Neural Network Battery Capacity Estimation Using Partial Charge Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Power storage devices, such as secondary batteries and uninterruptible power supplies, face challenges in accurately estimating their remaining capacity and degradation state, leading to unpredictable performance and potential failures, especially in critical applications like electric vehicles and data centers, due to the time-consuming and cumbersome processes of charge and discharge cycle analysis.
Innovation Solution
A capacity estimation method using a neural network that analyzes discharge and charge data from lithium-ion secondary batteries, specifically through self-discharge, mid-discharge to full charge, and constant load conditions, allowing for rapid degradation state assessment and prediction, even in large capacity batteries, without requiring a full discharge cycle.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional charge and discharge cycle analysis is used to estimate battery capacity, then measurement precision is improved, but loss of time increases significantly
Solution Approach 1:
The patent applies partial action by performing only a partial charge cycle (from 0% to 80% SOC) rather than a complete charge-discharge cycle. This partial charging process, combined with neural network analysis of the voltage-time characteristics during this reduced cycle, provides sufficient data for accurate capacity estimation while significantly reducing the time required compared to traditional full cycle methods.
2Measurement precision
If full charge-discharge cycle is performed for accurate capacity estimation, then measurement precision is improved, but productivity decreases
Solution Approach 1:
The patent applies preliminary action by performing the charging process up to 80% SOC as a preliminary step that provides sufficient information for capacity estimation without requiring the time-consuming final 20% charge and subsequent discharge. The neural network is trained to accurately predict capacity based on this preliminary charging data alone, thereby improving productivity while maintaining measurement precision.
3Ease of operation
If traditional neural network methods are used for capacity estimation, then ease of operation is improved, but measurement precision deteriorates for large capacity batteries
Solution Approach 1:
The patent applies parameter changes by modifying the charging cutoff parameter from the traditional 100% SOC to 80% SOC. This parameter change, combined with specific adjustments to the neural network training parameters (using voltage-time characteristics during the 0-80% charging phase), enables accurate capacity estimation for large capacity batteries while maintaining ease of operation. The method adapts the estimation parameters to match the partial charging process.
Data Source
AI summary
It is difficult to know the remaining amount and the degradation state of a power storage device, and it is also difficult to estimate how long the power storage device can be used. Data obtained through midway discharge and mid-to-full charge is used as the learning data to calculate the degradation state and the capacity. In other words, the learning data includes both a discharge curve of midway discharge and a charge curve of mid-to-full charge, and neural network processing is performed with the use of the learned data.