Battery State of Charge Estimation with Gaussian Process Regression

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Accurate estimation of battery state of charge (SoC) is challenging due to complex chemical and physical processes in batteries, which conventional methods fail to capture, leading to unreliable and inaccurate SoC estimation.

Innovation Solution

Employing a Gaussian process regression (GPR) framework to model the relationship between battery measurements and SoC, utilizing joint Gaussian distributions to determine SoC probabilistically, incorporating a Central Limit Theorem (CLT) to account for unpredictable environmental variations, and optionally combining with a Kalman filter for improved accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional model-based methods are used to estimate SoC, then the estimation process is simplified, but accuracy and reliability deteriorate due to inability to capture physics-based non-linear dynamics and parametric uncertainties

Engineering Contradiction:
ImproveSoC estimation accuracyVSAvoidestimation system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transforms the SoC estimation problem from a deterministic parameter estimation task to a probabilistic inference problem. By modeling SoC as a random variable with Gaussian distribution characterized by mean and variance, the method captures non-linear dynamics and parametric uncertainties through statistical parameters rather than attempting to model complex physical chemistry directly.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces complex physics-based chemical models with a data-driven Gaussian process regression framework. Instead of solving differential equations representing battery electrochemistry, the system uses statistical learning to map measurements to SoC, substituting mechanical/chemical modeling with probabilistic inference.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Ease of operation

If simplified models are used for SoC estimation, then ease of operation is improved, but measurement precision deteriorates due to inability to capture complex battery processes

Engineering Contradiction:
Improveestimation method simplicityVSAvoidSoC estimation accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent simplifies the estimation operation by changing from deterministic parameter fitting to probabilistic inference with Gaussian distributions. The simplified operation of computing mean and variance from joint Gaussian distributions provides both ease of calculation and accuracy by naturally handling non-linearities through the probabilistic framework.

Inventive Principle:
Principle #35Parameter changes

3Adaptability or versatility

If data-driven methods are used to map measurements to SoC, then adaptability to different battery conditions is improved, but reliability deteriorates due to simplified mappings that cannot accurately represent complex battery processes

Engineering Contradiction:
Improveadaptability to operating conditionsVSAvoidSoC estimation reliability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent enhances reliability by changing from a single-point estimation approach to a full probabilistic distribution representation. By outputting both mean and variance of the Gaussian distribution, the system provides not only the estimated SoC but also a measure of uncertainty, making the adaptation to different conditions more reliable and interpretable.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3465241B1Method and system for estimating state of charge of a battery using gaussian process regression
Publication Date: 2025.07.16 MITSUBISHI ELECTRIC CORP
  • EP3465241B1 patent drawingFigure 1A
  • EP3465241B1 patent drawingFigure 1B
  • EP3465241B1 patent drawingFigure 1C

AI summary

Method and system for estimating a state of charge (SoC) of a battery are disclosed. A method determines a first joint Gaussian distribution of values of the SoC given a set of historical measured physical quantities of the state of the battery and a corresponding set of historical values of the SoC of the battery. The method determines a second joint Gaussian distribution of SoC using the set of historical measured physical quantities and the corresponding set of historical values of the SoC, current measurement physical quantities of the battery and the first joint Gaussian distribution. Finally, the method determines a mean and a variance of the current value of the SoC of the battery from the second joint Gaussian distribution. The mean is an estimate of the current SoC of the battery, and the variance is a confidence of the estimate.