Lithium iron phosphate battery packs are widely employed for energy storage in electrified vehicles and power grids. However, their flat voltage curves rendering the weakly observable state of charge are a critical stumbling block for charge equalization management. This paper focuses on real-time active balancing of series-connected lithium iron phosphate batteries. In the absence of accurate in-situ state information in the voltage plateau, a balancing current ratio (BCR) based algorithm is proposed for battery balancing. Then, BCR-based and voltage-based algorithms are fused, responsible for the balancing task within and beyond the voltage plateau, respectively. The balancing process is formulated as a batch-based run-to-run control problem, as the first time in the research area of battery management. The control algorithm acts in two timescales, including time-wise control within each batch run and batch-wise control at the end of each batch. Hardware-in-the-loop experiments demonstrate that the proposed balancing algorithm is able to release 97.1% of the theoretical capacity and can improve the capacity utilization by 5.7% from its benchmarking algorithm. Furthermore, the proposed algorithm can be coded in C language with the binary code in 118,328 bytes only and thus is readily implementable in real-time.
A battery’s state-of-power (SOP) refers to the maximum power that can be extracted from the battery within a short period of time (e.g., 10 s or 30 s). However, as its use in applications is growing, such as in automatic cars, the ability to predict a longer usage time is required. To be able to do this, two issues should be considered: (1) the influence of both the ambient temperature and the rise in temperature caused by Joule heat, and (2) the influence of changes in the state of charge (SOC). In response, we propose the use of a model-based extreme learning machine (Model-ELM, MELM) to predict the battery future voltage, power, and surface temperature for any given load current. The standard ELM is a kind of single-layer feedforward network (SLFN). We propose using a set of rough models to replace the active functions (such as logsig()) in the ELM for better generalization performance. The model parameters and initial SOC in these “rough models” are randomly selected within a given range, so little prior knowledge about the battery is required. Moreover, the identification of the complex nonlinear system can be transferred into a standard least squares problem, which is suitable for online applications. The proposed method was tested and compared with RLS (Recursive Least Square)-based methods at different ambient temperatures to verify its superiority. The temperature prediction accuracy is higher than ±1.5 °C, and the RMSE (Root Mean Square Error) of the power prediction is less than 0.25 W. It should be noted that the accuracy of the proposed method does not rely on the accuracy of the state estimation such as SOC, thereby improving its robustness.
Injection molding (IM) quality optimization involves coupled process variables, nonlinear quality responses, and a limited budget for physical trials. Numerical optimizers can use measured process data effectively, but they do not directly represent troubleshooting rules and other experience-based knowledge that is recorded in text. This study proposes large language model-based extended data-driven optimization (LLM-EDO), a domain-knowledge-guided framework for surrogate-based candidate generation. The framework converts process variables, quality curve discrepancies, feasible bounds, and retrieved expert cases into a structured request. A language model then proposes a bounded parameter vector, and a fixed long short-term memory surrogate evaluates the resulting quality curve. The framework is examined in a surrogate-in-the-loop environment using repeated and multi-start tests and is compared with Bayesian optimization, simultaneous perturbation stochastic approximation, and knowledge-informed gradient-based SPSA. The reported runs show reductions in curve distance for the tested initial configurations and competitive terminal objective values. They also show run-to-run variability, and iteration counts are reported as decision rounds rather than wall-clock measurements. The results support the feasibility of domain-knowledge-guided candidate generation for surrogate-based quality curve matching under the tested process configurations.