ESS State Estimation Using Deep Learning Time-Series Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for estimating the state of health (SOH) and safety (SOS) of energy storage systems (ESS) provide inaccurate results due to reliance on statistical-based features with inconsistencies between input features and remaining capacity, failing to accurately predict cell lifespan and capacity.
Innovation Solution
A deep learning module that processes real-time time series data of current, voltage, and temperature using models like DFFN, DCNN, LSTM, and ConvLSTM to estimate SOH and remaining capacity, eliminating the need for feature extraction and providing accurate predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If statistical-based features are used for SOH estimation, then the system is simpler to implement, but the prediction accuracy deteriorates
Solution Approach 1:
The patent replaces statistical-based feature extraction methods with deep learning models that directly process raw time series data. This substitution eliminates the need for manual feature engineering while significantly improving prediction accuracy through automated feature learning from voltage, current, and temperature sequences.
Solution Approach 2:
The patent transforms the input data representation from extracted statistical features to raw time series data with multiple parameters (voltage, current, temperature) over time. This parameter change enables deep learning models to capture temporal dependencies and patterns that statistical methods miss, thereby improving SOH estimation accuracy.
2Device complexity
If feature extraction is performed before modeling, then the processing complexity is reduced, but the information completeness deteriorates
Solution Approach 1:
The patent removes the feature extraction step from the traditional pipeline and directly feeds raw time series data into deep learning models. This extraction of the intermediate processing step eliminates information loss while the models themselves perform automated feature learning, maintaining manageable complexity through end-to-end learning.
Solution Approach 2:
The deep learning models perform preliminary feature learning automatically from raw data before SOH estimation. This preliminary action of automated feature extraction within the model preserves all original information while reducing overall processing complexity by consolidating steps into a single trainable system.
3Use of energy by moving object
If conventional methods are used for SOH estimation, then the computational resources required are less, but the reliability of prediction deteriorates
Solution Approach 1:
The patent changes the computational approach from statistical feature analysis to deep learning with multiple parameters (voltage, current, temperature) processed through LSTM and ConvLSTM layers. This parameter-rich approach improves prediction reliability by capturing temporal patterns and interactions that conventional methods cannot detect.
Data Source
AI summary
Systems and methods for energy storage system (ESS) real-time state estimation and management. A system can include a deep learning (DL) module, which receives real-time measurement of time series data for current, voltage, and temperature of an ESS under test. The DL module can reshape the received time series data with a preprocessing component and then process the reshaped data with a DL component. The DL component can include various DL models such as DFFN, DCNN, LSTM, and ConLSTM, and utilize one or more of these DL models in estimating the SOH and/or remaining capacity for an ESS based on, for example, information about the ESS so as to generate more accurate estimations. The DL module can provide the estimations for the ESS under a variety of charging protocols. Such estimations can be utilized to control aspects of the ESS, such as optimizing its performance and extending its lifespan.