Analysis of State-of-Health Estimation Approaches and Constraints for Lithium-Ion Batteries in Electric Vehicles

Table 3

Merits and drawbacks of the data-driven methods [86–103].

Model name	Merits	Drawbacks and limitation
Neural network (NN)	Ideal for parameter estimation, requires less data, accurate	Needs a large set of training data as it depends on historic data
Recurrent neural network (RNN)	Efficient for sequential data	Gradient explosion or disappearance
Support vector machine (SVM)	Suitable for both high and nonlinear dimensional model, fast, and accurate	Sophisticated computational technique
ELM	Requires less computation; Learning model is fast	Delicate to the number of hidden neurons
RVM	Better sparsity; Avoids overfitting and underfitting of data	Lack of stability; Not feasible for long-term prediction
Elman NN	Adaptive to time fluctuating features, quick approaching speed	Easy to fall in local optimality