Modeling and prediction for discharge lifetime of battery systems using hybrid evolutionary algorithms

doi:10.1016/S0097-8485(00)00099-1

Computers & Chemistry

Volume 25, Issue 3, May 2001, Pages 251-259

https://doi.org/10.1016/S0097-8485(00)00099-1 Get rights and content

Abstract

A hybrid evolutionary modeling algorithm (HEMA) is proposed to build the discharge lifetime models with multiple impact factors for battery systems as well as make predictions. The main idea of the HEMA is to embed a genetic algorithm (GA) into genetic programming (GP), where GP is employed to optimize the structure of a model, while a GA is employed to optimize its parameters. The experimental results on lithium–ion batteries show that the HEMA works effectively, automatically and quickly in modeling the discharge lifetime of battery systems. The algorithm has some advantages compared with most existing modeling methods and can be applied widely to solving the automatic modeling problems in many fields.

Introduction

Regardless of the primary and the secondary battery systems, the discharge lifetime is one of the most important indicators concerned by users. Especially with the development of electric vehicles, it presses for an instantaneous, non-invasive and remote method to measure and predict the discharge lifetime of sealed batteries. One answer to this is to build the discharge lifetime model of battery systems based on observed data.

The traditional modeling methods include data fitting, regression analysis and some other approximation methods (He, 1995). Some discharge lifetime models have been obtained by using these methods, such as the Lifetime Equation of Fast Detection for the button type Li/MnO₂ cell (Liu and Gui, 1993, Liu and Gui, 1996) and the Peukert Equation for the lead-acid battery (Bode, 1977). The usual steps of these methods are to choose a model structure for the system initially, to determine the parameters contained in the model subsequently, and finally, to test the validity of the model (Xiang et al., 1988, Gan, 1991). However, due to the complexity and nonlinearity of electrochemical reactions of battery systems, it is usually difficult for people to choose a suitable model structure without having sufficient domain details and human expertise, and the determination of parameters also requires the modeler to have a rich mathematical knowledge and professional skills. In Qi et al. (1995), Qi and his coworkers applied the back propagation (BP) algorithm and the multilayer perceptron in neural networks to build ‘black box’ model for battery systems. But the performance of neural networks depends largely upon the topology and weight values, and it is usually an arduous task to design a neural network, including the topology and connection weights, for a specific problem. Recently Salkind et al. (1999) used the fuzzy logic methodology to predict the states of charge and health of battery systems, but their predicted values did not coincide with the measured values very well.

To overcome the drawbacks of the available methods mentioned above, we propose a hybrid evolutionary modeling algorithm (HEMA) to implement the automatic modeling of discharge lifetime with multiple impact factors for battery systems as well as make predictions. The experimental results on lithium–ion batteries testify the effectiveness of the algorithm.

Section snippets

Hybrid evolutionary modeling method

Genetic Programming (GP) (Banzhaf et al., 1997) is a new branch of Evolutionary Computation (Eiben and Michalewicz, 1999) which was introduced by John Koza and co-workers (Koza, 1992, Koza et al., 1994, Koza et al., 1999) in the 1990s. GP is an extension of John Holland's Genetic Algorithm (GA) (Holland, 1975) in which the genetic population consists of computer programs of varying sizes and shapes. In standard GP, computer programs are represented as parse trees rather than bit strings, where

Hybrid evolutionary modeling algorithm

Taking the evolutionary modeling of discharge lifetime of battery systems as an example, we give the detailed description of the hybrid evolutionary modeling algorithm in this section.

