Battery SoH Estimation With RL-Adjusted Filter Gain
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing battery management systems face challenges in accurately estimating the state-of-health of batteries in vehicles due to variations in cell properties over time and changing operating conditions, particularly temperature, which affects internal resistance and degradation rates, and require a more adaptive filter gain to correct for drifting estimates.
Innovation Solution
A method using a reinforcement learning agent to modify the filter gain factor K based on current operating conditions, specifically temperature, to update the state-of-health estimation of battery cells, incorporating a predetermined model for calculating state-of-health and accounting for variations between cells, with the gain factor adjusted dynamically to minimize estimation error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a predetermined model is used to calculate state-of-health, then the calculation can be performed efficiently, but the estimation accuracy deteriorates due to model dependency on limited operating conditions
Solution Approach 1:
The filter gain factor K is transformed from a static predetermined value to a dynamic variable that adapts to changing operating conditions. The reinforcement learning agent continuously adjusts K based on current temperature and battery state, allowing the estimation system to maintain high accuracy across diverse operating conditions while preserving computational efficiency.
Solution Approach 2:
The system changes the parameter K (filter gain factor) dynamically based on operating conditions rather than using a fixed value. The reinforcement learning agent modifies K according to temperature and battery state, enabling the predetermined model to achieve accurate estimates across varying conditions without requiring multiple temperature-specific models.
2Measurement precision
If the filter gain factor K is set to correct for drifting estimates, then the estimation accuracy improves, but the system complexity increases due to need for adaptive adjustment
Solution Approach 1:
The reinforcement learning agent enables the system to self-adjust the filter gain factor K autonomously based on observed operating conditions and estimation errors. The agent learns optimal K values through interaction with the battery system, eliminating the need for manual calibration or complex external adjustment mechanisms while maintaining high estimation accuracy.
Solution Approach 2:
The system implements a feedback loop where the reinforcement learning agent continuously monitors the difference between predetermined model estimates and actual battery behavior, then adjusts K accordingly. This feedback mechanism corrects drifting estimates automatically while keeping the overall system architecture relatively simple by building the adaptation capability into the existing estimation framework.
3Measurement precision
If laboratory measurements are performed to determine battery parameters, then the initial model accuracy is improved, but the system cannot adapt to cell property variations over time and between cells
Solution Approach 1:
The reinforcement learning agent enables the battery management system to self-adapt to cell property variations without requiring external recalibration or manual intervention. The agent continuously learns from operational data and adjusts the filter gain factor to account for degradation and cell-to-cell variations, maintaining accuracy throughout the battery lifecycle.
Solution Approach 2:
Laboratory measurements are performed in advance to establish the predetermined model and initial battery parameters. This preliminary characterization provides a solid foundation for accurate estimation, while the subsequent reinforcement learning adaptation handles time-varying variations, combining the benefits of both approaches.
Data Source
AI summary
A method and system for estimating the state-of-health of a battery The invention relates to a method for estimating the state-of-health, SoHest, of a battery in a vehicle, the method comprising: determining (100) a first battery property and a battery temperature of a battery in a vehicle; calculating (102) a state-of-health, SoHcalc, of a battery cell based on the determined first battery property using a predetermined model; providing (104) a function f estimating a battery cell degradation rate; updating (106) a state-of health estimated in a previous time step according to SoHest←SoHest+ƒ·dt+K·(SoHcalc−SoHest), where K is a gain factor which is dependent on operating conditions of the vehicle, and wherein K is modified (108) for each time step using a reinforcement learning agent.