Battery State-of-Health Prediction With Uncertainty Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the state of health of electrical energy storage units in battery-operated machines, such as electric vehicles, are inaccurate, particularly using physical aging models, which fail to predict the state of health based on usage behavior and machine parameters, leading to unreliable assessments of remaining battery capacity and service life.
Innovation Solution
A data-based state of health model is trained on operating variables and features to predict the state of health, incorporating model uncertainty, using Monte Carlo simulations and nearest neighbor methods to select real state of health characteristics, providing a confidence interval for a reliable prediction of the state of health and remaining service life.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a physical aging model is used to determine the state of health, then the present state of health can be ascertained, but the prediction accuracy is low with model errors exceeding 5%
Solution Approach 1:
The patent introduces an intermediary data processing layer between the physical aging model and the final prediction. This layer uses collected operating data (temperature, charge/discharge cycles, current, voltage) to train machine learning models that correct and refine the outputs of the physical aging model, thereby improving prediction accuracy while maintaining the interpretability of physical models.
Solution Approach 2:
The patent transforms the approach by changing from purely physical parameters to a hybrid approach that incorporates data-driven parameters. Operating features such as temperature profiles, charge/discharge rates, and cycle patterns are extracted and used as additional parameters to enhance the prediction model's accuracy beyond what traditional physical models can achieve alone.
2Measurement precision
If sensors are added to directly measure state of health, then measurement accuracy improves, but device complexity and space requirements increase
Solution Approach 1:
The patent enables the battery system to self-diagnose its state of health by utilizing data already collected during normal operation. The system processes its own operating data (temperature, voltage, current, state of charge) through machine learning models to determine SOH, eliminating the need for external sensors or interventions while maintaining high measurement accuracy.
Solution Approach 2:
The patent replaces the mechanical/sensor-based measurement approach with a computational approach. Instead of using physical sensors to directly measure SOH, the system uses software-based machine learning models that process electrical and thermal parameters to infer SOH, thereby reducing hardware complexity and space requirements.
3Measurement precision
If more operating data is collected for training, then prediction accuracy improves, but data processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary data processing and feature extraction during normal operation, preparing the data in advance for model training and prediction. Operating features are continuously collected and pre-processed, so when prediction is needed, the computational workload is already minimized, reducing real-time processing time while maintaining high accuracy.
Solution Approach 2:
The patent extracts only the most relevant operating features from the complete set of collected data for model training and prediction. By selecting key features such as temperature extremes, charge/discharge rates, and cycle patterns rather than processing all raw data, the system achieves high prediction accuracy with reduced computational effort and faster processing times.
Data Source
AI summary
A method for determining a predicted state of health of an electrical energy storage unit in a machine includes providing a data-based or hybrid state of health model, the data-based state of health model is trained, depending on operating variables of the electrical energy storage unit and/or operating features derived from the operating variables, to indicate a state of health and to indicate a model uncertainty of the indicated state of health, ascertaining a state of health characteristic and the associated model uncertainty of the energy storage unit based on the operating variables using the state of health model, and generating at least one random constructed state of health characteristic candidate that corresponds to constructed state of health characteristics within characteristic of confidence intervals, the characteristic of confidence intervals defined by the model uncertainties of the states of health of the ascertained state of health characteristic.