نحوه شارژ باتری یون لیتیوم
Abstract
1. Introduction
2. Neural network for battery modeling
3. Extended Kalman filtering algorithm for SoC estimation
4. Experiments and data preprocessing
5. Verification and discussion
6. Hardware-in-Loop verification
7. Conclusions
Acknowledgments
References
Abstract
Accurate state-of-charge (SoC) estimation is remarkably difficult due to nonlinear characteristics of batteries and complex application environment in electric vehicles (EVs), particularly low temperature and low SoC. In this paper, an improved battery model is first built using a feedforward neural network (FFNN) by introducing newly defined inputs. Based on the FFNN model and the extended Kalman filter algorithm, a FFNN-based SoC estimation method is designed, and its robustness is verified and discussed using the experimental data obtained at different temperatures. Finally, a hardware-in-loop test bench is built to further evaluate the real-time and generalization of the designed FFNN model. The results show that the SoC estimation can converge to the reference value at erroneous settings of an initial SoC error and an initial capacity error, and the SoC estimation errors can be stabilized within 2% after convergence, which applies to all the cases discussed in this paper, including low temperature and low SoC. This indicates that the FFNN-based method is an effective method to estimate SoC accurately in complex EV application environment.
Introduction
With the improvement of performance requirements for electric vehicles (EVs), such as longer driving range and faster speed, more powerful energy sources are needed. At present, lithium ion batteries are the most commonly used energy sources in EVs due to their advantage of higher energy density and longer lifetime than other batteries with different chemistries (Subburaj et al., 2015), but they still cannot provide sufficient energy to drive EVs as far as fossil fuels in traditional vehicles. Under these circumstances, it is crucial to fully utilize the energy stored in batteries for EVs through battery management systems (BMSs). Battery state-of-charge (SoC) estimation is one of the main tasks of BMSs and its accuracy influences performances of other functions in BMSs, including charging control (Di Yin et al., 2016), balancing control (Ma et al., 2018), thermal management (Zhu et al., 2015), and safety management (Xiong et al., 2019a). Due to nonlinear characteristics and complex operation environments of batteries in EVs, it is very difficult to obtain accurate SoC, so a welldesigned SoC estimation method is necessary for any BMSs.