Neural Network Parameter Identification for Lithium-Ion Batteries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current battery management systems (BMS) for lithium-ion batteries, based on equivalent circuit models, have limited predictive ability and require lengthy parameter identification processes, especially when obtaining OCV-SOC curves, which are difficult to measure during battery use.
Innovation Solution
A method and system for working condition sensitivity analysis and data processing using a neural network model to analyze electrochemical model parameters, involving normalization of voltage data and selection of a reference voltage interval with a maximum electric quantity change slope, to improve the identification of high-sensitivity parameters affecting battery output voltage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If electrochemical model parameters are identified using traditional methods (heuristic algorithms, neural networks, Kalman filter), then parameter identification can be performed, but the identification process takes a long time and requires measuring the OCV-SOC curve in advance which is difficult to obtain during battery use
Solution Approach 1:
The patent pre-processes voltage data by normalizing it and selecting a reference voltage interval with maximum electric quantity change slope before neural network input. This preliminary data preparation enables faster convergence during parameter identification without requiring advance OCV-SOC curve measurements, thus reducing identification time while maintaining accuracy.
Solution Approach 2:
The patent extracts and focuses on the reference voltage interval where the electric quantity change slope is maximum. By isolating this critical voltage range for sensitivity analysis and neural network training, the method concentrates computational resources on the most informative data portion, significantly reducing overall identification time while preserving identification precision.
2Reliability
If electrochemical models with dozens of physical parameters are used to fully reflect internal battery state, then management ability is greatly improved, but the complexity of the model limits practical applications
Solution Approach 1:
The patent segments the complex electrochemical model parameters into high-sensitivity and low-sensitivity groups based on working condition sensitivity analysis. By identifying and focusing on high-sensitivity parameters that most significantly affect battery output voltage, the method maintains reliable battery state estimation while reducing the effective number of parameters that need to be identified and managed.
Solution Approach 2:
The patent transforms the model parameters from a static identification problem into a dynamic sensitivity-based selection process. By analyzing how different parameters respond to working condition changes and selecting only the high-sensitivity parameters for active identification, the system achieves reliable battery management with reduced model complexity.
3Productivity
If voltage data is used directly for neural network training without normalization and sensitivity analysis, then the process is simpler, but the neural network performance and parameter identification accuracy are reduced
Solution Approach 1:
The patent applies preliminary normalization to voltage data, converting it to a standard range (0-1) using the formula v'=(v-vmin)/(vmax-vmin). This pre-processing step ensures uniform data distribution and accelerates neural network convergence, improving both processing efficiency and parameter identification accuracy simultaneously.
Solution Approach 2:
The patent applies different processing treatments to different portions of the voltage data based on their sensitivity characteristics. By identifying the reference voltage interval with maximum electric quantity change slope and focusing processing efforts on this specific range, the method optimizes neural network training efficiency and accuracy for the most critical parameter identification tasks.
Data Source
AI summary
The invention provides a method and a system of working condition sensitivity analysis and data processing for parameter identification and/or for training a parameter identification neural network. The method includes according to a selected reference voltage interval, obtaining a voltage data set corresponding to electrochemical model parameters in the reference voltage interval; normalizing voltage values of the voltage data set to obtain a characteristic voltage data set, wherein the number of voltage values corresponding to different electrochemical model parameters in the characteristic voltage data set is equal; and inputting the characteristic voltage data set into a neural network model, and outputting initial values of the electrochemical model parameters to analyze the working condition sensitivity by taking the electrochemical model parameters as labels. The electrochemical model parameters include high sensitivity parameters.