State of charge estimation combining physics-based and artificial intelligence models for Lithium-ion batteries

This paper presents a sequential model based on Physic Based Models (PBM) and Artificial Intelligence Models (AI) focused on the estimation of the State of Charge (SoC). The PBM can provide interesting information about the internal physical variables of the battery, which relate directly with the momentary SoC of the cell. This way, we can use this information to feed the AI model alongside with application measurements to obtain an accurate SoC estimation. By their nature, PBMs can potentially provide immense numbers of internal variables, some of which are irrelevant and redundant for the AI model. To solve this, a feature selection technique based on the regression score is introduced between both models. Selecting the most relevant variables we can build a model that reduces the computational cost of the model while improving the performance compared to using every feature. With this baseline, a Reduced-Order Model (ROM) with parameters from literature has been implemented in the PBM part and a Long-Short Term Memory (LSTM) network on the ML side. With this configuration and the PBM simplification the model needs little laboratory tests and low computational cost to outperform alternative solutions, which has been experimentally validated in different operating conditions of the cell.

Automated summary

This paper presents a sequential model based on Physic Based Models (PBM) and Artificial Intelligence Models (AI) focused on the estimation of the State of Charge (SoC). By selecting the most relevant variables, reducing computational cost and improving performance. The model has been experimentally validated to outperform alternative solutions with laboratory tests and low computational cost.