Determination of equivalent circuit parameters of a lithium-ion battery using hybrid pulse power characterization test results

Approaches based on the Equivalent Circuit Model (ECM) are widely used for State of Charge (SoC) estimation in lithium-ion batteries (LIBs). The Thevenin Equivalent Circuit Model (TECM) is often preferred due to its simplicity, the ease of analysis through electrical equations, and the ability to determine its parameters from experimental data. Thevenin circuit models are used in various configurations, with the most common being 1RC, 2RC, and 3RC structures. As the number of parallel RC branches increases in the model, the computational complexity in the SoC estimation process also increases. However, this comes with a decrease in SoC estimation error. Still, the reduction in error is not proportional to the increase in computation. The most critical factor affecting the accuracy of battery analysis via ECM is the correct estimation of model parameters. These parameters are determined either through calculations or estimations based on specific test data obtained from the battery. However, for models containing two or more RC branches, it is difficult to determine the R and C values through direct calculation. Nonetheless, the internal resistance (R₀) can be determined both by calculation and estimation, regardless of the number of RC branches. In this thesis, to analyze the relationship between computational load and estimation accuracy for different numbers of RC branches, the parameters of 1RC, 2RC, and 3RC models were estimated. Additionally, the effect on SoC estimation was comparatively analyzed in cases where only R₀ was calculated and the remaining parameters were estimated. All estimations were performed using the Nonlinear Least Squares (NLS) method. For validation, experimental data obtained using the US06 and UDDS standard driving cycles were used. The comparison criteria included error distributions (box plot graphs), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and computation time. As a result of the study, an accuracy rate of up to 98% was achieved.