Preparation of batteries and collection of sample data

The UR2025 lithium–ion battery had a sandwich structure using LiCoO₂ and modified petroleum coke as positive and negative materials respectively. The separator was a Celgard 2400 microporous polylene membrane. The electrolyte was 1 M LiClO₄ dissolved in a 1:1 mixture of propylene carbonate (PC) and dimethoxyethane (DME). We chose nine UR2025 lithium–ion batteries randomly as experimental batteries. Every battery was charged to 4.2 V at the current of 1.0 mA, then discharged under different

Results and discussion

The average fitting errors and prediction errors of 40 runs on different (D, F) are listed in Table 3. From Table 3 we can see that as far as the AFE is concerned, on the same function set, an increase in D leads to a lower AFE, but with a diminishing gradient. For example, on F₁, the AFE of D=4 is more than double that of D=5, while the difference between the AFEs of D=6, 7 is trivial. For the same D, the disparity in the AFEs on different function sets is not obvious. As far as the APE is

Conclusions

A hybrid evolutionary modeling algorithm HEMA is proposed in this paper. Its main idea is to embed a genetic algorithm GA into genetic programming GP, where GP is employed to optimize the structure of a model, while a GA is employed to optimize its parameters. The algorithm proves to work quickly and effectively in building the discharge lifetime models with multiple impact factors for lithium–ion batteries. It has the following advantages compared with most existing modeling methods:

1.
The

Acknowledgements

This work was supported in part by National Natural Science Foundation of China (No. 70071042, No. 60073043 and No. 29833090) and National Laboratory for Parallel and Distributed Processing. The authors would like to thank the anonymous referees for their helpful comments on the paper.

References (19)

R. Moros et al.
Comput. Chem. Eng.
(1996)
A.J. Salkind et al.
J. Power Sources
(1999)
T.J. VanderNoot et al.
J. Electroanal. Chem
(1998)
Banzhaf, W., Nordin, P., Keller, R.E., Francone, F.D., 1997. Genetic Programming: An Introduction on the Automatic...
H. Bode
Lead-Acid Batteries
(1977)
A.E. Eiben et al.
Evolutionary Computation
(1999)
Gan, R.-C., 1991. The Statistical Analysis of Dynamic Data. Beijing University of Science and Technology Press, Beijing...
He, J.-X., 1995. System Modelling and Mathematical Models. Fujian Science & Technology Press, Xiamen (in...
Hinterding, R., Michalewicz, Z., Eiben, A., 1997. Proceedings of the 4th International Conference on Evolutionary...

There are more references available in the full text version of this article.

Cited by (10)

