accuracy at the pack level. A deep neural network-based model was then developed for state of charge estimation, an‎d different network architectures were eva‎luated in terms of estimation accuracy an‎d computational complexity. The results showed that combining derived features with the representative cell selec‎tion strategy improves the robustness an‎d performance of the algorithm under various operating conditions, including noisy measurement environments. Finally, a transfer learning-based framework was proposed for estimating the state of health of battery packs. In this framework, a recurrent neural network model was first pretrained using large-scale datasets obtained from well-known single-cell battery datasets to learn the general degradation behavior of lithium-ion batteries. The pretrained model was subsequently fine-tuned using limited data collected from the representative cells of the target battery pack. The proposed framework enables accurate an‎d stable state of health estimation without requiring extensive long-term aging data from battery packs. The results indicate that the proposed approach provides an effective foundation for the development of intelligent battery management systems in advanced energy storage applications.

جعفر قيصري

عليرضا بخشايي

ايمان ايزدي , احمدرضا تابش , حميد خالوزاده