Deep learning approach towards accurate state of charge estimation for lithium-ion batteries using self-supervised transformer model

Figures

• 

  Figure 1. SOC estimation at room temperature. (a) LA92 drive cycle at 25 °C. (b) UDDS drive cycle at 25 °C. (c) US06 drive cycle at 25 °C.

• 

  Figure 2. SOC estimation at above zero ambient temperatures. (a) LA92 drive cycle at 10 °C. (b) UDDS drive cycle at 10 °C. (c) US06 drive cycle at 10 °C. (d) LA92 drive cycle at 25 °C. (e) UDDS drive cycle at 25 °C. (f) US06 drive cycle at 25 °C. (g) LA92 drive cycle at 40 °C. (h) UDDS drive cycle at 40 °C. (i) US06 drive cycle at 40 °C.

• 

  Figure 3. SOC estimation at below zero ambient temperatures. (a) LA92 drive cycle at 0 °C. (b) UDDS drive cycle at 0 °C. (c) US06 drive cycle at 0 °C. (d) LA92 drive cycle at − 10 °C. (e) UDDS drive cycle at − 10 °C. (f) US06 drive cycle at − 10 °C. (g) LA92 drive cycle at − 20 °C. (h) UDDS drive cycle at − 20 °C. (i) US06 drive cycle at − 20 °C.

• 

  Figure 4. Estimation plot on the test drive cycles of the Panasonic cell with various training combinations. (a) HWFET cycle at 0 °C. (b) US06 cycle at 0 °C. (c) HWFET cycle at 10 °C. (d) US06 cycle at 10 °C. (e) HWFET cycle at 25 °C. (f) US06 cycle at 25 °C.

• 

  Figure 5. Data preprocessing pipeline from raw values to positional encoding. (a) Raw data sample plot from the UDDS drive cycle at − 20 °C ambient temperature. (b) Input-label pairs constructed by sliding a fix width window across the train, validation, and test dataset. (c) Dataset converted into positional encoding to be ingested into Transformer.

• 

  Figure 6. Framework and architecture of the proposed transformer model.