Abstract:

Influenced by the usage conditions， the state of health （SOH） declining process of lithium-ion power battery of electric vehicles has a lot of fluctuations， which leads to the decrease of model prediction accuracy. In the short-term prediction of the remaining useful life （RUL） of lithium-ion batteries， the SOH fluctuations cannot be ignored， and in order to accurately predict the SOH fluctuations in the short term， effective health indicators need to be extracted from the lithium-ion battery operation data transmitted from real vehicles. A joint distribution feature input and sequence decomposition fusion method for lithium-ion battery RUL prediction is established， using K-means clustering method to construct joint distribution features of vehicle lithium-ion battery operation process， and using S-G filter for sequence decomposition of SOH decline curve. Long-short term memory neural network （LSTM） and multilayer perceptron （MLP） is used respectively for trend part and fluctuation part. The final prediction results are obtained by fusion. The theoretical analysis and the validation of the real vehicle collection data show that the fusion model can predict the short-term decline trend of the vehicle lithium-ion battery RUL while predicting the fluctuation of SOH， and has a high short-term prediction accuracy.

Key words: lithium-ion power battery, remaining useful life prediction, data driven, deep learning