Click to enlarge image.
Accurate capacity estimation is critical to ensure the safe and reliable usage of lithium-ion batteries, and datadriven methods are a promising technique for this task. However, the existing studies require the whole charging curve for feature extraction and usage of sophisticated machine learning methods, which are not suitable for online applications. This paper proposes a simple machine learning technique, partial least squares regression, for online battery capacity estimation based on the partial incremental capacity curve. The features can be easily obtained by interpolation of the measured charging profile without data smoothing, leading to a low computational cost. The proposed method is realized and tested on three battery datasets (#5, #7, #18) provided by NASA. Experimental results show that the model trained on 80% of the data samples of cell #5 can achieve a 0.01053Ah root mean squared error for the remaining 20% data of cell #5. The model is further verified on the other two battery datasets without changing model weights, and the test root mean squared error is 0.02046Ah for cell #7 and 0.02700Ah for cell #18, indicating the generality of the proposed capacity estimation method
Read or Download: PDF