Li-Ion Battery State Estimation Using RNN and CNN Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating State-of-Charge (SOC) and State-of-Health (SOH) of batteries, particularly Li-ion batteries, are cumbersome and inaccurate due to erratic dynamics during vehicle operation, non-linear dependencies on variables like temperature and discharge, and require computationally intensive models.
Innovation Solution
Utilizing Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) networks, and Convolutional Neural Networks (CNNs) to process time series data of battery attributes like voltage, current, and temperature, enabling accurate estimation of SOC and SOH without the need for complex hand-engineered models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning techniques (SVM, Neural Networks) are used for SOC and SOH estimation, then estimation capability is provided, but computational intensity and complexity increase
Solution Approach 1:
The patent replaces complex hand-engineered battery models and conventional machine learning approaches with deep neural networks (DNNs) that automatically learn features from raw voltage, current, and temperature measurements. This substitution eliminates the need for manual feature extraction and complex model formulation, reducing overall system complexity while maintaining or improving estimation accuracy.
Solution Approach 2:
The patent uses DNNs to create a virtual copy of the battery's electrochemical behavior through learned patterns from training data. Instead of relying on complex physical models that require numerous parameters, the network learns to replicate battery dynamics directly from measurement data, simplifying the estimation process while preserving accuracy.
2Measurement precision
If complex hand-engineered models are used to capture electrochemical processes, then estimation accuracy may improve, but computational overhead increases
Solution Approach 1:
The patent replaces computationally intensive hand-engineered electrochemical models with deep neural networks that perform calculations through learned weight matrices and activation functions. Once trained, the DNNs execute efficiently with minimal computational overhead, eliminating the need for complex real-time solution of differential equations and parameter estimation required by traditional models.
Solution Approach 2:
The patent performs the computationally intensive work of learning battery dynamics during an offline training phase using historical data. The trained network weights are then deployed for real-time estimation, shifting the computational burden from operation time to training time, thereby reducing real-time computational overhead while maintaining accuracy.
3Adaptability or versatility
If traditional estimation methods are used, then basic SOC estimation is achieved, but adaptability to varying ambient conditions and dynamic driving conditions is insufficient
Solution Approach 1:
The patent designs a universal deep neural network architecture that simultaneously estimates both SOC and SOH from the same set of inputs (voltage, current, temperature measurements). The network is trained on diverse data covering various ambient temperatures and driving conditions, enabling it to adapt to different operating scenarios without requiring separate models or complex operational procedures.
Solution Approach 2:
The patent employs DNNs with learnable parameters (weights and biases) that automatically adjust to capture the non-linear relationships between inputs and battery states under varying conditions. The network learns to adapt its internal parameters during training to handle different ambient temperatures, driving patterns, and battery aging stages, providing versatility without increasing operational complexity.
Data Source
AI summary
An approach to control or monitoring of battery operation makes use of an artificial neural network (ANN), which receives one or more battery attributes for a Lithium ion (Li-ion) battery, and determines, based on the received one or more battery attributes, a state-of-charge (SOC) and/or a state-of-health (SOH) estimate for the Li-ion battery. The ANN includes at least one of a recurrent neural network (RNN) and a convolutional neural network (CNN), and the series of values of the battery attributes includes at one of battery voltage values, battery current values, and battery temperature values.