Battery State Estimation Using LSTM Voltage Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing battery management systems fail to accurately detect unusual abnormal behaviors in batteries that are not detectable through simple voltage or current deviations, necessitating a more sophisticated method for identifying battery defects.
Innovation Solution
A battery management apparatus and method that utilizes a long short-term memory (LSTM) algorithm to estimate future voltage information based on current and voltage data over specific periods, determining battery states by comparing estimated and measured voltage information, and updating algorithms to adapt to changing health states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If simple voltage or current deviation comparison methods are used for battery defect detection, then the detection system remains simple and easy to implement, but unusual abnormal behaviors that do not manifest as simple deviations cannot be detected
Solution Approach 1:
The patent replaces simple electrical comparison methods with a machine learning-based prediction system. An LSTM neural network model predicts future voltage values based on historical data, and deviations between predicted and actual voltages indicate defects. This substitution enables detection of subtle abnormal behaviors that simple threshold comparisons cannot detect.
Solution Approach 2:
The patent introduces voltage prediction information as an intermediary element between direct voltage measurement and defect determination. By comparing measured voltage against predicted voltage (rather than against simple thresholds), the system can identify unusual abnormal behaviors. The prediction model acts as a mediator that translates historical patterns into actionable defect detection criteria.
2Measurement precision
If voltage prediction based on historical data is implemented to detect unusual abnormal behaviors, then detection accuracy for subtle defects improves, but computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary training of the LSTM model during periods when the battery is in normal operation. Historical voltage and current data are collected and used to train the prediction model in advance. Once trained, the model can quickly predict future voltage values during operation, reducing real-time processing requirements while maintaining high detection precision.
Solution Approach 2:
The patent implements a dynamic processing approach where the model training and updating occurs adaptively based on battery operation conditions. The system adjusts between training mode (when computational resources are available) and prediction mode (during critical monitoring periods), optimizing the balance between precision and processing time based on real-time needs.
Data Source
AI summary
A battery management apparatus according to an embodiment disclosed herein includes a measuring unit for measuring a current and a voltage of a battery and a controller for generating voltage estimation information by estimating the voltage of the battery for a second period of time after a first period of time based on first current information related to the current of the battery measured for the first period of time and first voltage information related to the voltage of the battery measured for the first period of time, determining a state of the battery based on the voltage estimation information and second voltage information related to the voltage of the battery measured for the second period of time.