Vehicle Battery Life Prediction Using Segmented Deterioration Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing battery life prediction techniques are prone to significant fluctuations in predicting vehicle battery life due to temporary changes in deterioration characteristics like internal resistance, leading to inaccurate remaining life estimates.
Innovation Solution
A battery life learning device and method that utilizes a learned prediction model from time-series data of vehicle batteries that have reached the end of their life, dividing data into partial sequences with labels for each period to suppress erroneous predictions from temporary fluctuations, and stopping predictions when unmeasured portions exceed a certain frequency, thereby improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the prediction function is switched when the deterioration characteristic temporarily changes, then the prediction can adapt to the current state, but the prediction value fluctuates greatly and accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training the prediction model using time-series data from batteries that have reached end-of-life. This allows the model to learn the relationship between deterioration characteristics and remaining life in advance, so when predicting for a current battery, the model can accurately estimate remaining life even when deterioration characteristics temporarily change, without needing to switch prediction functions.
2Adaptability or versatility
If time-series data including temporary fluctuations is used for learning, then the model can handle real-world variations, but erroneous predictions occur when temporary changes are misinterpreted as end-of-life
Solution Approach 1:
The patent implements feedback by using the actual remaining life values (labels) from end-of-life batteries to train the prediction model. The model learns from historical data where the ground truth is known, allowing it to distinguish between temporary fluctuations and genuine end-of-life conditions. This feedback mechanism enables the model to improve its prediction reliability while maintaining adaptability to real-world variations.
3Loss of information
If the prediction model is trained on complete time-series data, then comprehensive information is utilized, but calculation load increases significantly
Solution Approach 1:
The patent applies segmentation by dividing the time-series data into multiple sequences of different lengths and training multiple prediction models for each sequence type. This allows the system to select an appropriate model based on the available data length, utilizing comprehensive information when needed while reducing calculation load by using smaller models when data is limited or processing resources are constrained.
Data Source
AI summary
A battery life learning device including a learning section configured to obtain a learned prediction model for predicting a remaining life of a vehicle battery from time-series data of a deterioration characteristic of the vehicle battery, the learned prediction model being obtained by learning a prediction model from the time-series data of the deterioration characteristic of the vehicle battery based on learning data, the learning data including time-series data of the deterioration characteristic at a predetermined time point in the past of a vehicle battery for learning that has reached an end of life and the remaining life at the predetermined time point of the vehicle battery for learning.