A new battery capacity indicator for lithium-ion battery powered electric vehicles using adaptive neuro-fuzzy inference system

K. T. Chau, Kai Wu, C.c. ChanEnergy 2003 · DOI

TL;DR

This paper introduces a novel Adaptive Neuro-Fuzzy Inference System (ANFIS) model to accurately estimate the remaining capacity of lithium-ion (Li-ion) batteries in electric vehicles (EVs). By incorporating discharged/regenerative capacity distributions and temperature distributions as inputs, and using realistic EV driving profiles, the model achieves highly accurate State of Available Capacity (SOAC) estimations, crucial for extending EV range and usability.

What Problem Does This Paper Address?

Electric vehicles (EVs) are increasingly relying on advanced battery technologies like lithium-ion (Li-ion) for their superior energy and power density. A critical challenge for widespread EV adoption is accurately predicting the remaining battery capacity, often referred to as Battery Residual Capacity (BRC) or State of Available Capacity (SOAC). This metric directly impacts an EV's driving range and overall usability. Previous methods often simplified EV operating conditions, such as using constant discharge currents or generalized temperature inputs, leading to inaccurate estimations. Specifically, existing approaches failed to adequately account for the complex, dynamic nature of EV discharge currents, including regenerative braking, and the significant impact of temperature variations on Li-ion battery performance. This paper aims to bridge this gap by developing a more sophisticated model that considers these crucial factors.

Methods

The core of this paper is the development of a new ANFIS model for SOAC estimation. ANFIS combines the learning capabilities of neural networks with the rule-based reasoning of fuzzy logic. The proposed model utilizes a specific ANFIS architecture with five layers.

  • Inputs: Unlike previous methods that used instantaneous voltage or simplified current profiles, this model takes a more comprehensive approach. The inputs are:
  • Discharged/Regenerative Capacity Distributions: The paper divides the EV discharge current into four ranges and considers regenerative braking as a separate input. These are then represented as distributions rather than single values.
  • Temperature Distributions: Instead of a single temperature reading, the model uses two temperature ranges (above and below 20°C) to capture the significant temperature sensitivity of Li-ion batteries.
  • Output: The model's output is the State of Available Capacity (SOAC), which is the normalized BRC.
  • Learning Algorithm: A hybrid algorithm combining the Least Square Method (LSM) and Gradient Descent Method (GDM) is employed to tune the ANFIS parameters for optimal performance.

Datasets

The authors collected experimental data using a Digatron Battery Testing System BTS-600. This involved testing a commercially available Li-ion battery (3.6V nominal voltage, 15 Ah nominal capacity) under various conditions:

  • Discharge Current Profiles: Four realistic EV discharge current profiles were used, based on standard driving cycles: European Driving Cycle (ECE), US Federal Urban Driving Schedule (FUDS), US Federal Highway Driving Schedule (FHDS), and Japan Driving Mode 10.15 (JM10.15). These profiles were applied at different discharge rates, expressed in C-rates.
  • Temperatures: Experiments were conducted at 10°C, 20°C, and 30°C.

The collected data was divided into a training set and a validation set for model development and evaluation.

Key Insights and Ideas

  • Temperature Sensitivity is Crucial: Li-ion batteries exhibit significant temperature sensitivity, much more so than Ni-MH batteries. This necessitates using temperature distributions rather than instantaneous values as model inputs.
  • Dynamic Current Profiles Matter: Realistic EV discharge and regenerative braking profiles are essential for accurate capacity estimation. Simplified or constant current profiles are insufficient.
  • Distributions Over Single Values: Representing discharged/regenerative capacity and temperature as distributions (rather than single points) allows the ANFIS model to better capture the complex relationships and dynamics influencing battery capacity.
  • Terminal Voltage is a Disturbance: The instantaneous battery terminal voltage fluctuates significantly with current changes, making it a poor indicator of SOAC and potentially degrading estimation accuracy.

Key Results

The proposed ANFIS model demonstrated highly accurate SOAC estimation:

  • Training Data: The model achieved an Average Percentage Error (APE) of only 0.16% on the training data set, indicating excellent agreement between estimated and actual SOAC.
  • Validation Data: On the validation data sets, the APEs for all 12 validation data sets were within 1%. This is a significant improvement compared to previous methods, which reported APEs of 5.5% [10] and 2.7% [9] for similar tasks. The model's accuracy was confirmed across different realistic EV discharge current profiles.

Limitations and Future Work

  • Limitations:
    • The paper notes that while the number of divisions for input parameters (e.g., four for discharged capacity, two for temperature) improves accuracy, it also increases computational complexity. The chosen number of seven input parameters was determined through trial and error.
  • The model's performance is evaluated at specific temperatures (10°C, 20°C, 30°C). While it accounts for temperature ranges, further investigation into a wider temperature spectrum might be beneficial.
  • Future Work:
    • The paper suggests that the proposed SOAC model can be practically realized using a low-cost microcontroller. The realization diagram (Fig. 10) outlines how the classification, integration, and ANFIS units can be implemented on a microcontroller. This points towards the practical deployment of the model as a BRC indicator in actual EVs.
  • Further exploration of the optimal number of input parameter divisions for a better trade-off between accuracy and computational cost could be a future direction.

Summary updated

BibTeX
@inproceedings{chau2003new,
  author = {K. T. Chau and Kai Wu and C.C. Chan},
  title = {{A new battery capacity indicator for lithium-ion battery powered electric vehicles using adaptive neuro-fuzzy inference system}},
  booktitle = {Energy},
  year = {2003},
  doi = {10.1016/j.enconman.2003.09.031},
}
Figure