Predictive Battery Modeling

Chris Dan Gore

PREDICTIVE BATTERY MODELING

Sunseeker Solar powered electric car  Two parallel sets of 260 cell batteries  Gallium arsenide array (solar cells) 

Some History

1995 Car

2008 Car

2003 Car

Car Electrical Layout Sun

Motor Controller

Motors + Wheels

Throttle Solar Cells Power Point Tracker

Battery

Our Battery Pack (520 LiPolymer Cells)

2008 Sunseeker Battery Pack

Lithium-Polymer Battery Cell

LixMO2 + 6C <-> Li(1-x) MO2 + LiC6

Batteries are Nondeterministic

Peukert’s Law: Capacity = C/p(n-1)

Current State of Charge Estimation Methods

Battery Discharge

One Method for Determining EMF

Important Variables for SOC Estimation

State of charge estimation based on evolutionary neural network

Artificial Neural Networks  Inspired

by biological neural networks in the brain.  Supervised learning  Able to learn nonlinear systems  Inherently parallelizable

Biological Basis

Biological Neural Network

Artificial Neural Network

Activation Functions

Backpropagation  Gradient

Descent  Levenberg-Marquart  Rprop  Others

Feed Forward Neural Network Output:

Weight update:

Activation Function:

Weight Updating Strategy  Pattern

mode strategy: update weights after each training pattern  Batch mode strategy: update weights after a batch of training patterns

(Training patterns are just sets of samples)

Parallelization of Neural Networks  Parallelization

of neurons  Parallelization of neural layers  Parallelization of training sets

Parallelization of Neurons

Parallelization of Layers

Parallelization of Training Sets

Ability to Generalize  One

of the tougher goals of neural networks  Means that the algorithm works for a wide range of inputs

4 Layer Design

Online Training  Continuous

feedback system  Makes decisions based on the current state of the system  Incremental

Our Network Vn I It Tn Tenv

Vn+1 Tn+1

Test Bench

Test Bench

EMF Results

What Next?  More

test bench data  Data from the car  Improvement of generalizing algorithms

SOC Estimation Using Support Vector Machines

Thank You