Abstract:The state of health (SOH) of a Li-ion battery is a critical parameter that determines the remaining capacity and the remaining lifetime of the battery. In this paper, we propose SambaMixer a novel structured state space model (SSM) for predicting the state of health of Li-ion batteries. The proposed SSM is based on the MambaMixer architecture, which is designed to handle multi-variate time signals. We evaluate our model on the NASA battery discharge dataset and show that our model outperforms the state-of-the-art on this dataset. We further introduce a novel anchor-based resampling method which ensures time signals are of the expected length while also serving as augmentation technique. Finally, we condition prediction on the sample time and the cycle time difference using positional encodings to improve the performance of our model and to learn recuperation effects. Our results proof that our model is able to predict the SOH of Li-ion batteries with high accuracy and robustness.

What problem does this paper attempt to address?

### The Problem Addressed by the Paper This paper aims to address the problem of State of Health (SOH) prediction for lithium-ion batteries. Lithium-ion batteries are widely used in various applications, including consumer electronics and electric vehicles. However, over time, battery performance gradually declines, affecting its remaining capacity and lifespan. Accurately predicting the SOH of batteries is crucial for ensuring their safety, reliability, and efficiency. Traditional Battery Management Systems (BMS) typically rely on Equivalent Circuit Models (ECM) and Electrochemical Models (EM), but these models have limitations in terms of complexity and sensitivity to different operating conditions. In recent years, deep learning models have emerged as powerful tools for predicting the health state of lithium-ion batteries due to their ability to learn complex nonlinear relationships directly from data. This paper proposes a new model based on the Mamba State Space Model (SSM) called SambaMixer, designed to predict the SOH of lithium-ion batteries. The model is designed to handle long-term time dependencies in multivariate time series data and improve performance through positional encoding to capture sample jitter, varying lengths of time signals, and the recovery effects of lithium-ion batteries. ### Main Contributions 1. **Introduction of the Mamba State Space Model**: Applying the Mamba SSM to the problem of SOH prediction for lithium-ion batteries. 2. **Development of an Anchor-based Resampling Scheme**: Ensuring that time signals have the same number of samples, also serving as a data augmentation method. 3. **Application of Sample Time-based Positional Encoding Scheme**: Handling sample jitter, varying lengths of time signals, and the recovery effects of lithium-ion batteries. ### Method Overview 1. **Problem Definition**: The SOH prediction problem for lithium-ion batteries is defined, where the input is multivariate time series data of a single discharge cycle (current, voltage, temperature, and sample time), and the output is the SOH of that cycle. 2. **Model Architecture**: - **Resampling**: Using anchor-based resampling techniques to ensure that all cycles and battery time signals have the same number of samples. - **Input Projection**: Projecting the resampled voltage, current, and temperature signals into a high-dimensional feature space. - **Positional Encoding**: Using positional encoding based on sample time and cycle time difference to capture temporal information. - **Encoder Backbone**: An encoder based on MambaMixer, consisting of time-mixing and channel-mixing modules, to process multivariate time series data. - **Regression Head**: Predicting the SOH of a given cycle based on the encoded sequence. ### Experimental Results The authors evaluated the SambaMixer model on the NASA lithium-ion battery discharge dataset and demonstrated its superior performance over existing state-of-the-art models on this dataset. By introducing anchor-based resampling techniques and positional encoding, the model showed higher accuracy and robustness in SOH prediction.

SambaMixer: State of Health Prediction of Li-ion Batteries using Mamba State Space Models

State-space Modeling for Electrochemical Performance of Li-ion Batteries with Physics-Informed Deep Operator Networks

A Hybrid Data-Driven Approach for State of Health Estimation in Lithium-Ion Batteries

MambaLithium: Selective state space model for remaining-useful-life, state-of-health, and state-of-charge estimation of lithium-ion batteries

State of Health Prediction of Lithium-ion Batteries Based on SSA Optimized Hybrid Neural Network Model

State of health estimation of lithium-ion batteries based on Mixers-bidirectional temporal convolutional neural network

State of Health Prediction of Lithium-Ion Batteries Using Combined Machine Learning Model Based on Nonlinear Constraint Optimization

Battery State of Health Estimation Using the Sliding Interacting Multiple Model Strategy

Lithium-Ion Battery Health State Estimation Based on Feature Reconstruction and Optimized Least Squares Support Vector Machine

SOC and SOH Prediction of Lithium‐Ion Batteries Based on LSTM–AUKF Joint Algorithm

Data-Driven ICA-Bi-LSTM-Combined Lithium Battery SOH Estimation

Prediction of Li-ion battery state of health based on data-driven algorithm

State-of-Health Prediction for Li-ion Batteries for Efficient Battery Management System Using Hybrid Machine Learning Model

Data-Driven State of Health (SOH) Prediction for Lithium-Ion Batteries

A novel state-of-health prediction method based on long short-term memory network with attention mechanism for lithium-ion battery

State of Health Estimations for Lithium-Ion Batteries Based on MSCNN

State‐of‐Health Prediction for Lithium‐Ion Batteries Based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise‐Gate Recurrent Unit Fusion Model

State of Health Estimation for Lithium-Ion Batteries Based on Fusion Health Features and Adaboost-GWO-BP Model

Energy storage battery state of health estimation based on singular value decomposition for noise reduction and improved LSTM neural network

Research on Multi-Scale Signal Processing and Efficient Model Construction Strategies in Lithium-Ion Battery State Prediction

Lithium-ion battery SOH prediction based on VMD-PE and improved DBO optimized temporal convolutional network model