Towards Green AI: A Novel Hybrid Filter Based AI Approach for Energy Efficient State of Charge and Energy Estimation in Li-Ion Batteries Under Various Drive Cycles

Artificial intelligence models usually need a hefty dataset, which indeed requires more computational time, energy consumption, and laborious work to train the models to achieve accurate results. These models consume more energy to deliver output. Therefore, in this work proposes a novel filter technique (FT) based on an average reduction method. This approach improves the quality of the dataset, i.e., reduces volume and averages similar data to find the precision. This novel FT was incorporated with the convolutional neural network with bidirectional long short-term memory network (FT-CNN-Bi-LSTM) model and tested on four dynamic drive cycle circumstances. The key objective of this modelling technique is to enhance the performance and computational efficiency of the state of charge (SOC) and state of energy (SOE) of rechargeable battery systems such as lithium-ion batteries. A comparison examination was performed by proposed models configured with FT and traditional models, including FT-CNN, FT-LSTM, FT-BiLSTM, and FT-CNN-LSTM. The comparison examines SOC and SOE estimation efficacy and the model's capabilities and limitations. The FT-CNN-BiLSTM model was significantly tested on multiple dynamic drive cycles, including HWFET, LA92, UDDS, and US06. These drive cycle datasets were collected from our laboratory, where experiments were conducted at ambient temperature. The FT-CNN-Bi-LSTM approach achieved percentage improvements, 99.98% of RMSE, 99.97% of MSE, and 99.96% of MAE. The significant percentage improvement in computational time and energy consumption ranges from 59 to 66% and 61 to 68%. These results demonstrate superior efficiency for rapid SOC and SOE estimation in all dynamic drive cycles.