Battery Cluster SOC Estimation Using Hybrid Model Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating battery state of charge (SOC) are inaccurate due to battery nonlinear characteristics, requiring extensive time, complex calculations, or are unsuitable for real-time monitoring, and are affected by factors like temperature and discharge rate, leading to cumulative errors and aging-related inaccuracies.
Innovation Solution
A method combining ampere-hour integration with a state-of-charge prediction model, utilizing gradient boosting decision trees, to estimate SOC, and a system comprising data acquisition, first and second estimation modules, and a charge determination module to improve accuracy by weighted summation and model retraining based on distance metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If discharge experiment method is used to estimate SOC, then accuracy is improved, but time consumption increases significantly and it cannot be used on batteries in operation
Solution Approach 1:
The patent creates a virtual copy of the battery's charge-discharge process through neural network simulation. Instead of physically discharging the battery to measure SOC, the system uses historical data to train a neural network model that replicates battery behavior, allowing accurate SOC estimation without actual discharge operations.
Solution Approach 2:
The patent performs preliminary data collection and neural network training during periods when the battery is not in use or during normal operation. By pre-processing and storing charge-discharge data, the system prepares the neural network model in advance, enabling rapid SOC estimation without requiring time-consuming discharge experiments at the moment of measurement.
2Productivity
If ampere-hour integration method is used to estimate SOC, then real-time monitoring is enabled, but cumulative errors occur affecting accuracy
Solution Approach 1:
The patent implements a feedback mechanism where the neural network model continuously receives actual battery data (current, voltage, temperature) and compares predicted SOC with measured values. The system uses this feedback to correct cumulative errors by adjusting the model's internal parameters and recalibrating predictions, maintaining accuracy over extended operation periods.
Solution Approach 2:
The patent dynamically changes the parameters used in SOC estimation by incorporating multiple variables (current, voltage, temperature, charge-discharge rate) into the neural network model. This multi-parameter approach compensates for the limitations of single-parameter ampere-hour integration, reducing cumulative errors through comprehensive state assessment.
3Measurement precision
If OCV method is used to estimate SOC, then direct SOC measurement is achieved, but standing time of more than two hours is required making it unsuitable for real-time monitoring
Solution Approach 1:
The patent performs preliminary data collection during battery operation, accumulating charge-discharge data, voltage, and temperature information. By pre-processing this data and training the neural network model in advance, the system eliminates the need for lengthy standing periods, enabling accurate SOC estimation during normal battery operation without requiring the battery to be at rest.
Solution Approach 2:
The patent replaces the physical waiting process required by OCV method with a computational neural network model. Instead of mechanically waiting for the battery to reach open-circuit voltage after standing, the system uses mathematical modeling and data processing to instantly estimate SOC based on operational data, substituting time-consuming physical processes with rapid computational analysis.
4Measurement precision
If Kalman filter method is used to estimate SOC, then optimal state estimation with minimum variance is achieved, but significant matrix computations require microcontroller with high computational capability
Solution Approach 1:
The patent creates a simplified virtual model of the battery system using neural networks that replicates the complex electrochemical behavior without requiring complex mathematical computations. This virtual copy processes data through learned patterns rather than extensive matrix operations, reducing computational complexity while maintaining estimation accuracy.
Solution Approach 2:
The patent transforms the estimation problem from one requiring complex matrix computations to one solved by neural network pattern recognition. By changing the computational approach from traditional Kalman filter mathematics to neural network inference, the system achieves similar accuracy with significantly reduced computational requirements suitable for embedded microcontrollers.
5Stability of the object's composition
If combined voltage correction method is used to estimate SOC, then stability and simplicity are achieved, but accuracy depends on precision of charging curve which changes with battery aging
Solution Approach 1:
The patent implements a dynamic neural network model that adapts to battery aging by continuously learning from new data. Unlike static charging curves that become inaccurate over time, the neural network dynamically adjusts its parameters and patterns based on actual battery performance, maintaining accuracy throughout the battery's lifecycle while preserving estimation stability.
Solution Approach 2:
The patent enables the system to self-calibrate and self-improve by automatically learning from operational data. The neural network continuously refines its own parameters through ongoing data collection and training, eliminating the need for external recalibration or manual updates of charging curves, thus maintaining both stability and accuracy autonomously.
Data Source
AI summary
A method and a system for estimating a state of charge of a battery cluster, an electronic device, and a storage media are provided. The method comprises acquiring target data related to the state of charge; estimating, by an ampere-hour integration method, the state of charge based on the target data, to obtain a first estimated value; inputting the target data into a state-of-charge prediction model to estimate the state of charge and obtain a second estimated value, wherein the prediction model is obtained by training based on sample data; and determining a final estimated value of the state of charge based on the first estimated value, the second estimated value, and a first distance between the target data and the sample data. The method combines the ampere-hour integration method and the prediction model to estimate the state of charge, effectively improving the accuracy of the state of charge estimation.