Battery State Estimation Using Gradient-Based Inputs Across Cell Types
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks struggle to accurately estimate the state of charge (SOC) and deterioration degree (SOH) of secondary batteries with varying electrical characteristics from different manufacturers and models, limiting their applicability across diverse battery types.
Innovation Solution
A learning method for a state estimation model that preprocesses terminal current and voltage data to calculate difference gradients and open circuit voltage change amounts, using machine learning to generate input data for a Recurrent Neural Network (RNN) or Convolutional Neural Network (CNN) models, enabling accurate estimation of SOC and SOH across different battery types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional neural networks are trained with direct input of measured voltage, current, and internal impedance values, then estimation accuracy is improved for batteries of the same manufacturer and model, but adaptability to batteries with different electrical characteristics deteriorates
Solution Approach 1:
The patent transforms the input parameters from direct measured values (voltage V, current I, internal impedance Z) to their difference values (ΔV, ΔI, ΔZ) and gradient values (d(ΔV)/d(ΔI)). This parameter transformation makes the neural network inputs invariant to the absolute electrical characteristics of different battery types, enabling the same network to accurately estimate SOC and SOH across batteries from various manufacturers and models while maintaining high estimation accuracy
Solution Approach 2:
The patent creates a composite feature space by combining multiple transformed parameters (difference values and gradient values) as inputs to the neural network. This composite approach captures both the magnitude changes and rate of change of electrical characteristics, providing a robust representation that works across different battery types while maintaining precise estimation
2Measurement precision
If one neural network is trained for a specific manufacturer and model, then estimation accuracy is improved for that specific battery type, but device complexity increases when multiple networks are needed for different battery types
Solution Approach 1:
The patent creates a universal neural network that can estimate SOC and SOH for batteries from any manufacturer and model by using transformed input parameters (difference and gradient values). This single multi-functional network replaces the need for multiple specialized networks, reducing device complexity while maintaining high estimation accuracy across all battery types
Data Source
AI summary
A learning method of a state estimation model includes training the state estimation model to learn a relationship of state estimation input data obtained by preprocessing terminal currents and terminal voltages of a secondary battery with a charge rate or a deterioration degree of the secondary battery. The state estimation input data includes time-series data of: difference gradients of terminal voltage differences with respect to terminal current differences; open circuit voltages; open circuit voltage change amounts; and integrated current values of the terminal currents.