Battery State of Charge Estimation via Adaptive Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for accurately determining the state of charge (SOC) of batteries are either noise-sensitive and simple or complex and computationally intensive, failing to effectively cope with noise and nonlinear battery behavior.
Innovation Solution
A system and method that utilize a battery model with adaptive weights, combining a coulomb counter and back propagation algorithm to iteratively modify weights based on voltage errors, providing an accurate SOC estimation without the need for complex neural networks or off-line training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If simple SOC estimation methods are used, then device complexity is reduced, but measurement precision deteriorates due to noise sensitivity and inability to cope with nonlinear battery behavior
Solution Approach 1:
The patent transforms the SOC estimation problem from directly estimating SOC to estimating battery voltage through a neural network model, then deriving SOC from the voltage estimate. This parameter transformation allows the system to achieve high accuracy by modeling voltage-SOC relationship while keeping the estimation algorithm computationally efficient
Solution Approach 2:
The patent replaces complex mechanical/mathematical SOC estimation methods (Coulomb counting, voltage translation) with a neural network-based voltage estimation system. The neural network learns the nonlinear voltage-SOC relationship from training data, substituting complex analytical methods with a data-driven approach that is both accurate and computationally efficient
2Measurement precision
If complex regression algorithms like nonlinear least squares are used, then measurement precision is improved, but productivity deteriorates due to high computational effort and ECU runtime requirements
Solution Approach 1:
The patent performs the computationally intensive work of learning the voltage-SOC relationship offline during a training phase using nonlinear least squares regression. Once trained, the neural network model can be deployed in the ECU where it requires minimal computational resources for real-time SOC estimation, thus separating the heavy computational task from the real-time operation
Solution Approach 2:
The patent creates a simplified copy of the complex nonlinear least squares algorithm in the form of a trained neural network model. The neural network replicates the behavior of the complex algorithm but with much lower computational requirements during execution, enabling real-time deployment in resource-constrained ECU environments
3Measurement precision
If complex model-based systems like Kalman filter are used, then measurement precision is improved, but device complexity increases due to intensive calculation requirements
Solution Approach 1:
The patent uses a simple neural network model that can be trained offline and then deployed as a lightweight, pre-computed solution in the ECU. This 'disposable' training phase produces a compact model that replaces complex real-time calculations with simple forward propagation operations, significantly reducing runtime computational complexity while maintaining accuracy
Data Source
Figure 1~2
Figure 3~6
Figure 4
AI summary
The present invention provides a system for determining the state of charge of a battery. The system includes a battery model having an input configured to receive an estimated state of charge value. The battery model is configured to generate at least one voltage loss output from which an estimated battery voltage is determined, the or each voltage loss output being representative of the contributions to voltage loss due to a parameter. The system also includes a comparator configured to determine a voltage error between the estimated battery voltage and the actual battery voltage. The battery model is configured to modify an adaptive state of charge weight based on the determined voltage error, and the state of charge is determined by modifying the estimated state of charge with the adaptive state of charge weight.