Autoencoder Battery Assessment via Reconstruction Error
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the performance, state, and load condition of battery storage systems, especially in large-scale facilities like primary control reserve batteries, face challenges due to impractical full charge and discharge requirements and laboratory-based measurement limitations, which fail to accurately emulate real-world operating conditions.
Innovation Solution
A method using an autoencoder to reconstruct battery storage measurements, calculating a reconstruction error, and determining an assessment indicator based on this error, allowing for continuous monitoring of battery performance and state without laboratory measurements, applicable to all battery storage types, including large facilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full charge and discharge measurements are performed to determine SOH, then measurement accuracy is improved, but operational complexity and time consumption increase significantly
Solution Approach 1:
The patent applies partial action by using only a segment of the charge-discharge cycle (e.g., only discharge phase or a partial charge phase) rather than requiring complete cycles. This allows SOH estimation to be performed during normal operation without waiting for full cycles, significantly reducing time loss while maintaining measurement accuracy through selective measurement of critical phases.
Solution Approach 2:
The patent implements preliminary action by performing measurements during routine charge-discharge operations before the battery reaches extreme states. By capturing data during normal operational ranges rather than waiting for complete cycles, the system obtains sufficient information for SOH estimation without the time penalty of full charge-discharge protocols.
2Measurement precision
If laboratory measurements are used for fine tuning battery assessment methods, then measurement precision is improved, but adaptability to real-world operating conditions deteriorates
Solution Approach 1:
The patent applies self-service by enabling the battery assessment system to automatically adapt to different operating conditions using data from the battery's own operational history. The method uses recurrent neural networks that learn from sequential measurements during actual use, allowing the system to self-calibrate and maintain precision across varying conditions without requiring external laboratory intervention for each scenario.
Solution Approach 2:
The patent implements parameter changes by allowing the assessment model to dynamically adjust its parameters based on observed operating conditions. The recurrent neural network modifies its internal state and weighting parameters in response to varying temperature, load, and charge-rate conditions, maintaining measurement precision across diverse real-world scenarios rather than relying on fixed laboratory-calibrated parameters.
3Device complexity
If traditional measurement methods are used for battery state determination, then device complexity is reduced, but measurement precision and reliability under varying conditions deteriorate
Solution Approach 1:
The patent applies mechanics substitution by replacing complex physical measurement systems with an information-processing approach. Instead of using additional sensors or complex electrical measurement apparatus, the system uses a recurrent neural network that processes standard voltage, current, and temperature measurements through learned patterns, achieving high measurement precision with minimal additional hardware complexity.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A method and a system for determining an assessment indicator for assessing at least one of a performance, a state, and a load condition of a battery storage (10). A series (70) of measurements of at least one physical parameter of the battery storage (10), including at least one of voltage and current, is obtained. The measurements correspond to successive points of time of a time segment (72). The series (70) of measurements is input to an encoder (400) of an autoencoder (40). A reconstructed series (80) of measurements is received from a decoder (420) of the autoencoder (40). A value of a reconstruction error (x) is calculated based on the series (70) of measurements and the reconstructed series (80) of measurements, and the assessment indicator (BI) is determined based on the calculated value of the reconstruction error (x).