Battery Capacity Estimation Model Using Modular Self-Organizing Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the state of storage batteries require detailed material information and are computationally intensive, necessitating high time and cost for model generation, while seeking an efficient way to estimate battery state from operational data.
Innovation Solution
A learning apparatus and method using a modular Self Organizing Map neural network to generate an estimation model for maximum and remaining capacity of storage batteries by acquiring voltage and capacity data under various conditions, allowing interpolation across conditions and estimating capacity based on differential capacity and voltage characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a modeling is performed at a material level of the battery using detailed parameters, then the estimation accuracy is improved, but the computational load increases
Solution Approach 1:
The patent segments the battery modeling into two distinct approaches: a material-level model for accurate capacity estimation and an equivalent circuit model for low computational load during operation. The segmented approach allows each model to serve its specific purpose - the material model provides accuracy while the equivalent circuit model enables real-time computation with minimal processing power.
Solution Approach 2:
The patent introduces an intermediary approach by using the material-level model to generate training data, which then trains a simplified equivalent circuit model. This intermediary training process allows the simple model to achieve high accuracy without requiring detailed material parameters during actual battery operation, thus resolving the contradiction between accuracy and computational load.
2Measurement precision
If detailed material information is used for battery state estimation, then the estimation accuracy is improved, but the device complexity increases
Solution Approach 1:
The patent segments the modeling complexity by separating the detailed material-level modeling (used only for training) from the operational model (equivalent circuit model). This segmentation allows detailed information to be used where needed for accuracy while keeping the operational system simple and easy to implement in real-time battery management.
Solution Approach 2:
The patent creates a simplified copy of the material-level model through the equivalent circuit model. The detailed material model generates training data that is used to train the simpler equivalent circuit model, which then serves as a copy that maintains accuracy without requiring the complex detailed material parameters during operation.
3Measurement precision
If a probabilistic neural network is used for battery state estimation, then the estimation accuracy is improved, but the time and cost for model generation increase
Solution Approach 1:
The patent performs preliminary action by using the material-level model to generate comprehensive training data in advance. This pre-generated training data is then used to train the equivalent circuit model, eliminating the need for time-consuming probabilistic neural network training during operation. The preliminary data generation enables fast, accurate estimation without repeated complex model generation.
Solution Approach 2:
The patent replaces the expensive, time-consuming probabilistic neural network with a simpler, cheaper equivalent circuit model for the actual estimation task. The complex material-level model is used only once for training data generation, while the simple equivalent circuit model performs the repeated estimation operations efficiently and低成本.
Data Source
AI summary
A learning apparatus is provided, which comprises: a characteristic acquisition unit for acquiring one of a voltage and a capacity of a storage battery and a characteristic corresponding to a capacity change and a voltage change of the storage battery under a plurality of conditions during at least one of charging and discharging of the storage battery; and a learning unit for learning a relationship between one of the voltage and the capacity of the storage battery and the characteristic under each of the plurality of conditions to generate an estimation model for estimating at least one of a maximum capacity and a remaining capacity of the storage battery from one of the voltage and the capacity of the storage battery and the characteristic.