A self-correction single particle model of lithium-ion battery based on multi-population genetic algorithm
2023, Journal of Energy Storage
At present, the Single Particle (SP) model and the Extended Single Particle (ESP) model become a serious concern because they have real usage potentials for the battery management system. However, utilization of the models relies too much on accurate model parameters. Measuring accurately the parameters is a tough task, since they change with the battery condition during the actual use. In this study, Multi-Population Genetic Algorithm (MPGA), a certain kind of GA's improved algorithm, has been adopted to propose a Self-Correction Single Particle Model (SC-SPM). The SC-SPM can self-correct the measured values by sensing the terminal voltage change, which gives more accurate results for the solid phase diffusion process. To verify the proposed SC-SPM, four groups of simulations are performed, indicating that the SC-SPM is more accurate at observing the solid phase diffusion process when the battery condition changes beyond a certain range, and helping BMS better estimating other battery states. At the same time, the single step calculation time is <0.5 s which can meet the requirement of on-line application.
Lithium-Ion Battery State of Charge (SoC) Estimation with Non-Electrical parameter using Uniform Fiber Bragg Grating (FBG)
2021, Journal of Energy Storage
Citation Excerpt :
These methods are also adaptive to various stochastic systems and their algorithms can benefit from parallel processing. The algorithm is generally based on artificial neural networks (ANN) [8-10] although there are other models such as fuzzy controller [11], support vector machine [12], or combination of aforementioned models [13,14]. ANN can learn without supervision (unsupervised learning) which is one of the attractive features of machine learning.
Conventional SoC estimation methods mainly rely on electrical parameters such as the current and voltage of the battery. However, recent studies have shown that the non-electrical parameters such as strain and temperature that have non-linear relationships with the battery SoC can be adopted for SoC estimation. In this work, the use of non-electrical parameters for SoC estimation using Deep Neural Network (DNN) was proposed. Fiber Bragg Grating (FBG) sensors are employed for the simultaneous measurement of strain and temperature of the battery. Besides, Pseudohigh-Resolution interrogation (PHRI) method is adopted for demodulating the output spectra to improve the detection accuracy of the small wavelength signals from the FBG sensors. Our findings have shown a great improvement in the FBG signal quality based on a high up-sampling rate, k in the spectral processing using PHRI. This has a significant impact on the performance of the SoC estimation using DNN. In the comparison with SoC estimation based on electrical parameters, the proposed model based on non-electrical parameters has a better estimation performance.
LBM prediction of effective electric and species transport properties of lithium-ion battery graphite anode
2016, Solid State Ionics
Citation Excerpt :
Experimental and numerical studies [4–6] both imply that the complex microstructure of electrodes has a significant effect on the performance of lithium-ion batteries. Numerical models play a very important role in the scientific understanding of the transportation of lithium-ions over a porous electrode [6–13]. Lithium-ion batteries are generally analyzed using a macro-homogeneous porous electrode model.
Numerical models play a vital role in the developing and performance optimization of lithium-ion batteries. The key factor to the prediction accuracy of macro-scale models is the specification of effective transport properties. This study, based on the anisotropic microstructure of graphite anode reconstructed by an ellipsoid-based simulated annealing method (SAM), established a mesoscopic model of diffusion process to predict the effective electric and species transport properties of lithium-ion battery graphite anode via lattice-Boltzmann (LB) method. The effect of particle size on the transport properties of graphite anode was discussed in detail. In the electrode through-plane direction, if the ellipsoidal particles are thinner and flatter, both the effective electric and species transport properties decrease; in the other two directions, the effective electronic charge transport properties barely change with the change of particle size while the effective species transport properties increase along with the increase of the size of particles. In addition, to get a more accurate replica of the real graphite anode, we assumed the sizes of solid particles follow a normal distribution and reconstructed the microstructure of electrode. The LB calculation results reveal that the normal distribution of particle size increases the electronic charge conductivity in the electrode through-plane direction and decreases in the other two directions, compared to the electrode of constant-sized particles; the effective species diffusivities (or ionic charge conductivities) in electrode through-plane direction for different microstructures are closer.
Kinetics research for the synthesis of branch ether using genetic-simulated annealing algorithm withmulti-pattern evolution
2003, Chemical Engineering Journal
To obtain the kinetics parameters for the complex reaction system, a more intellectualized hybrid evolutionary algorithm was developed. In this new method, the normal genetic algorithm (GA) was modified with adaptive multi-annealing crossover and mutation strategies instead of simple strategy. The multi-pattern evolution was also adapted to improve the search efficiency. This hybrid algorithm not only avoids the problem of local optimum, but shows a higher estimating precision and a better convergence than that the normal GA. By using this algorithm, the kinetics parameters for the synthesis of ethyl tert-amyl ether from ethanol and tert-amyl alcohol over strongly acidic cation exchange resin (Ambelyst 15 (A-15)) as catalyst were successfully estimated. The calculated results using the parameters obtained from the hybrid evolutionary algorithm agree well with the experimental data.
Critical Review on the Battery State of Charge Estimation Methods for Electric Vehicles
2017, IEEE Access
On-line monitoring of capacity fade in lithium-ion batteries
2012, Journal of Chemical Engineering of Japan

View all citing articles on Scopus

View full text

Modeling and prediction for discharge lifetime of battery systems using hybrid evolutionary algorithms

Abstract

Introduction

Section snippets

Hybrid evolutionary modeling method

Hybrid evolutionary modeling algorithm

Preparation of batteries and collection of sample data

Results and discussion

Conclusions

Acknowledgements

Comput. Chem. Eng.

J. Power Sources

J. Electroanal. Chem

Lead-Acid Batteries

Evolutionary Computation