# Lithium battery health diagnosis and remaining life prediction-Part 2

**Experimental result**

Figure 6 shows the capacity estimation results of the model trained with 0.5C discharge (test 1) in the case of dynamic current rate discharge (test 2). The estimated capacity is close to the reference value. In most cycles, the estimated error is within 1%, and the root mean square error is 0.80%, indicating that the capacity estimation method has high accuracy and has application value in real vehicles.

Figure 6 Capacity estimation results

Figure 7 shows the effect of two different classification algorithms based on Random Forest and Neural Network (NN) on the identification of the aging stage. It can be seen that under different aging paths, the identification effect of random forest is better than that of neural network.

Figure 7 Identification of aging stage based on random forest and neural network (a) 0.5C; (b) 1C; (c) 1.5C; (d) 2C

Figure 8 shows the random forest model trained with two battery cells discharged at 0.5C, and the identification results of the aging stage under the condition of dynamic rate discharge. The identified initial cycle of the second stage of battery aging was the 107th cycle, and only three cycles were erroneously identified.

Figure 8 Identification result of aging stage

Figure 9 shows the prediction results of RUL under different aging paths by RLS under different values of the forgetting factor (μ). The variable forgetting factor is defined as follows:

In this paper, α=0.995 and μ0=0.95. The error threshold is 50 cycles, that is, the difference between the RUL and the true value within 50 cycles is regarded as an accurate prediction.

Figure 9 RUL prediction results (a) 0.5C; (b) 1C; (c) 1.5C; (d) 2C

Fig. 10 shows the prediction of RUL in the case of dynamic current magnification using RLS with a variable forgetting factor, and the prediction result is based on the SOH estimation result in Fig. 6. It can be seen that after 200 cycles, the predicted RUL falls within the allowable error range, and after 300 cycles, the prediction error is less than 10 in most cases. The probability density function of RUL obtained by Monte Carlo simulation, the 95% confidence interval of RUL is [265, 285], [181, 197] and [88, 108] at the 257th, 357th and 457th cycles, respectively Cycles are all within the scope of accurate prediction. The calculation time for each RUL prediction is about 0.3 seconds, and the amount of calculation is suitable for real vehicles.

Figure 10 Prediction of RUL under dynamic current rate using RLS with variable forgetting factor

**In Conclusion**

The author of this article has developed a lithium-ion battery health status diagnosis and remaining life prediction algorithm based on aging characteristics. First, six health indicators were extracted from the common segments in the charging voltage curve; then the neural network model and random forest model were trained with offline data, and the relationship between health indicators, SOH and aging stages was constructed; finally, the SOH estimates and aging Stage recognition results, use the RLS algorithm with forgetting factor to predict the RUL of the battery, and combine Monte Carlo simulation to generate the probability density function of the RUL prediction result.

Experimental results show that the algorithm can accurately estimate SOH regardless of constant or dynamic rate discharge, and the estimation error is less than 1% in most cases. After 200 cycles, the algorithm can achieve accurate RUL prediction for the battery cell under the dynamic discharge rate in 0.3 seconds. The experimental results show to a certain extent that the algorithm developed by the author of this article has high accuracy and small calculation amount, and it can be applied to actual electric vehicles.