Neural Network SOC Estimation Using Voltage-Current Trajectory Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In vehicles equipped with secondary batteries, accurate estimation of state of charge (SOC) is challenging, especially during overcharging or overdischarging, and the accuracy decreases with battery degradation, requiring a method for high-accuracy SOC estimation and capacity measurement at a low cost.
Innovation Solution
A neural network-based system that processes normalized time-series voltage and current data, removing idle periods and using superimposed data to construct a database for SOC estimation, which can be applied to various battery types, including lithium-ion and all-solid-state batteries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional SOC estimation methods are used, then the system is simple to implement, but the estimation accuracy decreases with battery degradation
Solution Approach 1:
The patent transforms 1D time-series voltage and current data into 2D images by plotting voltage versus current trajectories. This dimensional transformation enables the application of 2D image processing techniques and convolutional neural networks, which can capture complex nonlinear relationships and degradation patterns that 1D methods miss, thereby maintaining high accuracy despite battery aging.
Solution Approach 2:
The patent replaces traditional mechanical/electrical measurement systems with an information-processing system based on neural networks. Instead of using complex physical sensors and measurement circuits, the system uses software-based deep learning models that process voltage-current trajectory images, achieving high SOC estimation accuracy through algorithmic intelligence rather than hardware complexity.
2Measurement precision
If full discharge method is used to calculate maximum capacity, then the capacity measurement is accurate, but the time required is very long
Solution Approach 1:
The patent performs preliminary actions by collecting and storing voltage-current trajectory data during normal battery operation before capacity measurement is needed. This ongoing data collection during regular charging/discharging cycles builds up a database that can be used for rapid SOC estimation without requiring time-consuming full discharge tests, thus achieving accurate capacity assessment in real-time.
Solution Approach 2:
The patent creates a virtual copy of the battery's electrical behavior through voltage-current trajectory images and neural network models. Instead of physically discharging the battery to measure capacity, the system uses the learned patterns from trajectory images to estimate capacity and SOC, providing a rapid non-invasive measurement that avoids the time-consuming full discharge process while maintaining accuracy.
3Reliability
If traditional neural network approaches are used, then the model is simple to construct, but it cannot maintain accuracy under battery degradation
Solution Approach 1:
The patent applies 2D image processing techniques and convolutional neural networks to voltage-current trajectory data, transforming 1D time-series into 2D visual representations. This enables the use of powerful 2D pattern recognition algorithms that can capture degradation-induced changes in battery behavior, maintaining high SOC estimation accuracy even as the battery ages, at the cost of increased processing complexity.
Data Source
AI summary
A capacity measurement system of a secondary battery that estimates an SOC with high estimation accuracy in a short time at low cost is provided. The capacity measurement system of a secondary battery is an estimation system of a state of charge of a power storage device that includes a unit for acquiring time-series data of a voltage measured value and a current measured value of a first power storage device; a unit for normalizing the time-series data of the voltage measured value; a unit for normalizing the time-series data of the current measured value; a database creation unit for creating a database where an SOC of the first power storage device is linked to superimposed data of time-series data of a time axis corresponding to a vertical axis and time-series data of a time axis corresponding to a horizontal axis; and a neural network unit.